WO2019025901A1 - Systèmes et procédés de prédiction de l'apparition d'une sepsie - Google Patents

Systèmes et procédés de prédiction de l'apparition d'une sepsie Download PDF

Info

Publication number
WO2019025901A1
WO2019025901A1 PCT/IB2018/055498 IB2018055498W WO2019025901A1 WO 2019025901 A1 WO2019025901 A1 WO 2019025901A1 IB 2018055498 W IB2018055498 W IB 2018055498W WO 2019025901 A1 WO2019025901 A1 WO 2019025901A1
Authority
WO
WIPO (PCT)
Prior art keywords
values
sepsis
vital sign
dynamic
time interval
Prior art date
Application number
PCT/IB2018/055498
Other languages
English (en)
Inventor
Pierre SINGER
Yehudit APERSTEIN
Jonathan Cohen
Eli Bloch
Tammy ROTEM
Original Assignee
Mor Research Applications Ltd.
Afeka Yissumim Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Mor Research Applications Ltd., Afeka Yissumim Ltd filed Critical Mor Research Applications Ltd.
Publication of WO2019025901A1 publication Critical patent/WO2019025901A1/fr

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271—Specific aspects of physiological measurement analysis
    • A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/01—Measuring temperature of body parts ; Diagnostic temperature sensing, e.g. for malignant or inflamed tissue
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A61B5/024—Detecting, measuring or recording pulse rate or heart rate
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/08—Detecting, measuring or recording devices for evaluating the respiratory organs
    • A61B5/0816—Measuring devices for examining respiratory frequency

Definitions

  • the present invention in some embodiments thereof, relates to sepsis diagnosis and, more specifically, but not exclusively, to systems and methods of predicting sepsis onset.
  • the sepsis syndrome occurs when chemicals released into the bloodstream to fight an infection trigger inflammatory responses through the body. This condition may progress to severe sepsis with the presence of multiple organ dysfunction and further to septic shock in which sepsis is additionally complicated by a profound decrease in systemic blood pressure. Both these latter conditions are associated with significant morbidity and mortality and sepsis remains the most expensive condition treated in the hospital. Timely intervention with appropriate antibiotic administration and hemodynamic optimization has been shown to improve outcomes and decrease costs. This in turn requires early diagnosis which is dependent on the vigilance of the treating personnel responsible to identify symptoms heralding the onset of the syndrome. However, many demands are made on the staff of busy intensive care units, where these patients are typically treated, so that overlooking sepsis symptoms and/or delays in the administration of life-saving treatments may invariably occur.
  • a computer implemented method of providing a client terminal with a prediction of a likelihood of impending onset of sepsis based on measured values of different types of vital sign measurements of a target individual obtained over a recent monitoring time interval comprises: executing a code by at least one hardware processor of a computing device for: receiving via a network, a set of values for each of a plurality of different types of vital sign measurements each outputted by a respective vital sign sensor over a recent monitoring time interval, wherein each values of each set of values is associated with a different time stamp, each value of each set being associated to a respective different type of vital sign, accessing a plurality of dynamic features, wherein each dynamic feature is a combination of a certain dynamic function selected from a plurality of different dynamic functions and applied on a certain vital sign measurement selected from the plurality of different types of vital sign measurements, each one of the plurality of dynamic functions is based on local variability of values measured over the recent monitoring time interval, computing a plurality of extracted dynamic features
  • a system for providing a client terminal with a prediction of a likelihood of impending onset of sepsis based on measured values of different types of vital sign measurements of a target individual obtained over a recent monitoring time interval comprises: a non-transitory memory having stored thereon a code for execution by at least one hardware processor of a computing device, comprising: code for receiving via a network, a set of values for each of a plurality of different types of vital sign measurements each outputted by a respective vital sign sensor over a recent monitoring time interval, wherein each values of each set of values is associated with a different time stamp, each value of each set being associated to a respective different type of vital sign, code for accessing a plurality of dynamic features, wherein each dynamic feature is a combination of a certain dynamic function selected from a plurality of different dynamic functions and applied on a certain vital sign measurement selected from the plurality of different types of vital sign measurements, each one of the plurality of dynamic functions is based on local variability of values measured
  • a computer implemented method of generating a classifier for predicting a likelihood of impending sepsis onset for a target individual based on sets of different types of vital sign measurements of a target individual obtained over a recent monitoring time interval comprises: executing a code by at least one hardware processor of a computing device for: receiving via a network, for each of a plurality of sampled individuals, a set of values for each of a plurality of different types of vital sign measurements each outputted by a respective vital sign sensor over a recent monitoring time interval, wherein each values of each set of values is associated with a different time stamp, each value of each set being associated to a respective different type of vital sign, associating each of the sampled individuals with a label indicative of a diagnosis of sepsis or no sepsis obtained within a upcoming time interval consecutively following the end of the recent monitoring time interval, computing, by the computing device, for each of the sampled individuals, a plurality of extracted dynamic features according to a pluralit
  • a system for generating a classifier for predicting a likelihood of impending sepsis onset for a target individual based on sets of different types of vital sign measurements of a target individual obtained over a recent monitoring time interval comprises: a non-transitory memory having stored thereon a code for execution by at least one hardware processor of a computing device, comprising: code for receiving via a network, for each of a plurality of sampled individuals, a set of values for each of a plurality of different types of vital sign measurements each outputted by a respective vital sign sensor over a recent monitoring time interval, wherein each values of each set of values is associated with a different time stamp, each value of each set being associated to a respective different type of vital sign, code for associating each of the sampled individuals with a label indicative of a diagnosis of sepsis or no sepsis obtained within a upcoming time interval consecutively following the end of the recent monitoring time interval, code for computing, by the computing device, for each of
  • the systems and/or methods and/or code instructions described herein provide a technical solution to the technical problem of predicting onset of sepsis in a target individual and/or improving accuracy of prediction of the onset of sepsis.
  • the technical problem may relate to predicting onset of sepsis in a target individual based on routinely collected vital sign measurement data, for example, outputs of vital sign sensors, and/or vital sign measurements stored in an electronic medical record (EMR).
  • EMR electronic medical record
  • vital sign measurements that are collected as part of routine monitoring of the patient without requiring taking blood samples to a lab and conduct time-consuming analysis.
  • the systems and/or methods and/or code instructions described herein may relate to computer-related technology, namely, performance of a computing device and/or system for classifying values of vital sign measurements for prediction of sepsis onset.
  • a special purpose computer is designed rather than relating to an existing system.
  • Exemplary computational efficiency is based on one or more of:
  • Additional collection of vital sign measurements executively for prediction of sepsis and/or additional monitoring executively for prediction of sepsis is not necessarily required.
  • vital sign measurements of patients admitted to the ICU are routinely collected and stored in the patient's EMR record.
  • the existing collected EMR data may be processed for prediction of sepsis onset, without necessarily requiring collection of additional data.
  • the dynamic features described herein may be extracted directly from the raw vital sign measurements outputted by the vital sign sensors without significant additional processing. Additional pre-processing, for example, conversion of the data into a defined format, and/or aggregation computations on the data is not necessarily required. It is noted that some pre-processing may be required.
  • the dynamic features may be extracted from the vital sign measurements with relatively reduced processing requirements based on the methods described herein. Processing may be performed dynamically, and/or using existing computational hardware. The modification of the collected data to obtain data analyzable for the prediction of SEPSIS onset is not required. For instance, there is no need to identify and discard statistical outliers.
  • At least one dynamic function of the plurality of dynamic functions performs a computation according to a plurality of local minimum and maximum points identified for the respective set of values.
  • At least one dynamic function of the plurality of dynamic function performs a computation according to at least one of: an amount of local changes of the respective set of values, and magnitude of local changes of the respective set of values.
  • the certain dynamic function is selected from the group comprising: number of trend changes in the respective set of values, mean intensity of change in the respective set of values, median intensity of change in the respective set of values, minimum intensity change, maximum intensity change or any combination of the aforesaid.
  • the number of trend changes are computed according to the number of local minimum and maximum points of the certain set of values, wherein the mean intensity of change in the certain set of values is computed according to the mean magnitude of change between values of local minimum and maximum points, wherein the median intensity of change in the certain set of values is computed according to value in which the lower 50% of magnitude changes between values of local minimum and maximum points stop when sorted, minimum value of the magnitude of change between values of local minimum and maximum points for the certain set of values, maximum value of the magnitude of change between values of local minimum and maximum points for the certain set of values.
  • the plurality of different types of vital sign measurements exclude laboratory based investigations of biological samples of the target patient.
  • the certain vital sign measurements of the plurality of dynamic features are selected from the group comprising: blood pressure measured by a blood pressure sensor, heart rate measured by a heart rate sensor, body temperature measured by a temperature sensor, respiratory rate computed from output of a respiratory rate sensor or any combination of the aforesaid.
  • the plurality of dynamic features include: number of trend changes in respiratory rate measurements, number of trend changes in arterial pressure measurements, the minimal change in respiratory rate measurements and the median change in heart rate measurements.
  • the classifier is selected from the group comprising: logistic regression, linear support vector machine, support vector machine with radial kernel, support vector machine with polynomial kernel, artificial neural network or any combination of the aforesaid.
  • the upcoming time interval is about 4 hours starting consecutively from the end of the recent monitoring time interval.
  • the recent monitoring time interval during which the sets of values are obtained is about 8 hours.
  • the classifiers computes a likelihood of impending onset of sepsis as a binary classification into a first category indicative of statistically significant risk of onset of sepsis during the upcoming time interval or into a second category indicative of no statistical significant risk of sepsis onset during the upcoming time interval.
  • the method further comprises and/or the system further comprises code instructions for accessing a database mapping recommended sepsis treatments to prediction of sepsis onset, and outputting the recommended sepsis treatment corresponding to the computed prediction of sepsis onset in the target individual for presentation on the display.
  • the method further comprises and/or the system further comprises code instructions for generating an alert when the classifier computes a prediction of impending sepsis onset and transmitting the alert to a mobile device.
  • the set of values for each of the plurality of different types of vital sign measurements are obtained from a record of the target patient stored in a medical database, and the receiving, the computing values, and the computing are dynamically iterated based on new sets of values in the record of the target patient of the medical database using a sliding window of a size corresponding to the recent time monitoring interval.
  • the method further comprises and/or the system further comprises code instructions for providing the prediction of likelihood of impending sepsis onset to at least one automated treatment device for automated treatment of the patient according to the prediction of impending sepsis onset.
  • the at least one automated treatment device is selected from the group comprising: an automated antibiotic delivery mechanism that automates injection of antibiotics by, an intravenous (IV) flow rate mechanism that automatically adjusts IV fluid infusion rate to a target rate, an automated mechanical ventilator that automated increases the percent of oxygen being administered to a target percentage, an automated drug delivery mechanism that automatically administers vasopressors or any combination of the aforesaid.
  • an automated antibiotic delivery mechanism that automates injection of antibiotics by, an intravenous (IV) flow rate mechanism that automatically adjusts IV fluid infusion rate to a target rate
  • an automated mechanical ventilator that automated increases the percent of oxygen being administered to a target percentage
  • an automated drug delivery mechanism that automatically administers vasopressors or any combination of the aforesaid.
  • the method further comprises and/or the system further comprises code instructions for outputting the prediction of impending sepsis onset in the target individual.
  • sampled individuals who exhibited 2 of 4 SIRS criteria over the consecutive time interval are associated with a label indicative of suspicion of sepsis
  • sampled individuals who are associated with a documented infection over the consecutive time interval are associated with a label indicative of sepsis
  • sampled individuals who were not diagnosed with an infection during the consecutive time interval are associated with a label indicative of no sepsis.
  • the method further comprises and/or the system further comprises code instructions for selecting a subset of the plurality of dynamic features according to a separation requirement between a sub-population of the sampled individuals denoted as sepsis and another sub-population of the sampled individuals denoted as control, and generating the classifier according to the subset of the plurality of dynamic features.
  • generating comprising generating a plurality of classifier types, and further comprising evaluating the predictive performance of each of the plurality of classifier types, and selecting one of the plurality of classifier types according to the evaluated predictive performance.
  • FIG. 1 is a flowchart of a method of predicting impending sepsis onset by a classifier applied to dynamic features each a combination of a certain dynamic function and a certain types of vital sign measurement of a target individual, in accordance with some embodiments;
  • FIG. 2 is a flowchart of a method of generating a classifier for computing a prediction of impending onset of sepsis by classifying extracted dynamic feature values computed based on dynamic features each a combination of a certain dynamic function and a certain types of vital sign measurement, in accordance with some;
  • FIG. 3 is a block diagram of components of a system for predicting impending sepsis onset by a classifier and optionally for generating the classifier, in accordance with some embodiments;
  • FIG. 4 is a schematic depicting time intervals for prediction likelihood of sepsis onset, in accordance with some embodiments;
  • FIG. 5 is a flowchart depicting a process of selection of patients for inclusion in a study, in accordance with some embodiments;
  • FIG. 6 is a table comparing the target group of the study that developed sepsis and the control group of the study that did not develop sepsis, in accordance with some embodiments;
  • FIG. 7 is a table summarizing data depicting separation of the septic and control patient populations of the study based on all extracted combinations, in accordance with some embodiments;
  • FIG. 8 is a graph of a patient of the study that developed sepsis and another graph of another patient of the study that did not develop sepsis, demonstrating the behavior of Mean Arterial Pressure (MAP) during the first 8 hours of the 12 hour period of the study prior to sepsis time-point; and
  • MAP Mean Arterial Pressure
  • FIG. 9 is a graph of combinations of dynamic functions and vital sign types ranked according to an importance scale, for selection of the top four combinations for prediction of sepsis onset, in accordance with some embodiments;
  • FIG. 10 includes graphs that visually depict the distinction between the sepsis and non-sepsis population for the selected 4 combinations of dynamic functions and vital sign measurement, in accordance with some embodiments;
  • FIG. 11 is a table summarizing the performance results of each classifier type in classifying a testing set, which was not a part of the training set, in accordance with some embodiments.
  • FIG. 12 is a graph depicting the receiver operating characteristic curve (ROC) plots of the 5 generated and tested classifier types, in accordance with some embodiments.
  • ROC receiver operating characteristic curve
  • the present invention in some embodiments thereof, relates to sepsis diagnosis and, more specifically, but not exclusively, to systems and methods of predicting sepsis onset.
  • aspects of disclosed embodiments relate to systems, apparatuses, methods, and/or code instructions (e.g., stored in a data storage device executable by one or more hardware processors) for predicting a likelihood of impending onset of sepsis based on sets of values of different types of vital sign measurements of a target individual obtained over a recent monitoring time interval.
  • the onset of sepsis is predicted for an upcoming time interval consecutive to the end of the recent monitoring time interval.
  • Each set of values is outputted by (and/or computed based on the output of) a respective vital sign sensor that performs the measurement of the respective vital sign type.
  • Each value of each set of values is associated with a different time stamp.
  • Extracted dynamic features are computed according to defined dynamic features.
  • Each dynamic feature defines a combination of a certain dynamic function (from multiple available dynamic functions) and a certain type of vital sign measurement (from multiple available types of vital sign measurements).
  • Each dynamic function is based on local variability of the values measured over the recent monitoring time interval.
  • Each extracted dynamic feature is computed according to the respective defined dynamic feature, by applying the certain dynamic function to the respective set of values of the certain vital sign measurement, according to the combination defined by the certain dynamic feature.
  • a prediction of a likelihood of impending sepsis onset in the target individual within an upcoming time interval consecutive to the recent monitoring time interval is computed by a classifier classifying the computed extracted dynamic features.
  • the classifier may classify the extracted dynamic features into a binary classification indicative of statistically significant risk (i.e., high risk) of sepsis onset and no statistically significant risk (i.e., no or low risk) of sepsis onset.
  • the classifier classifies the extracted dynamic features into a statistically significant indication that there is no or low risk to develop sepsis.
  • Statistically significant may be qualified by a criterion such as a threshold value, which may be predefined, for example, a threshold of 60%, 75%, or other probability values.
  • the local variability (e.g., local min/max values) are calculated for the time span of the recent monitoring period, for example, the last 4, 6, 8, 12, 16 hours, or other time spans.
  • Summary values e.g., average
  • the length of the time span and/or the length of the monitoring period and/or the total number of time spans are selected to provide a minimum number of summary values for making a statistically significant prediction.
  • the dynamic features are indicative of the amount of instability of the respective vital sign during the recent time interval.
  • Each dynamic feature may be computed based on local variability of the certain set of values over the recent monitoring time interval.
  • Each dynamic feature may be computed based on local minimum and maximum values of the certain set of values over the recent monitoring time intervals.
  • the dynamic features may be computed based on the amount of local changes and/or the magnitude of the local changes.
  • Prediction of impending onset of sepsis, before the patient actually develops sepsis, allows a healthcare provider (e.g., attending physician) to initiate early antibiotic treatment, which may improve patient outcomes.
  • a healthcare provider e.g., attending physician
  • Prediction of impending onset of sepsis improves healthcare outcomes, for example, a shortened ICU admission.
  • An aspect of some embodiments of the present invention relates to systems, an apparatus, methods, and/or code instructions (stored in a data storage device executable by one or more hardware processors) for computing a classifier for predicting impending onset of sepsis based on a set of values of different types of vital sign measurements of a target individual obtained over a recent time interval. The onset of sepsis is predicted for an upcoming time interval consecutive to the end of the recent time interval.
  • Sets of values of different types of vital sign measurements are collected from sampled individuals. Each set of values is outputted by (and/or computed based on the output of) a respective vital sign sensor that performs the measurement of the respective vital sign.
  • Each set of values is per vital sign of all measured vital signs of each sampled individual.
  • Each value of each set of values is associated with a different time stamp.
  • Each sampled individual is then associated with a label indicative of whether the respective sampled individual developed sepsis during the upcoming time interval, or did not develop sepsis during the upcoming time interval.
  • Extracted dynamic features i.e., values
  • Each dynamic feature defines a combination of a certain dynamic function (from multiple available dynamic functions) and a certain type of vital sign measurement (from multiple available types of vital sign measurements).
  • Each dynamic function is based on local variability of the values measured over the recent monitoring time interval.
  • Each extracted dynamic feature is computed according to the respective defined dynamic feature, by applying the certain dynamic function to the respective set of values of the certain vital sign measurement, according to the combination defined by the certain dynamic feature.
  • One or more classifiers are trained for predicting the onset of sepsis within another upcoming time interval based on the extracted dynamic features and the corresponding label.
  • the classifier may classify the extracted dynamic features, for example, into a binary classification indicative of statistically significant risk of sepsis onset and no statistically significant risk of sepsis onset, or for example into a statistically significant indication that there is no/low risk to develop sepsis.
  • the systems and/or methods and/or code instructions described herein provide a technical solution to the technical problem of predicting onset of sepsis in a target individual and/or improving accuracy of prediction of the onset of sepsis.
  • the technical problem may relate to predicting onset of sepsis in a target individual based on routinely collected vital sign measurement data, for example, outputs of vital sign sensors, and/or vital sign measurements stored in an electronic medical record (EMR).
  • EMR electronic medical record
  • vital sign measurements that are collected as part of routine monitoring of the patient without requiring taking blood samples to a lab and conduct time-consuming analysis.
  • the systems and/or methods and/or code instructions described herein may relate to computer-related technology, namely, performance of a computing device and/or system for classifying values of vital sign measurements for prediction of sepsis onset.
  • a special purpose computer is designed rather than relating to an existing system. Exemplary computational efficiency is based on one or more of:
  • Additional collection of vital sign measurements executively for prediction of sepsis and/or additional monitoring executively for prediction of sepsis is not necessarily required.
  • vital sign measurements of patients admitted to the ICU are routinely collected and stored in the patient's EMR record.
  • the existing collected EMR data may be processed for prediction of sepsis onset, without necessarily requiring collection of additional data.
  • the dynamic features described herein may be extracted directly from the raw vital sign measurements outputted by the vital sign sensors without significant additional processing. Additional pre-processing, for example, conversion of the data into a defined format, and/or aggregation computations on the data is not necessarily required. It is noted that some pre-processing may be required.
  • the dynamic features may be extracted from the vital sign measurements with relatively reduced processing requirements based on the methods described herein. Processing may be performed dynamically, and/or using existing computational hardware. The modification of the collected data to obtain data analyzable for the prediction of SEPSIS onset is not required. For instance, there is no need to identify and discard statistical outliers.
  • the systems and/or methods and/or code instructions described herein, for example, may relate to computer-related technology, namely, performance of a computing device and/or system for automated administration of treatments.
  • the prediction of impending onset of sepsis may trigger automated administration of treatment of the patient by one or more automated treatment delivery devices.
  • automated injection of antibiotics by an automated antibiotic delivery mechanism automated adjustment of intravenous fluid infusion rate to a target rate by an IV flow rate mechanism, automated increase in percent of oxygen being administered to a target percentage by an automated mechanical ventilator, and/or automated administration of vasopressors by an automated drug delivery mechanism.
  • the automated administration can be performed dynamically (forcefully changing parameter values over time according to a predetermined protocol), adaptively (e.g., depending on the measured vital signs etc.), or according to a constant rate of delivery.
  • the systems and/or methods and/or code instructions described herein generate new data in the form of the values computed according to combinations of dynamic features and relating to vital sign measurement types.
  • the values computed according to the combinations relate to performance of the computing device in predicting onset of sepsis based on computationally efficient (e.g., in terms of processing time, processor utilization, and/or memory requirements) extraction of the dynamic features from the values of the vital sign measurements.
  • the dynamic features increase accuracy of the onset of sepsis predicted by the computing device.
  • the systems and/or methods and/or code instructions described herein generate new data in the form of the classifier that predicts sepsis onset based on values computed according to combinations of dynamic features and pertaining to sets of values of different types of vital sign measurements.
  • the systems and/or methods and/or code instructions described herein relate to an underlying process within the technical field of machine learning, in particular, within the field of automated prediction by a trained classifier.
  • the systems and/or methods and/or code instructions described herein do not simply describe prediction of sepsis onset using a mathematical operation and receiving and storing data, but combine the acts of computing extracted dynamic features according to defined dynamic features, wherein each dynamic feature is defined as a combination of a certain dynamic function and a certain type of vital measurement outputted by respective vital sign sensor, applying a classifier to the extracted dynamic features, and outputting the prediction of sepsis onset for presentation on a display of a client terminal.
  • the systems and/or methods and/or code instructions stored in a storage device executed by one or more processors described here go beyond the mere concept of simply retrieving and combining data using a computer.
  • the systems and/or methods and/or code instructions described herein are tied to physical real-life components, including one or more of: vital sign sensors, network equipment, physical user interfaces (e.g., display), a data storage device storing patient data, and a hardware processor(s) that execute code instructions.
  • the present invention may be a system, a method, and/or a computer program product.
  • the computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.
  • the computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device.
  • the computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
  • a non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD- ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, and any suitable combination of the foregoing.
  • RAM random access memory
  • ROM read-only memory
  • EPROM or Flash memory erasable programmable read-only memory
  • SRAM static random access memory
  • CD- ROM compact disc read-only memory
  • DVD digital versatile disk
  • memory stick a floppy disk, and any suitable combination of the foregoing.
  • Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network.
  • the network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers.
  • a network adapter card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing/processing device.
  • Computer readable program instructions for carrying out operations of the present invention may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages.
  • the computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.
  • the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).
  • electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present invention.
  • These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
  • These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function/act specified in the flowchart and/or block diagram block or blocks.
  • the computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
  • each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s).
  • the functions noted in the block may occur out of the order noted in the figures.
  • two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.
  • sepsis onset means a new diagnosis of sepsis in a patient that has not been diagnosed with sepsis.
  • impending sepsis means a future diagnosis of sepsis during an upcoming consecutive time interval.
  • impending sepsis and sepsis onset are sometimes interchangeable, and/or may sometime be combined, for example, impending onset of sepsis means a future diagnosis of sepsis during an upcoming consecutive time interval, in a patient that is not currently diagnosed with sepsis during the monitoring time interval during which vital sign measurements are collected.
  • Known methods require laboratory test results, which is in contrast to the systems, methods, and/or code instructions described herein, that predict onset of sepsis from (optionally standard) vital sign measurements collected based on routine patient monitoring without requiring additional sensor data and/or laboratory investigations.
  • such methods are based on an aggregation of the entire dataset, for example, a measure of central tendency (e.g., average) and/or an analysis of the corpus of the dataset (e.g., maximum and/or minimum value in the dataset, standard deviation of the values in the dataset, overall trend of the data) in contrast to the systems, methods, and/or code instructions described herein, that extract dynamic features based on local variability of the values.
  • FIG. 1 is a flowchart of a method of predicting impending sepsis onset by a classifier applied to dynamic features each a combination of a certain dynamic function and a certain types of vital sign measurement, in accordance with some embodiments of the present invention.
  • FIG. 2 is a flowchart of a method of generating a classifier for computing a prediction of impending onset of sepsis by classifying extracted dynamic feature values computed based on dynamic features each a combination of a certain dynamic function and a certain types of vital sign measurement, in accordance with some embodiments.
  • FIG. 3 is a block diagram of components of a system 300 for predicting impending sepsis onset by a classifier and optionally for generating the classifier, in accordance with some embodiments.
  • System 300 may implement the acts of the method described with reference to FIG. 1 and/or FIG. 2, by processor(s) 302 of a computing device 304 executing code instructions stored in a storage device (e.g., memory) 306.
  • a storage device e.g., memory
  • one or more features stored as code instructions may be implemented as circuitry and/or hardware and/or firmware.
  • Computing device 304 receives one or more set of values of different types of vital sign measurements of a target individual and/or of multiple sampling individuals, for example, from a client terminal 308 that collects the values from the patient (e.g., bedside monitoring unit) via vital sign sensor 311, from a storage device and/or server 310 in communication with a medical database 310A (e.g., including electronic medical records (EMR)) storing the values measured by vital sign sensor 311, and/or directly from output of the vital sign sensors 311.
  • a client terminal 308 that collects the values from the patient (e.g., bedside monitoring unit) via vital sign sensor 311, from a storage device and/or server 310 in communication with a medical database 310A (e.g., including electronic medical records (EMR)) storing the values measured by vital sign sensor 311, and/or directly from output of the vital sign sensors 311.
  • EMR electronic medical records
  • the set(s) of values of vital sign measurements may be provided to computing device 304 over a network 312 via a data interface 314.
  • Exemplary data interfaces 314 include, for example, a network interface card, an internal data bus, a wire connection, a wireless connection, other physical interface implementations, and/or virtual interfaces (e.g., software interface, application programming interface (API), software development kit (SDK)), a physical interface for connecting to a cable for network connectivity, network communication software providing higher layers of network connectivity, and/or other implementations.
  • API application programming interface
  • SDK software development kit
  • Network 312 may include, for example one or more of: the internet, a local area network, a wireless network, a cellular network, a virtual private network, and a point to point connection with another computing device.
  • Computing device 304 may be integrated, for example, with an existing server (e.g., medical database server storing EMR 310A), with a bedside patient monitoring unit, with an ICU client terminal acting as a monitoring station.
  • Computing device 304 may be implemented as a separate standalone unit, as code instructions stored in a data storage device of another computing device (e.g., server, client terminal), executed by processor(s) of the other computing device, and/or as a hardware unit that is integrated with the other computing device (e.g., hardware card or chip that is plugged into the hardware of the other computing device).
  • computing device 304 may be implemented as, for example, a client terminal, a server, a computing cloud, a mobile device, a desktop computer, a thin client, a Smartphone, a Tablet computer, a laptop computer, a wearable computer, glasses computer, and a watch computer.
  • Client terminal(s) 308 may receive the computed prediction of onset of sepsis, and/or may receive alerts from computing device 304, for example, indicative of a prediction of sepsis onset in the target patient, as described herein.
  • Exemplary client terminals 308 include a server, a computing cloud, a mobile device, a desktop computer, a thin client, a Smartphone, a Tablet computer, a laptop computer, a wearable computer, glasses computer, and a watch computer.
  • Processor(s) 302 of computing device 304 may be implemented, for example, as a central processing unit(s) (CPU), a graphics processing unit(s) (GPU), field programmable gate array(s) (FPGA), digital signal processor(s) (DSP), and application specific integrated circuit(s) (ASIC).
  • processors 302 may include one or more processors (homogenous or heterogeneous), which may be arranged for parallel processing, as clusters and/or as one or more multi core processing units.
  • a non-transitory storage device e.g., a memory
  • 306 stores code instructions implementable by processor(s) 302, for example, a random access memory (RAM), read-only memory (ROM), and/or a storage device, for example, non-volatile memory, magnetic media, semiconductor memory devices, hard drive, removable storage, and optical media (e.g., DVD, CD-ROM).
  • Storage device 306 stores classifier training code 306A that executes one or more acts of the method described with reference to FIG. 2 for generating the classifier, and/or sepsis prediction code 306B that executes one or more acts of the method described with reference to FIG. 1 that includes applying the classifier to compute the prediction of onset of sepsis.
  • Computing device 304 may include a data repository 316 for storing data, for example, a classifier repository 316A that stores one or more trained classifiers, and/or a feature repository 316B that stores the extracted dynamic features computed from the set of values of vital sign measurements. It is noted that the classifier and/or features may be stored in storage device 306 when being processed by processor(s) 302.
  • Data repository 316 may be implemented as, for example, a memory, a local hard-drive, a removable storage unit, an optical disk, a storage device, and/or as a remote server and/or computing cloud (e.g., accessed via a network connection).
  • Computing device 304 and/or client terminal(s) 308 include and/or are in communication with a user interface 318 that includes a mechanism for a user to enter data (e.g., configure monitoring of the prediction of sepsis onset for a patient) and/or view presented data (e.g., the computed prediction of sepsis onset).
  • exemplary user interfaces 318 include, for example, one or more of, a touchscreen, a display, a keyboard, a mouse, and voice activated software using speakers and microphone.
  • Computing device 304 may be in communication with one or more automated treatment administration devices 350 (e.g., via network 312, directly in communication with, and/or integrated therein using code instructions and/or hardware components), for example, a controller that controls and/or performs one or more of: automated delivery of antibiotics to the patient (e.g., injection, activation of intravenous lines), automated delivery of vasopressor(s) medications to the patient (e.g., injection, control of intravenous flow), automated delivery of intravenous fluids (e.g., normal saline) to the patient (e.g., adjusted IV flow rate), and/or automated adjustment of percentage of oxygen being administered to the patient, for example, by a ventilation machine.
  • Treatment administration device 350 may include and/or be in communication with a clinical decision support system.
  • FIG. 4 is a schematic depicting time intervals for prediction of sepsis onset, in accordance with some embodiments of the present invention.
  • Monitoring time interval 402 denotes the time interval during which vital sign measurements are collected for prediction of sepsis onset in the target patient, for example, about 4 hours, about 8 hours, about 12 hours, about 16 hours, about 24 hours, or other time intervals.
  • the target patient is not diagnosed with sepsis during monitoring time interval 402.
  • Monitoring time interval 402 denotes a start 404 of the monitoring and an end 406 of the monitoring.
  • the prediction of impending onset of sepsis is performed at end 406 of the monitoring interval based on the vital sign measurements collected during monitoring time interval 402.
  • monitoring time interval 402 may define a sliding window of a fixed duration.
  • the prediction of impending sepsis may be performed adaptively, for example, every 10 minutes, every 15 minutes, every 30 minutes, every hour, or other intervals, based on the measurements of the vital signs collected during the preceding sliding windows of the monitoring time intervals 402 (e.g., the previous 8 hours).
  • the prediction performed at the time denoted by 406 is for onset of sepsis during the upcoming time interval 408, for example, within the next hour, 2 hours, 4 hours, 8 hours, or other interval of time.
  • Upcoming time interval 408 may be consecutive to monitoring time interval 402.
  • the sample patients are evaluated for the presence or absence of sepsis at end 410 of upcoming time interval 408.
  • the classifier(s) may be generated by computing device 304, and/or obtained from an external computing device.
  • the generated classifier(s) may be stored in classifier repository 316A by data repository 316 associated with computing device 304.
  • the generated classifier(s) predict a likelihood of impending sepsis onset for a target patient based on sets of different types of vital sign measurements of the target individual obtained over the recent monitoring time interval.
  • the classifier(s) receive as input values of extracted dynamic features computed according to dynamic features, each dynamic feature defining a combination of a certain dynamic function and a set of a certain types of vital sign measurement (to which the certain dynamic feature is applied), and outputs a computed prediction of the likelihood of onset of sepsis during the upcoming time interval.
  • the prediction may be outputted as a binary classification into a first category indicative of statistically significant risk of onset of sepsis during the upcoming time interval, or into a second category indicative of no statistically significant risk of sepsis onset during the upcoming time interval.
  • Statistically significant risk may be qualified by a criterion such as a threshold value, which may be predefined, for example, a threshold of 60%, 75%, or other values.
  • the prediction may be performed for additional and/or other classification categories, optionally according to established clinical guidelines, for example, severe sepsis, and septic shock.
  • the prediction may include a computed probably of the target patient being diagnosed with sepsis during the upcoming time interval.
  • the statistically significant indication is indicative of no/low risk of developing sepsis.
  • FIG. 2 is a flowchart of a method of generating the classifier(s) for prediction of sepsis onset.
  • a set of values each value of the set pertaining to a different type of vital sign measurements outputted by a respective vital sign sensor over the recent monitoring time interval is received for sampled individuals.
  • Each values of each set of values is associated with a different time stamp.
  • the sets of values may be obtained, for example, from EMR 310A, from a client terminal 308 storing the set of values (e.g., patient bedside monitoring unit, nurse station server), and/or from the vital sign sensors 311.
  • the sets of values may be received over network 312 via data interface 314.
  • Each set of values may be stored as summary values associated with respective start and end times of time intervals, for example, measurements collected every 5 minutes, 10 minutes, 15 minutes, or 30 minutes, or other time intervals.
  • the set of values may be stored as continuously collected data, for example, analogue signals and/or digital samples of analogue signals. Continuously collected data may be processed to arrive at a summary value for the time interval, for example, multiple measurements collected over 15 minutes may be averaged into a single summary value for the 15 minute interval.
  • a summary measurement may be computed based on data collected every 10 minutes (i.e., 6 an hour) for a total of 48 summary measurements.
  • Each set of values of each vital sign measurement may be stored as a vector, for example, mathematically denoted as X t e R" , wherein X denotes a vector, i denotes the respective vital sign measurement, R denotes a real number, and n denotes the number of measurements performed over the monitoring time interval.
  • the vital sign measurements include monitoring data that would otherwise be collected routinely, for example, for a patient admitted to the ICU.
  • the vital sign measurements exclude laboratory based investigations of biological samples of the target patient and/or measurements by sensors for exclusive prediction of sepsis onset.
  • Exemplary vital sign measurements include:
  • Blood pressure measured by a blood pressure sensor For example, mean arterial pressure measured via an arterial line.
  • Heart rate measured by a heart rate sensor for example, based on electrocardiogram (ECG) and/or pulse oximeter.
  • Body temperature measured by a temperature sensor for example, a thermometer.
  • each of the sampled individuals is associated with a label indicative of a diagnosis of sepsis or no sepsis.
  • the sepsis state or no sepsis is evaluated for each sampled individual at the end of the consecutive time interval following the end of the recent monitoring time interval.
  • the sepsis or not sepsis state may be determined, optionally automatically from the EMR, for example, by detecting an indication of administration of (or an order to administer) antibiotics, and/or by detecting a diagnosis of sepsis (e.g., in a medical diagnosis field of the EMR), and/or by automatic evaluation of medical criteria for diagnosis of sepsis.
  • the label may be a metadata tag associated with the respective sampled individual.
  • sampled individuals who exhibited 2 of 4 SIRS criteria by the end of (i.e., over) the consecutive time interval are associated with a label indicative of suspicion of sepsis.
  • Sampled individuals who are associated with a documented infection over the consecutive time interval are associated with a label indicative of sepsis.
  • Sampled individuals who were not diagnosed with an infection during the consecutive time interval are associated with a label indicative of no sepsis.
  • each dynamic feature is defined as a combination of a certain dynamic function (selected from multiple dynamic functions) which is applied to a certain types of vital sign measurement (selected from the available types of vital sign measurement).
  • the certain dynamic function processes the set of values of the associated vital sign measurement to output a value denoted as the extracted dynamic feature.
  • the total number of possible combinations is denoted as the number of different types of vital sign measurements multiplied by the number of dynamic functions.
  • Each dynamic function is based on local variability of the values of a certain vital sign measurement measured over the recent monitoring time interval.
  • Each dynamic function may be computed based on, for example, local minimum and local maximum points (i.e., values) identified for the set of values of the respective vital sign measurement, and/or other defined points (e.g., the number of times the values cross a certain threshold within a given time period).
  • the local minimum and maximum points and/or threshold crossing events may be indicative of an event of trend change.
  • Each dynamic function may be computed, for example, according to an amount of local changes of the respective set of values, and/or magnitude of local changes of the respective set of values.
  • the local minimum and local maximum points in each set 3 ⁇ 4 are identified, and denoted as members of corresponding vectors Y
  • the points in Y are sorted according to appearance in the time-series X, mathematically denoted as
  • Each dynamic function processes a set of values.
  • a certain dynamic function may processes a first set of heart rate values, and the certain dynamic function may process other sets of values such as a second set of respiratory rate values.
  • Exemplary dynamic functions include: Number of trend changes in the respective set of values. Mathematically denoted IFI, which equals to the size of set Y The number of trend changes are computed according to the number of local minimum and maximum points of the set of values of the respective vital sign measurement. The number of trend changes compares instability of a certain vital sign. Relatively more trend changes in a certain vital sign is indicative of less satiability than fewer changes in the certain vital sign.
  • the mean intensity of change function is indicative of the mean magnitude of changes in the vital sign. Relatively higher mean intensity of changes in a certain vital sign is indicative of less stability than lower mean intensity of changes in the certain vital sign.
  • intensity of change in the set of values of the respective vital sign measurement is computed according to value in which the lower 50% of magnitude changes between values of local minimum and local maximum points stop when sorted.
  • the median intensity of change is indicative of the median of changes in a vital sign, the value in which the lower 50% of measurements stop at.
  • Minimum intensity change Mathematically denoted as ⁇ min ⁇ y i+1 - .
  • the minimum intensity change is indicative of the minimal magnitude of change of a certain vital sign during the monitoring time interval.
  • Maximum intensity change Mathematically denoted as max max ⁇ y i+l - >>, ⁇ ⁇ ⁇ The maximum value of the magnitude of change between values of local minimum and maximum points of the set of values of the respective vital sign measurement.
  • the maximum intensity change is indicative of the maximal magnitude of change of a certain vital sign during the monitoring time interval.
  • a subset of the dynamic features is selected according to a separation requirement between a sub- population of the sampled individuals denoted as septic and another sub-population of the sampled individuals denoted as control.
  • the separation requirement may be based on, for example, a t-statistic test that statistically examines the difference between the mean of the sampled individuals denoted as septic and the mean of the sampled individuals denoted as control, to determine whether the different is statistically significant or random.
  • the subset of combinations may be selected according to the combinations that provide the most statistically significant separation between the populations, and/or to exclude combinations based on random separation.
  • the classifier is generated according to the subset of dynamic features (i.e., subset of combinations). For example, 4, 6, 8, 10, or other numbers of the subset of dynamic features.
  • the subset of dynamic features reduces the data storage requirements for storing the data computed based on the dynamic features, and/or reduces processor utilization and/or reduces processing time for computing the data based on the dynamic features and/or classification by the classifier.
  • An exemplary subset of dynamic features includes: number of trend changes in respiratory rate measurements, number of trend changes in arterial pressure measurements, the minimal change in respiratory rate measurements, and the median change in heart rate measurements.
  • one or more classifiers are generated according to the dynamic features (optionally the subset of dynamic features) and corresponding labels. Each classifier predicts impending onset of sepsis for a target individual.
  • classifiers of multiple types are generated. Exemplary classifier types include: logistic regression, linear support vector machine, support vector machine with radial kernel, support vector machine with polynomial kernel, artificial neural network, or any combination of the aforesaid.
  • Logistic regression is tool of medical data analysis, including mortality or morbidity outcomes prediction. LR may serve as a benchmark with other more advanced machine learning models. LR builds a function which describes the correlation between input and output variables. The function's output is forwarded as input to a sigmoid function which assures the overall output is a probability. LR may be mathematically represented as:
  • Support Vector Machine is a classification model utilized for medical applications.
  • the SVM is based on computing a hyperplane of the form w T x + b— 0 which provides the best separation between the population of patients with sepsis and the population of patients without sepsis.
  • f ( x , ) si 8 n x , + b ) [00127] where each x i which fulfils w T x + b > 0 is classified as 1, and x i which fulfill w T x + b ⁇ 0 are classified as -1.
  • the SVM's output is passed to a sigmoid function.
  • a kernel function is used.
  • a kernel function is a mathematical manipulation which maps the problem characteristics in the plane to a higher dimension in which the required hyperplane is found. The input is replaced by a kernel function ⁇ ( ):
  • Two exemplary kernels include: a polynomial basis function is of the form: [00131] A radial basis function is of the form:
  • An artificial neural network is a multilayered mathematical representation of a learning network which maps the correlation between inputs and outputs by backtracking to evaluate and minimize errors.
  • (k) denotes input of the k th neuron where i
  • w t (k) denotes a value of correlation between the k ' and k—l neurons.
  • F denotes a propagation function for classification (e.g., a sigmoid function)
  • b denotes a bias of the respective neuron.
  • y(k) denotes the output of the k' h neuron.
  • New examples are then run through the ANN from input neurons to outputs.
  • the predictive performance of each of the classifier types is evaluated, for example, according to area under curve (AUC) of the receiver operating characteristic curve (ROC), sensitivity, specificity, accuracy positive predictive value (PPV), and/or negative predictive value (NPV).
  • AUC area under curve
  • ROC receiver operating characteristic curve
  • PPV accuracy positive predictive value
  • NPV negative predictive value
  • One or more classifier types are selected according to one or more of the evaluated predictive performance, optionally at least based on the AUC (i.e., classifier types that are associated with the best performance prediction metrics).
  • the ROC is a graphical plot that illustrates the diagnostic ability of a binary classifier as a discrimination threshold is varied.
  • the ROC curve is created by plotting the true positive rate (TPR) against the false positive rate (FPR) at various threshold settings.
  • the AUC denotes the probability of the classifier ranking a randomly chosen positive instance higher than a randomly chosen negative one (assuming positive ranks higher than negative).
  • the AUC may be mathematically represented as:
  • AUC ⁇ °° TPR(T)FPR (T)dT
  • the generated classifier(s) for predicting impending sepsis are outputted.
  • the generated classifier(s) receive values computed according to the selected subset of combinations, and computes a prediction of impending onset of sepsis for the target individual.
  • sets of values are received for each of the different types of vital sign measurements.
  • the vital sign measurements are defined according to the generated classifier(s).
  • the sets of values are outputted (and/or computed from output of) respective vital sign sensors over a recent monitoring time interval.
  • the different vital sign measurements exclude laboratory based investigations of biological samples of the target patient.
  • the different vital sign measurements exclude data collected for exclusively predicting onset of sepsis.
  • the sets of values may be obtained, for example, from records of the patient stored in a medical database (e.g., EMR), from a client terminal that collects the sensor outputs, and/or from the sensors themselves.
  • a medical database e.g., EMR
  • client terminal that collects the sensor outputs, and/or from the sensors themselves.
  • the sets of values may be collected dynamically, continuously and/or at predefined intervals, for dynamic prediction of the imminent onset of sepsis according to a sliding window having a size corresponding to the recent monitoring time interval.
  • extracted dynamic feature values are computed according to the (subset of) dynamic features (i.e., combinations) associated with the classifier(s).
  • Each dynamic features defines a combination of a certain dynamic function (selected from available multiple dynamic functions) applied to a certain type of vital sign measurement (selected from multiple available types of vital sign measurements).
  • the extracted dynamic features denotes the value computed from the certain dynamic function applied to the certain type of vital sign measurement data.
  • the (subset of) dynamic features are defined according to the classifier(s), based on the (sub) combinations used to train the classifier(s).
  • the classifier(s) receive as input the extracted dynamic feature values calculated according to the dynamic features (i.e., combinations).
  • the classified s) computes a prediction of impending sepsis onset in the target individual within the upcoming time interval consecutive to the recent monitoring time interval.
  • the prediction may be outputted as a classification into a category
  • a probability e.g., between 0 and 1 of the target individual developing sepsis.
  • the prediction of impending sepsis onset (or optionally the prediction of no sepsis) in the target individual is outputted.
  • the prediction may be, for example, presented on a display of an bedside monitoring unit, presented on a display of a nurse monitoring station server, presented on a display of a mobile device, further processed by code instructions, stored in a record of the target individual in a medical database (e.g., EMR), and/or forwarded to another computing device.
  • a medical database e.g., EMR
  • an alert is generated when the classifier computes an indication of impending sepsis onset.
  • the alert may, for example, be transmitted to a mobile device of a healthcare provider (e.g., on-call physician), and/or act as a trigger of an alarm of a bedside unit (e.g., beeping sound, flashing light).
  • patients may be automatically identified for a research study, according to defined inclusion criteria that include prediction of sepsis. Records of the patients may be flagged, stored in a database, and/or extracted, and forwarded to a server of the researcher.
  • a treatment priority score may be computed based on the prediction of sepsis and/or other data.
  • the treatment priority score may be an absolute score (e.g., emergency treatment required, urgent treatment required) and/or a relative score (e.g., ranking of patients in treatment priority).
  • the treatment score may be presented on a display, and/or provided as input to automated treatment machine(s) 350 that perform automated treatment of the patients.
  • the prediction of impending sepsis (and/or prediction of no sepsis) is provided as input for one or more automated treatment administration devices 350, for example, a controller that controls and/or is in communication with one or more of: a clinical decision support system, an intravenous (IV) fluid control mechanism, a ventilation machine, and a drug injection machine.
  • Automated treatment devices 305 may automatically administer one or more treatments according to the prediction of impending sepsis (and/or optionally according to the prediction of no sepsis).
  • the treatments may be automated based on a database mapping recommended sepsis treatments to prediction of sepsis onset (e.g., based on clinical guidelines).
  • the sepsis treatment corresponding to the prediction is obtained from the database, and treatment is automatically administered accordingly.
  • Exemplary treatments automatically administered by automated treatment administration devices 350 based on the prediction of sepsis onset include: automated injection of antibiotics by an automated antibiotic delivery mechanism, automatic adjustment of intravenous fluid infusion rate to a target rate by an IV flow rate mechanism, automated increase in percent of oxygen being administered to a target percentage by an automated mechanical ventilator, and automated administration of vasopressors by an automated drug delivery mechanism.
  • the prediction of impending sepsis may be provided as a decision support input for use in other hospital departments, for example, internal medicine wards. Such departments may use the sepsis onset alerts as a motivation to move a patient with high probability of sepsis onset to the intensive care unit.
  • the prediction of impending sepsis may be applied to non-conventional hospitals, for example, field hospitals or hospices, where only minimum bedside vitals are collected from patients.
  • Various embodiments and aspects of the systems, methods, and/or code instructions described herein and as claimed in the claims section below find computational support in the following examples.
  • EMRs General Intensive Care Unit
  • RMC Rabin Medical Center
  • Petah Tikva Israel
  • the EMRs document in real-time all clinical data as well as laboratory data, drug administration and medical notes for all patients admitted to the ICU.
  • the data was anonymized prior to analysis to exclude all specifics of patient identity. The trial was approved by the hospital's institutional review board with a waiver of informed consent as the study did not affect clinical care and all data were anonymized.
  • Step 1 Patients who exhibited 2 of the 4 SIRS criteria over a consecutive 24 hour interval were marked as "Suspicion of Sepsis”.
  • Step 2 Of these, all patients with a documented infection according to positive bacterial cultures were recorded as "septic".
  • FIG. 5 which is a flowchart depicting the process of selection of patients for inclusion in the study, in accordance with some embodiments of the present invention.
  • FIG. 6 is a table comparing the target group that developed sepsis (sometimes termed herein the Sepsis population) and the control group that did not develop sepsis (sometimes termed herein the non-Sepsis population), in accordance with some embodiments of the present invention. No significant differences in age or gender were noted between the groups.
  • patient data were analyzed as a casual time-series.
  • the analysis of the selected 4 vital sign stems from the fact that these parameters are typically available in all ICUs, and are collected at frequent intervals.
  • vectors X were constructed to represent the following vital sign measurements: mean arterial pressure, heart rate, respiratory rate and temperature.
  • the extracted dynamic features provide a good separation between target and control populations, as depicted in FIG. 7, which is a table summarizing data depicting separation of the septic and control patient populations based on all extracted dynamic features (i.e., based on all combinations), in accordance with some embodiments of the present invention.
  • FIG. 8 depicts a graph 802 of a patient of the study that developed sepsis and a graph 804 of another patient of the study that did not develop sepsis, demonstrating the behavior of Mean Arterial Pressure (MAP) during the first 8 hours of the 12 hour period of the study prior to sepsis time -point.
  • MAP Mean Arterial Pressure
  • FIG. 9 is a graph of combinations of dynamic functions and vital sign types 902 ranked according to an importance scale 904, for selection of the top four combinations for prediction of sepsis onset, in accordance with some embodiments of the present invention.
  • the selected combination of dynamic functions and corresponding vital sign measurements were: number of trend changes in respiratory rate, and number of trend changes in arterial pressure, the minimal change in respiratory rate, and the median change in heart rate.
  • a compact model consisting of the subset of 4 combination of dynamic functions and vital sign measurements extracted from the target patient is created.
  • FIG. 10 includes graphs 1002, 1004, 1006, and 1008 that visually depict the distinction between the sepsis and non-sepsis population for the selected subset of 4 combinations of dynamic functions and vital sign measurement, in accordance with some embodiments of the present invention.
  • Graph 1002 depicts results for the combination of the mean arterial pressure (vital sign) and the number of trend changes (dynamic function).
  • Graph 1004 depicts results for the combination of the respiratory rate (vital sign) and the number of trend changes (dynamic function).
  • Graph 1006 depicts results for the combination of the heart rate (vital sign) and the median change (dynamic function).
  • Graph 1008 depicts results for the combination of the respiratory rate (vital sign) and the minimal change (dynamic function).
  • each graph denotes the computed value of the combination of the dynamic function and vital sign for the non-Sepsis population.
  • the B label of each graph denotes the computed value of the combination of the dynamic function and vital sign for the Sepsis population.
  • the input for generating the classifiers is the dataset of 600 vectors which include both the study (i.e., Sepsis) and control (i.e., non-Sepsis) groups.
  • the dataset was divided into a training set of 75% (450 records) and a test set of 25% (150 records). The ratio between positive (septic patients) and negative (non-septic patients) examples was maintained in both sets.
  • the 600 patients were partitioned into mutually exclusive sets for training and testing the prediction algorithm.
  • the final classifier is selected according to the classifier that produces the best Area under Curve (AUC) which is used to examine predictive performance of machine learning in medical applications.
  • AUC Area under Curve
  • FIG. 11 is a table summarizing the performance results of each classifier type in classifying a validation portion of the testing set, which was not a part of the training set, in accordance with some embodiments of the present invention. From the results summarized in the table of FIG. 11, it is evident that SVM with the radial basis function provided the highest AUC of 88.38%, the highest PPV of 0.9524 (i.e., the probability that a given sepsis prediction is accurate), and the highest specificity of 0.9615 (i.e., the ability to correctly identify a non-septic case).
  • FIG. 12 is a graph depicting the receiver operating characteristic curve (ROC) plots of the 5 generated and tested classifier types, in accordance with some embodiments of the present invention. It is noted that based on the graph, the SVM-RBF classifier covers the largest area, and therefore has the highest diagnosis ability to discriminate between septic and non- septic.
  • ROC receiver operating characteristic curve
  • composition or method may include additional ingredients and/or steps, but only if the additional ingredients and/or steps do not materially alter the basic and novel characteristics of the claimed composition or method.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Veterinary Medicine (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Biophysics (AREA)
  • Physiology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Data Mining & Analysis (AREA)
  • Cardiology (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

L'invention concerne un procédé consistant : à recevoir un ensemble de valeurs pour chaque type de différents types de mesures de signe vital délivrées chacune par un capteur de signe vital respectif sur une période de surveillance récente, à accéder à des caractéristiques dynamiques, chaque caractéristique dynamique étant une combinaison d'une fonction dynamique donnée et étant appliquée sur une mesure de signe vital donnée, chacune des fonctions dynamiques étant fondée sur une variabilité locale des valeurs mesurées sur la période de surveillance récente, à calculer des caractéristiques dynamiques extraites par application de la fonction dynamique donnée à l'ensemble respectif de valeurs de la mesure de signe vital donnée en fonction de la caractéristique dynamique donnée, et à calculer une prédiction de probabilité d'apparition de sepsie imminente chez un individu cible dans une période à venir consécutive à la période de surveillance récente par un classificateur classant les caractéristiques dynamiques extraites.
PCT/IB2018/055498 2017-08-02 2018-07-24 Systèmes et procédés de prédiction de l'apparition d'une sepsie WO2019025901A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US201762540101P 2017-08-02 2017-08-02
US62/540,101 2017-08-02

Publications (1)

Publication Number Publication Date
WO2019025901A1 true WO2019025901A1 (fr) 2019-02-07

Family

ID=65232435

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/IB2018/055498 WO2019025901A1 (fr) 2017-08-02 2018-07-24 Systèmes et procédés de prédiction de l'apparition d'une sepsie

Country Status (1)

Country Link
WO (1) WO2019025901A1 (fr)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110123274A (zh) * 2019-04-29 2019-08-16 上海电气集团股份有限公司 一种脓毒血症的监测系统
CN110123277A (zh) * 2019-05-17 2019-08-16 上海电气集团股份有限公司 一种脓毒血症的数据处理系统
WO2019165004A1 (fr) * 2018-02-21 2019-08-29 Patchd, Inc. Systèmes et procédés de surveillance de sujet
CN110517775A (zh) * 2019-08-12 2019-11-29 上海艾莎信息科技有限公司 疾病风险预测方法、装置、计算机设备和存储介质
US20210052218A1 (en) * 2019-08-20 2021-02-25 Patchd, Inc. Systems and methods for sepsis detection and monitoring
EP3785611A1 (fr) * 2019-08-30 2021-03-03 Hill-Rom Services, Inc. Système de surveillance de la septicémie
EP3799068A1 (fr) * 2019-09-24 2021-03-31 Siemens Healthcare GmbH Système et procédé de notification de maladies infectieuses
WO2023128863A3 (fr) * 2021-12-28 2023-09-14 National University Of Singapore Procédés et systèmes de stratification du risque de sepsis

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090149724A1 (en) * 2007-12-05 2009-06-11 Massachusetts Institute Of Technology System and method for predicting septic shock
US20110112442A1 (en) * 2007-05-02 2011-05-12 Earlysense Ltd. Monitoring, Predicting and Treating Clinical Episodes
US20120245439A1 (en) * 2008-11-20 2012-09-27 David Andre Method and apparatus for determining critical care parameters
WO2016040295A1 (fr) * 2014-09-09 2016-03-17 Lockheed Martin Corporation Procédé et appareil de détection de maladies
US20170049403A1 (en) * 2010-03-15 2017-02-23 Singapore Health Services Pte Ltd System and method for predicting acute cardiopulmonary events and survivability of a patient

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110112442A1 (en) * 2007-05-02 2011-05-12 Earlysense Ltd. Monitoring, Predicting and Treating Clinical Episodes
US20090149724A1 (en) * 2007-12-05 2009-06-11 Massachusetts Institute Of Technology System and method for predicting septic shock
US20120245439A1 (en) * 2008-11-20 2012-09-27 David Andre Method and apparatus for determining critical care parameters
US20170049403A1 (en) * 2010-03-15 2017-02-23 Singapore Health Services Pte Ltd System and method for predicting acute cardiopulmonary events and survivability of a patient
WO2016040295A1 (fr) * 2014-09-09 2016-03-17 Lockheed Martin Corporation Procédé et appareil de détection de maladies

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019165004A1 (fr) * 2018-02-21 2019-08-29 Patchd, Inc. Systèmes et procédés de surveillance de sujet
CN110123274A (zh) * 2019-04-29 2019-08-16 上海电气集团股份有限公司 一种脓毒血症的监测系统
CN110123277A (zh) * 2019-05-17 2019-08-16 上海电气集团股份有限公司 一种脓毒血症的数据处理系统
CN110517775A (zh) * 2019-08-12 2019-11-29 上海艾莎信息科技有限公司 疾病风险预测方法、装置、计算机设备和存储介质
US20210052218A1 (en) * 2019-08-20 2021-02-25 Patchd, Inc. Systems and methods for sepsis detection and monitoring
EP3785611A1 (fr) * 2019-08-30 2021-03-03 Hill-Rom Services, Inc. Système de surveillance de la septicémie
US20210059597A1 (en) * 2019-08-30 2021-03-04 Hill-Rom Services, Inc. Sepsis monitoring system
CN112447289A (zh) * 2019-08-30 2021-03-05 希尔-罗姆服务公司 败血症监测系统
EP3799068A1 (fr) * 2019-09-24 2021-03-31 Siemens Healthcare GmbH Système et procédé de notification de maladies infectieuses
CN112635046A (zh) * 2019-09-24 2021-04-09 西门子医疗有限公司 用于感染病通告的系统和方法
US11798691B2 (en) 2019-09-24 2023-10-24 Siemens Healthcare Diagnostics Inc. System and method for infectious disease notification
WO2023128863A3 (fr) * 2021-12-28 2023-09-14 National University Of Singapore Procédés et systèmes de stratification du risque de sepsis

Similar Documents

Publication Publication Date Title
WO2019025901A1 (fr) Systèmes et procédés de prédiction de l'apparition d'une sepsie
Zeiberg et al. Machine learning for patient risk stratification for acute respiratory distress syndrome
US11961621B2 (en) Predicting intensive care transfers and other unforeseen events using machine learning
Bolourani et al. A machine learning prediction model of respiratory failure within 48 hours of patient admission for COVID-19: model development and validation
US11631497B2 (en) Personalized device recommendations for proactive health monitoring and management
Bloch et al. Machine learning models for analysis of vital signs dynamics: a case for sepsis onset prediction
Shashikumar et al. DeepAISE–An interpretable and recurrent neural survival model for early prediction of sepsis
Mao et al. Medical data mining for early deterioration warning in general hospital wards
US20200395129A1 (en) Systems and methods for identification of clinically similar individuals, and interpretations to a target individual
US11450434B2 (en) Implementation of machine-learning based query construction and pattern identification through visualization in user interfaces
Guarneros-Nolasco et al. Identifying the main risk factors for cardiovascular diseases prediction using machine learning algorithms
van Wyk et al. How much data should we collect? A case study in sepsis detection using deep learning
Castiñeira et al. Adding continuous vital sign information to static clinical data improves the prediction of length of stay after intubation: a data-driven machine learning approach
Gill et al. Artificial intelligence to enhance clinical value across the spectrum of cardiovascular healthcare
Guillame-Bert et al. Learning temporal rules to forecast instability in continuously monitored patients
Dagliati et al. Careflow mining techniques to explore type 2 diabetes evolution
Al-Mualemi et al. A deep learning-based sepsis estimation scheme
Li et al. Real-time sepsis severity prediction on knowledge graph deep learning networks for the intensive care unit
Payrovnaziri et al. Enhancing prediction models for one-year mortality in patients with acute myocardial infarction and post myocardial infarction syndrome
Carlin et al. Predicting individual physiologically acceptable states at discharge from a pediatric intensive care unit
Sami et al. The role of data pre-processing techniques in improving machine learning accuracy for predicting coronary heart disease
Tsiklidis et al. Predicting risk for trauma patients using static and dynamic information from the MIMIC III database
US11335461B1 (en) Predicting glycogen storage diseases (Pompe disease) and decision support
Fu et al. Utilizing timestamps of longitudinal electronic health record data to classify clinical deterioration events
Goldstein et al. Classifying individuals based on a densely captured sequence of vital signs: An example using repeated blood pressure measurements during hemodialysis treatment

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18841758

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18841758

Country of ref document: EP

Kind code of ref document: A1