AU2019280053B2

AU2019280053B2 - Hierarchical adaptive closed-loop fluid resuscitation and cardiovascular drug administration system

Info

Publication number: AU2019280053B2
Application number: AU2019280053A
Authority: AU
Inventors: Behnood GHOLAMI
Original assignee: Autonomous Healthcare Inc
Current assignee: Autonomous Healthcare Inc
Priority date: 2017-05-12
Filing date: 2019-12-12
Publication date: 2020-05-07
Anticipated expiration: 2038-05-11
Also published as: AU2020264409B2; AU2020213388B2; AU2020203918A1; AU2020202414A1; AU2020203914B2; AU2020203914A1; AU2020203918B2; AU2020264408A1; AU2020264408B2; AU2020213388A1; AU2020264409A1; AU2020202414B2; AU2019280053A1

Abstract

The present disclosure describes a closed-loop fluid resuscitation and/or cardiovascular drug administration system that uses continuous measurements and adaptive control architecture. The adaptive control architecture uses a function approximator to identify unknown dynamics and physiological parameters of a patient to compute appropriate infusion rates and to regulate the endpoint of resuscitation. Resuscitation Fluid (Crystalloid/Colloid) Fluid E xchange Circulation lw Interstitial (Blood) Tissue Urine Blood Evaporation

Description

HIERARCHICAL ADAPTIVE CLOSED-LOOP FLUID RESUSCITATION AND CARDIOVASCULAR DRUG ADMINISTRATION SYSTEM

Background of the Invention [0001] Any discussion of the background art throughout the specification should in no way be considered as an admission that such background art is prior art, nor that such background art is widely known or forms part of the common general knowledge in the field in Australia or worldwide.

[0002] All references, including any patents or patent applications, cited in this specification are hereby incorporated by reference. No admission is made that any reference constitutes prior art. The discussion of the references states what their authors assert, and the applicants reserve the right to challenge the accuracy and pertinence of the cited documents. It will be clearly understood that, although a number of prior art publications are referred to herein, this reference does not constitute an admission that any of these documents forms part of the common general knowledge in the art, in Australia.

[0003] Reliable and consistent fluid resuscitation (i.e., intravenous administration of fluids) is critical for perioperative and intensive care unit (ICU) fluid management in human and animal populations. The goal of fluid resuscitation is to restore blood volume in the circulatory system to an acceptable level to ensure adequate tissue perfusion. However, large intra-patient and inter-patient variability in physiological parameters, and the effects of different illnesses and medications can result in under-resuscitation and over-resuscitation of patients.

Incorporation by Reference [0004] Each patent, publication, and non-patent literature cited in the application is hereby incorporated by reference in its entirety as if each was incorporated by reference individually.

Summary of the Invention [0005] It is an object of the present invention to overcome or ameliorate at least one or more of the disadvantages of the prior art, or to provide a useful alternative.

[0006] Throughout this specification, unless the context requires otherwise, the words “comprise”, “comprises” and “comprising” will be understood to imply the inclusion of a stated step or element or group of steps or elements or stated integer or group of integers but not the

2019280053 01 Apr 2020 exclusion of any other step or element or group of steps, elements or integers.

[0007] In the claims, as well as in the summary above and the description below, all transitional phrases such as “comprising”, “including”, “carrying”, “having”, “containing”, “involving”, “holding”, “composed of”, and the like are to be understood to be open-ended i.e. to mean “including but not limited to”. Only the transitional phrases “consisting of” and “consisting essentially of” alone shall be closed or semi-closed transitional phrases, respectively.

[0008] Any one of the terms “including” or “which includes” or “that includes” as used herein is also an open term that also means including at least the elements/features that follow the term, but not excluding others. Thus, “including” is synonymous with and means “comprising”.

[0009] In some embodiments, the disclosure provides a method for fluid resuscitation and/or cardiovascular drug administration comprising:

(a) initiating an infusion rate of a fluid and/or a cardiovascular drug into a subject at an infusion rate;

(b) receiving hemodynamic sensor data of the subject from at least one medical monitoring device attached to the subject by a hierarchical control architecture system, wherein the hierarchical control architecture system comprises:

at least one adaptive controller; and a logic-based controller;

wherein the subject’s hemodynamic data is received by the logic-based controller and the at least one adaptive controller; and wherein the at least one adaptive controller and the logic-based controller are in communication with one another;

(c) generating by the at least one adaptive controller an altered infusion rate based on the subject’s hemodynamic data, previous infusion rates, and internal parameters and states of the at least one adaptive controller;

(d) verifying by the logic-based controller that the subject’s hemodynamic data and the at least one adaptive controller states do not violate rules governing the operation of the at least one adaptive controller;

(e) sending the altered infusion rate from the logic-based controller to at least one

2019280053 01 Apr 2020 infusion pump;

(f) automatically administering by the at least one infusion pump the fluid and/or cardiovascular drug to the subject at the altered infusion rate or infusion rates upon receipt of the infusion rate or infusion rates; and (g) repeating Steps (b) to (f) at set time intervals.

[0010] In some embodiments, the disclosure provides a method for fluid resuscitation and/or cardiovascular drug administration clinical decision support comprising:

(a) asking a user to provide an initial infusion rate of a fluid and/or a cardiovascular drug;

at least one adaptive controller; and a logic-based controller;

(d) verifying by the logic-based controller that the hemodynamic data and the at least one adaptive controller states do not violate rules governing the operation of the at least one adaptive controller;

(e) verifying by the logic-based controller whether the altered infusion rate meets the requirements to notify the user and if not the infusion rate is kept at the previous value;

(f) displaying the altered infusion rate from the at least one adaptive controller if the altered infusion rate meets the requirement in Step (e);

2019280053 01 Apr 2020 (g) asking the user to either accept the altered infusion rate or change the new infusion rate to a different value or different values; and (h) repeating Steps (b) to (g) at set time intervals.

[0011] In embodiments, this disclosure provides for a clinical decision support (semi-automated) system for notifying a user through audio-visual alarms if certain infusion therapy performance criteria are violated. A notification can be issued by a higher-level logic-based controller. In some embodiments, the notification asks the user to change the therapy type from one type to another. For example, in some embodiments, if the subject was being administered with fluid, the user can be asked to consider administering a vasopressor.

[0012] In another embodiment, this disclosure provides for a clinical decision support system for issuing a prompt when a change in infusion therapy is implicated based at least on a background infusion rate. An appropriate notification message and/or an appropriate change in infusion therapy recommendation message may be determined. Further, the appropriate notification message and/or the appropriate recommendation message may be displayed.

[0013] In another embodiment, this disclosure provides for an apparatus for displaying infusion therapy information which can be used to optimize infusion therapy. The apparatus comprises a memory, a display, and a processor in operable communication with the memory and the display. The display displays at least one appropriate notification message and appropriate change in infusion therapy recommendation when the processor issues the notification message and/or the recommendation message upon implication for a change in infusion therapy based at least on an active infusion rate stored in memory and a background infusion rate. In some embodiments, the background infusion rate is calculated by a lower-level controller based at least on received sensor measurement from a hemodynamic monitoring device.

[0014] According to a first aspect of the invention, there is provided a computer-implemented method for providing clinical decision support for regulating hemodynamic measurements of a subject, the method comprising:

(a) receiving by way of a user interface at least one initial infusion rate;

(b) recording the at least one initial infusion rate as a last user-approved infusion rate in a database;

(c) receiving from a sensor hemodynamic measurements of a subject by a hierarchical control architecture system;

wherein the hierarchical control architecture system comprises:

2019280053 01 Apr 2020 (a) at least one lower-level controller; and (b) a higher-level logic controller;

wherein the subject’s hemodynamic measurements are received by the higher-level logic controller and the at least one lower-level controller;

wherein the at least one lower-level controller and the higher-level logic controller are in communication with one another; and wherein the at least one lower-level controller comprises:

(a) a dynamic observer comprising a set of dynamic observer states for estimating a current value of physiological parameters of the subject; and (b) a dynamical system using a function approximator comprising a set of basis functions and a set of weights to approximate unknown dynamics of the subject;

(d) initializing the value of the set of weights of the at least one lower-level controller based at least in part on the at least one initial infusion rate;

(e) updating the set of weights of the at least one lower-level controller based at least in part on current values of the set of weights, the set of dynamic observer states, and current value of the hemodynamic measurements;

(f) generating by the at least one lower-level controller an altered infusion rate based at least in part on the subject’s hemodynamic measurements and the set of weights of the at least one lower-level controller;

(g) verifying by the higher-level logic controller whether the altered infusion rate meets requirements to notify the user;

(h) displaying the altered infusion rate from the at least one lower-level controller if the altered infusion rate meets the requirements in Step (g);

(i) asking the user to either accept the altered infusion rate or change the altered infusion rate to a different value;

(j) replacing by the higher-level logic controller the value of the last user-approved infusion rate in the database with the altered infusion rate accepted by the user or the different value of the altered infusion rate if changed by the user in Step (i);

(k) resetting the value of at least one element of the set of weights of the at least one lower-level controller by the higher-level logic controller to a new value based at least in part on the different value of the altered infusion rate if changed by the user in Step (i), otherwise the set of weights of the at least one lower-level controller are kept the same; and

2019280053 01 Apr 2020 (1) administering the fluid and/or cardiovascular drug at the last user-approved infusion rate in the database to the subject, thereby regulating the subject’s hemodynamic measurements.

[0015] Preferably, the function approximator is a neural network with sigmoidal basis functions.

[0016] Preferably, the user is notified if at least one performance criterion is violated.

[0017] Preferably, violation of the at least one performance criterion is determined by comparing a set threshold with the difference between the hemodynamic measurements of the subject and target hemodynamic data in a set time interval.

[0018] Preferably, the hierarchical control architecture system further comprises a second lower-level controller.

[0019] Preferably, the first lower-level controller is engaged through the higher-level logic controller, and wherein the higher-level logic controller recommends additional engagement of the second lower-level controller when at least one performance criterion is violated.

[0020] Preferably, the performance criterion comprises improvement in the hemodynamic measurements of the subject while administering the one or more fluid or cardiovascular drug.

[0021] Preferably, the last user-approved infusion rate is communicated to an infusion pump for administering the fluid and/or drug at the last user-approved infusion rate.

[0022] According to a second aspect of the invention, there is provided a method for fluid and/or cardiovascular drug administration, the method comprising:

(a) receiving by way of a user interface at least one initial infusion rate;

(b) recording the at least one initial infusion rate as a last user approved infusion rate in a database;

wherein the hierarchical control architecture system comprises:

(a) at least one lower level controller; and (b) a higher level logic controller;

wherein the subject’s hemodynamic measurements are received by the higher level logic controller and the at least one lower level controller;

wherein the at least one lower level controller and the higher level logic

2019280053 01 Apr 2020 controller are in communication with one another; and wherein the at least one lower level controller has an internal dynamical system model that adapts to a dynamical system model of the subject;

(d) initializing parameters of the internal dynamical system model of the at least one lower level controller based at least in part on the at least one initial infusion rate such that the lower level controller generates an infusion rate equal to the at least one initial infusion rate;

(e) generating a new infusion rate by the lower level controller at set time intervals based at least in part on the hemodynamic measurements and the parameters of the internal dynamical system model of the at least one lower level controller;

(f) verifying by the higher level logic controller upon generation of the new infusion rate whether the new infusion rate meets requirements to notify the user;

(g) notifying the user if the new infusion rate meets the requirement in Step (f);

(h) asking the user to either accept the new infusion rate or change the new infusion rate to a different value;

(i) replacing by the higher level logic controller the value of the last user approved infusion rate in the database with the altered infusion rate accepted by the user or the different value of the altered infusion rate if changed by the user in Step (h);

(j) resetting at least one parameter of the internal dynamical system model of the at least one lower level controller to a new value based at least in part on the different value of the altered infusion rate if changed by the user in Step (h), otherwise the set of weights of the at least one lower level controller are kept the same; and (k) administering the fluid and/or cardiovascular drug at the last user approved infusion rate to the subject.

[0023] Preferably, the internal dynamical system model of the at least one lower-level controller uses a function approximator comprising a set of basis functions and a set of weights to approximate an unknown function describing distribution of the fluid and/or drug in the body of the subject.

[0024] Preferably, the function approximator is a neural network with sigmoidal basis functions.

[0025] Preferably, at least one element of the set of weights is reset to the new value based at least in part on the last user-approved infusion rate if the new infusion rate is changed to the different value.

2019280053 01 Apr 2020 [0026] Preferably, the hierarchical control architecture system further comprises a second lower-level controller.

[0027] Preferably, the first lower-level controller is initially engaged through the higher-level controller, wherein the higher-level controller recommends additional engagement of the second lower-level controller when at least one performance criterion is violated.

[0028] Preferably, the performance criterion comprises improvement in hemodynamic measurements of the subject while administering one or more fluid or cardiovascular drug.

[0029] Preferably, the internal dynamical system model adapts to the dynamical system model of the subject by updating the parameters of the internal dynamical system of the at least one lower-level controller based at least in part on current values of the parameters of the internal dynamical system of the at least one lower-level controller and the hemodynamic measurements of the subject.

[0030] According to a third aspect of the invention, there is provided a clinical decision support system for fluid resuscitation or drug administration comprising:

a processing module configured to receive sensor-obtained hemodynamic measurements of a subject from a hemodynamic monitoring device;

at least one lower-level controller, the at least one lower-level controller comprising the infusion rate computation engine and configured to compute infusion rates based at least upon the sensor-obtained hemodynamic measurements of the subject and internal parameters;

at least one higher-level logic-based controller communicatively coupled to the at least one lower-level controller, configured to:

verify whether infusion rates of the at least one lower-level controller meet infusion rate requirements;

determine timing of engaging with the at least one lower-level controller; process end-user responses to suggested infusion rates; and determine whether to reset internal states of the at least one lower-level controller based at least in part on the end-user responses;

a user interface configured to:

display suggested infusion rates;

communicate user interactions to the at least one higher-level logic-based controller; and notify a user if at least one performance criterion is violated.

2019280053 01 Apr 2020 [0031] Preferably, the at least one lower-level controller comprises:

a dynamic observer comprising a set of dynamic observer states for estimating a current value of one or more physiological parameters of the subject; and a dynamical system using a function approximator comprising a set of basis functions and a set of weights to approximate unknown dynamics of the subject.

[0032] Preferably, the internal states of the at least one lower-level controller comprises the set of dynamic observer states and the set of weights.

[0033] Preferably, the function approximator is a neural network with sigmoidal basis functions.

[0034] Preferably, at least one element of the set of weights is reset to a new value based at least in part on the end-user responses.

[0035] Preferably, the at least one lower-level controller has an internal dynamical system model that adapts to the dynamical system model of the subject.

[0036] Preferably, violation of the at least one performance criterion to notify the user by the user interface is determined by comparing a set threshold with the difference between the hemodynamic measurements of the subject and target hemodynamic data in a set time interval.

[0037] According to a fourth aspect of the invention, there is provided a system for displaying infusion therapy information for optimizing infusion therapy, the apparatus comprising:

a memory;

a display; and one or more processor in operable communication with the memory and the display, the processor capable of:

receiving sensor-obtained hemodynamic measurements of a subject from a hemodynamic monitoring device; and issuing to the display a prompt upon determining an appropriate notification message and an appropriate change in infusion rate recommendation message wherein the notification message and the recommendation message are based upon detection of a change in infusion therapy by a higher-level logic-based controller, the detection based at least on:

the sensor-obtained hemodynamic measurements of the subject;

an active infusion rate stored in the memory; and a background infusion rate computed by a lower-level controller;

ίο

2019280053 01 Apr 2020 communicating user interactions with the issued prompt to the higher-level logic-based controller;

wherein the higher-level logic-based controller is communicatively coupled to the lower-level controller.

[0038] Preferably, the active infusion rate stored in the memory is a last user-approved infusion rate.

[0039] Preferably, the lower-level controller comprises:

a dynamic observer comprising a set of dynamic observer states capable of estimating a current value of one or more physiological parameters of the subject; and a dynamical system comprising a function approximator with a set of basis functions and a set of weights capable of approximating unknown dynamics of the subject.

[0040] Preferably, the higher-level logic-based controller further causes the one or more processors to reset to a new value at least one element of the set of weights of the dynamical system of the lower-level controller.

[0041] Preferably, the system further comprises an infusion pump for administering one or more fluid and/or cardiovascular drug.

[0042] Preferably, the lower-level controller has an internal dynamical system model that adapts to a dynamical system model of the subject.

[0043] Preferably, the prompt is issued to the display only if the absolute value of the difference between the active infusion rate and the background infusion rate is larger than a pre-determined threshold.

[0044] According to a fifth aspect of the invention, there is provided a computer-implemented method for providing closed-loop control for regulating hemodynamic measurements of a subject, the method comprising:

receiving by way of a user interface at least one initial infusion rate;

receiving from a sensor hemodynamic measurements of a subject by a hierarchical control architecture system;

wherein the hierarchical control architecture system comprises:

at least one lower-level controller; and a higher-level logic controller;

2019280053 01 Apr 2020 wherein the subject’s hemodynamic measurements are received by the higher-level logic controller and the at least one lower-level controller;

wherein the at least one lower-level controller and the higher-level logic controller are in communication with one another;

wherein the at least one lower-level controller comprises:

a dynamic observer comprising a set of dynamic observer states for estimating a current value of physiological parameters of the subject; and a dynamical system using a function approximator comprising a set of basis functions and a set of weights to approximate unknown dynamics of the subject;

initializing the value of the set of weights of the at least one lower-level controller based at least in part on the at least one initial infusion rate;

updating the set of weights of the at least one lower-level controller based at least in part on current values of the set of weights, the set of dynamic observer states, and current value of the hemodynamic measurements;

generating by the at least one lower-level controller an altered infusion rate based at least in part on the subject’s hemodynamic measurements and the set of weights of the at least one lower-level controller;

notifying the user and disengaging the at least one lower-level controller if at least one performance criterion is violated; and communicating to an infusion pump to administer the fluid and/or drug at the altered infusion rate, thereby regulating the subject’s hemodynamic measurements.

[0045] Preferably, the function approximator is a neural network with sigmoidal basis functions.

[0046] Preferably, violation of the at least one performance criterion is determined by comparing a set threshold with the difference between the hemodynamic measurements of the subject and target hemodynamic data in a set time interval.

[0047] Preferably:

the at least one lower-level controller comprises a first lower-level controller initially engaged by the higher-level logic controller; and the higher-level logic controller automatically engages a second lower-level controller when at least one patient-related performance criterion is violated.

[0048] Preferably, the patient-related performance criterion comprises improvement in the hemodynamic measurements of the subject while administering the one or more fluid or

2019280053 01 Apr 2020 cardiovascular drug.

[0049] Preferably, the higher-level logic controller verifies that the altered infusion rate is in a safe range specified by the user and if not the higher-level logic controller changes at least one value of the set of weights of the at least one lower-level controller.

[0050] According to a sixth aspect of the invention, there is provided a method for fluid and/or cardiovascular drug administration, the method comprising:

receiving by way of a user interface at least one initial infusion rate;

wherein the hierarchical control architecture system comprises:

at least one lower-level controller; and a higher-level logic controller;

wherein the at least one lower-level controller and the higher-level logic controller are in communication with one another; and wherein the at least one lower-level controller has an internal dynamical system model that adapts to a dynamical system model of the subject;

initializing parameters of the internal dynamical system model of the at least one lower-level controller based at least in part on the at least one initial infusion rate such that the lower-level controller generates an infusion rate equal to the at least one initial infusion rate;

generating a new infusion rate by the lower-level controller at set time intervals based at least in part on the hemodynamic measurements and the parameters of the internal dynamical system model of the at least one lower-level controller;

notifying the user and disengaging the at least one lower-level controller if at least one performance criterion is violated; and administering the fluid and/or cardiovascular drug at the new infusion rate to the subject.

[0051 ] Preferably, violation of the at least one performance criterion is determined by comparing a set threshold with the difference between the hemodynamic measurements of the subject and target hemodynamic data in a set time interval.

[0052] Preferably, the internal dynamical system model of the at least one lower-level controller uses a function approximator comprising a set of basis functions and a set of weights to

2019280053 01 Apr 2020 approximate an unknown function describing distribution of the fluid and/or drug in the body of the subject.

[0053] Preferably, the function approximator is a neural network with sigmoidal basis functions.

[0054] Preferably, at least one element of the set of weights is reset to a new value if the new infusion rate falls outside the safe range.

[0055] Preferably:

the at least one lower-level controller comprises a first lower-level controller initially engaged by the higher-level controller; and the higher-level controller automatically engages a second lower-level controller when at least one patient-related performance criterion is violated.

[0056] Preferably, the patient-related performance criterion comprises improvement in hemodynamic measurements of the subject while administering one or more fluid or cardiovascular drug.

[0057] Preferably, the internal dynamical system model adapts to the dynamical system model of the subject by updating the parameters of the internal dynamical system of the at least one lower-level controller based at least in part on current values of the parameters of the internal dynamical system of the at least one lower-level controller and the hemodynamic measurements of the subject.

[0058] Preferably, the higher-level logic controller verifies that the new infusion rate is in a safe range specified by the user and if not the higher-level logic controller changes at least one parameter of the internal dynamical system model of the at least one lower-level controller.

[0059] According to a seventh aspect of the invention, there is provided a closed-loop system for fluid resuscitation or drug administration comprising:

at least one lower-level controller, the at least one lower-level controller comprising the infusion rate computation engine and configured to compute infusion rates based at least upon the sensor-obtained hemodynamic measurements of the subject and internal parameters; and at least one higher-level logic-based controller communicatively coupled to the at least one lower-level controller, configured to:

2019280053 01 Apr 2020 verify whether infusion rates of the at least one lower-level controller meet infusion rate requirements;

determine timing of engaging with the at least one lower-level controller; and determine timing of disengaging the at least one lower-level controller based on violation of a performance criterion;

a user interface configured to:

display computed infusion rates;

communicate initial infusion rate by the end-user to the at least one higher-level logic-based controller; and notify the end-user if at least one performance criterion is violated.

[0060] Preferably, the at least one lower-level controller comprises:

[0061] Preferably, the internal states of the at least one lower-level controller comprises the set of dynamic observer states and the set of weights.

[0062] Preferably, the function approximator is a neural network with sigmoidal basis functions.

[0063] Preferably, at least one element of the set of weights is reset to a new value if the infusion rate falls outside a user-specified safe range.

[0064] Preferably, the at least one lower-level controller has an internal dynamical system model that adapts to the dynamical system model of the subject.

[0065] Preferably, the criterion for the disengagement of the at least one lower level controller is determined by comparing a set threshold with the difference between the hemodynamic measurements of the subject and target hemodynamic data in a set time interval.

Brief Description of the Figures [0066] It should be noted in the following description that like or the same reference numerals in different embodiments denote the same or similar features.

[0067] Notwithstanding any other forms which may fall within the scope of the present invention,

2019280053 01 Apr 2020 a preferred embodiment/preferred embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings in which:

FIG. 1A is a diagram that illustrates a two-compartment model of fluid exchange in the body for non-burn patients;

FIG. IB is a diagram that illustrates a three-compartment model of the microvascular exchange system for burn patients;

FIG. 1C is a diagram that illustrates a two-compartment model of cardiovascular drug exchange in the body;

FIG. 2A is a diagram that illustrates the overall architecture of a fully automated closed-loop fluid resuscitation or cardiovascular drug administration system;

FIG. 2B is a diagram that illustrates the overall architecture of a fully automated closed-loop fluid resuscitation and cardiovascular drug administration system;

FIG. 3A is a diagram that illustrates the overall architecture of a partially automated (clinical decision support) fluid resuscitation or cardiovascular drug administration system;

FIG. 3B is a diagram that illustrates the overall architecture of a partially automated (clinical decision support) fluid resuscitation and cardiovascular drug administration system;

FIG. 4A is a diagram that illustrates components of a fully automated, closed-loop fluid resuscitation or cardiovascular drug administration system;

FIG. 4B is a diagram that illustrates components of a fully automated, closed-loop fluid resuscitation and cardiovascular drug administration system;

FIG. 5A is a diagram that illustrates components of a partially automated clinical decision support fluid resuscitation or cardiovascular drug administration system;

FIG. 5B is a diagram that illustrates components of a partially automated clinical decision support fluid resuscitation and cardiovascular drug administration system;

FIG. 6 is a flow chart that illustrates the lower-level adaptive control architecture according to embodiments;

FIG. 7 is a graph that depicts exemplary stroke volume variation versus time;

2019280053 01 Apr 2020

FIG. 8 is a graph that depicts exemplary fluid infusion rates computed by the adaptive control framework;

FIG. 9 is a graph that depicts exemplary plasma volume in circulation versus time;

FIG. 10 is a graph that depicts exemplary changes in filtered SW(%) in 2 canine subjects experiencing controlled hemorrhage;

FIG. 11 is a graph that depicts exemplary changes in infusion rates in 2 canine subjects experiencing controlled hemorrhage;

FIG. 12 is a graph that depicts exemplary changes in filtered SW(%) in 3 canine subjects experiencing uncontrolled hemorrhage;

FIG. 13 is a graph that depicts exemplary changes in infusion rates in 3 canine subjects experiencing uncontrolled hemorrhage;

FIG. 14 is a graph that depicts exemplary changes in filtered SW(%) in 2 canine subjects that were hypotensive as a result of administration of sodium nitroprusside (S4-1) and increase in isoflurane (S5-1);

FIG. 15 is a graph that depicts exemplary changes in infusion rates in 2 canine subjects that were hypotensive as a result of administration of sodium nitroprusside (S4-1) and increase in isoflurane (S5-1);

FIG. 16 is a flow chart that illustrates components of the higher-level controller for the closed-loop fluid resuscitation system in embodiments;

FIG. 17 is a flow chart that illustrates components of the higher-level controller for the clinical decision support case in embodiments;

FIG. 18 is a graph that depicts exemplary mean arterial pressure versus time;

FIG. 19 is a graph that depicts exemplary cardiovascular drug infusion rates computed by the adaptive control framework;

FIG. 20 is a graph that depicts exemplary mean arterial pressure versus time;

FIG. 21 is a graph that depicts exemplary fluid infusion rates computed by the adaptive control framework;

2019280053 01 Apr 2020

FIG. 22 is a graph that depicts exemplary stroke volume variation versus time;

FIG. 23 is a graph that depicts exemplary fluid infusion rates computed by the fluid resuscitation adaptive control framework;

FIG. 24 is a graph that depicts exemplary mean arterial pressure versus time;

FIG. 25 is a graph that depicts exemplary vasopressor infusion rates computed by the cardiovascular drug administration adaptive control framework;

FIG. 26 is a diagram that depicts an embodiment of a fluid resuscitation and cardiovascular drug administration system implemented on hardware;

FIG. 27 is a block diagram illustrating an embodiment of the apparatus for displaying infusion therapy information for optimizing infusion therapy;

FIG. 28 is a diagram that depicts an embodiment of the apparatus for displaying infusion therapy information for optimizing infusion therapy;

FIG. 29 is a flow chart that illustrates the administration of pre-approved ranges of infusion rates according to embodiments; and

FIG. 30 is a flow chart that illustrates the alarm function to prevent under- or over-administration of fluid and/or drug according to embodiments.

Detailed Description [0068] Reliable and consistent fluid resuscitation (i.e., intravenous administration of fluids) is critical for perioperative and intensive care unit (ICU) fluid management in human and animal populations. Fluid management is required for surgical patients as well as patients suffering from hypovolemia, sepsis, severe sepsis, septic shock, burn, and other conditions. The goal of fluid resuscitation is to restore blood volume in the circulatory system to an acceptable level in order to ensure adequate tissue perfusion (i.e., blood delivery to tissue). However, large intra-patient and inter-patient variability in physiological parameters, and the effects of different illnesses and medications can result in under-resuscitation and over-resuscitation of patients.

[0069] Under-resuscitation results in hypovolemia, which can lead to hypoperfusion and organ failure. Over-resuscitation results in fluid overload, which can lead to complications such as pulmonary edema. Fluid overload is associated with higher rates of morbidity and mortality.

2019280053 01 Apr 2020

Restrictive fluid resuscitation protocols reduce the number of mechanical ventilation days and the length of hospital stays.

[0070] Cardiovascular drug administration can be used independently to address a clinical condition (e.g., vasopressors are used to increase blood pressure to a clinically acceptable value or inotropic agents are used to change the contractility of the heart). Cardiovascular drugs, such as vasopressors, are administered independent of fluids in critical care for hypotensive patients. Cardiovascular drugs can also be used in combination with fluid resuscitation. For example, to address hypotension and hypovolemia in sepsis and during surgery, a vasopressor and fluid can be administered simultaneously.

[0071] The present disclosure describes a reliable and consistent closed-loop fluid resuscitation system, a clinical decision support fluid resuscitation system, a closed-loop cardiovascular drug administration system, a clinical decision support cardiovascular drug administration system, a closed-loop fluid and cardiovascular drug administration system, and a clinical decision support fluid and cardiovascular drug administration system. The system uses continuous measurements from standard operating room (OR) or ICU hemodynamic monitoring devices or sensors or a built-in or add-on modules to measure such continuous measures to compute the required fluid and/or cardiovascular drug infusion rates for patients receiving continuous infusion. Adaptive control architecture is used to compute the required infusion rates to regulate an endpoint of fluid or drug administration including, but not limited to, static indicators of fluid responsiveness and dynamic indicators of fluid responsiveness for the fluid resuscitation module, and hemodynamic measures for the cardiovascular drug administration module. In some embodiments, the static indicators of fluid responsiveness include mean arterial pressure, central venous pressure, heart rates, cardiac output, stroke volume, cardiac index, and urine output rates of patients. In some embodiments, dynamic indicators of fluid responsiveness include stroke volume variation, pulse pressure variations, systolic pressure variation, dynamic arterial elastance, and pleth variability indices of patients. In some embodiments, hemodynamic measures include mean arterial pressure, central venous pressure, systolic pressure, diastolic pressure, heart rate, cardiac output, cardiac index, systemic vascular resistance, stroke volume, and urine output. The closed-loop fluid resuscitation and/or cardiovascular drug administration system described herein comprises an adaptive controller or two adaptive controllers that use a function approximator, such as a neural network, Fourier functions, or wavelets, to identify the unknown dynamics and physiological parameters of a patient to compute appropriate infusion rates and to regulate the endpoint of fluid resuscitation or cardiovascular drug administration.

2019280053 01 Apr 2020 [0072] The developed fluid resuscitation system can use either crystalloids or colloids during resuscitation and use vasoactive cardiovascular drugs (e.g., vasopressor) or inotropic agents. The fluid infusion rate (e.g., in mL/hour) is highly dependent on a patient needs. In some embodiments, the fluid infusion rate can range from 0 to 3,000 mL/hr or more in humans, and 0 to 40,000 mL/h or more in animals. The cardiovascular drug infusion rate (e.g., in mcg/kg/min) is highly dependent on a patient’s needs. In some embodiments, the cardiovascular drug infusion rate (e.g., vasopressor or inotropic agents) can range from 0 mcg/kg/min to 40 mcg/kg/min, or can exceed 40 mcg/kg/min.

[0073] Any method or algorithm described herein can be embodied in software or set of computer-executable instructions capable of being run on a computing device or devices. The computing device or devices can include one or more processor (CPU) and a computer memory. The computer memory can be or include a non-transitory computer storage media such as RAM which stores the set of computer-executable (also known herein as computer readable) instructions (software) for instructing the processor(s) to carry out any of the algorithms, methods, or routines described in this disclosure. As used in the context of this disclosure, a non-transitory computer-readable medium (or media) can include any kind of computer memory, including magnetic storage media, optical storage media, nonvolatile memory storage media, and volatile memory. Non-limiting examples of non-transitory computer-readable storage media include floppy disks, magnetic tape, conventional hard disks, CD-ROM, DVD-ROM, BLU-RAY, Flash ROM, memory cards, optical drives, solid state drives, flash drives, erasable programmable read only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), non-volatile ROM, and RAM. The computer-readable instructions can be programmed in any suitable programming language, including JavaScript, C, C#, C++, Java, Python, Perl, Ruby, Swift, Visual Basic, and Objective C. Embodiments of the invention also include a non-transitory computer readable storage medium having any of the computer-executable instructions described herein.

[0074] A skilled artisan will further appreciate, in light of this disclosure, how the invention can be implemented, in addition to software and hardware, using one or more firmware. As such, embodiments of the invention can be implemented in a system which includes any combination of software, hardware, or firmware. In the context of this specification, the term “firmware” can include any software programmed onto the computing device, such as a device’s nonvolatile memory. Thus, systems of the invention can also include, alternatively or in addition to the computer-executable instructions, various firmware modules configured to perform the algorithms

2019280053 01 Apr 2020 of the invention.

[0075] According to embodiments, the computing device or devices can include a mainframe computer, web server, database server, desktop computer, laptop, tablet, netbook, notebook, personal digital assistant (PDA), gaming console, e-reader, smartphone, or smartwatch, which may include features such as a processor, memory, hard drive, graphics processing unit (GPU), and input/output devices such as display, keyboard, and mouse or trackpad (depending on the device). Embodiments can also provide a graphical user interface made available on one or more client computers. The graphical user interface can allow a user on a client computer remote access to the method or algorithm.

[0076] Additional embodiments of the invention can include a networked computer system for carrying out one or more methods of the invention. The computer system can include one or more computing devices which can include a processor for executing computer-executable instructions, one or more databases, a user interface, and a set of instructions (e.g., software) for carrying out one or more methods of the invention. According to other embodiments, the computing device or devices can be connected to a network through any suitable network protocol such as IP, TCP/IP, UDP, or ICMP, such as in a client-server configuration and one or more database servers. The network can use any suitable network protocol and can be any suitable wired or wireless network including any local area network, wide area network, Internet network, telecommunications network, Wi-Fi enabled network, or Bluetooth enabled network.

Compartmental Modeling for Characterizing Fluid Distribution [0077] The present disclosure models the microvascular exchange system to characterize the distribution of fluids in the body.

[0078] A two-compartment dynamic system model is used for all patient populations except for patients with burn injuries. The compartments for two-compartment dynamic system models include circulation (blood) and interstitial tissue. A 4th-order nonlinear state space model representation of the dynamic system can provide a simplified model of the microvascular exchange system for non-burn patients. The states of the two-compartment dynamic system include the volume of fluids and albumin mass in each compartment (a total of four states).

[0079] A three-compartment model is used for patients with burn injuries. The states of the three-compartment dynamic system model include circulation (blood), injured tissue, and uninjured tissue. A 6th-order nonlinear state space model representation of the dynamic system

2019280053 01 Apr 2020 can provide a simplified model of the microvascular exchange system for bum patients. The states of the three-compartment dynamic system include volume of fluids and albumin mass in each compartment (a total of six states).

[0080] The parameters of the two- and three-compartment models of the microvascular exchange system characterize fluid and mass exchange between different compartments. These parameters are generally unknown and are different from patient to patient.

[0081] FIG. 1A illustrates a two-compartment model of the fluid exchange in the body for non-burn patients. FIG. IB illustrates a three-compartment model of the microvascular exchange system for burn patients.

[0082] In some embodiments, a higher number of compartments are used to model the microvascular exchange system of a patient population. In some embodiments, a model dynamic system consists of 2, 3, 4, or 5 compartments. In some embodiments, a 1, 2, 3, or 4-compartment dynamic system model is used for a patient population. In some embodiments, a two-compartment dynamic system model is used for a patient population. In some embodiments, a three-compartment dynamic system model is used for a patient population. In some embodiments, a four-compartment dynamic system model is used for a patient population.

Compartmental Modeling for Characterizing Cardiovascular Drug Distribution [0083] The present disclosure models the cardiovascular drug distribution in the body using a compartmental model.

[0084] Cardiovascular drug distribution can be modeled using a two-compartment model. The compartments for two-compartment dynamic system models include circulation (blood) and tissue. A 2nd-order nonlinear state space model representation of the dynamic system can provide a simplified model of the cardiovascular drug distribution for patients. The states of the two-compartment dynamic system include the concentration of cardiovascular drug in each compartment.

[0085] The parameters of the two-compartment model of the cardiovascular drug distribution characterizes cardiovascular drug mass exchange between different compartments. These parameters are generally unknown and are different from patient to patient. FIG. 1C illustrates a two-compartment model of the cardiovascular drug distribution in the body.

[0086] In some embodiments, a higher number of compartments are used to model the

2019280053 01 Apr 2020 cardiovascular drug distribution for a patient. In some embodiments, a model dynamic system consists of 2, 3, 4, or 5 compartments. In some embodiments, a 1, 2, 3, or 4-compartment dynamic system model is used for a patient population. In some embodiments, a two-compartment dynamic system model is used for a patient population. In some embodiments, a three-compartment dynamic system model is used for a patient population. In some embodiments, a four-compartment dynamic system model is used for a patient population.

Process for Computing Fluid Infusion Rate and/or Cardiovascular Drug Infusion Rate for a Closed-Loop System or a Clinical Decision Support System [0087] The disclosure provides an adaptive control architecture framework for a closed-loop fluid resuscitation and/or cardiovascular drug administration system. The fluid resuscitation and/or cardiovascular drug administration system of the disclosure receives data from a monitoring system or a set of monitoring systems. In some embodiments, the fluid resuscitation and/or cardiovascular drug administration system receives data from an existing monitoring system or systems. In some embodiments, the fluid resuscitation and/or cardiovascular drug administration system receives data from a built-in monitoring system or systems. In some embodiments, the fluid resuscitation and/or cardiovascular drug administration system receives data from an add-on monitoring system or systems.

[0088] The received data comprises one of: blood pressure, heart rate, stroke volume variation, pulse pressure variation, dynamic arterial elastance, pleth variability index, urine output rate, systolic pressure variation, central venous pressure, mean arterial pressure, cardiac output, cardiac index, systolic pressure, diastolic pressure, systemic vascular resistance, or stroke volume. In some embodiments, the received data comprises a combination of blood pressure, heart rate, stroke volume variation, pulse pressure variation, dynamic arterial elastance, pleth variability index, urine output rate or urine output, systolic pressure variation, central venous pressure, mean arterial pressure, cardiac output, cardiac index, systolic pressure, diastolic pressure, systemic vascular resistance, or stroke volume. In some embodiments, a built-in or add-on monitoring system or systems are used and the input data includes one or a combination of an invasive blood pressure signal or a non-invasive pressure signal collected from a patient’s arm or leg using a blood pressure cuff.

[0089] The disclosed fluid resuscitation and/or cardiovascular drug administration system can transmit data to a receiver or a set of receivers. In some embodiments, the disclosed fluid resuscitation and/or cardiovascular drug administration system can transmit data to an external or

2019280053 01 Apr 2020 built-in infusion pump or infusion pumps, a user, an electronic medical record, or a remote location. In some embodiments, the disclosed fluid resuscitation and/or cardiovascular drug administration system can transmit data to a combination of receivers. In some embodiments, the disclosed fluid resuscitation and/or cardiovascular drug administration system can transmit data to an infusion pump or infusion pumps and a user or users. In some embodiments, the disclosed fluid resuscitation system can transmit data to an infusion pump or infusion pumps, a user or users, and an electric medical record. In some embodiments, the disclosed fluid resuscitation system can transmit data to an infusion pump or infusion pumps, a user or users, an electronic medical record, and a remote location.

[0090] The adaptive architecture of the disclosure can be implemented in a fully automated architecture or a partially automated architecture.

[0091] Full automation—In some embodiments, the adaptive architecture of the disclosure is implemented in a fully automated architecture, wherein the infusion rate of the infusion pump or infusion rates of the infusion pumps are updated automatically by the system using the most recent value of the infusion rate or infusion rates.

[0092] Partial automation (also referred to as clinical decision support)—In some embodiments, the adaptive architecture of the disclosure is implemented in a partially automated architecture. In some embodiments, the framework of the disclosure is used within a clinical decision support context, where recommended infusion rates are displayed to the end-user (clinician) for approval. In some embodiments, the system can request for approval before changing the infusion rate of the pump or pumps. In some embodiments, the end-user can use recommended infusion rate changes and manually change the infusion rate on the pump or pumps based on his/her clinical judgment and the recommendation provided by the clinical decision support system. In some embodiments, the end-user can enter whether the recommended infusion rate was accepted or rejected. In some embodiments, the end-user can enter a new manually changed infusion rate or infusion rates. In some embodiments, the system can automatically change the infusion rate of the pump or infusion rates of the pump after the user approves or modifies the recommended infusion rate or infusion rates.

Architecture of the Disclosure [0093] The present disclosure is able to determine suitable infusion rates based on an input from patient-monitoring devices, built-in monitoring devices, add-on monitoring modules, or sensors. Infusion rates can be adjusted based on the input from patient-monitoring devices, built-in

2019280053 01 Apr 2020 monitoring devices, add-on monitoring modules, or sensors. The adaptive control framework of the disclosure does not require any patient-specific information (e.g., age, gender, weight, diagnosis). Furthermore, the framework does not require an accurate model of the patient dynamics and the patient-specific physiological parameters.

[0094] The present disclosure describes a closed-loop or clinical decision support fluid resuscitation and/or cardiovascular drug administration system predicated on hierarchical adaptive control architecture. In some embodiments, the hierarchical control architecture is composed of one or two lower-level adaptive controllers and a higher-level, logic-based controller. In some embodiments, the higher-level, logic-based controller is a rule-based expert system.

[0095] If the system is used for fluid resuscitation, the hierarchical control architecture is composed of an adaptive controller fluid module and a higher-level, logic-based controller. If the system is used for only cardiovascular drug administration, the hierarchical control architecture is composed of an adaptive controller cardiovascular drug module and a higher-level, logic-based controller. If the system is used for fluid resuscitation and cardiovascular drug administration, the hierarchical control architecture is composed of an adaptive controller fluid module, an adaptive controller cardiovascular drug module, and a higher-level, logic-based controller.

[0096] The lower-level controller focused on fluid resuscitation uses an adaptive control framework to regulate a measure of fluid responsiveness to a desired value by adjusting the fluid infusion rate. In some embodiments, the lower-level controller can regulate mean arterial pressure, systolic pressure, diastolic pressure or a measure of fluid responsiveness including stroke volume variation, pleth variability index, pulse pressure variation, dynamic arterial elastance, central venous pressure, urine output rate or urine output, or systolic pressure variation. While the goal of this lower-level controller is to regulate a measure of fluid responsiveness to a desired value, the controller may achieve a measurement that is close to the desired value (with some error). Lower-level controller design parameters can be changed to adjust the value of this error.

[0097] The lower-level controller focused on cardiovascular drug administration uses an adaptive control framework to regulate a hemodynamic measure to a desired value by adjusting the cardiovascular drug infusion rate. In some embodiments, the lower-level controller can regulate mean arterial pressure, systolic pressure, diastolic pressure, heart rate, cardiac output, stroke volume, systemic vascular resistance, or cardiac index. In some embodiments the administered cardiovascular drug is a vasopressor and the lower-level controller can regulate mean arterial pressure, systolic pressure, systemic vascular resistance, or diastolic pressure. In some

2019280053 01 Apr 2020 embodiments, the administered cardiovascular drug is an inotropic agent and the lower-level controller can regulate cardiac output, cardiac index, mean arterial pressure, systemic vascular resistance, or heart rate. In some embodiments, the administered cardiovascular drug is a chronotropic agent, and the lower-level controller can regulate heart rate or cardiac output. While the goal of this lower-level controller is to regulate a hemodynamic measure to a desired value, the controller may achieve a measurement that is close to the desired value (with some error). Lower-level controller design parameters can be changed to adjust the value of this error.

[0098] The role of the higher-level, logic-based controller is different depending on whether the system is used to fully automate or partially automate fluid resuscitation and/or cardiovascular drug administration. The higher-level, logic-based controller monitors the performance of the lower-level adaptive controller(s) and the patient’s response to therapy. In case of fluid management, a measure of fluid responsiveness or tissue perfusion is monitored (e.g., mean arterial pressure, stroke volume variation, pulse pressure variation, systolic pressure variation, dynamic arterial elastance, pleth variability index etc.) and in case of cardiovascular drug administration, a measure of hemodynamic function is monitored (e.g., mean arterial pressure, heart rate etc.). If certain performance criteria are violated, then the higher-level controller can “disengage” the lower-level controlled s) (if the system is used for full automation of fluid resuscitation) or the higher-level controller will stop providing suggested infusion rates and will alert the end-user (if the system is used for partial automation of fluid resuscitation). Time stamps, infusion rates and measurement data can be sent to an internal of external database by the higher-level controller for archiving purposes.

[0099] The higher-level controller can also determine the timing of engaging each lower-level controller. In some embodiments, the higher-level controller engages the fluid resuscitation module first, and if some performance criteria is not met after a period of time, the drug administration module is also engaged. The higher-level controller (in clinical decision support) can also determine the when to notify the user. In some embodiments, the higher-level controller notifies the user only if the difference between the newly computed infusion rate by the lower-level controller and the last user-approved infusion rate is higher than some threshold.

[0100] In some embodiments, in a partial automation application, the higher-level controller can use end-user response (to accept or reject the suggested infusion rate) to update the internal state of the lower-level controller. In addition, in both full automation and partial automation, the higher-level controller can change the internal states of the lower-level controlled s) if the computed infusion rate is out of a “safe” range defined by the end-user. In some embodiments, if

2019280053 01 Apr 2020 the infusion rate computed by the lower-level controller exceeds the maximum allowable infusion rate then the higher-level controller resets the internal parameters and variables of the controller to preset default values.

[0101] The higher-level controller can also provide recommendation to the user to engage the cardiovascular drug administration module if a patient’s hemodynamic variable of interest is not improved after a certain period of time after fluid resuscitation. In some embodiments, the higher-level controller asks the user to engage the fluid resuscitation module first, and if some performance criteria is not met after a period of time, asks the user to engage the drug administration module. In some embodiments, the higher-level controller asks the user to engage the cardiovascular drug administration module first, and if some performance criteria is not met after a period of time, asks the user to engage the fluid resuscitation module.

[0102] FIG. 2A illustrates the overall architecture of a fully automated closed-loop fluid resuscitation system or a fully automated closed-loop cardiovascular drug administration system of the disclosure. A sensor (e.g., hemodynamic monitor) sends data to the lower-level controller and the higher-level controller. The higher-level controller monitors the performance of the lower-level controller and the response of the patient to fluids or cardiovascular drugs by monitoring measurements from the sensor, internal state of the lower-level controller, and infusion rate computed by the lower-level controller (which is sent to the infusion pump by the higher-level controller). The lower-level controller can send or receive data from the higher-level logic controller.

[0103] The higher-level controller can send or receive data from the lower-level controller. The human user (clinician) can interact with the closed-loop system through a user interface to set a target value for the measurement (e.g., set target stroke volume variation of 13% or set mean arterial pressure of 65mmHg), set the range of “safe” infusion rates (e.g., between 0 and 3,000 mL/hr for fluids or 0 and 0.5 mcg/kg/min for a vasopressor), start and stop the system, or set a backup infusion rate in case of loss of sensor signal (e.g., 1,000 mL/hr for fluid or 0.2 mcg/kg/min for vasopressor). The lower-level controller processes a patient’s data received from a sensor or a hemodynamic monitoring device (external or built-in), and sends computed infusion rates to the higher-level controller. The higher-level controller ensures that the infusion rate meets all requirements (e.g., it is in the safe range) and if so sends a command to an infusion pump (external or built-in), which administers fluids or cardiovascular drugs to the patient.

[0104] FIG. 2B illustrates the overall architecture of a fully automated closed-loop fluid

2019280053 01 Apr 2020 resuscitation system and cardiovascular drug administration system of the disclosure. In this system, the closed-loop system provides both fluid and cardiovascular drug simultaneously. One or two sensors (e.g., hemodynamic monitor and vital sign monitor) send data to the lower-level controller fluid module and lower-level controller cardiovascular drug module and the higher-level controller. The higher-level controller monitors the performance of the lower-level controllers and the response of the patient to fluid and cardiovascular drug by monitoring measurements from the sensors, internal state of the lower-level controllers, and infusion rates computed by the lower-level controllers (which are sent to two different infusion pumps by the higher-level controller: a fluid infusion pump and a cardiovascular drug infusion pump). Lower-level controllers can send or receive data from the higher-level logic controller.

[0105] The higher-level controller can send or receive data from the lower-level controllers. The human user (clinician) can interact with the closed-loop system through a user interface to set a target value for the measurement or measurements (e.g., set target stroke volume variation of 13% and mean arterial pressure of 65mmHg), set the range of “safe” infusion rates (e.g., between 0 and 3,000 mL/hr for fluids and 0 and 0.5 mcg/kg/min for a vasopressor), start and stop the system, or set a backup infusion rate in case of loss of sensor signal (e.g., 1,000 mL/hr for fluid and 0.2 mcg/kg/min for vasopressor). The lower-level controllers process patient’s data received from one or more sensors or a hemodynamic monitoring devices (external or built-in), and sends computed infusion rates to the higher-level controller. The higher-level controller ensures that the infusion rate meets all requirements (e.g., they are in the safe range) and if so sends a commands to infusion pumps (external or built-in), which administer fluids and cardiovascular drugs to the patient. Data measured from sensors used by the lower-level fluid module and lower-level cardiovascular drug administration module may be the same (e.g., mean arterial pressure for both modules) or different (e.g., stroke volume variation for fluid module and mean arterial pressure for cardiovascular drug module).

[0106] FIG. 3A illustrates the overall architecture of a partially automated (clinical decision support) fluid resuscitation or cardiovascular drug administration system of the disclosure. A monitoring device or sensor (e.g., hemodynamic monitor, vital sign monitor, urometer, etc.) sends data to the lower-level controller and the higher-level controller. The higher-level controller monitors the performance of the lower-level controller and the response of the patient to fluids or cardiovascular drugs by monitoring measurements from the sensor, internal state of the lower-level controller, and infusion rate computed by the lower-level controller and the action taken by the human user (clinician). The lower-level controller can send or receive data from the higher-level

2019280053 01 Apr 2020 logic controller. The higher-level controller can send or receive data from the lower-level logic controller.

[0107] The human user can interact with the partially automated (clinical decision support) system through a user interface to set a target value for the measurement (e.g., set target stroke volume variation of 13% or set target mean arterial pressure of 65mmHg), set the range of “safe” infusion rates (e.g., between 0 and 3,000 mL/hr for fluid or 0 to 0.5 mcg/kg/min for vasopressor), and start and stop the system. The lower-level controller can process a patient’s data received from a sensor or a hemodynamic monitoring device (external or built-in), and send recommended infusion rates to the user interface to be displayed. The human user can then either accept or change the recommended rate to an acceptable value. The human user can then manually change the infusion rate on the pump (if pump is not built-in or a pump that is not connected to the system by wire or wireless connection) or instruct the system to change the infusion rate (for built-in or a pump that is connected to the system by wire or wireless connection).

[0108] FIG. 3B illustrates the overall architecture of a partially automated (clinical decision support) fluid resuscitation and cardiovascular drug administration system of the disclosure. In this architecture fluid and cardiovascular drug is administered simultaneously. A monitoring device or sensor or two monitoring devices and sensors (e.g., hemodynamic monitor, vital sign monitor, urometer) sends data to the lower-level controller fluid module and a lower-level controller cardiovascular drug module and the higher-level controller. The higher-level controller monitors the performance of the lower-level controllers and the response of the patient to fluids and cardiovascular drugs by monitoring measurements from the sensor or sensors, internal state of the lower-level controllers, and infusion rate computed by the lower-level controllers and the action taken by the human user (clinician). Lower-level controllers can send or receive data from the higher-level logic controller. The higher-level controller can send or receive data from the lower-level logic controllers.

[0109] The human user can interact with the partially automated (clinical decision support) system through a user interface to set a target value for the measurement or measurements (e.g., set target stroke volume variation of 13% and set target mean arterial pressure of 65mmHg), set the range of “safe” infusion rates (e.g., between 0 and 3,000 mL/hr for fluid and 0 to 0.5 mcg/kg/min for vasopressor), and start and stop the system. Lower-level controllers can process patient’s data received from one or more sensors or hemodynamic monitoring devices (external or built-in), and send recommended infusion rates to the user interface to be displayed. The human user can then either accept or change the recommended rate to an acceptable value. The human user can then

2019280053 01 Apr 2020 manually change the infusion rate on the pump, if the pump is not built-in or a pump is not connected to the system by wire or wireless connection. The human user can also manually instruct the system to change the infusion rate for a built-in pump or a pump that is connected to the system by wire or wireless connection.

[0110] FIG. 4A illustrates components of a fully automated, closed-loop fluid resuscitation or cardiovascular drug administration system of the disclosure. A hemodynamic monitor or sensor sends a patient’s data to a sensor measurement database. An infusion rate computation engine, which is embedded in the lower-level controller, retrieves sensor measurements and computes infusion rates. The infusion rates (and the corresponding sensor measurements) are communicated with an infusion rate database. The infusion rate computation engine can send data to the infusion rate database and an infusion rate verification system. The infusion rate verification system, which is embedded in the higher-level logic-based controller, ensures the computed rates meet the requirements and if acceptable sends the newly computed infusion rate to the infusion pump controller. The infusion pump controller then automatically changes the infusion rate of the infusion pump to administer an amount of fluid or cardiovascular drug based on commands received by the infusion pump controller.

[0111] FIG. 4B illustrates components of a fully automated, closed-loop fluid resuscitation and cardiovascular drug administration system of the disclosure. The overall system is composed of two subsystems: a subsystem for fluid management and a subsystem for cardiovascular drug management. In each subsystem, a hemodynamic monitor or sensor sends a patient’s data to a sensor measurement database. An infusion rate computation engine, which is embedded in the lower-level controller, retrieves sensor measurements and computes infusion rates. The infusion rates and the corresponding sensor measurements are communicated with an infusion rate database. The infusion rate computation engine can send data to the infusion rate database and an infusion rate verification system. The infusion rate verification system, which is embedded in the higher-level logic-based controller, ensures the computed rates meet the requirements and if acceptable sends the newly computed infusion rate to the infusion pump controller. The infusion pump controller then automatically changes the infusion rate of the infusion pump to administer an amount of fluid or cardiovascular drug based on commands received by the infusion pump controller.

[0112] FIG. 5A illustrates components of a partially-automated clinical decision support fluid resuscitation or cardiovascular drug administration system of the disclosure. A hemodynamic monitor or sensor sends a patient’s data to a sensor measurement database. An infusion rate

2019280053 01 Apr 2020 computation engine, which is embedded in the lower-level controller, retrieves sensor measurements and computes infusion rates. The infusion rates and the corresponding sensor measurements are communicated with an infusion rate database. The infusion rate computation engine sends data to the infusion rate verification system, which is embedded in the higher-level logic-based controller. The infusion rate verification system ensures the computed rates meet the requirements including the requirement to notify the user of significant changes in infusion rate and if acceptable notifies the clinician using a user interface. The recommended new infusion rate is presented to the clinician, who either approves the recommended infusion rate or requests a modification of the rate based on the clinician’s qualitative judgement. The approved or modified infusion rate is sent to a database archiving clinician approved infusion rates. The clinician administers the approved or modified fluid or cardiovascular drug by either changing the infusion rate manually, if a pump is not built-in or the pump is not connected to the system by wire or wireless connection. The clinician can also instruct the system to change the infusion rate to the approved value if a pump is built-in or a pump is connected to the system by wire or wireless connection.

[0113] FIG. 5B illustrates components of a partially-automated clinical decision support fluid resuscitation and cardiovascular drug administration system of the disclosure. The overall system is composed of two subsystems: a subsystem for fluid management and a subsystem for cardiovascular drug management. In each subsystem, a hemodynamic monitor or sensor sends a patient’s data to a sensor measurement database. An infusion rate computation engine retrieves sensor measurements and computes infusion rates. The infusion rates and the corresponding sensor measurements are communicated with an infusion rate database. The infusion rate computation engine, which is embedded in the lower-level controller, can send data to the infusion rate verification system. The infusion rate verification system, which is embedded in the higher-level logic-based controller, ensures the computed rates meet the requirements including the requirement to notify the user of significant changes in infusion rate and if acceptable notifies the clinician using a user interface. The recommended new infusion rates are presented to the clinician, who either approves the recommended infusion rates or requests a modification of the rates based on the clinician’s qualitative judgement. The approved or modified infusion rates are sent to a database archiving clinician approved infusion rates. The clinician administers the approved or modified fluid and cardiovascular drug by changing the infusion rates manually if a pump is not built-in or a pump that is not connected to the system by wire or wireless connection. The clinician can also administer the approved or modified fluid and cardiovascular drug by instructing the system to change the infusion rates to the approved value if a pump is built-in or a pump is

2019280053 01 Apr 2020 connected to the system by wire or wireless connection.

Examples

Example 1—Computing Fluid or Drug Infusion Rates Using Lower-Level Adaptive Control Architecture [0114] The present disclosure describes a process of computing the fluid or cardiovascular drug infusion rate using lower-level adaptive control architecture. The lower-level adaptive control architecture is applied to a problem involving only fluid administration, only cardiovascular administration, or fluid and cardiovascular drug administration. In the case of a combined fluid and cardiovascular drug administration, two lower-level adaptive controllers running in parallel are implemented to compute infusion rates for fluid and cardiovascular drug. The fluid or cardiovascular drug infusion rate is computed using lower-level adaptive control architecture through the following steps:

(1) Select a sensor that can measure an endpoint for fluid or cardiovascular drug administration. For fluid administration, the endpoint includes variables such as stroke volume variation, pulse pressure variation, mean arterial pressure, dynamic arterial elastance, urine output rate, pleth variability index, dynamic arterial elastance, central venous pressure, systolic pressure, diastolic pressure, or systolic pressure variation. For cardiovascular drug administration, the endpoint includes variables such as mean arterial pressure, cardiac output, cardiac index, heart rate, systolic pressure, diastolic pressure, systemic vascular resistance, or central venous pressure. The values of the sensor are recorded and smoothed using window averaging or other noise reduction techniques. The measurements could be performed invasively or non-invasively.

(2) The controller architecture assumes that a patient is modeled by an augmented dynamical system model consisting of a 2-, 3-, or n-compartment model (characterizing fluid or cardiovascular drug distribution), as well as a fictitious state. The fictitious state follows the same trend as of the fluid volume in circulation (in the case of fluid management) or cardiovascular drug mass in circulation (in the case of cardiovascular drug management) with some time lag.

(3) A dynamic observer (estimator) is used to estimate deviation of the fictitious state and the fluid volume in circulation (in the case of fluid management) or

2019280053 01 Apr 2020 cardiovascular drug mass in circulation (in the case of cardiovascular drug management) from steady state (equilibrium) values.

(4) A function approximator (e.g., neural network or wavelets etc.) is used to approximate the unknown dynamics of the patient and the parameters of the patient, which are modeled by an augmented dynamical system.

(5) The function approximator weights (or coefficients), sensor measurement values, and infusion rate computed by the adaptive controller at time t~ Δί are used to compute a new infusion rate. Alternatively, the function approximator weights (or coefficients), sensor measurement values, and infusion rate computed by the adaptive controller at times t~ Δί and t~ 2Δί are used to compute a new infusion rate.

(6) The new infusion rate is sent to higher level controller for further processing.

(7) The function approximator weights (or coefficients) are updated using estimates provided by the dynamic observer, sensor measurement values, and the infusion rate computed by the adaptive controller at time t - Δί. Alternatively, the function approximator weights are updated using estimates provided by the dynamic observer, sensor measurement values, and the infusion rate computed by the adaptive controller at times t - Δί and t - 2Δί.

(8) A delay of T seconds/minutes (e.g., 1 second, 10 seconds, 1 minute, 2 minutes, etc.) is introduced.

(9) Step (3) is repeated to close the loop.

[0115] FIG. 6 details a flow chart of the lower-level adaptive control architecture of the disclosure outlining Steps (1) to (9).

Example 2—Function Approximation for Modeling Unknown Dynamics and Patient Parameters for Fluid Distribution [0116] A two-compartment model is used to build the closed-loop fluid resuscitation architecture. The same approach is applied for compartmental models with 3 or more compartments.

[0117] A mass balance equation for a two-compartment dynamical system model is given by:

2019280053 01 Apr 2020

V_c(t) = u(t) +/l(O -Jt (t) -/urine(t) “/bloodl/h t ⁰a_c(t) = - a_bi_ood(t), bfit) Jt(t) /_LW /evaporation (/)» d_t(t) = -Q_L(t) + Qt(t), where u(t) is the rate of fluid infusion, Jy(t) is the rate of volume transfer from interstitial tissue to circulation, J_t (/) is the rate of volume transfer from circulation to interstitial tissue, Jurine(f), Jbiood(f), and Jeva_Poration(/) denote loss of fluid volume due to urine production, hemorrhage, or evaporation (and other types of insensible water loss), respectively. Furthermore, abiood(f) denotes the rate of loss of protein (albumin) mass from hemorrhage, Qt(f) is the rate of albumin mass transfer from circulation to interstitial tissue, and Quif) is the rate of albumin mass transfer from interstitial tissue to circulation.

[0118] The above equation can be rewritten in state space form, namely:

v_c(0 = A(v(t)) +^(v(t))u(t), v_c(0) = v_c>0, t > 0(1) d_c(t) = /2(^)), a_c(0) = a_c0(2) = /(^(0), v_t(0) = v_t>0(3) d_t(t) =/4(^(0), a_t(0) = a_t>0(4) where for t > 0, v(/) = [v_c(t), a_c(/), vt(0, a//)] , and v_c(/), a_c(t), and denote fluid volume in circulation, albumin mass in circulation, fluid volume in interstitial tissue, and albumin mass in interstitial tissue, respectively. In addition, f\(y) and/3(v) denote functions characterizing the rate of change of fluid in circulation and tissue compartments, respectively, and//(v) and///v) denote functions characterizing the rate of change of albumin in circulation and tissue compartments, respectively, g(y (/)) characterizes the effect of fluid infusion on the compartmental model. Note that variables indicating exchange of volume or mass with the outside environment including, J_urine(f), Jbiood(t), Jeva_Poration(/), and <2biood(/) have been incorporated into functions /](v), ffiv), and ffiy). Note that v_{c 0}, a_{c 0}, v_{t 0}, and a_{c O}denote volume and albumin mass in circulation and interstitial tissue at /=0, respectively. Also note that the functions/1(^,/2(/),/3(/),/4(/), and g(v) are generally unknown for each individual patient.

2019280053 01 Apr 2020 [0119] The original two-compartment model is then modified to introduce a certain structure in the dynamics. A fictitious state, x|(t), is added such that the fictitious state follows the same trend as the volume of fluid in circulation, v_c(f), with some time lag. The dynamics of the fictitious state is given by the linear differential equation

Xf(t) = CiVcCt) + c₂Xf(t), %_f(0) = 0, t > 0, (5) where cj, E R are design parameter. Next, error variables are defined that quantify the deviation of each variable from its equilibrium state (steady state value). The error variables are defined by equations (6) -(10):

e_f(t) =	%_f(t) - x_f>e	(6)
	V_c(t) - V_c,_e	(7)
	«cW ^c,e	(8)
	V_t(t) - V_t,e	(9)
e_a,t(f) =	a_t(t) - a_t>e	(10)

where ef(f) denotes the deviation of the fictitious state from fictitious state equilibrium value xf _e, e_v?c(f) denotes the deviation of fluid volume in circulation from its equilibrium value ^vc,e’ ^ea,c(0 denotes the deviation of albumin mass from its equilibrium value a_c?e, e_yj(t) denotes the deviation of fluid volume in tissue from its equilibrium value v^_e, and e_at(t) denotes the deviation of albumin mass from its equilibrium value a^_e.

[0120] It follows from (5) that:

v_c,_e = [0121] Hence, (1) to (5) can be rewritten as:

e_f(t) = c^/t) + c₂e_f(t)(11) e_v,c(t) = /i(e_v(t)) + g(e_v(t))u(t), e_v>c(0) = e_v>c>0, t > 0(12) e_a,c(t) = e_a,c(0) = e_a,₀,(13)

2019280053 01 Apr 2020 e_v,t(t) = A(evCt)), e_v,t(°) = ^ev,t,O> (¹⁴)

MO = A(e_v(t)), e_a,t(0) = e_a,t,o, (15) where for t > 0, e_v(t) = [e_VjC(t),e_ajC(t),e_Vjt(t),e_a>t(t)], and/i(v(0),/2MO), ΛΜΟλ /4(v(r)), g(v(/)) are rewritten as/₁(e_v(t)),/₂(e_v(t)),/₃(e_v(t)),/₄(e_v(t)),^(e_v(t)), respectively, after performing a change of variables using equations (6)-(10). Parameters a] and «2 ^{are 14,0,1}introduced, where a], G R, and equation (12) is rewritten as:

e_v>c(t) = + a₂e_f(t) + h (At)} + i?(e_v(t))u(t), e_v>c(0) = e_v>c>0, t > 0, where:

/ι(Αθ) = - aie_v,c(t) - a₂ef(t), (16) and ⁼ [^ev,c(^)i ^ef (^), ^ea,c(i); ^ev,t(^)> ^ea,t(^)]^T [0122] Next, a series of basis functions are used, such as neural networks basis functions (e.g., radial basis functions or sigmoids), wavelets, or Fourier functions, to approximate Λ(ξ) and #(ξ). Specifically:

AD = + ^εΡ (ΙΌ

AD = + ^£q (18) where:

i·¹®.....^® Μ·.·, _{are two sets of basis functions>} ρ(ξ) = [ρι(<), ••.Ρ»,(ί)]^Τ, ¢(0 = [Ql(£), ·· · , tfnq(£)]^T, ^wp(t) - [^wp,l (*), · · , ^wP,np(t)]^T, and Wq(t) = [w_q,l(t), ..., Wq_inq(t)]^TaiC time varying weights (or coefficients) corresponding to the basis functions, and 8p and 8q are the approximation errors.

[0123] The presented framework is general; however, for illustration purposes, sigmoidal neural β(ξ) — ;—A pi, σ, to e R, network basis functions of the form can be used. oq, «2, ^cb ^an<l ^c2

2019280053 01 Apr 2020 can be selected such that:

Ά <>’) is asymptotically stable (i.e., the real parts of all the eigenvalues of A are negative). One specific choice for these parameters are cq = 100, = —100, oq = 0, and «2 ⁼ “100.

Example 3—Function Approximation for Modeling Unknown Dynamics and Patient Parameters for Cardiovascular Drug Distribution [0124] A two-compartment model is used to build the closed-loop cardiovascular administration architecture disclosed herein. The same approach is applied for compartmental models with 3 or more compartments.

[0125] A drug mass balance equation for a two-compartment dynamical system model is given by:

d_c(t) = u(t) -/_M (t) + /_c(t) -/otherCO» ^{ϋ 0} where w(/) is the rate of drug infusion, Jm(/) is the rate of drug mass elimination from circulation as a result of metabolism, (/) is the rate of drug mass transfer from circulation to tissue, Λ (/) is the rate of drug mass transfer from tissue to circulation, J_other(/) is the rate of drug elimination from circulation other than metabolism (e.g., as a result of hemorrhage).

[0126] The above equation can be rewritten in state space form, namely:

d_c(0 = fi(d(tf) + ^(d(t))u(t), d_c(0) = d_c0, t > 0 (20) d_t(O = /2(^(/)), d_t(0) = d_t,o (21)

T where for t > 0, ¢/(/) = [¢4(1), ¢/((/)] , and d_c(t) and ¢/((/) denote cardiovascular drug mass in circulation and tissue, respectively. In addition,/((¢/) and/2(d) denote functions characterizing the rate of change of drug mass in circulation and tissue compartments, respectively, £(¢/(/)) characterizes the effect of cardiovascular drug infusion on the compartmental model. Note that variables indicating exchange of mass with the outside environment, that is, Jother(Z), and Ai(t), have been incorporated into function f\(d), and/2(v). Note that d_{c 0}, and d_{t 0}, denote drug mass in

2019280053 01 Apr 2020 circulation and tissue at /=0, respectively. Also note that the functions f\(d), /2(d), and g(d) are generally unknown for each individual patient.

[0127] The original two-compartment model is then modified to introduce a certain structure in the dynamics. A fictitious state, xf(/), is added such that the fictitious state follows the same trend as the cardiovascular drug mass in circulation, d_c(f), with some time lag. The dynamics of the fictitious state is given by the linear differential equation x_f(t) = c₂d_c(t) + c₁x_f(t), x_f(0) = 0, t > 0,(22) where c}, C2 G R are design parameters. Next, error variables are defined that quantify the deviation of each variable from its equilibrium state (steady state value). The error variables are defined by equations (23)-(26):

e_f(t) = x_f(t) - x_f>e(23) e_d,_c(t) = d_c(t) - d_c,e(24) e<i,t(t) = ^dt(O - d_t,e(25) where cqf/) denotes the deviation of the fictitious state from fictitious state equilibrium value xf _e, e^ _c(t) denotes the deviation of cardiovascular drug mass in circulation from its equilibrium value d_c,_e, and eft t(0 denotes the deviation of cardiovascular drug mass in tissue from its equilibrium value [0128] It follows from (22) that d_{c e} = — — x_{f e}. Hence, (20) - (22) can be rewritten as: ' e_f(t) = c₂e_d>c(t) + c₁e_i(t)(26) ea,c(t) = /i(e_d(t)) + ^(e_d(t))u(t), e_d>c(0) = e_d>c>0, t > 0(27) ed,t(f) = A (t) ), e_d,t(°) = e_d,t,o,(28) where for t > 0, e_d(t) = [e_dc(t), e_d>t(t)], and f\(d(t)\ f2(d(t)), g(d(t)) are rewritten as /i(e_d(t)),/₂(e_d(t)),.g(e_d(t)), respectively, after performing a change of variables using equations (23) - (26). Parameters a\ and «2 ^are then introduced, where a), cq G R, and equation (12) is rewritten as:

2019280053 01 Apr 2020 ea,c(t) = aie_d,cW + a₂efW + ^h + ^(e_d(t))u(t), e_d>c(0) = e_d>c>0, t > 0, where:

ft(KO) = /i(e_d(t)) - tfie_d,_c(t) - a₂ef(t), (29) and ^(t) = [e_d>c(t),e_f(t),e_d>t(t)].

[0129] Next, a series of basis functions are used, such as neural networks basis functions (e.g., radial basis functions or sigmoids), wavelets, or Fourier functions, to approximate ^(ξ) and ^-specifically:

h(D = ^wp (t)p(D + ^£p (³°)

XD = + ^£q (³¹) where:

Pi(£)>·· >Ρη_ρ(ξ) andgi(£), ·· ><fri_q(£) _{are tw0 sets Q}p basis functions, ρ(ξ) = [ρΐ(ξ),.--,Ρη_ρ(ξ)]^Τ, ¢(£) = [¢1 (ξ), · ·, 9n_q(C)]^T, Wp(t) = [wP;1(t), ..., Wp,np(t)]^T, and wq(t) = [w_q,i(t), ..., w_q,_nq(t)]^TalC time varying weights (or coefficients) corresponding to the basis functions, and 8p and 8q are the approximation errors.

[0130] The presented framework is general; however, for illustration purposes, sigmoidal neural ^s(£) ~ ι_+εσ(ξ--ξ₀) > ^σ> ^{6 R}>

network basis functions of the form can be used. 04, «2, ^cb and Q can be selected such that:

- IS Ά ™ is asymptotically stable (i.e., the real parts of all the eigenvalues of A are negative). One specific choice for these parameters are cj = 0.2, c-χ = 0.013, 04 = 0, and «2 ⁼ “0.2.

Example 3—Computing the Value of Continuous Infusion [0131] At each time instant t, the infusion rate of the controller (for fluid or cardiovascular drug)

2019280053 01 Apr 2020 is given by:

, _λ ( 0, u(t) < 0, ku(t), u(t) > 0, where:

and z(t) = [m(t — Δί), m(t — 2Δί), u(t — Δί), u\t — 2Δί)]^τ, r 1 1 — Q(a(t)) — ’ · ’ ] _{+ e}-ff„_nodo(m(t-Ai)-m_ta[g_CC) ’ _______________1_______________ _________________1_________________ ]_ + g—ΟΊ (m(£—2Δί)—miargct) ’ ’ 1 I g2Δί)~THtargot) ^?

1

-------------------,-----------------------1 1 β-σι«(ί-2Δ0 1 _j_ J ^σι> ·>^σηοάθ, ^are sigmoid parameters (e.g., ranging from -100 to 100), «node represents the number of nodes of the neural network (e.g., «node ⁼ 8), ^and ^m(0 represents the smoothed (denoised) sensor measurement used as an endpoint for fluid or drug administration (e.g., stroke volume variation, urine output rate, mean arterial pressure, central venous pressure, systolic pressure variation, etc. for fluid resuscitation; and mean arterial pressure, heart rate, systolic pressure, diastolic pressure, systemic vascular resistance, central venous pressure, etc. for cardiovascular drug administration) at time t. u(t) represents computed infusion rate at time t. For example, smoothed (denoised) sensor measurement can be obtained by a moving window average where the mean value of the sensor measurement in a time window (e.g., 2 minutes) is computed and assigned to m(f). Sensor values can be preprocessed to drop measurements that appear to be noisy or invalid and only acceptable values are included in the window averaging.

[0132] Alternatively:

z(t) = — At), tt(i — At)]^T,

1 1 + _e-<nu(t-At) ’ · · ’ ! _e-a„_nodcu(t-Ai)

2019280053 01 Apr 2020 [0133] The infusion rate gets updated in regular time intervals and each new value of the infusion rate may be referred to as a new infusion rate or an altered infusion rate. In some embodiment, the lower-level controller comprises an internal dynamical system represented by the states such as weights and dynamic observer states. The update laws for the weights, that is, equations to periodically recompute the value of the weights based on previous values of the weights, hemodynamic sensor measurements, and other variables are given by the set of ordinary differential equations:

w_p(t) = βίΓ^ρ^,-Ρίζ^χ^^ΡΒο), w_p(0) = w_Pj0, t > 0,

W_q(t) = ^₂r(lV_q(t),-Q(z(t))u(t)xJ(t)PB₀), W_q(0) = W_q0 %_c(t) = Ax_c(t) + L(m(t) - m_target - y_c(t)), x_c(0) = 0, y_c(t) = Cx_c(t),

T where the states of the dynamic observer (estimator) x_c(t) = [x_c 1 (0, x_c,2(0] represents the estimated values of ef(f) and e_v?c(f) (in the case of fluid resuscitation) or q(t) and e^ _c(i) (in the case of cardiovascular drug administration), β\, βγ > 0 denote design parameters (e.g., β\ =

0.02 and β% = 0.04), and representative values for other parameters are given by L = [-1, 0] , Bq = [0, 1] , and C = [-1, 0] . In addition, «target is ^l^^c desired value for the end point of fluid resuscitation or cardiovascular drug administration (e.g., for fluid resuscitation if the goal is to maintain a stroke volume variation of 13%, then «target ⁼ 13; similarly, for cardiovascular drug administration if the goal is to maintain a mean arterial pressure of 65mmHg, then «target ⁼ 65), and P satisfies the Lyapunov equation given by:

(A - LC)^TP + P(A - LC) + R = 0, (33) where R > 0.

[0134] For example, if & = /2x2 and the parameter values above are used, then:

_β [0.4116 0.0039] _f ,. ,, , . . _{+ r} , ₅ [0.0005 0.003] _f ^{P =} Laam or ^for cardiovascular drug administration, and P = for

LU.UUDh? Z.D J L U.UUd U.UUoJ fluid resuscitation. In addition, the function f(0,y) is used to ensure that controller parameters

2019280053 01 Apr 2020 remain bounded. This function is defined as:

y, y, ^y ~ iWil^PTiP if/(0) < 0, if f(fi) > 0, and V/^Ty < 0, if /(#) > 0 ^and V/^Ty > 0, where/: R⁷⁷ —> R is defined as f(a\ A (^£g + 1)0^T# - Cax ^£Aax ’ (34) $max ^> θ’ ^εθ ^> θ’ ) represents the gradient operator, and represents the Euclidean norm. For example, 7_max =le6 and εθ =le-5. The internal dynamical system of the lower-level controller involving the function approximator would approximate the unknown dynamics of the patient. This is achieved by changing the values of the weights in set time intervals and adapting the function approximator model to the model of the patient over time. Although the exact structure and parameters of the dynamical system model of the subject (patient) is not known, we can use a function approximator, which is a flexible model, to approximate the dynamical system model of the patient and over time adapt the model such that our approximation is close enough to the actual patient model.

Example 4—Computer Simulations for Fluid Resuscitation [0135] Adaptive control framework is used to simulate fluid resuscitation of a 70kg patient losing blood at a rate of 2ml/kg/min at t = 0. The goal is to maintain the stroke volume variation measurements at 15%. A At of 0.001 hour (3.6 seconds) is used for the simulations and // = 2, /2 = 4, and «node ⁼ 8· The patient model involved a compartmental model to model fluid distribution and the relationship between volume in circulation and SAD/ was based on a nonlinear relationship based on experimental results on dogs.

[0136] FIG. 7 shows stroke volume variation (SVV(%)) versus time. The target stroke volume variation is 15%. The SVV(%) starts at about 9 %, and changes with the introduction of fluid resuscitation. The SW(%) increases to the target value of 15%. At t = 1 hr, the blood loss increases to 3mL/kg/min, and the controller increases the infusion rate to drive stroke volume variation measurements to the target value of 15%.

2019280053 01 Apr 2020 [0137] FIG. 8 shows infusion rates computed by the adaptive control framework. The infusion rate is about 700 mL/h at /=0 (start of the simulation). The infusion rate is rapidly increased to about 1500 mL/h to maintain an SVV(%) of 15%. At t = 1 hr, the blood loss increases to 3mL/kg/min, and the controller increases the infusion rate to about 2250 mL/h to maintain the target SVV(%) of 15%.

[0138] FIG. 9 shows plasma volume in circulation versus time. The plasma volume in circulation rapidly decreases due to loss of blood in spite of fluid resuscitation until reaching an equilibrium value of about 450 mL. At t = 1 hr, the blood loss increases to 3mL/kg/min, and the controller increases the infusion rate to drive stroke volume variation measurements to the target value of 15%. The plasma volume in circulation remains about the same even after an increase in blood loss.

Example 5—Animal Study for Fluid Resuscitation [0139] The adaptive control framework of the disclosure was used to provide automated and semi-automated (clinical decision support) fluid resuscitation to five dogs in different hemorrhaging/hypovolemic scenarios.

[0140] Five mature intact Beagle dogs, determined to be healthy based on a physical examination and hemogram, were included in the experiment. The dogs were individually housed and provided commercial dry dog food and water ad libitum. Each dog and experiment was identified (Table 1). For example, SI-2 represents the second experiment performed on Subject SI. Studies on subjects were performed on different days. Individual trials on the same subject were performed on the same day and a stabilization period was used between trials. All dogs were euthanized with sodium pentobarbital (100 mg/kg, IV) upon completion of the experiments.

Table 1

Study subject information and experiment summary. CH= controlled hypovolemia, UH= uncontrolled hypovolemia, RH= relative hypovolemia, RAH= relative and absolute hypovolemia

Subject	Weight (kg)	Study 1	Study 2
SI	11.6	CH	CH
S2	7.1	RAH	RAH
S3	10.8	UH	-
S4	12.2	RH	UH

2019280053 01 Apr 2020

S5

9.3

RH

UH

[0141] Animal Preparation—Food but not water was withheld for 6 hours before each experiment. An intravenous catheter was transcutaneously positioned in the cephalic vein for administration of hydromorphone, 0.15 mg/kg IV. Anesthesia was produced ten minutes later by administering 3.5 to 6mg/kg IV propofol to facilitate orotracheal intubation, and initially maintained at a vaporizer setting of 1.5-2% isoflurane in oxygen. The dogs were positioned on their right side and mechanically ventilated at 10-12 breaths/min and 10-14 mL/kg tidal volume in order to maintain the end-tidal partial pressure of carbon dioxide (ETco₂) between 38 and 48mmHg. To avoid potential changes in SW, tidal volume settings for each subject was not changed during the study. Esophageal temperature was maintained (37°C) with temperature-controlled warm air blankets.

[0142] Vascular catheters were surgically placed in the left jugular vein and right carotid and femoral arteries after perivascular administration of 0.5-1.0 mL 2% lidocaine. The carotid or femoral artery catheter was connected to a FloTrac sensor with low-compliance fluid filled tubing. The FloTrac sensor was connected to a Vigileo monitor for determination and continuous monitoring of SW. The FloTrac sensor was positioned and zeroed at the level of the sternum. The pressure line of the FloTrac sensor was flushed with 4 mL/hr of lactated ringer’s solution (LRS). Heart rate was determined from a Lead II electrocardiogram (ECG). Criteria for obtaining accurate SW recordings during mechanical ventilation were employed.

[0143] A 5 Fr Swan-Ganz catheter was percutaneously advanced via the right jugular vein (2 dogs) into the pulmonary artery under fluoroscopic guidance for measurement of cardiac output by thermodilution. Alternatively, cardiac output was determined by a previously implanted flow probe (3 dogs) positioned around the ascending aorta, proximal to brachio-cephalic trunk for continuous recording of cardiac output.

[0144] Experimental Procedures—Five dogs were subjected to 9 experiments. Lactated Ringer ’ s solution (LRS) was administered as fluid resuscitation. A variety of experimental hypovolemic conditions were created in order to mimic various clinical conditions (Table 1). Absolute controlled hypovolemia (2 trials) was produced during 1.5 minimum alveolar concentration (MAC: 1.27% used throughout) of isoflurane anesthesia by withdrawing 15 mL/kg/15 minutes from either the right carotid or right femoral artery (S1 -1). There was a 30-minute stabilization period between the end of the closed-loop fluid resuscitation (Sl-1) and the beginning of the second hemorrhage, 40 mL/kg/30 minutes (SI-2). Closed-loop fluid administration was initiated within 10 minutes of

2019280053 01 Apr 2020 completion of each blood withdrawal and continued until SW reached a predetermined target range equal to or less than 13 ± 3%.

[0145] Absolute uncontrolled hypovolemia (S3-1, S4-2, S5-2; 3 trials), designed to simulate blood loss from a severed artery was produced by withdrawal of approximately 50% (40 mL/kg) of the dogs estimated blood volume (80 mL/kg) from the right carotid or right femoral artery over one hour. Five successive 8.0 mL/kg increments of blood were withdrawn continuously in increments that were completed at approximately 7-8, 18-20, 30-32, 43-45, and 60 minutes after initiating hemorrhage. Closed-loop fluid resuscitation was initiated 10 minutes after initiation of absolute uncontrolled hypovolemia (i.e., beginning of Stage 2 of blood withdrawal) and continued until SVV reached a predetermined target range equal to or less than 13 ± 3%.

[0146] Relative hypovolemia (2 trials) was produced by either increasing the inspired concentration of isoflurane to 2.0-2.5 MAC (S5-1, 1 trial) or administering sodium nitroprusside (5-10 mcg/kg/min; S4-1, 1 trial) until mean arterial blood pressure was < 50 mmHg. The target range SVV was set at 13 ± 3% for S4-1 and 5 ± 3% for S5-1. Relative and controlled absolute hypovolemia were produced by increasing the concentration of isoflurane (0.25-0.5%, 1.5-2.0 MAC multiples) in order to decrease MAP by >30% (S2-1, 1 trial) or administering sodium nitroprusside (1-15 mcg/kg/min; S2-2, 1 trial) followed by withdrawal of 15 mL/kg/minutes of blood. The target range SW was set at 13 ± 3% for S2-1 and S2-2. The subject was resuscitated to the target SW value in S2-1 and stabilized before initiating the second study (S2-1). Fluid resuscitation started 15 minutes after achieving relative and controlled absolute hypovolemia.

[0147] The closed-loop fluid resuscitation system was employed in a “partial automation ” mode (clinical decision support) in two experimental trials, one involving absolute uncontrolled hemorrhage (S4-2) and one involving relative hypovolemia from sodium nitroprusside administration (S4-1), where the system displayed the recommended infusion rate every minute and the user manually changed the infusion rate. A Horizon NXT Modular Infusion System pump was used in the partial automated mode. While the system was able to provide infusion rate recommendations more frequently, an update interval of 1 minute was chosen to allow sufficient time for the clinical staff to adjust the pump settings manually. Measured SVV values were filtered in all studies, including automated and partially automated modes, using a 1-minute moving window averaging to remove noise. The adaptive controller was implemented using a neural network with sigmoidal basis functions to use as the function approximator.

[0148] The subject continuously received a fluid infusion and the control system changed the

2019280053 01 Apr 2020 infusion rate every few seconds in response to changes in SW. Vigileo transmitted the most recent value of the SVV measurement every 2 seconds using serial communication. The adaptive control framework was implemented on a laptop. The laptop was connected to the Vigileo using a Serial to USB cable. The closed-loop fluid resuscitation system used a 1-minute moving window averaging to remove noise. The control algorithm was implemented in the Python programming language running on a laptop with the Linux operating system. Measurements from Vigileo were recorded by the laptop using serial communication. The closed-loop fluid resuscitation algorithm computed an infusion rate every 11 seconds using the average SW values in the past 1-minute. The infusion rate was sent by the laptop to an infusion pump (supporting flow rates from 0.06 to 4200 mL/hr) using a USB connection.

[0149] Performance Metrics—Performance metrics that are of clinical relevance were defined in order to assess the performance of the closed-loop fluid resuscitation system. Specifically, Ttarget was defined as the duration from start of fluid administration to restoration of an acceptable SVV target range.

[0150] We defined the acceptable SVV target range to be equal to 13 ± 3% with the exception of two experiments, namely, S5-2 (uncontrolled hypovolemia) and S5-1 (relative hypovolemia), where the acceptable SW target range was 10 ± 3% and 5± 3%, respectively. The Rin range was defined as the percentage of time that SW target value stayed in the acceptable range (i.e., duration of time SVV stayed in the acceptable range divided by total duration of resuscitation). Other performance metrics included minimum and maximum infusion rates denoted by u_min and Umax, respectively, total infused volume to reach the acceptable SW target range denoted by Vtarget, and, in absolute hypovolemia experiments, the ratio of total fluid volume infused to total blood loss denoted by V_ratio. To ensure that the subject can be maintained at the acceptable SW target range, the resuscitation continued after reaching the acceptable SW target range, and hence, Vtarget and total infused volume are not equal. Continued resuscitation after reaching Vtarget lasted approximately 15 minutes on average. The stabilization period between different trials on the same subject started after the completion of the fluid resuscitation of the previous trial.

[0151] Controlled Hypovolemia—Absolute hypovolemia during 1.5 MAC isofhirane anesthesia decreased MAP from 100 to 86 mmHg (dog S-l) and from 109 to 54 mmHg (Sl-2) after withdrawal of 15 mL/kg and then 40 mL/kg of blood respectively (Table 2). Heart rate increased and cardiac output decreased after withdrawal of 15 mL/kg and 40 mL/kg of blood (Table 2). In addition, the SW increased after withdrawal of 15 mL/kg and 40 mL/kg of blood. The SW returned to the target range (13% ± 3) after the administration of 7 mL/kg and 66 mL/kg of LRS,

2019280053 01 Apr 2020 respectively (Table 2). The total infused volumes were 189 mL (S1-1: Vratio = LI) and 925 mL (SI-2: Vratio = 2), respectively.

Table 2

Hemodynamic data and performance metrics for controlled hypovolemia study. T_target denotes the duration from start offluid administration to restoration of an acceptable SW target range, Umin and Umax denote minimum and maximum infusion rates, respectively, and Vt_arget denotes total infused volume to reach the acceptable SW target range

	Baseline	Before Resuscitation	After Resuscitation
ml/kg bled	15	40	15	40	15	40
CO (L/min)	1.5	2.1	1.3	0.8	2.1	1.6
MAP (mmHg)	100	109	86	54	109	63
HR (bpm)	85	108	107	187	108	154
SW(%)	6	15	19	42	13	16

Performance Metrics

Study	Ttarget (min)	Umin (ml/kg/hr)	Umax (ml/kg/hr)	V target (ml/kg)
Sl-1 (15ml/kg)	6.6	26	64	7
Sl-2 (40 mL/kg)	50.3	70	87	66

[0152] Uncontrolled Hypovolemia—Closed-loop fluid administration was initiated during the second stage of blood withdrawal, and continued throughout hemorrhage (S3-1, S4-2, S5-2). Closed-loop fluid administration was stopped approximately 30 minutes after the end of the last (fifth) stage of hemorrhage as this exceeds the average time for fluid equilibration with the interstitial fluid compartment. Mean arterial blood pressure and cardiac output decreased and heart rate increased during the initial 3-4 stages of uncontrolled hypovolemia (Table 3). Heart rate increased throughout hemorrhage and remained elevated throughout hemorrhage and fluid administration (Table 3). The SW increased during the first stage of uncontrolled hypovolemia and returned toward baseline values thereafter (Table 3). Total infused volumes for S3, S4, and S5 were 1,092 mL (Vratio = 2.5), 1,243 mL (Vratio = 2.8), and 548 mL (Vratio = 1.4), respectively. The fluid administration system was used in a partially-automated (human-in-the-loop) mode for S4, where the recommended infusion rate was displayed to the user every minute and the user

2019280053 01 Apr 2020 manually changed the infusion rate to the recommended value.

2019280053 01 Apr 2020

Table 3

Hemodynamic data and performance metrics for uncontrolled hypovolemia study. u_min and Umax denote minimum and maximum infusion rates, respectively, and Rm range the percentage of time that SW target value stayed in the acceptable range

	Baseline	Stage 1	Stage 2	Stage 3	Stage 4	Stage 5	End
CO (L/min)	1.5 (1.1-1.7)	0.8 (0.5-1.5)	0.8 (0.3-1.4)	1.0 (0.4-1.4)	0.9 (0.4-1.7)	1.0 (0.7-1.7)	1.3 (1.1-2.0)
MAP (mmHg)	81 (62-85)	59 (55-91)	66 (35-87)	65 (45-67)	68 (50-72)	68 (53-78)	59 (44-83)
HR (bpm)	118 (113-121)	134 (120-171)	132 (108-170)	132 (117-174)	135 (117-170)	139 (116-168)	137 (107-167)
SW(%)	8 (6-12)	21 (17-25)	17 (12-19)	15 (11-16)	14 (11-16)	15 (10-15)	10 (8-18)

Performance Metrics

Dog	Umin (ml/kg/hr)	Umax (ml/kg/hr)	Rin range (%)
S3-1	44	115	54
S4-2	40	82	71
S5-2	39	64	100

[0153] Relative Hypovolemia from Vasodilatation—Dogs were made hypotensive by increasing the inspired concentration of isoflurane (S5-1 dog) or by administering sodium nitroprusside (S4-1) until mean arterial blood pressure was less than or equal to 50 mmHg. The target range SW was set at 13 ± 3% for S4-1 and 5 ± 3% for S5-1. The administration of isoflurane or sodium nitroprusside decreased mean arterial blood pressure and increased heart rate and either did not change or increased cardiac output, respectively (Table 4). SW increased in both dogs (Table 4). Closed-loop fluid resuscitation increased cardiac output and decreased SW in both dogs. Arterial blood pressure increased during fluid administration in the dog administered sodium nitroprusside (S4-1) but not in the dog administered isoflurane (S5-1) (Table 4). Total infused volumes were 187 mL (nitroprusside: 15 mL/kg) and 265 mL (isoflurane: 28 mL/kg).

2019280053 01 Apr 2020

Table 4

Hemodynamic data and performance metrics for the relative hypovolemia study. T_target denotes the duration from start offluid administration to restoration of an acceptable SW target range, Umin and Umax denote minimum and maximum infusion rates, respectively, Vt_arget denotes total infused volume to reach the acceptable SW target range, and Rm _range the percentage of time that SW target value stayed in the acceptable range. I = isoflurane; SN = sodium nitroprusside

	Baseline	Before Resuscitation	After Resuscitation
CO (L/min)	0.81; 1.4 SN	0.81; 1.6 SN	1.4 I; 2.0 SN
MAP (mmHg)	62 I; 69 SN	45 I; 47 SN	35 I; 94 SN
HR (bpm)	84 I; 89 SN	87 I; 115 SN	89 I; 126 SN
SW(%)	7 I; 9 SN	10 I; 17 SN)	6 I; 8 SN

Performance Metrics

Study	Ttarget (min)	Umin (ml/kg/hr)	Umax (ml/kg/hr)	V target (ml/kg)	Rin range (%)
S4-1	3	11	67	3	100
S5-1	4	63	73	5	90

[0154] Relative Hypovolemia and Controlled Hypovolemia—Increasing the isoflurane concentration from 1.5 to 2.5 MAC (S2-1) or administration of sodium nitroprusside (S2-2) was followed by blood withdrawal (15 mL/kg/15 minutes). The sodium nitroprusside infusion rate started at 1 mcg/kg/min and was increased to 15 mcg/kg/min. The target range SW was set at 13 ±3%. Absolute hypovolemia (15 mL/kg/15 min) during 2.5 MAC isoflurane anesthesia decreased MAP from 60 mmHg before blood withdrawal to 39 mmHg after blood withdrawal (S2-1). In addition, MAP increased from 39 mmHg to 46 mmHg after a 15-minute stabilization period. Heart rate changed minimally (110 vs. 112 bpm) and cardiac output decreased 20% (1.0 vs. 0.8 L/min) after the production of relative hypovolemia combined with controlled hypovolemia (15 mL/kg/15 min). The SW increased from 13% at 1.5 MAC to 21% at 2.5 MAC and then to 41% after withdrawal of 15 mL/kg of blood. The SW decreased to 26% after a 15-minute stabilization period. The SW returned to the target range (13% ± 3) at 43 minutes (Ttarget) after the administration of 78 mL/kg of LRS. Total infused volume was 573 mL (V_ratio = 5.4). The maximum fluid administration rate (u_max) was 113 mL/kg/hr. Heart rate decreased (112 to 99) and cardiac output increased to above the baseline value (1.0 vs. 1.2) after LRS administration but MAP remained relatively unchanged (43 mmHg) until the isoflurane concentration was decreased.

2019280053 01 Apr 2020 [0155] Absolute hypovolemia (15 mL/kg/15 min) during 1.5 MAC and sodium nitroprusside administration decreased MAP from 113 to 92 mmHg prior to blood withdrawal (S2-2). MAP was 101 mmHg after blood withdrawal. Heart rate changed minimally and cardiac output decreased 25% (1.3 vs. 1.0 L/min) after the production of relative hypovolemia combined with controlled hypovolemia (15 mL/kg/15 min). The SW increased from 10% to 15% after sodium nitroprusside administration and then to 22% after withdrawal of 15 mL/kg of blood. The SW returned to the target range (13% ± 3) at 26 minutes (Ttarget) after the administration of 46 mL/kg of LRS. The maximum fluid administration rate (u_max) was 108 mL/kg/hr. Cardiac output increased to above the baseline value (1.3 vs. 2.1) after LRS administration and MAP increased to near the baseline value (Table 5). Total infused volume was 400 mL (V_ratio = 3.7).

Table 5

Hemodynamic data and performance metrics for the relative hypovolemia and controlled hypovolemia study. T_target denotes the duration from start of fluid administration to restoration of an acceptable SW target range, u_min and Umax denote minimum and maximum infusion rates, respectively, Vt_arget denotes total infused volume to reach the acceptable SW target range, and Rin range the percentage of time that SW target value stayed in the acceptable range. I = isoflurane; SN = sodium nitroprusside

	Baseline	Before Resuscitation	After Resuscitation
CO (L/min)	1.01; 1.3 SN	0.81; 1.0 SN	1.2 I; 2.1 SN
MAP (mmHg)	93 I; 113 SN	46 I; 101 SN	40 I; 115 SN
HR (bpm)	1101; 134 SN	1121; 114 SN	99 I; 108 SN
SW(%)	13 1; 10 SN	26 I; 22 SN	15 I; 21 SN

Performance Metrics

Study	Ttarget (min)	Umin (ml/kg/hr)	Umax (ml/kg/hr)	V target (ml/kg)	Rin range (%)
S2-1	43	88	113	78	100
S2-2	26	66	108	46	53

[0156] The closed-loop fluid resuscitation system used the compartmental modeling framework to compute fluid infusion rates. The two-compartment model of the microvascular exchange system involved parameters that are generally unknown and different from patient to patient. In addition, the two-compartment model was only an approximation of the fluid distribution in the

2019280053 01 Apr 2020 body, resulting in modeling error.

[0157] The adaptive control fluid resuscitation algorithm uses the compartmental characteristics of the fluid distribution in the body. To address modeling error and unknown parameters of the compartmental model governing fluid distribution, the adaptive algorithm used a “function approximator”. The function approximator was characterized by a set of parameters, which were continuously estimated by the closed-loop system in real-time. The closed-loop fluid resuscitation system re-computes the fluid infusion rate periodically. The function approximator used the values of infusion rates and SW measurements to estimate the dynamics of fluid distribution. The controller performance was evaluated with computer simulations on a two-compartment model prior to conducting the animal study. The results presented are the first attempt to use the disclosure in live subjects.

[0158] Data from this study confirmed that an adaptive closed-loop fluid administration system based on a compartmental model of fluid distribution provided targeted goal-directed fluid therapy in dogs subjected to experimental conditions that mimicked clinical scenarios of absolute (controlled, uncontrolled), relative hypovolemia or a combination of relative hypovolemia and absolute controlled hypovolemia. Stroke volume variation was restored and maintained to within a preselected normal target range in less than one hour after initiating fluid administration. Larger volumes of blood loss (40 vs. 15 mL/kg) increased the V_ratio required to restore SW to the target range but remained below amounts based upon lactated ringer’s solution (LRS) distribution.

[0159] The adaptive control algorithm was based on physiology, and the parameters of the model were estimated in real-time. The framework provided a mechanism to account for inter-patient and intra-patient variability in the fluid resuscitation process.

[0160] FIGS. 10 and 11 show the results for 2 canine subjects SI and S2 (total of 4 studies) experiencing controlled hemorrhage. FIG. 10 shows changes in filtered SW(%) versus time, and FIG. 11 shows changes in infusion rate (mL/hr) versus time. The target SW was 13%. In study S1-1, the infusion rate dropped from 750 mL/hr to about 400 mL/hr to maintain an SW of 13%. Once reaching the target SW(%), the infusion rate fluctuated between 300 - 400 mL/hr to maintain an SW of 13% In study SI-2, the infusion rate started from about 800 mL/hr increased to about 1000 mL/hr and decreased to about 850 mL/hr. In study S2-1, the infusion rate started from about 800 mL/hr and in the end reached about 650 mL/hr. In study S2-2, the infusion rate started from about 750 mL/hr and in about 30 minutes reached about 500 mL/hr but increased to 750 mL/hr.

2019280053 01 Apr 2020 [0161] FIGS. 12 and 13 shows the results for 3 canine subjects S3, S4, and S5 experiencing uncontrolled hemorrhage. FIG. 12 shows changes in filtered SW(%) versus time, and FIG. 13 shows changes in infusion rate versus time. The target SW was 13% in studies S3-1 and S4-2 and was 10% in study S5-2. In study S4-2, the partial automation (clinical decision support) was used where every 1 minute the clinician used fluid rates recommended by the system to manually change the infusion rate on the pump. In the other two studies S3-1 and S5-2 the fully automated closed-loop system was used. In study S3-1, the infusion rate started from about 600 mL/hr and increased to aboutl200 mL/hr to drive SW to 13% and then dropped to 750 mL/hr. In study S4-2, the infusion rate started from about 750 mL/hr increased to 999 mL/hr and then decreased to about 600 mL/hr to drive SVV to 13%. In study S5-2, the infusion rate started from about 500 mL/hr and fluctuated between 400-600 mL/hr to drive SW to the target value of 10%.

[0162] FIGS. 14 and 15 shows the fluid resuscitation of 2 canine subjects S4 and S5 that were hypotensive as a result of administration of sodium nitroprusside (in S4-1) and increase in the inhalant anesthetic isofhirane (in S5-1). In study S5-1, the fully automated closed-loop system was used while in study S4-1 the partially automated (clinical decision support) system was used where the recommended infusion rate was displayed every 1 minute to the user and the user changed the infusion rate on the pump manually. FIG. 14 shows changes in filtered SW(%) versus time, and FIG. 15 shows changes in infusion rate versus time. The target SW was 13% in S4-1 and 5% in S5-1. In S4-1, the infusion rate was started at about 650 mL/hr and decreased to between 150-350 mL/hr once the target 13% was reached. In S5-1, the infusion rate started at about 700 mL/hr and was kept close to this value until 25 minutes into the study, when the study was terminated as SW approached the desired SW of 5%.

Example 6—Higher-Level Controller [0163] The higher-level controller was designed as a rule-based expert system and served to:

(i) monitor the lower-level controller (fluid resuscitation module and/or cardiovascular drug administration module) function for possible anomalies;

(ii) monitor the patient’s general status and response to fluid therapy and/or cardiovascular drug administration;

(iii) in the case of closed-loop mode decide to engage fluid resuscitation module, cardiovascular drug administration module, or both, and the timing of engagement of the modules;

2019280053 01 Apr 2020 (iv) handle scenarios related to sensor failure or infusion rate exceeding maximum safe limit;

(v) modify the lower-level controller (fluid resuscitation module or cardiovascular drug administration module) states in the case of clinical decision support if the user disagrees with the computed infusion rate or when the user specifies an initial infusion rate;

(vi) provide clinical decision support to address potential problems; and (vii) in clinical decision support mode, notifying the user with the new infusion rate when the infusion rate needs to be updated.

[0164] Maintaining Infusion Rate in Safe Range—If the computed infusion rate by the lower-level controller (fluid resuscitation module and/or cardiovascular drug administration module) was larger than the maximum safe rate specified by the user, the higher-level controller reset the lower-level controller associated with unallowable infusion rate, that is, it reinitialized the internal states of the lower-level controller including function approximator weights (or coefficients) to the default values (i.e., changed w_p, w_q, andx_c to their values at /=0) or it reset the weights of the function approximator such that the infusion rate computed by the lower-level controller matched a preset infusion rate previously provided by the user using the user interface. Given the user specified infusion rate, the weights of the function approximator was chosen such that:

Wq,new(0) = W_q(0) wjnew(0)) = (preset infusion rate)(l + w_q (Ο)φ(ζ(Ο))) [0165] Warning the User—If the total volume delivered to the patient exceeded a set threshold (e.g., 2000 mL or 10,000 mcg), the higher-level controller warned the user of the risk of complications.

[0166] Monitoring the Lower-Level Controller—If performance degradation in the closed-loop or clinical decision support system was detected, the higher-level controller notified the user through an audio-visual alarm, and the higher-level controller stopped the lower-level controller. Performance degradation was defined as:

(i) a rapid change in weights (or coefficients) of the function approximator, that is,

2019280053 01 Apr 2020 ^^>threshold (e.g., 0.1, or 1, 10 etc.); or zz) the number of resets of the lower-level controller by the higher-level controller exceeding a threshold value (e.g., 1, 2, 3, 4, etc.); or (ii) the absolute difference between the measured value (e.g., SW or mean arterial pressure) and the target value did not decrease in a period of time set by user while the infusion was continuously increased. In case of sensor failure or unavailability of measurements, the higher-level controller disabled the lower-level controller and set the infusion rate to a backup the infusion rate set by the user.

[0167] Deciding When to Engage Fluid Resuscitation and Cardiovascular Drug Administration Modules—In certain scenarios such as sepsis, it is preferable to provide fluid resuscitation first, and if unsuccessful in achieving an acceptable condition, administer a cardiovascular drug such as a vasopressor. The higher-level controller first engaged the fluid resuscitation module and monitored the patient. After a period of time when a clinically relevant hemodynamic variable (e.g., mean arterial pressure) was not improved in spite of fluid resuscitation (e.g., the mean arterial pressure was not in the acceptable range of 65-80 mmHg after 30 minutes), the higher-level controller engaged the cardiovascular drug administration module to administer a vasopressor. In some embodiments, the clinician instructed the higher-level controller to engage the cardiovascular drug administration module while the fluid resuscitation module was already running. In some embodiments, the clinician instructed the higher-level controller to engage the fluid resuscitation module while the cardiovascular administration module was already running. In this scenario, stroke volume variation was not improved by only administering a cardiovascular drug (e.g., a vasopressor) and the higher-level controller engaged the fluid resuscitation module.

[0168] FIG. 16 illustrates the components of the higher-level controller for the closed-loop fluid resuscitation and/or cardiovascular drug administration system. The higher-level controller can monitor the fluid resuscitation module and/or the cardiovascular drug administration depending on whether the system is used for fluid resuscitation only, for cardiovascular drug administration only, or combined fluid resuscitation and cardiovascular drug administration.

[0169] In the clinical decision support system case, the lower-level controller (fluid resuscitation module or cardiovascular drug administration module or both) sent newly computed infusion rate to the higher-level controller. Specifically, we considered 3 scenarios:

2019280053 01 Apr 2020 (i) fluid resuscitation module only;

(ii) cardiovascular drug administration module only; and (iii) fluid resuscitation module and cardiovascular drug administration module working concurrently. In each scenario, the lower-level controller sent newly computed infusion rate to the higher-level controller. The higher-level controller compared the new rate with the last user approved rate (i.e., for a fluid resuscitation, the newly computed fluid infusion rate was compared with the last user-approved fluid infusion rate, and for cardiovascular drug administration, drug infusion rate was compared with the last user-approved drug infusion rate).

[0170] If the difference between the new infusion rate and the last user-approved rate was less than the threshold value (e.g., 25% of the last approved rate), then the infusion rate was not changed, the user was not notified, and the lower-level controller continued its operation. If the difference between the new infusion rate and the last user-approved rate was higher than some threshold (e.g., 25% of the last approved rate), then the recommended infusion rate was displayed to the user through the graphical user interface. If the user accepted the infusion rate, the higher-level controller allowed the lower-level controller to continue its operation (none of the weights and internal parameters of the lower-level controller were changed) and the infusion rate was updated to the new rate and the accepted infusion rate was recorded as the last-approved infusion rate in the database. However, if the user changed the recommended infusion rate to a different value, the higher-level controller reset the controller and set the weights of the function approximator such that the infusion rate computed by the lower-level controller matched the infusion rate entered by the user. The infusion rate was also updated to the rate provided by the user, where the infusion rate provided by the user was recorded as the last user-approved infusion rate in the database. Given the user specified new infusion rate, the weights of the function approximator was chosen such that

Wq,new(0) = W_q(0) wJnewCOji’CXO)) = (new infusion rate)(l + w_q (0)(?(z(0))) [0171] FIG. 17 illustrates the components of the higher-level controller for the clinical decision support case.

[0172] Similarly, when the user specifies an initial infusion rate when the clinical decision

2019280053 01 Apr 2020 support system started, the higher-level controller recorded that value as a last user-approved infusion rate in the infusion rate database and used this value to initialize the parameters of the lower-level controller by setting the weights of the function approximator such that the infusion rate computed by the lower-level controller matched the infusion rate entered by the user. Given the user specified new infusion rate, the weights of the function approximator was chosen such that

Wq,new(0) = W_q(0) wjnew= (new infusion rate)(l + w_q (Ο)φ(ζ(Ο)))

Example 7—Computer Simulations for Vasopressor Administration [0173] Adaptive control frameworks were used to simulate cardiovascular drug (vasopressor) administration of a 70 kg patient experiencing sepsis and the associated hypotension. The goal was to maintain the mean arterial pressure at 65. A Δί of 0.1 minute (6 seconds) was used for the simulations and β\ = 6e-5, /fy ⁼ 13e-6, and «node ⁼ 8· Only the vasopressor was administered to maintain a mean arterial pressure of 75 mmHg. The patient model involved a cardiovascular model to model hemodynamics and a compartmental model to model fluid distribution.

[0174] FIG. 18 shows mean arterial pressure (MAP) versus time. The target MAP was 75 mmHg, and the 65-75 mmHg region is highlighted on the graph. MAP started at below 60 mmHg, and changes with the introduction of vasopressor epinephrine. The MAP increased to the target value of 75mmHg.

[0175] FIG. 19 shows infusion rates computed by the adaptive control framework. The initial infusion rate was chosen by the user to be 0.12 mcg/kg/min at /=8 (start of the vasopressor administration). The infusion rate gradually increased and then started to decrease as MAP started to approach the target MAP of 75 mmHg reaching a low infusion rate of 0.85 mcg/kg/min. The infusion rate then started to gradually increase to maintain MAP of 75 mmHg.

Example 8—Computer Simulations for Fluid Administration Using Mean Arterial Pressure [0176] Adaptive control frameworks were used to simulate fluid resuscitation of a 70 kg patient with hypotension as a result of sepsis. The goal was to maintain the mean arterial pressure at 75 mmHg. A Δ/ of 0.1 minute (6 seconds) was used for the simulations and β\ = 0.02, /fy ⁼ θ·θ4, and «node ⁼ 8· The patient model involved cardiovascular modeling to model hemodynamics and

2019280053 01 Apr 2020 a compartmental model to model fluid distribution.

[0177] FIG. 20 shows mean arterial pressure (MAP) versus time. The target MAP is 75 mmHg and the 65-75mmHg region is highlighted on the graph. MAP starts at approximately 45 mmHg, and changes with the introduction of crystalloid fluid. The MAP increases to the target value of 75 mmHg.

[0178] FIG. 21 shows infusion rates computed by the adaptive control framework. The initial infusion rate was chosen was approximately 140 mL/min. The infusion rate gradually increased to 160 mL/min and then decreased to around 80 mL/min.

Example 9—Computer Simulations for Fluid Administration and Cardiovascular Drug Administration [0179] Adaptive control frameworks were used to simulate fluid resuscitation and cardiovascular drug (vasopressor epinephrine) administration for a 70 kg patient with hypotension as a result of sepsis. The goal was to maintain the mean arterial pressure at 75 mmHg and stroke volume variation (SW) of 12%. A Δ/ of 0.1 min (6 seconds) was used for the simulations and β\ = 0.02, β2 = 0.04, and n_nOdc = 8 for the fluid resuscitation module and β\ = 6e-5, /?2 ⁼ 13e-6, and «node = 8 for the cardiovascular drug administration module. The simulation involved crystalloid fluid and epinephrine (vasopressor). The fluid resuscitation module used SW data to compute fluid infusion rates, and the cardiovascular drug administration module used mean arterial pressure (MAP) to compute the vasopressor infusion rates. The higher-level controller first engaged the fluid resuscitation module to provide fluid infusion. After 30 minutes, the higher-level controller engaged the cardiovascular administration module as mean arterial pressure was not improved after fluid infusion. The patient model involved cardiovascular modeling to model hemodynamics and compartmental model to model fluid distribution. The start of simulation is when the patient condition rapidly deteriorates.

[0180] FIG. 22 shows stroke volume variation (SW(%)) versus time. The target stroke volume variation was 12%. The SW(%) started at about 18 %, and was reduced to 10% after fluid resuscitation. SW momentarily increased due to the administration of epinephrine (a transient vasodilatory effect). In the end of the simulation, SW was approximately 11% and remained close to the target SW value of 12%.

[0181] FIG. 23 shows fluid infusion rates computed by the fluid resuscitation adaptive control framework. The infusion rate started at approximately 130 mL/min and was reduced to 40 mL/min

2019280053 01 Apr 2020 after SVV was at an acceptable range. The sudden increase in SW at t=30 min resulted in an increase in infusion rate which then reduced back to 40 mL/min after SW was reduced.

[0182] FIG. 24 shows mean arterial pressure (MAP) versus time. The target MAP was 75 mmHg and the 65-75 mmHg region is highlighted on the graph. MAP dropped to 50 mmHg as the patient condition deteriorated. After 30 minutes, when the cardiovascular drug administration module was engaged by the higher-level controller, MAP started to increase and approached the set value of 75 mmHg. At the end of the simulation, MAP was approximately 70 mmHg and remained close to the target MAP value of 75 mmHg.

[0183] FIG. 25 shows vasopressor infusion rates computed by the cardiovascular administration adaptive control framework. The initial infusion rate was chosen to be approximately 0.12 mcg/kg/min. After a slight increase, the rate decreased to approximately 0.9 mcg/kg/min and then stabilized around 0.12 mcg/kg/min as MAP is maintained close to 70 mmHg.

Example 10—Fluid Resuscitation and Cardiovascular Drug Administration System Implemented on Hardware [0184] A hardware platform was developed to implement the fluid resuscitation system and the cardiovascular drug administration system. The system is composed of a processing module and an associated touchscreen. The processing module is able to receive data from a hemodynamic monitor (invasive or non-invasive measurement) or a blood pressure module (either invasive measurement through a pressure transducer connected to an arterial line or non-invasive measurement) through a wired connection. In addition, the processing module is able to send real-time infusion rate data through a wired connection to an infusion pump. The touchscreen is used for displaying graphs such as received measurements over time and infusion rate over time. The main dashboard also displays current measurement, current infusion rate, and total volume of fluid/drug infused. The touchscreen allows the user to specify different parameters required by the system. For example, the user sets the target measurement value, initial infusion rate, backup infusion rate (in case of missing sensor data, the closed-loop system is disengaged and transitions to backup infusion rate until the user takes over) maximum infusion rate and other variables. In the case of clinical decision support, the touchscreen is used to communicate the new infusion rate value to request user’s approval or give the user the ability to modify the rate. Once the user approves the infusion rate, the processing module updates the infusion rate of the infusion pump to the infusion rate approved by the user.

[0185] FIG. 26 is a system diagram of an embodiment of the hardware platform.

2019280053 01 Apr 2020 [0186] FIG. 27 is a block diagram illustrating an embodiment of the apparatus for displaying infusion therapy information for optimizing infusion therapy.

[0187] FIG. 28 depicts an embodiment of the apparatus for displaying infusion therapy information for optimizing infusion therapy. The apparatus comprises a memory, a display, and one or more processor in operable communication with the memory and the display. The processor is capable of receiving sensor-obtained hemodynamic measurements of a subject from a hemodynamic monitoring device. The processor is further capable of issuing a prompt to the display upon determining an appropriate notification message and an appropriate change in infusion rate recommendation message, which are implicated upon detection of a change in infusion therapy. The detection of a change in infusion therapy is based at least on hemodynamic measurements of a subject, an active infusion rate stored in the memory, and a background infusion rate computed by a lower-level controller.

[0188] In some embodiments, an active infusion rate is a last user-approved infusion rate. The last-approved infusion rate can be a last infusion rate that was recommended through a prompt and accepted by a user. In other embodiments, an active infusion rate can be an infusion rate obtained from an infusion pump that is actively administering fluid into a subject. A background infusion rate is an infusion rate that upon updates to its value does not actively change the infusion rate of fluid or drug administered to a subject, but rather it updates in the background. In some embodiments, a background infusion rate is computed by a lower-level controller. In other embodiments, the background infusion rate can take on the value of an active infusion rate. This may be through mechanisms such as a rejection of a recommended infusion rate prompted to a user which permits the user to manually override the recommendation with a desired infusion rate. In this example, the desired infusion rate would become an active infusion rate and the background infusion rate, at least momentarily, would take on the value of this set desired infusion rate.

[0189] An appropriate notification message can alert a user that a new recommended infusion rate is available. In other embodiments, the notification message can alert a user that subject being treated with the infusion therapy is not improving based upon hemodynamic measurements. In yet another embodiment, the notification message can alert a user of a failure in receiving hemodynamic measurements from a monitoring device. In other embodiments, this notification message can alert a user of a failure in administering fluid, for instance, due to a status message received from an infusion pump. In other embodiments, the notification message can alert the user that received hemodynamic measurements exceed, equal, or drop below some threshold.

2019280053 01 Apr 2020 [0190] An appropriate change in infusion rate recommendation message can be a recommendation to increase or decrease the current active infusion rate. In other embodiments, an appropriate change in infusion rate recommendation message can be a specific value for infusion rate (e.g., 700 mL/hr, 300 mL/hr, 0 mL/hr). In one embodiment, this specific value can be 0 mL/hr, that is, to stop infusion of fluids or drugs to a patient, if an associated notification message alerts the user that the hemodynamic condition of a subject is not improving; in another embodiment, the recommendation to stop infusion of fluids or drugs may be due to an associated notification message that alerts the user that some hemodynamic measurement exceeds or drops below some threshold.

[0191] In some embodiments, the notification message is generated only if the absolute difference between the active infusion rate and the background infusion rate is larger than a pre-determined threshold. In case the absolute difference between the active infusion rate and the background infusion rate is less than the pre-determined threshold then the value of the background infusion rate is ignored and the lower-level controller continues with computing new background infusion rates until the criterion for notifying the user is met. In some embodiments, the pre-determined threshold is a percentage of the active infusion rate (e.g., 1%, 2%, 5%, 10%). In other embodiments, the pre-determined threshold is an absolute value (e.g., 0.001 mL/hr, 0. Iml/hr, Iml/hr, lOml/hr, 100 mL/hr, 1000 mL/hr).

[0192] Referring again to FIG. 27, the display can be a touchscreen-based display in which a user can interact with any prompt through touch gestures. In other embodiments, the display can be a non-touchscreen display and the user can interact with any prompt with suitable input devices, such as physical buttons, a mouse/trackpad, a keypad/keyboard, etc.

[0193] A detection of a change in infusion therapy can be performed by the higher-level logic-based controller based upon the performance criteria discussed earlier. In some embodiments, the detection of a change in infusion therapy occurs when an active infusion rate and a background infusion rate differ in value by some percentage (e.g., 5%, 10%, 12%, etc.) of the active infusion rate, some percentage of the background infusion rate, or some percentage of the average of the active and background infusion rate; in other embodiments, the difference can be some absolute value (e.g., -50mL/hr, 50 mL/hr, lOOmL/hr, etc.) which would trigger the detection.

[0194] Referring to FIG. 28, in some embodiments of the apparatus, upon issuance of a prompt 2810 that contains a slide to accept element or tap to reject element to the display containing a notification message 2820 and a change in infusion rate recommendation message 2830, there is

2019280053 01 Apr 2020 a limited time for the user to respond with an acceptance or a rejection. If the user wishes to manually override the recommended infusion rate with a desired infusion rate, an infusion rate can be set with infusion meter slider 2840. However, if the user rejects using the reject/discard/trash button of 2810 or if the timer 2800 expires, the active infusion rate remains unchanged and the background infusion rate is set to the same value as the unchanged active infusion rate. If the recommended infusion rate is not accepted (e.g., due to manual override, rejection, or timer expiration), the higher-level logic controller resets at least one element of the set of weights of the lower-level controller to a new value based at least in part on the background infusion rate.

Fluid and/or Cardiovascular Drug Administration with Pre-Approved Ranges of Infusion Rates by the User [0195] The presented disclosure can be implemented in a way to minimize unnecessary interactions with the user regarding changes in infusion rates, which the user may deem acceptable or negligible. The advantage of this implementation is that the clinical decision support system has a specific level of autonomy to automatically change infusion rates as long as the infusion rates are within the pre-approved range specified by the user. In some embodiments, the clinical decision support system can be coupled to an internal or external infusion pump.

[0196] In some embodiments, the user can enter pre-approved ranges of infusion rates into the system with the intent that the system is authorized to change infusion rates automatically without requesting the user to approve the changes as long as the infusion rates fall within the pre-approved ranges. In some embodiments, the system automatically changes the infusion rate by instructing the pump and updating infusion rate on the graphical user interface as long as the new infusion rates are within the pre-approved infusion rates. In some embodiments, a pre-approved range can be entered using a lower limit and an upper limit (e.g., 50-100, 100-500, 500-1000 mL/hr).

[0197] In some embodiments, the approved range can be entered based on the last user-approved infusion rate. In this case, the user authorizes the clinical decision support system to internally approve infusion rates that are within a set distance (specified by the user) from the last user-approved infusion rate. For example, the user may authorize a pre-approved range of 500 mL/hr, in which case if the last user-approved infusion rate is 750 mL/hr the system is authorized to change the infusion rate as long as the new infusion rates are within 250 mL/hr to 1,250 mL/hr. In some embodiments, the user can specify a percentage change as a pre-approved range. For example, the user can approve a 10% change as a pre-approved range. In this case, if, for example, the last-approved infusion rate is 1,000 mL/hr, then the clinical decision support

2019280053 01 Apr 2020 system can internally approve new infusion rates in the range 900 mL/hr to 1,100 mL/hr without requesting user’s approval.

[0198] In some embodiments, the higher-level logic-based controller will compare the new infusion rate received from the lower-level controller with the pre-approved ranges. If the new infusion rate is within the pre-approved range (that is the new infusion rate is greater than or equal to the lower limit and less than or equal to the upper limit), then the higher-level logic can instruct the infusion pump to update the infusion rate to the new infusion rate value. However, if the new infusion rate falls outside the pre-approved range, the user is notified using the graphical user interface, where the user would approve, reject, or modify the new infusion rate.

[0199] Referring to FIG. 29, a flow chart of the process is presented. The process starts by the user entering the pre-approved infusion rates into the system using the user interface. The lower-level controller computes new infusion rates based at least in part on hemodynamic measurements. The higher-level logic-based controller receives the newly computed infusion rates from the lower-level controller and determines whether the new infusion rate is within the pre-approved range. If so, it instructs the pump to update the infusion rate. Otherwise, it asks the user to approve, reject, or modify the new infusion rate.

Alarm Function to Prevent Under- or Over-Administration of Fluid and/or Drug [0200] The hierarchical control system described in this disclosure can be used for monitoring therapy to a subject in order to flag cases where the subject would receive lower or higher than expected infusion of a fluid or cardiovascular drug. Both fluid and cardiovascular drug over-administration or under-administration is detrimental to the patient, and hence, a system that continuously monitors therapy including the infusion rate and patient’s hemodynamic measurements and alerts the user if the patient is receiving more than expected or less than expected fluid and/or cardiovascular drug is desirable. In some embodiments, the system continuously monitors the patient’s hemodynamic measurements, the type of therapy or therapies, and the infusion rate or infusion rates of the therapy or therapies, and notifies the user through a user interface if the patient is getting more than expected or less than expected fluid and/or cardiovascular drug. In some embodiment, this is implemented in a hemodynamic patient monitor at the bedside receiving information from the pump. In some embodiments, this is implemented as a software application as part of a remote patient monitoring platform, where information from hemodynamic monitors and infusion pumps are continuously analyzed and cases of over-administration or under-administration of fluid and/or cardiovascular drug is detected. In

2019280053 01 Apr 2020 some embodiments, data from multiple hemodynamic monitors and infusion pump for different patients are all sent to the same remote location for analysis. In some embodiments, the system is implemented inside an infusion pump. In this case, the infusion pump receives hemodynamic measurement data from a hemodynamic monitor or other source.

[0201] Referring to FIG. 30, in some embodiments, the lower-level controller runs in the background and computes new infusion rates periodically. The newly computed infusion rates are kept in the background, and analyzed by the higher-level logic-based controller. The internal states of the lower-level controller are initialized such that the starting infusion rate generated by the lower-level controller matches the actual infusion rate specified by the user. In some embodiments, the lower-level controller uses a function approximator with a set of weights, where the weights are determined in a way such that the initial infusion rate generated by the lower-level controller matches the actual infusion rate specified by the user. In some embodiments, the function approximator is a neural network. As an example, given the user specified infusion rate, the weights of the function approximator was chosen such that:

Wq,new(0) = W_q(0) wJ,new(0)T(z(0)) = (pump infusion rate)(l + w^(0)Q(z(0))) [0202] As discussed above, after initialization, the lower-level controller periodically computes a new infusion rate in the background and sends the background infusion rate to the higher-level logic-based controller. The higher-level controller records the current value of the background infusion rate and the current value of the actual infusion rate in a monitoring data base. A monitoring module periodically analyzes the background infusion rate and the actual infusion rate. If the difference between the background infusion rate and the actual infusion rate is more than a set threshold (e.g., the difference between the rates are 1%, 5%, 10% etc. more or less than the actual infusion rate), then the monitoring module sends a command to the higher-level logic controller to notify the user of over-administration or under-administration. The notification could be in the form of an audio/visual alarm or by generating a number or index indicating the difference between the actual infusion rate and the background infusion rate.

[0203] In some embodiments, the monitoring module sends a command to the higher-level logic-based controller to notify the user when it identifies at least one entry in the monitoring database with an absolute value of difference between actual infusion rate and background infusion rate greater than a set threshold. In some embodiments, the monitoring module sends a command

2019280053 01 Apr 2020 to the higher-level logic-based controller to notify the user when it identifies in the monitoring database at least two consecutive entries or two non-consecutive entries in a set time interval with an absolute value of difference between actual infusion rate and background infusion rate greater than a set threshold.

[0204] In some embodiments, alarm system is implemented on an infusion pump, where the infusion pump periodically receives data from a hemodynamic monitor. In some embodiments, the audio and/or visual alarm can be generated on the infusion pump using a graphical user interface. In some embodiments, a dedicated light of a specific color (e.g., red, or yellow, or blinking red, blinking yellow) could turn on accompanied by an audio alarm to indicate under-administration or over-administration of fluid and/or cardiovascular drug.

Claims

THE CLAIMS DEFINING THE INVENTION ARE AS FOLLOWS:

1. A computer-implemented method for providing clinical decision support for regulating hemodynamic measurements of a subject, the method comprising:

(a) receiving by way of a user interface at least one initial infusion rate;

(b) recording the at least one initial infusion rate as a last user-approved infusion rate in a database;

(c) receiving from a sensor hemodynamic measurements of a subject by a hierarchical control architecture system;

wherein the hierarchical control architecture system comprises:

(a) at least one lower-level controller; and (b) a higher-level logic controller;

wherein the subject’s hemodynamic measurements are received by the higher-level logic controller and the at least one lower-level controller;

wherein the at least one lower-level controller and the higher-level logic controller are in communication with one another; and wherein the at least one lower-level controller comprises:

(c) a dynamic observer comprising a set of dynamic observer states for estimating a current value of physiological parameters of the subject; and (d) a dynamical system using a function approximator comprising a set of basis functions and a set of weights to approximate unknown dynamics of the subject;

(d) initializing the value of the set of weights of the at least one lower-level controller based at least in part on the at least one initial infusion rate;

(e) updating the set of weights of the at least one lower-level controller based at least in part on current values of the set of weights, the set of dynamic observer states, and current value of the hemodynamic measurements;

(f) generating by the at least one lower-level controller an altered infusion rate based at least in part on the subject’s hemodynamic measurements and the set of weights of the at least one lower-level controller;

(g) verifying by the higher-level logic controller whether the altered infusion rate meets requirements to notify the user;

(h) displaying the altered infusion rate from the at least one lower-level controller if the altered infusion rate meets the requirements in Step (g);

2019280053 01 Apr 2020 (i) asking the user to either accept the altered infusion rate or change the altered infusion rate to a different value;

(j) replacing by the higher-level logic controller the value of the last user-approved infusion rate in the database with the altered infusion rate accepted by the user or the different value of the altered infusion rate if changed by the user in Step (i);

(k) resetting the value of at least one element of the set of weights of the at least one lower-level controller by the higher-level logic controller to a new value based at least in part on the different value of the altered infusion rate if changed by the user in Step (i), otherwise the set of weights of the at least one lower-level controller are kept the same; and (l) administering a fluid and/or cardiovascular drug at the last user-approved infusion rate in the database to the subject, thereby regulating the subject’s hemodynamic measurements.
2. The method of Claim 1, wherein the function approximator is a neural network with sigmoidal basis functions.
3. The method of Claim 1, wherein the user is notified if at least one performance criterion is violated.
4. The method of Claim 3, wherein violation of the at least one performance criterion is determined by comparing a set threshold with the difference between the hemodynamic measurements of the subject and target hemodynamic data in a set time interval.
5. The method of any one of Claims 1 to 4, wherein the hierarchical control architecture system further comprises a second lower-level controller.
6. The method of Claim 5, wherein the first lower-level controller is engaged through the higher-level logic controller, and wherein the higher-level logic controller recommends additional engagement of the second lower-level controller when at least one performance criterion is violated.
7. The method of Claim 6, wherein the performance criterion comprises improvement in the hemodynamic measurements of the subject while administering the fluid or cardiovascular drug.
8. The method of any one of claims 1 to 7, wherein the last user-approved infusion rate is communicated to an infusion pump for administering the fluid and/or drug at the last

2019280053 01 Apr 2020 user-approved infusion rate.
9. The method of any one of claims 1 to 8, wherein at least one requirement verified by the higher-level logic controller to notify the user is determined by comparing a set infusion rate threshold with the difference between the altered infusion rate and the last user-approved infusion rate.
10. The method of any one of claims 1 to 9, wherein the logic-based controller is a rule-based expert system.
11. The method of any one of claims 1 to 10, wherein steps (f) to (1) are repeated in set time intervals.
12. The method of any one of claims 1 to 11, wherein the hemodynamic measurement of the subject comprises the subject’s blood pressure, heart rate, stroke volume variation, urine output rate, central venous pressure, pulse pressure variation, dynamic arterial elastance, systolic pressure variation, mean arterial pressure, systolic pressure, diastolic pressure, cardiac output, cardiac index, systemic vascular resistance, or stroke volume.
13. The method of any one of claims 1 to 12, wherein the hemodynamic measurement of the subject is derived from sensor-obtained hemodynamic measurements within a time window of set length.
14. The method of any one of claims 1 to 13, wherein the hemodynamic measurements are measured invasively or non-invasively.
15. The method of any one of claims 1 to 14, wherein the fluid is a crystalloid, colloid, or a blood product, and the cardiovascular drug is a vasoactive drug or an inotropic drug used to improve cardiovascular function.
16. The method of any one of claims 4 to 15 when dependent upon claim 3, wherein the at least one performance criterion comprises rate of change of weights greater than a set threshold; and/or a sustained increase in infusion rate without improvements in the subject’s fluid resuscitation, hemodynamic, or cardiovascular status as measured by hemodynamic measurements.