JP2015512539A - System and method for transferring patient care from signal-based monitoring to risk-based monitoring - Google Patents

Abstract

A risk-based patient monitoring system for caring for critically ill patients combines data from multiple sources to assess current and future risks to the patient. Thereby, the health care provider can view the current risk profile of the patient, and can continuously track the medical trajectory. The physiology monitoring module in this system uses a plurality of measurements to estimate a probability density function (PDF) of a plurality of internal state variables (ISV) that represent the components of the physiology related to patient treatment and condition. The clinical trajectory interpreter module in this system uses the estimated PDF for the ISV to identify which patient can be classified into the assumed patient states and assigns a probability that the patient will be in each identified state. The combination of patient status and their probability is defined as the clinical risk to the patient.

Description

TECHNICAL FIELD The present invention relates to a risk-based patient monitoring system and a patient monitoring method. More particularly, the invention relates to a method for combining patient data from different sources to assess a patient's current and future risks.

  Medical medicine is becoming increasingly complex with the introduction of new sensors and treatments. As a result, clinicians face enormous patient data that needs to be evaluated and fully understood in order to prescribe optimal treatment from the many options available while reducing patient risk. Will do. One environment in which such a vast amount of information is more problematic is the intensive care unit (ICU). In the intensive care unit, the physician's experience as well as the physician's ability to absorb available physiological information has a strong impact on clinical outcomes. Hospitals that do not have trained intensive care physicians on a 24-hour basis have a mortality rate of 14.4% compared to 6.0% for a fully-centered center. It became clear that It is estimated that raising the level of care to that of an average physician trained for all ICUs can save 160,000 lives annually and save $ 4.3 billion. As of 2012, there is a shortage of intensive care physicians, and according to projections, this shortage will only worsen and is expected to reach 35% in 2020.

  The value of experience in critically ill patient care can be explained by the fact that clinical data in the ICU is provided at a rate much higher than what most competent physicians can absorb, and investigations have revealed This means that, under the condition of excessive information, the probability of error occurrence is 6 times higher, and 11 times higher if there is a serious lack of time. Moreover, treatment decisions in the ICU are highly dependent on clinical signs that cannot be measured directly but are deduced from other physiological information. Clinician expertise and background therefore play a more important role in the minute-by-minute decision making process. Naturally, this results in large variations in the estimation of hidden parameters. For example, in critically ill patient care, a number of parameters that substitute for cardiac output are continuously monitored, but the results of the investigation have been the subjective assessment by clinicians and thermodilution. This means that there is little correlation with objective measurement by law. Experienced intensive care practitioners effectively implement risk management and incorporate such inherent uncertainties into their decision-making process, which means that treatment is based on the most likely patient condition. As well as taking into account the patient's risk of being in the opposite situation. From this perspective, experienced intensive care physicians face excessive data in intensive care when a large number of extraneous signals are diverted from patient observations to risk assessment.

  Therefore, a clear need in the ICU is to achieve a paradigm shift from signal-based patient monitoring to risk-based patient monitoring, thereby helping physicians confronted with enormous data in the ICU It is a decision support system that enables it.

SUMMARY OF THE INVENTION In accordance with the present invention, a risk-based patient monitoring system for critical patient care is disclosed. The system combines data from any bedside monitor, electronic medical records or medical records, and other patient specific information to assess current and future risks to the patient. Furthermore, this system can also be realized as a decision support system that prompts the user for a specific action according to a standardized medical plan when a specific risk of the patient passes a predetermined threshold. Yet another implementation of the technology described herein is an outpatient monitoring system. According to this system, patient and family assessments are combined with information about drug dosing plans and physician assessments to generate patient risk profiles, patient clinical trajectories are continuously tracked, visits or additional Decision support is provided to the physician as to when a specific test should be scheduled.

  According to one embodiment, a risk-based monitoring application is executed in a system processor, which includes a data reception module, a physiological function observation module, a clinical trajectory interpreter module, and a visualization and user interaction module. According to one embodiment, the data receiving module can be configured to receive data from a bedside monitor, an electronic medical record, a treatment device, and for evaluation formation based on prior information regarding a patient's clinical risk. It can be configured to receive any other information that may be relevant, as well as any combination thereof.

  The physiology observation module uses a plurality of measurements to estimate a probability density function (PDF) of internal state variables (ISV) that represent components of physiology related to patient treatment and condition. The clinical trajectory module can be configured by multiple possible patient states, and given the estimated probability density function of internal state variables, how likely any of those patient states will occur It can be determined whether there is.

  In various embodiments, the clinical trajectory interpreter module determines a patient condition that can classify a patient, and given an estimated probability density of internal state variables, can now assume a patient that can be classified. The state can also be determined. In this way, each possible patient state is assigned a probability value between 0 and 1. The combination of patient status and their probability is defined as the clinical risk to the patient.

  The visualization and user interaction module receives i) continuous or intermittently acquired physiological function measurements and patient-specific identifiers such as condition, demographics, visual examination time series from the data receiving module; ii) A time series of probability density functions of estimated internal state variables is received from the physiology observation module, and a time series of probabilities that the patient is in a particular state and hazard levels of individual risks from the clinical trajectory interpreter module. The visualization and user interaction module then visualizes the data in a graph that represents the dependence of variables and time. These graphs can be visualized by plotting the data directly on the screen, or, in the case of a probability density function, the likelihood at a particular time and value can be encoded with a color pattern and the data Visualize by plotting. The visualization and user interaction module can visualize the current risks to the patient by representing them by boxes of different sizes and colors. In this case, the size of the box corresponds to the probability of the patient condition at a particular point in time, and the color of the box corresponds to their hazard level. In addition to these, the Visualization and User Interaction module allows users to set alarms based on patient status probabilities, share those alarms with other users, create notes related to patient risk, and create those notes. Can be shared with other users, and other requirements in the patient's medical history can be viewed.

According to one aspect of the invention, computer-implemented media and methods for patient risk-based monitoring include the following. That is,
A) obtaining, by a computer, data associated with a plurality of internal state variables each representing a parameter that is physiologically related to one of patient treatment and condition;
B) storing the data associated with the acquired plurality of internal state variables in a storage device accessible by a computer;
C) generating a probability density function estimated for the plurality of internal state variables by a computer;
D) From the probability density function in which the internal state variable is generated, a computer is identified as to which of the plurality of assumed patient states the patient can be classified at the present time, and the identified assumption Generating a probability value associated with each patient condition by a computer;

  According to one embodiment, the probability value associated with the identified assumed patient condition is 0-1. According to another embodiment, the method according to the invention further comprises the step of E) displaying on the screen the probability values and the identified individual assumed patient states associated with the probability values. In this case, a combination of the identified assumed patient state and individual probability values associated with the assumed patient state is defined as a clinical risk for the patient.

  According to another aspect of the present invention, a risk-based monitoring system for monitoring a patient has the following configuration. That is, a processor, a storage device connected to the processor, a data receiving module connected to cooperate with a plurality of information sources related to a patient, a physiological function observation module communicating with the data receiving module, A clinical trajectory interpreter module that communicates with the physiological function observation module is provided. The data receiving module obtains data associated with a plurality of internal state variables respectively representing parameters physiologically related to one of a patient's treatment and condition. The physiological function observation module generates a probability density function of the internal state variable, and the clinical trajectory interpreter module determines to which of the plurality of assumed patient states the patient can be classified at the present time. Identify and generate a probability value associated with each identified patient condition identified.

  According to one embodiment, the present invention is further configured as follows. That is, the clinical trajectory interpreter module, the data receiving module, and a user interaction module that communicates with the storage device are provided, and the user interaction module is identified and associated with the probability value and the probability value. Each assumed patient state is displayed, and a combination of the identified assumed patient state and the associated individual probability value is defined as a clinical risk for the patient.

According to yet another aspect of the present invention, the system can make use of specific measurements, such as hemoglobin, with an unknown amount of latency, i.e., these measurements can be made via the current time and data communication link. It is valid at a past time relative to the time of arrival. The physiology observation module can handle this type of measurement out of sequence using backpropagation. With backpropagation, the current estimate of the ISV is projected back to a valid point in time of the measurement so that information from potential measurements can be incorporated appropriately. Thus, according to another aspect of the present invention, a computer-implemented method for patient risk-based monitoring includes the following steps. That is,
A) Data associated with a plurality of internal state variables respectively representing parameters physiologically related to one of the patient's treatment and condition is acquired by a computer, provided that the data is associated with the plurality of internal state variables. Acquiring a part, not all, of the collected data in the same cycle;
B) storing the data associated with the acquired plurality of internal state variables in a storage device accessible by a computer;
C) generating a probability density function estimated for the plurality of internal state variables by a computer;
D) From the probability density function generated for the internal state variable, a first plurality of assumed patient states P (S ₁ ), P (S ₂ ), P (S ₃ ),. . . , P (S _n ) to which assumed patient state it was possible to classify the patient at the preceding time point by the computer, and the probability value associated with each identified preceding assumed patient state, Computer generated step.

According to one embodiment, said step of generating an estimated probability density function comprises: C1) calculating an estimated probability density function for said first plurality of internal state variables at a current time step t _k . and generating, C2) using the defined transition probability kernel, said by progress in the opposite direction from the time step t _k definitive probability estimation until another time step t _kN, said at time step t _kN said another Generating a probability density function for a plurality of internal state variables. However, N is an integer greater than 1.

  First, it should be understood that although exemplary techniques relating to one or more embodiments in the present disclosure are described below, the systems and / or methods disclosed herein will now be known. Regardless of whether or not it exists, it can be implemented using any number of techniques. Further, the disclosure of the present application is not limited to the technology described below, including the exemplified implementation means, drawings, and exemplary designs and techniques illustrated and described below, and is within the scope of the appended claims. Can be changed with the corresponding maximum range.

Schematic showing a risk-based medical monitoring environment according to the present invention Schematic showing the basic configuration of the risk-based physiological function observation module according to the present invention Schematic showing an exemplary graph of the probability density function for a selected ISV generated by a physiology monitoring module according to the present invention. Schematic showing an exemplary graph of the probability density function for a selected ISV generated by a physiology monitoring module according to the present invention. Schematic showing an exemplary graph of the probability density function for a selected ISV generated by a physiology monitoring module according to the present invention. Schematic showing a non-limiting example of a physiological function observation process according to the present invention Schematic showing a non-limiting example of a physiological function observation process according to the present invention Schematic showing a timeline where backpropagation is used to incorporate information according to the present invention Schematic showing an example of a process according to the invention relating to mean arterial pressure (ABPm) Schematic showing an example of resampling according to the invention Schematic illustrating a clinical trajectory interpreter module that performs state probability estimation using a combined probability density function of multiple ISVs to calculate the probability of various patient states in accordance with the present invention. Schematic diagram illustrating a non-limiting example of a patient state definition employed by a clinical trajectory interpreter module according to the present invention. A non-limiting example of how the clinical trajectory interpreter module can employ patient state definitions to assign a probability that a patient will be classified into each of four possible patient states at a particular point in time Schematic Schematic showing an alternative approach for estimating probabilities of various patient conditions according to the present invention Schematic diagram illustrating a non-limiting example of the definition of a patient state assigned a hazard level by a clinical trajectory interpreter module according to the present invention. Schematic showing patient states and their individual probabilities incorporated into three graphs referred to as etiology according to the present invention Schematic showing an example of an etiology tree for a predetermined set of patient states and physiological variables according to the present invention Schematic showing the method for calculating the utility of various measurements according to the invention Schematic diagram illustrating one possible approach for incorporating external computations generated from third party algorithms in accordance with the present invention. Schematic diagram showing an example of an external calculation built-in instruction according to the present invention Schematic diagram showing additional examples of external computation built-in instructions according to the present invention Schematic showing an example of the functionality of the visualization and user interaction module according to the present invention Schematic diagram showing an example of a summary view that can convey a risk profile for each patient in a single hospital according to the present invention on a single screen Schematic showing an example of a technique that can implement a view representing a patient's ongoing risk according to the present invention. Schematic diagram showing how the slider provided at the top of the patient view can be used to view patient risk history according to the present invention. A diagram illustrating how a user can manipulate an etiology tree by clicking on a synthesized patient state and viewing the constituent patient states according to the present invention. Schematic showing how a user can view the predicted risk for a patient within the same framework by sliding the slider to advance beyond the current time in accordance with the present invention. Figure Schematic showing how a user can select and set an alarm for a particular risk according to the present invention Schematic diagram illustrating a patient risk trajectory according to the present invention, i.e. yet another possibility to visualize the evolution of patient state probabilities associated with a particular patient risk. Schematic showing how the system can directly visualize the probability density function of various internal state variables in accordance with the present invention. Schematic diagram illustrating an example of a tagging function that can be implemented by a user interface according to the present invention. Schematic showing the news distribution screen of the user interface according to the present invention Schematic showing the condition summary screen of the user interface according to the present invention Schematic diagram illustrating the ability of a user interface to include and display reference material accessible through the Internet or stored in a system according to the present invention. Schematic showing a general purpose Dynamic Bayesian Network DBN that can be employed to capture a physiology model of a patient in a temporary relaxed state of HLHS stage 1 in accordance with the present invention. Schematic showing a plurality of equations that can be used to model physiological dynamics in HLHS stage 1 in accordance with the present invention. Schematic illustrating equations that can be used to extract the relationship between dynamic and derived variables in modeling in accordance with the present invention. Schematic diagram illustrating the potential of an observation model that can be used to correlate derived variables with available sensor data in accordance with the present invention. Schematic showing the attributes, patient status, and possible etiology tree that can be utilized by the clinical trajectory interpreter module in the HLHS stage 1 population in accordance with the present invention. Schematic illustrating one possibility for an environment in which a risk-based monitoring system can be employed to assist a clinician in deciding whether to apply a particular treatment according to the present invention. Schematic illustrating a non-limiting example of a set of patient conditions associated with a transfusion that can be used to notify a transfusion decision in accordance with the present invention. Schematic diagram illustrating another application of a risk-based monitoring system applied to a standardized medical plan according to the present invention Schematic illustrating one application for a risk-based monitoring system combined with a specific type of standardized clinical plan according to the present invention Schematic showing an example of risk stratification that can be employed by the system in the context of nitric oxide therapy in accordance with the present invention. Schematic showing one possible patient condition that can describe the clinical trajectory of ADHD patients according to the present invention. List of available patient evaluation modalities as M1, M2, M3 Schematic diagram showing a dynamic model of patient progression from one state to another extracted by a dynamic Bayesian network according to the present invention Schematic showing alternative embodiments with respect to two predictions about how patient status may transition in a single month when changing drugs or dosages in accordance with the present invention. Schematic diagram illustrating one embodiment of visualization embodiments and scenarios displaying patient clinical trajectories and risks in accordance with the present invention. Schematic showing the assessment of patient and patient trajectory at week 9, at which point the risk-based patient monitoring system determines the probability distribution function for the patient's condition for each of the past 9 weeks according to the present invention. Figure FIG. 3 shows a follow-up evaluation based on a teacher and parent Vanderbilt diagnosis according to the present invention. A diagram illustrating all the available measurements that are likely to produce significant improvements in accordance with the present invention, i.e. the resulting assessment based on workplace visits, parent and teacher assessments Diagram showing yet another follow-up according to the present invention when it is certain that the patient has been improved steadily in 9 minutes and a stable improvement is seen between the two assessments. Diagram showing the follow-up assessment according to the present invention for patients and patient trajectories in the absence of measured values, showing the state of increased uncertainty due to lack of recent observations Diagram showing the state of uncertainty described above when a complete patient assessment (all measurement modalities) is performed in accordance with the present invention. Demonstrates further visualization possibilities from the system output described above, and illustrates the possibility of patient status transitions when treatment plan changes, such as drug changes, are made in accordance with the present invention. Figure

DETAILED DESCRIPTION Disclosed herein is a technique that provides clinical staff with risk-based patient monitoring for individual patients. The technique described here can be implemented as a monitoring system for critical patient care. This system combines data from various bedside monitors, electronic medical records or medical records, and other patient specific information to assess current and future risks to the patient. Furthermore, this technique can also be implemented as a decision support system that prompts the user for specific actions according to a standardized medical plan when a patient's specific risk passes a predetermined threshold. Yet another embodiment of the described technique is an outpatient monitoring system. According to this system, patient and family assessments are combined with information about drug dosing plans and physician assessments to generate patient risk profiles, patient clinical trajectories are continuously tracked, visits or additional Decision support is provided to the physician as to when a specific test should be scheduled.

System Modules and Interactions Referring now to the drawings, FIG. 1 illustrates a risk-based medical monitoring environment 1010. This provides health-based providers, such as doctors, nurses or other health care providers, with risk-based monitoring according to various embodiments of the present invention. In this case, the patient 101 can be connected to one or more physiological sensors or bedside monitors 102, thereby monitoring various physiological parameters of the patient. Examples of these physiological function sensors include, but are not limited to, blood oximeters, blood pressure measuring devices, pulse measuring devices, glucose measuring devices, one or more analytical measuring devices, electrocardiogram recording devices, and the like. Can be included. In addition to these, routine tests and examinations can be performed on the patient, and data can be stored in an electronic medical record (EMR) 103. The electronic medical record 103 includes, but is not limited to, hemoglobin, arterial and venous oxygen content, lactic acid, body weight, age, sex, ICD-9 code, capillary refill time, and clinician subjective Information such as observations, patient's own evaluation, prescribed drugs, drug administration plans, genetic characteristics, etc. can be stored. Additionally, the patient 101 can be coupled to one or more treatment devices 104 that are configured to provide treatment to the patient. In various embodiments, the treatment device 104 can include an extracorporeal membrane oxygenator, a ventilator, a drug infusion pump, and the like.

  In accordance with the present disclosure, patient 101 can be provided with improved risk-based monitoring through existing methods. A patient specific risk-based monitoring system (generally referred to herein as system 100) may be configured to receive patient related information. Examples of the patient-related information include real-time information from the bedside monitor 102, EMR patient information from the electronic medical chart 103, information from the treatment device 104 such as settings, injection rates, and drug types. Other patient-related information includes the patient's medical history, previous treatment plan, previous and current laboratory test results and current laboratory test results, allergy information, predisposition to various conditions, and patient occurrence. An assessment based on prior information about the conditions and conditions to be obtained and the possibilities associated with them. For simplicity, the various types of information listed here are generally referred to as “patient-specific information” below. In addition, the system can be configured to determine the clinical risk using the received information, and then the determination can be presented to the health care provider. Note that healthcare providers include, but are not limited to, doctors, nurses, or other types of clinical staff.

  In various embodiments, the system includes one or more of a processor 111, a storage device 112 connected to the processor 111, and a network interface 113 configured to allow the system to communicate with other devices over a network. It is included. In addition, the system can include a risk-based monitoring application 1020. The application can include instructions executable by the computer, which when executed by the processor 111 allows the system to perform risk-based monitoring for the patient, eg, the patient 101.

  The risk-based monitoring application 1020 includes, for example, a data reception module 121, a physiological function observation module 122, a clinical trajectory interpreter module 123, and a visualization and user interaction module 124. According to one embodiment, the data receiving module 121 can be configured to receive data from the bedside monitor 102, the electronic medical record 103, the treatment device 104, and provides prior information regarding the patient's clinical risk. It can be configured to receive any other information that may be relevant to the evaluation formation based on, as well as any combination thereof.

The physiological function observation module 122 uses a plurality of measurements and, according to a predefined physiological function model, a probability density function (PDF) of an internal state variable (ISV) that represents a component of the physiological function related to treatment and condition of a patient. ). The ISV may be directly observable with noise (a non-limiting example is a heart rate is an observable ISV), or it may be invisible or hidden ( As a non-limiting example, the oxygen supply DO ₂ defined as aortic blood oxygen saturation cannot be measured directly and is therefore not visible) or may be measured intermittently (non-limiting As a typical example, the concentration of hemoglobin measured by a complete blood count test can be observed intermittently (ISV).

  According to one embodiment, instead of assuming that all variables can be deterministically estimated without error, according to the physiological function observation module 122 of the present invention, a probability density function is provided as an output. . Further details regarding the physiological function observation module 122 will be described later.

  The clinical trajectory module 123 can be configured, for example, by a plurality of possible patient states, and given the estimated probability density function of the internal state variables, what is the probability of any of those patient states occurring? Can be determined. A patient condition may be defined as a qualitative description of physiology at a particular point in the clinical trajectory that is recognizable by practice and may have implications for making clinical decisions. Examples of specific patient conditions include, but are not limited to: hypotension with sinus tachycardia, hypoxia with myocardial depression, compensatory circulatory shock, cardiac arrest, bleeding, etc. I will give you a list. In addition, these patient states can be specific to a particular medical condition, and the range of each patient state can be defined by various physiologic variables and data thresholds. In various embodiments, the clinical trajectory interpreter module 123 can determine the patient's condition using reference material, information provided by the healthcare provider, and any information collected from other sources, such as Patients can be categorized into conditions. For example, reference material may be stored in a database or other storage device 130 that is accessible to the risk-based monitoring application 1020 via the network interface 113. These reference materials should include reference materials, medical literature, expert research, information provided by physicians, and materials compiled from any other material that can be used as a reference for providing medical care to patients. Can do. According to some embodiments, the clinical trajectory interpreter module 123 may initially identify a patient population that is similar to the monitored patient. In this way, the clinical trajectory interpreter module 123 can utilize relevant historical data based on the identified patient population to help determine possible patient conditions.

  The clinical trajectory interpreter module 123 can also determine possible patient states, given the estimated patient state probability density function provided by the physiology observation module 122, and easily classify patients into those patient states. be able to. In this way, each possible patient condition is assigned a probability value between 0 and 1. The combination of patient status and their probability is defined as the clinical risk to the patient. Further details regarding the clinical trajectory interpreter module 123 will be described later.

  The visualization and user interaction module 124 is configured to receive the outputs of the data receiving module 121, the physiological function observation module 122, and the clinical trajectory interpreter module 123, and presents them to clinical staff. The visualization and user interaction module 124 was calculated by the two modules 122 and 123 as current patient risks, their evolution over time, a probability density function of time-dependent internal state variables, and by-products, Show other features that are beneficial to medical practice. In addition to this, the visualization and user interaction module 124 allows the user to set alarms based on patient status probabilities, share those alarms with other users, create notes related to patient risk, and You can share your notes with other users and view other requirements in the patient's medical history. More details regarding the visualization and user interaction module 124 will be described later.

Physiological Function Observation FIG. 2 shows a basic schematic diagram of the physiological function observation module 122. This module uses two models of patient physiology: a dynamic model 212 and an observation model 221. Dynamic model 212 to model captures the relationship that occurs between the internal state variables at time t _{k + 1} which approaches internal state variables and another at a certain point in time t _k, thereby the patient physiology as one system The current state of the system has information about possible future developments in this system. If the patient's physiology tends to be maintained in a homeostatic state by self-regulation, a clear rationale for introducing this kind of memory into internal state variables representing homeostatic stability, such as oxygen transport and oxygen consumption. Sex exists.

The observation model 221 can capture the relationship between the measured physiological function variable and other internal variables. Examples of this type of model are:
a) Dependence of the difference between systolic and diastolic arterial pressure (also referred to as pulse pressure) on stroke volume b) Relationship between heart rate, stroke volume and cardiac output c ) Relationship between hemoglobin concentration, cardiac output and oxygen transport d) Relationship between Vanderbilt Assessment Scale and clinical status of attention-deficit / hyperactivity disorder patients e) Measurable and therefore observable parameters And some other relationship between internal state variables.

The physiological function observation module 122 functions as a recursive filter, in which information based on previous measurement values is used to generate predictions of internal state variables and the likelihood of possible future measurement values. It is compared with the last measured value. In particular, the physiological function observation module 122 uses the dynamic model 212 in the prediction step or prediction mode 210, and uses the observation model 221 in the update step or update mode 220. During the prediction mode 210, the physiology observation module 122 creates the estimated PDFs of the ISV 213 at the current time step t _k and supplies them to the dynamic model 212, which is in the next time step T _k. An ISV prediction 211 for ₊₁ is generated. This is achieved by using the following equation:

ただし、ISVs(t_k) = {ISV₁(t_k),ISV₂(t_k),ISV₃(t_k),...lSV_n(t_k)}およびM(t_k)は、時点ｔ_kまでのすべての測定値の集合である。確率P(ISVs(t_k+ l)|ISVs(t_k))によって、ダイナミックモデル２１２を表す遷移確率カーネルが定義され、これによれば推定されたＰＤＦが時間の経過とともにどのように進展するのかが規定される。確率P(ISVs(t_k)|M(t_k))は推論エンジン２２２によって供給され、これは先行の時間ステップにおいて測定値が取得されている場合の、ＩＳＶの事後確率である。生理機能観測モジュール１２２のアップデートモード２１０中、予測されたＩＳＶ２１１が、データ受信モジュール１２１から受け取った測定値と、観測モデル２２１を用いて比較され、その結果、ＩＳＶがアップデートされて、新たに利用可能な情報を反映させる。モジュール１２２の推論エンジン２２２は、予測されたＰＤＦを事前確率として用いることによってこのアップデートを達成し、これは測定値の統計を用いてアップデートされて、現在のＩＳＶのＰＤＦ推定値を反映する事後確率を達成する。推論エンジン２２２は、ベイズの定理である次式を用いてアップデートステップ２２０を達成する。

Where ISVs (t _k ) = {ISV ₁ (t _k ), ISV ₂ (t _k ), ISV ₃ (t _k ), ... lSV _n (t _k )} and M (t _k ) A set of all measurements up to _k . The probability P (ISVs (t _k + l) | ISVs (t _k )) defines a transition probability kernel that represents the dynamic model 212, and how the estimated PDF evolves over time. Is defined. The probability P (ISVs (t _k ) | M (t _k )) is supplied by the inference engine 222, which is the posterior probability of the ISV when measurements are taken in the previous time step. During the update mode 210 of the physiological function observation module 122, the predicted ISV 211 is compared with the measurement value received from the data reception module 121 using the observation model 221, and as a result, the ISV is updated and newly available. Information. The reasoning engine 222 of the module 122 achieves this update by using the predicted PDF as a prior probability, which is updated with the measured statistics to reflect the current ISV's PDF estimate. To achieve. The inference engine 222 accomplishes the update step 220 using the following equation, which is Bayes' theorem.

ただしP(m₁(t_k+1), m₂(t_k+1),... m_n(t_k+1)|ISVs(t_k+1))は、観測モデル２２１によって供給される条件付き尤度カーネルであり、これによれば現時点で予測されているＩＳＶに対し、現時点で受け取った測定値がどの程度の尤度であるかが求められる。

Where P (m ₁ (t _{k + 1} ), m ₂ (t _{k + 1} ), ... m _n (t _{k + 1} ) | ISVs (t _{k + 1} )) is supplied by the observation model 221 This is a conditional likelihood kernel, which determines the likelihood of the currently received measurement value relative to the ISV currently predicted.

  If the current estimate for the ISV's PDF is not available at an initial time, eg, t = 0, the physiology observation module 122 can use the initial estimate 250, which is experienced with possible values for the ISV. Or from statistical analysis of pre-collected patient data.

FIG. 3 shows a non-limiting example of a model that enables physiological function observation according to the present invention. Management of the oxygen delivery DO ₂ is an important part of patient care, although not directly observable. Therefore, improved medical care can be performed by precise estimation of DO ₂ . In the example shown, this estimate hemoglobin concentration (Hg), heart rate (HR), arterial pressure diastolic and systolic, and is achieved by SpO _2. The dynamic model 212 assumes that oxygen transport is driven by a feedback process that stabilizes against stochastic disturbances. Similarly, the hemoglobin concentration is controlled around a nominal value of 15 mg / dL.

The observation model 221 includes arterial oxygen saturation SpO ₂ , hemoglobin concentration and arterial oxygen concentration CaO ₂ , dependency of the difference between systolic ABP _s and diastole ABP _d , arterial pressure (pulse rate) in stroke volume. Pressure), and the relationship among the heart rate HR, stroke volume SV, and cardiac output is considered. These two models are collected as a dynamic Bayesian network (DBN), and the physiological function observation module 122 uses this DBN to continuously track the oxygen transport amount. A dynamic Bayesian network is a systematic method that represents statistical dependency by a graph, in which a vertex of the graph represents a variable (observable and unobservable), and an edge represents a causal relationship. A detailed description of DBN as an example for DO ₂ estimation can be found in US Provisional Application No. 61 / 699,492, filed September 11, 2012, entitled “SYSTEMS AND METHODS FOR EVALUATING CLINICAL TRAJECTORIES AND TREATMENT STRATEGIES FOR OUTPATIENT CARE” ", Attorney Docket No. 44429-00107 PROV, and United States Provisional Application No. 61 / 684,241, filed August 17, 2012, title of invention" SYSTEM AND METHODS FOR PROVIDING RISK ASSESSMENT IN ASSISTING CLINICIANS WITH EFFICIENT AND EFFECTIVE BLOOD MANAGEMENT ", Agent Docket No. 44429-00106 PROV, this application claims priority to them, the disclosure of which is hereby incorporated by reference into this application. To do.

FIG. 4 shows a non-limiting example of the above physiology observation that tracks DO ₂ , but this figure shows tracking over a longer time interval, ie tracking over 4 steps. In this observation, the main hidden ISV is the oxygen carrying amount variable (DO ₂ ). Two types of measured values, namely hemoglobin (Hg) and oxygen (SpO ₂ ), are depicted in the dashed circle in FIG. SpO ₂ is a continuous or periodic measurement value, and the physiological function observation module 122 detects these measurement values, for example, sensors such as a bedside monitor 102 and a treatment device 104 that are connected to the patient 101 and continuously report information. Receive from. Hemoglobin (Hg) is an example of an intermittent or aperiodic measurement extracted from a patient's laboratory test, based on a sporadic and irregular basis, sometimes potentially relative to the current system time. The observation side can be used. The physiological function observation module 122 can handle both types of measurement values. Module 122 since is because with tracking hidden ISV example DO _2, even in the absence of measured values, the measured values of all types continue to estimate observations continuously. FIG. 4 depicts these estimates for the SpO ₂ and Hg cases. As can be seen from the figure, SpO ₂ measurements are regularly available at each time step, while Hg is only available at two time steps.

  As mentioned above, the system is able to use specific measurements such as hemoglobin with an unknown amount of latency, i.e. these measurements are relative to the current point in time, i.e. the point of arrival over the data communication link. It is valid at a relative past point in time. The physiological function observation module 122 can handle this type of measurement out of sequence using back-propagation. With backpropagation, the current estimate of the ISV is projected back to a valid point in time of the measurement so that information from potential measurements can be incorporated appropriately.

FIG. 5 illustrates such a timeline. As shown in FIG. 5, hemoglobin arrives at the current system time T _k , which is valid at time T _k−2 and is associated with the current ISV (DO ₂ ). Back-propagation is a method of updating the current ISV probability estimate P (ISVs (t _k ) | M (t _k )) with a potential measurement m (t _kn ) relative to the current time point. . Backpropagation is realized by the same method as the prediction method described above. In this case, there is a transition probability kernel P (ISVs (t _kn ) | ISVs (t _k )), which defines how the current probability develops in the reverse direction in time. If there is a current set of measurements excluding potential measurements, this can be used to calculate the probability of ISV at time t _kn as follows.

これらの確率が計算されると、ベイズのルールを用いることで、潜在的な測定情報が標準的なアップデータに組み込まれる。

Once these probabilities are calculated, potential measurement information is incorporated into the standard updater using Bayesian rules.

その後、前述の予測ステップを用いて、アップデートされた確率が現在時点ｔ_kにバックプロパゲートされる。情報を組み込むために、バックプロパゲーションを利用することができる。

Thereafter, the updated probability is back propagated to the current time t _k using the prediction step described above. Backpropagation can be used to incorporate information.

Another function of the physiological function observation module 122 is smoothing. A healthcare provider utilizing the system 100 may be interested in the patient status at some past time. By smoothing, the physiology monitoring module 122 can provide a more accurate estimate of the patient's ISV at a past point in time. In this case, all new measurements received by the system since that time are incorporated, resulting in a better estimate than the original filtered estimate of the entire patient state at that time, with P ( ISVs (t _kn ) | M (t _k )) is calculated. This is achieved by using the first step of backpropagation. According to this step, an estimate of the probability at this time, incorporating all the measurements up to time t _k , is developed backwards to the time of _interest t _kn using the defined transition probability kernel. . This is also depicted in FIG. 5, which indicates that the user is interested in the patient condition at time T _kn and the estimate is smoothed back to this time.

  The physiology observation module 122 maintains an estimate of each measurement value available to the system 100 based on the physiology model and the statistical model, so that the module 122 has a measurement value that is not related to the current information contained in the measurement value. Artifacts can be filtered. This is done by comparing the newly acquired measurement with the predicted likelihood of the possible measurement, if there is a previous measurement. If the model considers that new measurements cannot occur with considerable probability, those measurements are not incorporated into the estimation. The process of comparing measurements with their predicted likelihoods effectively filters artifacts and reduces noise.

  FIG. 6 shows an example of this type of process for mean arterial pressure (ABPm). Since ABPm is collected using a venous catheter, the measured signal is often degraded by artifacts when the catheter is used for medical procedures such as blood sampling or line flushing. In FIG. 6, raw ABPM measurements before being processed by the physiology observation module are shown along with the identified measurement artifacts, and further filtered after being processed by the physiology observation module 122. Measurements are shown. As can be seen, the measurement artifacts have been removed, leaving the correct signal.

  According to various embodiments, the physiology observation module 122 can utilize multiple algorithms for estimation or inference. Depending on the physiology model used, the physiology observation module 122 may use an accurate inference scheme, such as a junction tree algorithm or an approximate inference scheme using Monte Carlo sampling such as a particle filter, a Kalman filter, etc. A simple Gaussian approximation algorithm, or any variation thereof, can be used.

  As described above, the physiological function model used by the physiological function observation module 122 can be implemented using a probabilistic framework known as a dynamic Bayesian network. A dynamic Bayesian network is a graph of causal and probabilistic relationships between ISVs in a system, and these causal and probabilistic relationships are in a single time instance and over time. As it happens. Since this type of model expression is flexible, the physiological function observation module 122 can use a plurality of different inference algorithms. The choice of algorithm depends on the characteristics of the physiology model used, the inference accuracy required by the application, and the computational resources available in the system. As used herein, accuracy refers to whether an accurate inference scheme is used or an approximate estimation scheme is used. If the complexity of the physiological function observation model is limited, it is appropriate to use an accurate inference algorithm. In other cases, with more complex physiological function observation models, there is no closed-form reasoning solution, or if it exists, it cannot be computationally processed with available resources. In this case, an approximate reasoning scheme can be used.

  The simplest case where accurate reasoning can be used is when the ISVs in the physiology model are all continuous variables and the relationship between each ISV in the model is limited to a linear Gaussian relationship. In this case, a standard Kalman filter algorithm can be used to perform the inference. According to this type of algorithm, the probability density function for ISV is a multivariate Gaussian distribution, represented by means and covariance matrices.

  If all the ISVs in the model are discrete variables and the structure of the graph is limited to chains or trees, the physiology observation module 122 uses a forward-backward algorithm or a probability propagation algorithm for inference. Each can be used. The junction tree algorithm is a generalization of these two algorithms that can be used regardless of the underlying graph structure, and thus the physiological function observation module 122 can also use this algorithm for inference. The junction tree algorithm comes with an additional computational cost, which may not be acceptable for the application. In the case of discrete variables, the probability distribution function may be represented in a flat plate shape. To state here, if the model consists only of continuous variables with a linear Gaussian relationship, this algorithm can be used for inference, but in this case the algorithm is equivalent to a Kalman filter. Therefore, a Kalman filter is used as an example of the algorithm.

  If the physiology model consists of both continuous and discrete ISVs with non-linear relationships between variables, accurate inference solutions are not possible. In this case, the physiology module 122 can use an approximate reasoning scheme based on sampling techniques. The simplest version of this type of algorithm is a particle filter algorithm that uses sequential importance sampling. For more efficient sampling, a Markov chain Monte Carlo (MCMC) sampling method can also be used. For complex, non-linear physiological relationships, this type of approximate reasoning scheme provides the highest flexibility. As will be apparent to those skilled in the art, the models and inferences employed by the physiology observation module may be any combination of the modeling and inference techniques described above, or other equivalent modeling and inferences. Techniques may be included.

  If a particle filtering method is used, a resampling scheme is required to avoid degeneracy. The physiological function observation module can use an adaptive resampling scheme. As will be described in detail below, the ISV state space region may correspond to different patient conditions as well as different hazard levels for the patient. As the number increases, the patient's condition becomes more dangerous to the patient's health. A sufficient number of sampled particles in the region may be required to ensure an accurate estimate of the probability of a particular condition of the patient. It may be most important to maintain an accurate estimate of the probability of a region with a high hazard level, so that an adaptive resampling approach allows sufficient particles in a high hazard region in state space. Are reliably sampled. FIG. 7 shows an example of this resampling. State 1 and state 2 have the highest hazard level. The plot on the left shows the samples generated by standard resampling. It should be noted that state 1 and state 2 are most likely, and of course there are more particles in these states. The right plot shows the impact of adaptive resampling. Note that in this case, the number of samples has increased significantly in the highest risk area.

Clinical Trajectory Interpreter Referring now to FIG. 8, the clinical trajectory interpreter 123 receives the ISV joint probability density function from the physiology observation module 122 and performs state probability estimation 801 to calculate probabilities for various patient conditions. To do. The probability density function of the ISV can be defined in closed form and can be defined, for example, as a multidimensional Gaussian curve 260, or approximated by a histogram 280 of particles 270, as shown in FIGS. be able to. In both cases, the probability density function of ISV can be expressed as P (ISV1 (t), ISV2 (t),... ISVn (t)), where t is the time associated with them. Given internal state variables, patient states can be defined by a conditional probability density function:
P (S | ISV1, SV2, ..., ISVn) where _{S ∈ S 1, S 2,} ..., S n represents all possible patient state S _i. The determination of the probability of the patient in a particular state S _i can then be made according to the following equation:

If P (ISV1 (t), ISV2 (t), .., ISVn (t)) is defined by a closed-form function such as a multidimensional Gaussian curve 260, the integration is performed as it is. Can do. P (ISV1 (t), ISV2 (t),.., ISVn (t)) is approximated by a histogram 280 of particles 270, and the space over ISV ₁ , ISV ₂ , ..., lSV _n is shown in FIG. In the case where P (S | ISV ₁ , ISV ₂ , ..., ISV _n ) is defined by being divided into regions as shown, calculate the percentage of particles 270 for each region The probability P (S _i (t)) can be calculated by

  Once the patient state probabilities are estimated, the clinical trajectory interpreter module 123 can assign a different hazard level 802 for each patient state, or can incorporate each state into a different etiology 803. The clinical trajectory interpreter module 123 can cooperate with the physiological function observation module 122 to perform the measurement validity determination 804, whereby different views such as, for example, invasive blood pressure or invasive oxygen concentration monitoring. Determine the effectiveness of blood measurements. According to one embodiment, the clinical trajectory interpreter module 123 determines the probability that the patient is in a particular state, rather than the exact state in which the patient is placed.

FIG. 9 depicts a non-limiting example of a patient state definition that can be employed by the clinical trajectory interpreter module 123. In particular, the assumption here is that the function P (S | ISV ₁ , ISV ₂ , ..., ISV _n ) is defined by dividing the region over the internal state variables ISV ₁ , ISV ₂ , ..., ISV _n It can be done. According to a special example, a patient's physiology shall be defined by two variables: pulmonary vascular resistance (PVR) and cardiac output. The specific risks and individual etiologies that can be captured by these two ISVs are based on the effects of increased pulmonary vascular resistance on the circulatory system. In particular, high PVR can cause right heart failure and thus reduced cardiac output. Thus, by assigning thresholds to the two variables, PVR can be used to define PVR normal and PVR high attributes, and CO can be used to define CO normal and CO low attributes. By combining these attributes, four distinct states can be defined: State 1: CO low, PVR normal; State 2: CO low, PVR high; State 3: CO normal, PVR high; State 4: CO normal, PVR normal.

  FIG. 10 shows how the clinical trajectory interpreter module 123 can employ patient state definitions to assign the probability of a patient being classified into each of the four possible patient states at a particular point in time. A non-limiting example is shown. In this example, the clinical trajectory interpreter module 123 receives a joint probability density function of P (cardiac output (Tk), pulmonary vascular resistance (Tk)), integrates it over the region for each particular condition, and P (S1 (Tk)), P (S2 (Tk)), P (S3 (Tk)), and P (S4 (Tk)) are generated. In this way, the clinical trajectory interpreter module 123 assigns a probability that a particular patient condition will progress when information is provided by the physiological function observation module 122. If the output of the physiological function observation module 122 is not the closed-form function 260 but the histogram 280 of the particles 270, the clinical interpreter does not perform integration, but only calculates the individual proportions of the particles 270 in each region. I do.

FIG. 11 depicts an alternative approach for estimating probabilities for various patient conditions. According to this alternative approach, to calculate the probabilities P (S1), P (S2), P (S3) and P (S4), the clinical trajectory interpreter module 123 has two consecutive time windows T _{Use the} ISV joint probability function for _k and T _{k + 1} to calculate the moving average of the time window. Note that in this example, the size of the time window is doubled for two time instances, that is, the window can be any suitable size. As a result of performing the moving average of the time window in this manner, the clinical interpreter module 123 performs a dynamic analysis of the trajectory in the ISV. That is, according to this, the probability that the physiological trajectory described by ISV exists in a specific area in a specific time frame is measured. In other words, such a probability calculation makes an estimate of the probability that a particular patient condition will progress within the selected time frame, rather than at the selected time instance.

  The clinical trajectory interpreter module 123 also assigns a hazard level to each particular state. FIG. 12 depicts a non-limiting example regarding the definition of a patient condition assigned a hazard level by the clinical trajectory interpreter module 123. Hazard levels can be obtained from clinician examinations, references, or other clinical sources. In the specific embodiment, the clinical trajectory interpreter module 123 classifies four different hazard levels: 1-minimum risk, 2-mild risk, 3-moderate risk, 4-severe risk. The combination of the probability of a certain patient condition and its hazard level will hereinafter be referred to as “patient risk”.

  FIG. 13 shows how patient states and their individual probabilities can be incorporated into three graphs called etiology. Specifically, the attributes “normal” and “low” assigned to the cardiac output ISV are the base nodes of the graph. Each of these vertices has two child elements assigned to the pulmonary vascular resistance attribute. According to this tissue structure, the patient state is a leaf (and a vertex) in the tree. This special tree will be referred to as the etiology tree. The visualization and user interaction module 124 can use this etiology tree, which generates a stratified view of various patient risks, as described below.

  FIG. 14 shows that there may not be only one etiology tree for a given set of patient status and physiological function variables. In particular, FIG. 14 shows an etiology tree that is an alternative to the embodiment of FIG. The root of this alternative pathogenesis tree starts with an attribute assigned to pulmonary vascular resistance rather than cardiac output. In this case, it is preferable that different rules may be used for tree generation depending on various factors and usage situations. For example, a single etiology tree may be selected in preparation for different implementations depending on various clinical situations or user preferences. Moreover, the tree can be dynamically changed when the risk changes and as the clinical situation evolves.

Effectiveness of various measurements There are measurements that can harm patients or delay patient recovery during hospitalization. Examples of this type of dangerous measurement are all measurements derived from catheters, such as invasive blood pressure and blood oxygen saturation, which have been found to significantly increase the risk of infection. Accordingly, it may be beneficial to provide the clinician with an assessment of the effectiveness of each potentially harmful measurement during the treatment process. FIG. 15 shows a method for calculating the effectiveness of various measurements.

Referring to FIG. 15, the risk-based system 100 and the clinical trajectory interpreter module 123 can calculate the effectiveness of a particular measurement according to the illustrated procedure. Specifically, in step 9001, it is possible to select the measurement m _i. Given the measured value _mi and the current time point (t _current ), in step 9002, the clinical trajectory interpreter module 123 issues a command to the physiological function observation module 122, and outputs the physiological function observation module output (probability of the internal state variable). the density function), from any given point in time prior to the current time point (t _current-T) until the present time t _current, with the exception of the measurement m _i from the output of the algorithm can be simulated. The simulated output of the physiology module is generated and a set of patient state probabilities, ie P _sim (S ₁ (t _current )), P _sim (S2 (t _current )), ..., P _sim (S _{When n} (t _current )) arrives, the clinical trajectory interpreter module 123 can then simulate an estimation of state probabilities in step 9003. Thereafter, in step 9004, the state probabilities determined from all available measurements, ie, P (S ₁ (t _current )), P (S ₂ (t _current )), ..., P (S _n (t use _current)), clinical trajectory interpreter module 123 may calculate the effectiveness of the measurement m _i by the following equation.

これは、利用可能なすべての測定値が存在するときの患者状態分布と、測定値ｍ_iがタイムインターバルＴにわたり除かれたときの患者状態分布との間の、カルバック・ライブラー・ダイバージェンスでもある。

This is also the Cullback liberian divergence between the patient state distribution when all available measurements are present and the patient state distribution when the measurements _mi are removed over time interval T. .

As another option, in step 9005, the clinical trajectory interpreter module 123 can calculate the efficacy for m _i according to the following equation by using the hazard level r _i assigned to the patient condition S _i .

  In the same way, the clinical trajectory interpreter module 123 can calculate the validity calculation for any group of measurements, not just for one particular measurement. Effectiveness calculations can also include components that capture potential hazards associated with a particular measurement. For example, the open catheter measurement described above has a higher level of harm assigned to it. In this way, the hazard assigned to the measurement is subject to a trade-off for the value of the information supplying the measurement. An example of the effectiveness calculation thus changed is as follows.

ただしＨ（ｍ_i）は、利用可能な測定各々の害を表す関数を規定するものである。

Where H (m _i ) defines a function representing the harm of each available measurement.

The risk-based monitoring system 100 can also incorporate external calculations generated from third-party algorithms implemented on the same computing medium as the patient-based monitoring system or as part of an external device. FIG. 16 illustrates one possible approach for incorporating external computations generated from third party algorithms. Specifically, the output from the external calculation 9110 is provided to the clinical trajectory interpreter module 123, which executes the embedded instructions 9120. As a result, the state probability estimate 801 generates new states P (newS ₁ ), P (newS ₂ ), P (newS ₃ ),..., P (newS _{n + m} ), thereby n n An increased number of states n + m from the original number of states can be generated. Similarly, built-in instructions 9120 can be provided to hazard level assignment 802 and etiology organization 803.

FIG. 17 shows an example of an embedded instruction. In this example, information on a specific binary attribute A = a ₁ or A = a ₂ and details on how the supplied information is captured by the conditional probability P (EC | A) in the embedded instruction Are supplied by the external calculation EC. Also, if there are _four original states S ₁ , S ₂ , S ₃ , S ₄ , states S ₃ and S ₄ are assigned two additional attributes A = a ₁ and A = a ₂ by an embedded instruction. Allows you to specify how you can update and change to the four new states newS ₃ , newS ₄ , newS ₅ and newS ₆ . To perform this update, the built-in instructions can also use the previous probabilities P (A | S ₃ ) and P (A | S ₄ ). These previous probabilities can be derived based on a retrospective study by analyzing how much of the patients with state S ₃ or S ₄ simultaneously shake A = a ₁ or A = a _2. it can.

  Another way to derive previous probabilities is to seek out the clinician's views.

  By using the built-in instructions, a new state state probability estimate 801 can be derived from the following equation:

ただしｉは｛３，４｝にあり、ｊは｛１，２｝にあり、Ｐ（Ｓ_j）は、生理機能観測モジュール１２２の出力から導出された本来の患者状態確率である。

However, i is in {3, 4}, j is in {1, 2}, and P (S _j ) is the original patient state probability derived from the output of the physiological function observation module 122.

  FIG. 18 shows an additional example of an external calculation built-in instruction. Again, the risk-based monitoring system 100 is shown in FIG. 18 in both cases where the external calculation 9110 is generated on the same calculation medium as the patient-based monitoring system or as part of the external device. Perform the embedding as shown. The premise in this case is that the external calculation 9110 supplies direct information regarding a specific internal state variable estimated by the physiological function observation module 122 (or the enhanced physiological function observation module 9300). That's what it means. Therefore, in order to incorporate an external calculation, the physiological function observation module 122 can treat the external calculation 9110 as an additional measurement value, and can directly incorporate it into the observation model 221.

Visualization and user interaction [DB2]
FIG. 19 shows an example of the functions of the visualization and user interaction module 124. Specifically, the module 124 can receive all available patient information and data, including data from the data receiving module 121, the joint probability density function generated by the physiological function observation module 122, the clinical trajectory. The etiology tree estimated by the interpreter module 123, the risk, and the effectiveness of the open measurement are included. Using this information, the visualization and user interaction module 124 can generate:
1) Patient unit view 1501 describing patient risk, diagnosis, etc .;
2) View 1502 of the patient's electronic medical record including test results, prescribed medication, diagnosis, etc .;
3) View 1503 showing the patient's ongoing risk;
4) a patient risk trajectory, ie a view 1504 showing how the probability of a particular patient condition has evolved within a particular time frame;
5) View 1505 showing the effectiveness of the patient's measurement in the specific patient risk estimate;
6) A plot 1506 of the PDF against the estimated ISV of the patient representing the temporal evolution of the PDF against the ISV;
7) a view 1507 that allows navigating within the patient's etiology tree, i.e., allowing different levels of the tree to be visualized;
8) View 1508 showing the predicted risk of the patient;
9) A view 1509 that allows the clinician to view and set alarms based on patient risk;
10) View 1510 showing patient physiology monitoring data and its temporal evolution;
11) Any combination of the above.
In addition to these, the visualization and user interaction module 124 can also generate a patient tag view 1511, a patient condition summary 1512, reference and training material 1513, notes and tag settings 1514.

  FIG. 20 shows an example of a summary view 2000 that can transmit the risk profile for each patient in a specific hospital unit on a single screen. The risk profile represents what the cumulative probability of a patient at a particular hazard level is. This is calculated by summing the current probabilities of all states at a particular hazard level. For example, the total probability, hazard level, is represented by the height of four bars, each bar corresponding to a particular hazard level. In this particular embodiment, the hazard levels are green (diagonal hatching)-minimum risk, yellow (vertical hatching)-mild risk, orange (horizontal hatching)-medium risk, red ( Dot hatching)-a severe risk.

  FIG. 21 illustrates one possible approach for a view 2100 that represents a patient's ongoing risk. Each box with rounded corners corresponds to one specific risk. Colors correspond to green (diagonal hatching) hazard levels-minimum risk, yellow (vertical hatching)-light risk, orange (vertical hatching)-medium risk, red (dot hatching)-severe The height of the box corresponds to the probability of a particular patient condition. Risks are grouped based on their hazard level. This screen and the individual risks are updated in real time as new data becomes available.

  Still referring to FIG. 21, in addition to ongoing patient risk visualization, the system 100 can also provide information regarding the effectiveness of various invasive measurements in these risk determinations. In particular, the illustrated example provides the effectiveness of measuring open arterial pressure (ABP) and open central venous pressure (CVP). Effectiveness can be represented by a fill of bars 2110 and 2120, with maximum effectiveness corresponding to 6 filled bars. 6 bars that are filled can be displayed with color gradation, 1-dark green, 2-light green, 3-yellow, 4-red, 5-purple, 6-white or no fill, It can be. In this particular embodiment, the bar 2110 filled for ABP has all six colors, whereas two of the bars 2120 filled for CVP are each 1-dark. Green, 2-light green, and the remaining fill bar 2110 is 6-white or no fill.

  FIG. 22 depicts a view 2200 showing how a slider 2210 or other graphical element at the top of the patient view can be used to view a history of patient risk. In particular, according to this embodiment, slider 2210 is moved to show patient risk about 4 hours before the current time. This allows the clinician to view the continuous evolution of patient risks and compare those patient risks with the applied treatment or other external factors. In various embodiments, the slider 2210 can be moved in the user interface using a pointing device or command to specify a desired time zone, or a slider or other graphic if used with a touch panel display. It can be moved by touching and dragging the element.

  Referring now to both FIG. 21 and FIG. 22, the etiology tree is now used to show two states according to FIG. 21: state A2130: hypoxia with low cardiac output and state B2140: low Qp: Combined with hypoxia with Qs, they are represented by a single patient condition (hypoxia) 2220. According to this particular embodiment, this is used to fit text into the box of FIG. 22 that is smaller than FIG. As depicted in FIG. 23, the user can navigate the etiology tree in view 2300 by clicking on the synthesized patient state 2220 to display the constituent states 2130 and 2140. it can.

  FIG. 24 shows how the user can view the predicted risk for the patient within the same framework by moving the slider 2410 ahead of the current point in time. It is depicted in view 2400 [DB3].

  FIG. 25 depicts an interactive dialog box 2510 within the view 2500 through which the user can define conditions that set an alarm for a particular risk. To do this, the user selects a particular risk and sets an upper threshold and a lower threshold for the patient state probability assigned to that risk. As long as the patient state probability is between the upper threshold and the threshold, the alarm is not triggered. An alarm is activated when the patient condition exceeds a threshold. When the alarm is activated, the system 100 notifies a selected set of people or sends a notification to other clinical systems. Any of the modules 122-124 can store their individual threshold data ranges each time and can generate triggers in response to specific parameters.

  In FIG. 26, a view 2600 illustrates yet another possibility to visualize the patient risk trajectory, ie, the evolution of patient state probabilities associated with a particular risk. In this case, the user can select which time series of patient state probabilities they want to display and the system plots those probabilities on the time axis.

FIG. 27 shows how the system 100 can directly represent the probability density function of various internal state variables in the view 2700. In particular, in this example, the estimated oxygen carrying PDF is plotted on a graph 2710 as a function of time, with the likelihood increasing as the color becomes darker. Similarly, an estimated PDF of mixed venous oxygen saturation (SvO ₂ ) is plotted in graph 2720 while being compared to actual measurements (dark circles).

Information Sharing Between Users FIG. 28 shows an example of a definition interface 2810 for performing tagging in a view 2800. This interface allows the clinician to mark or tag specific time instances or specific time periods 2820 that represent points of interest or important points during the clinical course. Via dialog box 2830, the tag can be shared or sent to a designated recipient, or the tag can be included in a note or any other part of the user interface. The user can attach a unique comment or observation to the tag via the dialog box 2840, and further categorize the tag into categories based on the menu list 2850, eg, depending on the tag, medication dosage changes, interventions, A memo or the like related to monitoring or measurement of the device can be represented. Tags and their individual time series markings can be coded with colors to represent various attributes such as their categories. For example, a chronological green tag mark can represent a drug change, a red tag mark can represent an intervention, and a yellow tag mark can represent a period of increased concern. When a tag is set, the user can be prompted to define a time instance or period, tag category, tag annotation, and how the system should handle the tag. Furthermore, annotations can be suggested by utilizing natural language processing that converts the etiology of the condition to an annotation format.

  FIG. 29 depicts a news feed view 2910 in view 2900, which includes tags, notes, or information obtained from an external source, such as obtained from an electronic medical record (EMR). Blood sampling time etc. are included. News feed 2910 allows clinicians to view and fill in events, attention periods, interventions, notes, tags, etc. entered by other clinicians. At that time, the clinician may browse all news feeds or may sort them based on tag categories, hazard levels, and the like. In addition, the clinician can search for tags by keyword, intervention format, residence time, information source, and the like. Entries 2920-2926 in the newsfeed 2010 can be displayed in words or images such as risk profiles, whether in categories, tag sources, or patient overviews.

FIG. 30 depicts a condition summary view 3010 within view 3000 that allows a clinician to request a condition summary by selecting or clicking on a particular patient condition. In this case, the condition summary view 3010 can present a description of the patient condition to the clinician, where the system determines the definition in this window 3020 and the conclusion regarding the probability of this patient condition. It includes both information on whether it has arrived. As shown, this view 3010 can provide the likelihood and hazard level of the patient state, the definition of the patient state by ISV threshold, the likelihood for each attribute that defines the state, A natural language description of the contributed evidence and an evidence window 3030 can be provided, either by converting the ISV PDF into a quantitative text description or by showing the numerical information associated with the evidence as is. Done. As an example, FIG. 30 depicts a condition summary view 3010 for a patient condition “shock” resulting from low cardiac output. In this case, the condition summary view 3010 shows the possibility of a “shock due to low cardiac output” condition and the hazard level indicated by hatching with color or dots. In addition, the definition of shock (mixed venous oxygen saturation ≦ 45%) and low cardiac output (cardiac output <3.2 L / min / m ² ) are probable that each of these will be met (eg for shock 40%, 30% for CO low). The evidence window 3030 conveys information that leads to the evaluation of “shock due to low cardiac output”. According to this embodiment, the system has converted information about the etiology probabilities into text format, and in this particular case, the estimated probabilities are mainly pulse pressures below a nominal value representing a reduction in stroke volume ( (Systolic blood pressure−diastolic blood pressure).

  FIG. 31 illustrates in the view 3100 the ability of the user interface to capture and display reference material accessible through the Internet, stored in the system 100, or remotely accessible. The detailed view 3120 can be displayed by selecting the “detail” button 3110 in the condition summary view 3010, and reference information associated with the condition summary view 3010 is shown in this view. Reference information can include causes, interventions, general complications, anatomical structures, and related literature. In addition, this feature can be used as a training tool to familiarize clinicians with the management or treatment strategies of specific patient populations.

HLHS Stage 1 Example Next, it will be described how the system 100 and techniques according to the present invention can be applied to modeling the clinical course of a particular patient population. In this case, it is assumed that intensive treatment, that is, postoperative recovery after temporary relaxation in stage 1 of a patient with hypoplastic left heart syndrome is performed.

Hypoplastic left heart syndrome is a congenital heart disease that is manifested by underdeveloped left ventricle and left atrium. As a result, patients suffering from such symptoms do not have separate systemic and pulmonary blood flow, instead the right ventricle is responsible for pumping blood to both the whole body and the lungs. Yes. Therefore, in order to optimize hemodynamics during intensive care, it is necessary to manage the ratio of blood flow through the lungs (pulmonary blood flow Q _p ) and blood flow through the whole body (body blood flow Q _s ). The optimal hemodynamic state is obtained when the ratio of pulmonary blood flow to body blood flow, expressed as Q _p / Q _s , is 1, and an appropriate oxygen transport amount DO ₂ to the tissue is achieved. Is the case. To reach such an optimal state, the patient's physiology often goes through other less beneficial states, through proper identification of those states and the application of appropriate treatment strategies for each of those states. The quality of postoperative care is determined.

  Table 1 lists the state variables, variable descriptions, units, and variable types used in the HLHS physiology model after stage 1 temporary relaxation. As will be apparent to those skilled in the art, these variables include circulation, hemodynamics and oxygen exchange components in HLHS physiology, but additional changes such as breathing and metabolism may be made without changing the premise of the present invention. The model can be changed or expanded by various physiological function components.

  FIG. 32 shows a general-purpose Dynamic Bayesian Network DBN that can be employed to capture a physiology model of a patient in a temporary relaxed state of HLHS stage 1. The graphical model schematically depicted in FIG. 32 captures the causal and probability relationships between each variable in the model. According to this DBN, state variables are incorporated into three groups, that is, dynamic variables 3210, derived variables 3220, and observed variables 3230. The dynamic variable 3210 is a variable having a value that changes over time based on a dynamic probability model described below. The derived variable 3230 is an amount that depends on a dynamic variable with some functional relationship. These variables are calculated or derived from the latest dynamic variables on demand and are therefore dynamic in nature. Observation variable 3230 is a variable that is directly measured by one of the sensors connected to the system and the patient. The observation variable 3230 represents an actual dynamic state variable or derived state variable observed while receiving noise.

  FIG. 33 lists a plurality of formulas that can be used to model physiological dynamics in the HLHS stage 1. This model consists of four main probabilistic model types. The first model type is a stochastic feedback control model (Equations 1, 2, 7). These variables have reference values that maintain the body, but are disturbed by some random process and deviate from this reference value. The strength with which the body tries to maintain this variable is determined by the feedback constant. The second model type is a drift diffusion process (Equations 3 and 4). These variables are moved over time by a random white noise and drift rate process. The third model type is a simple random walk process (Equations 5, 6, 8). The last dynamic model type is a memoryless process model, in which the variable has no relation to the variable in the preceding period, but follows some defined variance that extends over the reasonable support points of the variable, i.e. the parameter It is simply a random variable (Equation 9, 10, 11) with a value that changes with each time instance according to the gamma distribution over the positive real line along with A and B. Except for Equations 9, 10, and 11, the operating noise of each dynamic model is independent white Gaussian noise.

  FIG. 34 shows formulas that can be used to extract the relationship between dynamic variables and derived variables in modeling. Some of these functional relationships are in fact true for general human physiology, but many are the result of parallel circulation physiology specific to the HLHS population. Equations 12-15 represent the relationship for the directly measured variables. Equations 16-18 represent the functional relationship of the most important variables in the management of post-surgical care for HLHS patients, particularly the functional relationship between cardiac output (CO) and pulmonary blood flow ratio (Qp: Qs). Represent. These variables cannot be measured as they are without complicated procedures.

Given these functional relationships and dynamic state definitions, FIG. 35 depicts one possibility of an observation model that can be used to correlate derived variables with available sensor data. Each observation model is a conditional Gaussian relationship. Measurements received from sensors in this model represent direct observations of hidden state variables that are corrupted by additional independent Gaussian white noise with some variance. In the drawing, the quantity observed as a hidden state variable is described with a wavy line on the variable name. According to this embodiment, different sensors can be mapped to the same hidden state variable, although they can possibly have different noise levels. For example, SpO ₂ supplied by a pulse oximeter measurement represents a potential hidden physiologic state, SaO ₂ or non-invasively represents arterial oxygen saturation. A venous catheter inserted directly into the arterial bloodstream also measures this amount, which is an invasive procedure. Catheter measurements should be more accurate than pulse oximeters. In this model, this is handled by a smaller measurement variance R.

  In HLHS physiological function observation, inference via DBN is performed using a particle filter. As already mentioned, a particle filter is an example of an approximate inference scheme that uses Monte Carlo sampling of internal state variables, whereby the probability density function of each state variable is approximated using an empirical distribution based on the number of particles. The This filter uses a process known as Sequential Importance Sampling SIS to continuously resample particles from the latest approximate probability distribution. In this filter, a weight is assigned to each particle. When a new observation value or measurement value arrives, the weight of each particle is updated based on the likelihood of the individual particle in the observation. The particles are then resampled based on their relative updated weights, where particles with the highest weight are more likely to be resampled than particles with a lower weight.

FIG. 36 depicts the attributes, patient status, and pathogenesis tree possibilities that can be utilized by the clinical trajectory interpreter module 123 in the HLHS stage 1 population. The variable of total cardiac output, defined as the sum of body blood flow and pulmonary blood flow, is used to define “low” and “normal” total cardiac output, and the Qp: Qs ratio is Qp : Qs ratio is used to derive “low”, “equilibrium”, “high”, and the value of hemoglobin concentration Hgb is used to derive “low” and “normal” of hemoglobin, and venous blood mixing The value of oxygen saturation SvO ₂ is used to derive the hemodynamic “shock” and “no shock” attributes. This results in eight possible states defined below:
1) When both the attribute “shocked” and the total cardiac output “low” exist, the shock caused by the total cardiac output “low”;
2) Shock caused by hemoglobin “low” when the attribute “shock”, total cardiac output “normal”, and hemoglobin “low” are involved.
3) When attribute “shocked”, total cardiac output “normal”, hemoglobin “normal”, circulatory “equilibrium”, shock due to unknown cause;
4) When attribute “shock”, total stroke volume “normal”, hemoglobin “normal”, Qp: Qs “low”, shock due to Qp: Qs “low”;
5) When attribute “shock”, total stroke volume “normal”, hemoglobin “normal”, Qp: Qs “high”, shock due to Qp: Qs “high”;
6) When the attribute is “no shock” and the circulation is “normal”, the circulation is “normal”;
7) Qp: Qs “low” when the state is defined by the attribute “no shock”, Qp: Qs “low”;
8) Qp: Qs “high” in a state defined by the attribute “no shock” and Qp: Qs “high”.
FIG. 35 also shows a feasible method of a pathogenesis tree representing the relationship between attributes and patient states. By using the particle approximation of the internal state variable to calculate the relative proportion of particles within each state, the probability of 8 states can be calculated.

Application examples Risk-based monitoring systems associated with assessment of outcomes of expected treatments Other application examples of risk-based monitoring systems are used to determine whether to apply specific treatments, for example, whether to apply blood transfusions. Is to help clinicians. Infusion of blood and blood products is generally a hospital procedure. Nevertheless, transfusion indicators and policies are not well established between hospitals and have not been applied consistently. Many studies have shown that transfusion practices are diverse in various hospitals, practitioners, and procedures. Such diversity exists even when applied to a single procedure (eg, coronary artery bypass grafting).

  Furthermore, blood transfusion has been increasingly recognized as an independent risk factor for morbidity and mortality. Specific events and outcomes associated with blood transfusions include sepsis, organ ischemia, increased respiratory support time, increased hospital stay, and short- and long-term morbidity. This relationship is proportional to the amount of blood transfused, and the suggestion is that high hematocrit values can be harmful. Of course, researchers routinely recommend a transfusion policy that achieves a risk-benefit trade-off based on information.

  Establishing a solid and effective transfusion policy proved to be a difficult task. The consensus among healthcare professionals is that simple policies such as guidelines based on hemoglobin thresholds are not appropriate indicators. This is due to the compensatory nature of hemodynamic physiology, and patients have a variable capacity to tolerate low hemoglobin. Therefore, in order to make an effective transfusion decision, factors such as compensatory reserve, intravascular volume, hemodynamic stability, treatment type, and other patient data need to be integrated. Thus, there is an indispensable requirement for a blood management policy, which utilizes the entire set of relevant clinical variables, which determines the risk-benefit ratio of transfusion. This is exactly achieved by applying a risk-based monitoring system.

  FIG. 37 illustrates one possibility of an environment in which a risk-based monitoring system can be employed to assist a clinician in determining whether a particular treatment should be applied. According to the present invention, the patient 101 is monitored intermittently and continuously by a plurality of measuring devices 3910. Listed as continuous measurements are mixed venous oxygen saturation (SvO2) 3911, systolic blood pressure, diastolic blood pressure and mean blood pressure (ABP s | d | m) 3913, heart rate 3916 Are monitored by a bedside monitor. Listed as intermittent measurements are blood pH 3915, hemoglobin concentration (Hgb) 3912, and blood lactate concentration 3914, which are monitored by periodic blood tests. These measurements 3910 are provided to a risk-based monitoring system 3940 that has been expanded by treatment assessment. This system is constituted by a plurality of other modules in addition to the physiological function observation module 122 and the clinical trajectory interpreter module 123 described above. The assumed treatment complication determination module 3942 receives information from the clinical trajectory interpreter module 123 along with information regarding patient demographics 3931 and treatment type 3932. Module 3942 uses this information to query the outcome database 3943 and returns to module 3942 information that there are various possible complications if: a) Patient Is in a specific patient state with a specific probability; b) the patient belongs to a specific demographic (age, gender, etc.); c) the patient is receiving a specific type of treatment; d) a) above , B), c). On the other hand, using a retrospective study 3991, a randomized clinical trial 3992, an institution-specific outcome 3993 determined based on previously collected patient data for a particular institution, and an outcome study 3990 derived from some combination of these requirements An outcome database 3943 can be constructed.

  If a possible complication is determined by the assumed treatment complication determination module 3942, the module 3942 feeds this information back to the extended visualization and user interaction module 3941. Extended visualization and user interaction module 3941 combines patient specific risk-based monitoring performed by physiology observation module 122 and clinical trajectory interpreter module 123 with assessment of possible complications. This provides the system with an excellent view, based on which the clinician 3920 can better recognize the risks and benefits of treatment, such as blood transfusions, and whether or not this treatment 3960 should be administered. Judgment can be made efficiently and effectively.

FIG. 38 provides a non-limiting example of a set of patient conditions associated with a transfusion that can be used to notify a transfusion decision. These conditions include hemoglobin dynamics (decrease / stable / increase) and hemodynamic compensation when blood oxygen carrying capacity decreases. In noncompensating patients, the blood flow self-regulatory mechanism cannot overcome the decline in blood oxygen carrying capacity, and the sign is the start of anaerobic metabolism. These seven states can be determined by three internal state variables: oxygen transport, hemoglobin, and hemoglobin production / loss rate. Specifically, if hemoglobin is higher than 13 mg / dL, no Hgb-related medical condition is present (4001). If Hgb is lower than 13 mg / dL, there will be 6 remaining states, which are defined by 5 different attributes. From the oxygen delivery ISV, the system can determine whether the patient is in a compensatory state, for example, this can be premised on 400 ml / min for an involuntary patient receiving ventilation. DO2 higher than / m ² represents a compensatory state, and lower than this value represents a patient who is not in a compensatory state. The other three attributes are derived from the Hgb rate ISV: stable 4013 and 4014 (rate is nearly zero), increase 4011 and 4012 (rate is positive), and decrease 4015 and 4016 (rate is negative).

Use of a risk-based monitoring system with a standardized clinical plan Yet another use of a risk-based monitoring system is in the case of using a standardized medical plan. FIG. 39 shows one possible embodiment for this application. Specifically, data from the clinical trajectory interpreter module 123 is supplied to the treatment inquiry module 4142. The treatment inquiry module 4142 makes an inquiry to the treatment plan database 4143 based on the determined patient risk. The treatment plan database 4143 designates mapping between patient risk and data. When the database 4143 feeds back any treatment plan, the extended visualization and user interaction module 4141 presents the plan to the clinician. Clinician 3920 can make a clinical decision 4190 for patient 101. The user's decision, and the situation where the decision was made (calculated patient risk, estimated ISV, and other patient data obtained at the time of the decision) are then recorded in the decision database 4144. The decision database 4144 can then be compared to patient outcomes and can be used to improve treatment plans.

  In FIG. 40, one application is shown for a risk-based monitoring system combined with a specific type of standardized clinical plan. In this particular example, a medical decision is considered as to whether the patient will be treated with nitric oxide. Nitric oxide is a pulmonary vasodilator and is used to treat high pulmonary vascular resistance and consequently pulmonary hypertension that can reduce cardiac output. In this example, the medical plan uses the risks calculated by the clinical trajectory interpreter module 123 and stratifies them into two categories, low risk and high risk (4230). If the risk is low (4201), the recommended decision is not treatment (4202), and if each patient is classified as high risk, the recommended decision is treatment (4203). The health care provider can make a decision whether to follow the recommendation (4250) or ignore the recommendation (4260). If a health care provider chooses to ignore treatment recommendations for high-risk patients, the health care provider needs to justify it (4220). Similarly, if a health care provider chooses treatment for a low-risk patient, the health care provider still needs to justify it (4210). Justifications 4210 and 4220 associated with patient outcomes can be utilized to further refine risk stratification 4230 and risk-based monitoring system 100.

  FIG. 41 shows an example of risk stratification that can be employed by the system in the context of nitric oxide therapy. Specifically, the patient may have four different states: State 1: CO low, PVR normal; State 2: CO low, PVR high; State 3: CO normal, PVR high; State 4 : CO normal, PVR normal. Patients at low risk can be defined as P (State 1) <10% and P (State 1) + P (State 2) <30%. Similarly, patients at high risk can be defined as P (State 1)> 10% and P (State 1) + P (State 2)> 30%.

Use of Clinical Risk Assessment System in Outpatient Care in Chronic Conditions According to yet another embodiment of the present invention, outpatient care can be followed by clinical trajectory. Because it is chronic patient care, it involves sporadic patient assessment based on intermittent visits, patient self-assessment, and observation by caregivers. The result is uncertainty in determining the patient's clinical course and the efficiency of the treatment strategies described. To achieve effective patient care, clinicians need to understand these uncertainties and must reduce them. The clinician has two major decisions at his discretion. In other words, 1) To improve clinician's understanding of clinical trajectories, visit scheduling, examination instructions, requests for self-assessment (or caregiver evaluation), and / or 2) further improvement of symptoms and possible side effects Formulating drug or drug metering changes to achieve good trade-offs. To inform this decision process, it is necessary to process available patient information, conveying the uncertainties of the clinical trajectory and its estimation, and the expected effects of different treatment strategies on the future progress of the clinical trajectory. Is done.

  As a non-limiting example of a risk-based monitoring system for clinical trajectory tracking of outpatients, here is a case of applying this system to outpatient care for attention deficit / hyperactivity disorder (ADHD) in pediatric patients I will consider it. FIG. 42 illustrates one possibility of a patient condition that can describe the clinical trajectory of an ADHD patient. These patient states are the same as those used by the Clinical global impression-improvement scale. 1) very marked deterioration; 2) significant deterioration; 3) deterioration; 4) unchanged; 5) slight improvement; 6) significant improvement; The patient state distribution (PSD) is a set of probabilities that a patient is in any of these seven states when all available information and observations are obtained.

  For assessment of patient status, clinicians can schedule workplace visits to investigate directly, or they can request a Vanderbilt diagnostic test from their family or teacher (this test is Depending on responder, teacher or parent). In FIG. 43, available patient evaluation modalities are listed as M1, M2, and M3. These models can be used to map test questions and answers to patient status. Both clinical and test-based assessments are accompanied by uncertainties that prevent accurate determination of which of the seven conditions a patient currently belongs to.

  As shown in FIG. 44, a dynamic model or patient progress from one state to another can be extracted by a dynamic Bayesian network (DBN). According to FIG. 44, the direction of the arc-shaped arrow represents a statistical dependency, that is, the connection from “patient state @ t1” 4601 to “M1” is a probability density function (PDF): P (M1 | patient State @ t1). Similarly, in the illustrated DBM, “patient state @ t2” 4602 (patient state at a specific time point t2) depends on “patient state @ t1” 4601 (patient state at time point t1 returned to before). It represents that. According to the idea of the present invention, this model can estimate the patient state even if some or all of the possible measurements are not obtained, i.e., lacking M1, as shown. The patient state can be estimated even at the time point t2 and at the time point t3 (“patient state @ t3” 4603) where all the measurement values are missing.

FIG. 45 shows an alternative embodiment with respect to two predictions about how patient status may change in a single month when changing drugs or dosages. . This prediction is based on a statistical model derived by the following method:
Step 1: Separating from the retrospective data the group of patients who have passed state A at any treatment time point and have undergone a treatment change (drug 1 dosage 1 → drug 2 dosage 2);
Step 2: For each patient, set the time when this particular event occurred to t0;
3) Step 3: For each patient, identify what the patient status is at time t0 + 1 month (1M) (or any desired time step unit);
4) Calculate the percentage of patients transitioning from state A to state i, where i represents all seven possible patient states;
5) Set percentage as transition probability under specific treatment change.

  FIG. 46 illustrates one possibility of a visualization embodiment and scenario that displays a patient's clinical trajectory and risk. This user interface represents that the patient is in an “unchangeable” state, which is established by three separate measurements: workplace visit, teacher-based Vanderbilt diagnosis, and parent-based Vanderbilt diagnosis. Yes. The solid line on the screen represents that a drug was prescribed to the patient in the first week of treatment (drug 1).

  FIG. 47 shows patient and patient trajectory assessments during the ninth week. At this point, the clinical risk assessment system determines a probability density function for each patient condition for the past six weeks. The measurements available at this point are the teacher and parent's Vanderbilt diagnosis. In the case of the illustrated example, since the probability of worsening of the patient condition is high, the clinician instructs the drug metering change, which is represented by a red dashed line. In addition, the user interface also shows side effects due to reports of patient headaches.

  FIG. 48 shows the follow-up evaluation based on the teacher and parent Vanderbilt diagnosis. According to this embodiment, the doctor determines the drug change indicated by the white line.

  FIG. 49 shows the resulting assessment based on all available measurements that are likely to produce significant improvements, namely workplace visits and parent and teacher assessments.

  In FIG. 50, it is certain that the patient will improve steadily, and another follow-up is shown when there is a steady improvement between the two assessments. Yes. Here, the PDF for the patient trajectory is continuously estimated based on the applied reasoning. However, the accuracy (PDF density) is even higher when measurements are present.

  FIG. 51 shows a follow-up assessment of a patient and patient trajectory when no measurement is present. In this case, uncertainty has increased due to lack of recent observations.

  FIG. 52 shows the state of uncertainty described above when a complete patient evaluation (all measurement modalities) has been performed. This uncertainty is backpropagated over time by the estimation engine to produce a more accurate patient trajectory estimate, thereby assisting the physician in deducing that the patient is in a steady state.

  FIG. 53 shows a further visualization possibility from the system output described above. This figure shows the transition of the patient state that can occur when changing the treatment plan, for example when changing the medication. In addition, it is possible to predict what side effects may occur. For all side effects, there are three expression stages: mild, moderate and severe, each represented by three boxes drawn next to the side effects in the figure. Each color is colored corresponding to a specific probability of occurrence for a specific side effect, and a darker color represents a higher probability.

  As described above, various examples and embodiments according to the present invention have been described in detail. However, it will be appreciated that these examples and embodiments of the present invention are given for illustrative purposes only and that one skilled in the art will depart from the disclosure scope of the present invention as defined in the appended claims. Without limitation, various improvements and modifications can be made to the embodiments disclosed herein.

Claims (28)

In a computer-implemented method for risk-based monitoring of a patient,
A) obtaining, by a computer, data associated with a plurality of internal state variables each representing a parameter that is physiologically related to one of patient treatment and condition;
B) storing the data associated with the acquired plurality of internal state variables in a storage device accessible by a computer;
C) generating a probability density function estimated for the plurality of internal state variables by a computer;
D) From the probability density function generated for the internal state variable, a first plurality of assumed patient states P (S ₁ ), P (S ₂ ), P (S ₃ ),. . . , P (S _n ) to which assumed patient state the patient can be classified at present is identified by the computer, and a probability value associated with each identified assumed patient state is generated by the computer Comprising the steps of:
A computer-implemented method for patient risk-based monitoring. The probability value associated with the identified assumed patient condition is 0-1.
The method of claim 1. E) further comprising the step of displaying via the user interface the probability value and the identified individual assumed patient condition associated with the probability value;
In this step, a combination of the identified assumed patient state and an individual probability value associated with the assumed patient state is defined as a clinical risk for the patient.
The method of claim 1. E) assigning a hazard level value associated with each of the identified assumed patient states;
F) further comprising representing the probability value and the hazard level value associated with each identified assumed patient state as a measure of risk to the patient.
The method of claim 1. Step C)
C1) generating an estimated probability density function for the first plurality of internal state variables at time step t _k ;
C2) From the probability density function generated in step C1) at time step t _k and the received measurements associated with each of the internal state variables, the plurality of internals at another time step t _{k + 1} Generating a probability density function for the state variable,
The method of claim 1. Each received measurement of each of the internal state variables is associated with the same time step,
The method of claim 5. The method of claim 5, wherein some but not all of the received measurements for each of the internal state variables are associated with the same time step. Step C)
C1) If a measurement value has been received in advance, a step of comparing a new received measurement value associated with one internal state variable with a predicted likelihood predetermined for a possible measurement value When,
C2) If the new received measurement value is not within the prediction likelihood predetermined for the measurement value that can be assumed for the associated internal state variable, the new received measurement value can be associated. And not incorporating into the estimated probability density function for the internal state variable,
The method of claim 1. Said step D)
D1) receiving external calculation data from the source as a probability value associated with a new attribute representing a parameter that is physiologically related to one of the patient's treatment and condition;
D2) From the probability density function generated for the internal state variable and the probability value associated with the new attribute, a plurality of second assumed patient states P (newS ₁ ), P (newS) ₂ ), P (newS ₃ ),. . . , P (newS _{n + m} ), the computer is identified as to which assumed patient state the patient can be classified at the present time, and the probability value associated with each identified assumed patient state is calculated by the computer. Including the step of generating by
The method of claim 1. Step C)
C1) generating a probability density function estimated for the first plurality of internal state variables at time step t _k ;
C2) receiving external calculation data associated with one specific internal state variable among the plurality of internal state variables from a source;
C3) The probability density function generated in step C1) at time step t _k , the received measurement value associated with each of the internal state variables, and one specific internal state among the plurality of internal state variables Generating a probability density function for the plurality of internal state variables at another time step t _{k + 1} from external computational data associated with the variable;
The method of claim 1. In a risk-based monitoring system for monitoring patients,
A processor;
A storage device connected to the processor;
A data receiving module connected to cooperate with a plurality of information sources associated with the patient;
A physiological function observation module that communicates with the data reception module;
A clinical trajectory interpreter module that communicates with the physiological function observation module;
The data receiving module acquires data associated with a plurality of internal state variables each representing a physiologically related parameter to one of a patient's treatment and condition,
The physiological function observation module generates a probability density function of the internal state variable;
The clinical trajectory interpreter module includes a first plurality of assumed patient states P (S ₁ ), P (S ₂ ), P (S ₃ ),. . . , P (S _n ) to which assumed patient state it is possible to classify the patient at this time, and a probability value associated with each identified assumed patient state is generated. ,
Risk-based monitoring system. A user interaction module is provided for communicating with the clinical trajectory interpreter module and the storage device, the user interaction module displaying the probability value and the identified individual assumed patient state associated with the probability value. Defining a combination of the identified assumed patient condition and the associated individual probability value as a clinical risk to the patient;
The system of claim 11. The physiological function observation module includes a dynamic model and an observation model stored in the storage device,
The system of claim 11. The physiological function observation module further includes an interface engine configured to cooperate with the dynamic model and the observation model stored in the storage device.
The system of claim 13. The physiological function observation module has an operation prediction mode,
Inside the said operating prediction mode, said probability density function estimated for the plurality of internal state variables at time step t _k, the supplied to the dynamic model, the plurality of different time step t _{k + 1} Generate an estimated probability density function for the state variables,
The system of claim 13. Some but not all of the received measurements of each of the internal state variables are associated with the same time step;
The system according to claim 14. The physiological function observation module compares a new received measurement value of one internal state variable with a predetermined predictive likelihood of an assumed measurement value if the measurement value is received in advance. If the new received measurement value is not within a predetermined predictive likelihood for a possible measurement value for the associated internal state variable, the new received measurement value is associated Do not incorporate into the estimated probability density function for the internal state variable,
The system of claim 11. The clinical trajectory interpreter module receives external calculation data as a probability value associated with a new attribute representing a physiologically related parameter to one of a patient's treatment and condition, and stores it in the internal state variable. A second plurality of assumed patient states P (newS ₁ ), P (newS ₂ ), P (newS ₃ ) from the probability density function generated for the new attribute and the probability value associated with the new attribute. ),. . . , P (newS _{n + m} ) to which assumed patient state the patient can be classified at present, and generates a probability value associated with each identified assumed patient state.
The system of claim 11. The physiological function observation module generates a probability density function estimated for the first plurality of internal state variables at a time step t _k ;
Corresponding to the probability density function generated in C1) at time step t _k , the received measurement value associated with each of the internal state variables, and one specific internal state variable among the plurality of internal state variables Generating a probability density function for the plurality of internal state variables at another time step t _{k + 1} from the calculated external calculation data;
The system of claim 11. In a computer program product comprising a non-transitory computer readable medium having executable instructions stored as computer program code,
A) computer program code for obtaining data associated with a plurality of internal state variables each representing a parameter that is physiologically related to one of patient treatment and condition;
B) computer program code for storing data associated with the plurality of acquired internal state variables;
C) computer program code for generating a probability density function estimated for the plurality of internal state variables;
D) From the probability density function generated for the internal state variable, it is identified to which assumed patient state among a plurality of assumed patient states that the patient can be classified at the present time, and the identified assumed patient Including computer program code for generating probability values associated with each state,
Computer program product. The probability value associated with the identified assumed patient condition is 0-1.
21. The computer program product of claim 20. E) further comprising computer program code for displaying the probability value and the identified individual assumed patient condition associated with the probability value;
In the computer program code, a combination of the identified assumed patient condition and an individual probability value associated with the assumed patient condition is defined as a clinical risk for the patient.
21. The computer program product of claim 20. E) computer program code for assigning a hazard level value associated with each identified assumed patient condition;
F) further comprising computer program code representing the probability value and the hazard level value associated with each identified assumed patient condition as a measure of risk to the patient;
21. The computer program product of claim 20. The computer program code C) further comprises
C1) computer program code for generating a probability density function estimated for the first plurality of internal state variables at time step t _k ;
C2) time and the probability density function generated by the computer program code C1) in step t _k, the color and the received measured value of the internal state variables respectively, said plurality of internal state variables in another time step t _{k + 1} Including computer program code for generating a probability density function for
21. The computer program product of claim 20. Some but not all of the received measurements of each of the internal state variables are associated with the same time step;
21. The computer program product of claim 20. In a computer-implemented method for risk-based monitoring of a patient,
A) Data associated with a plurality of internal state variables respectively representing parameters physiologically related to one of the patient's treatment and condition is acquired by a computer, provided that the data is associated with the plurality of internal state variables. Acquiring a part, not all, of the acquired data in the same cycle;
B) storing the data associated with the acquired plurality of internal state variables in a storage device accessible by a computer;
C) generating a probability density function estimated for the plurality of internal state variables by a computer;
D) From the probability density function generated for the internal state variable, a first plurality of assumed patient states P (S ₁ ), P (S ₂ ), P (S ₃ ),. . . , P (S _n ) to which assumed patient state it was possible to classify the patient at the preceding time point, the computer identifies the probability value associated with each identified preceding assumed patient state. A computer-generated step,
A computer-implemented method for patient risk-based monitoring. Step C)
C1) generating an estimated probability density function for the first plurality of internal state variables at the current time step t _k ;
C2) using the defined transition probability kernel, by progress in the opposite direction from the probability estimation in the time step t _k until another time step t _kN, the plurality of internal state variables at time step t _kN said another Generating a probability density function for N, where N is an integer greater than 1,
27. The method of claim 26. A second plurality of internal state variables each representing a physiologically relevant parameter in one of the patient's treatment and condition comprises acquired data associated with the second plurality of internal state variables; In step C),
C1) generating an estimated probability density function for said second plurality of internal state variables at time step t _k ;
C2) time and the probability density function generated in step C1) in step t _k, and a probability density function associated with the different internal state variables at time step t _k, in another time step t _{k + 1} Generating a probability density function for the second plurality of internal state variables.
The method of claim 1.

Images