A system and a method for developing a data model for medical use

Abstract

A method of developing a data model for medical use is provided. The method includes training a multi-parameter data model (302) on health parameters related to one or more of multiple organs (304) of individuals. One or more data models (306) are trained on health parameters related to one or more organ systems (308) of individuals. The data models (306) are trained for an organ system susceptible to sepsis or the failure of which indicates sepsis in the individuals. The learnings from the data models (306) and the multi-parameter data model (302) are merged to obtain an aggregated data model (310) used for predicting risk of sepsis. The data models (306) may be used for monitoring health parameters related to one or more organ systems (308) predicted to be at risk of impact from sepsis by the aggregated data model (310).

Description

A SYSTEM AND A METHOD FOR DEVELOPING A DATA MODEL FORMEDICAL USEFIELD

[0001] The present invention relates to healthcare, and specifically relates to a system and a method for developing a data model for predicting risk of sepsis of an individual and monitoring health of individual organ systems at risk of sepsis.BACKGROUND

[0002] Sepsis is a life-threatening organ dysfunction caused by a dysregulated immune response to infection. Sepsis is associated with significant morbidity, mortality, and healthcare expenses. Early detection of sepsis is considered one of the key aspects to improve the outcomes, and constitutes one of the most important and challenging problems in healthcare.

[0003] Development of organ dysfunction is the most important clinical event during sepsis, as development of organ dysfunction directly causes mortality and morbidity. For example, liver dysfunction after sepsis is an independent risk factor for multiple organ dysfunction and sepsis-induced death. Further, pre-existing liver dysfunction is a risk factor for the progression of infection to sepsis. The organ systems that are important for monitoring of sepsis may be sub-divided into the following categories: hepatic, renal, cardiovascular, coagulation, and respiration.

[0004] Many algorithms exist for monitoring organ damage and sepsis risk separately. For example, conventional approaches of monitoring organ damage predict a stage of liver disease, but such conventional approaches do not monitor risk of sepsis because of the stage of the liver disease. No conventional approach exists for monitoring the health of organs in the context of sepsis and for monitoring the risk of organ damage after an episode of sepsis. Parallel monitoring of organs and sepsis gives clinicians actionable items by zooming into the state of organ health.

[0005] Thus, there is a need of a method of determining risk of sepsis based on the health parameters of specific organs of an individual.OBJECT OF THE INVENTION

[0006] A general objective of the present invention is to develop a data model for predicting the risk of sepsis by monitoring the health of organs.

[0007] Another objective of the invention is to train a multi-parameter data model to detect sepsis of a person in early stage.

[0008] Another objective of the present invention is to develop data models that may be used for monitoring the health of individual organ systems after the risk of sepsis has been determined.

[0009] Another objective of the invention is to fine-tune one or more data models for accurate monitoring of health of organs.

[0010] Still another objective of the invention is to determine impact of health of organs on sepsis.SUMMARY

[0011] The summary is provided to introduce aspects related to a system and a method of developing a data model for medical use, and the aspects are further described below in the detailed description. This summary is not intended to identify essential features of the claimed subject matter nor is the summary intended for use in determining or limiting the scope of the claimed subject matter.

[0012] In one aspect, a method for predicting risk of sepsis includes training of a multiparameter data model on health parameters related to one or more of multiple organ systems, sepsis specific parameters, and routine parameters of a plurality of individuals. The method further includes training one or more data models, where each data model of the one or more data models is trained on a plurality of health parameters related to an organ system of the plurality of individuals. The organ system corresponding to which a data model is trained is an organ system that is affected from sepsis or an organ system whose failure indicates sepsis. The one or more data models may also be trained using intermediary learnings of the multi-parameter data model and an output of the multiparameter data model.

[0013] The learnings from the one or more data models are merged with health parameters used for training the multi-parameter data model, intermediary learning of the multiparameter data model, and the output of the multi-parameter data model to obtain anaggregated data model. The aggregated data model may be executed for providing an output indicative of a risk of sepsis in an individual of the plurality of individuals.

[0014] In one aspect, the health status related to one or more of the organ systems that may be affected by sepsis is monitored using the one or more data models based on the risk of sepsis predicted by the aggregated data model. The health status related to one or more of the organ systems is monitored based on learnings of the aggregated data model, user input, and availability of the health parameters related to the one or more organ systems. The monitoring of the health parameters related to one or more of the organ systems may be performed either at regular or irregular intervals. The monitoring of the health parameters related to one or more of the organ systems is performed for organ systems determined to be present at risk from sepsis.

[0015] In one aspect, the one or more data models are trained on hepatic system, renal system, cardiovascular system, coagulation system, and respiratory system.

[0016] In one aspect, the one or more data models may be used for monitoring the health status of individual organs systems.

[0017] In one aspect, the one or more data models, the multi-parameter data model, and the aggregated data model are trained using one or more Machine Learning (ML) techniques such as Bayesian models, probabilistic graphical models, neural networks, or tree-based algorithms including ensemble approaches using stacking, voting, and boosting, etc. The ensemble approaches used in tree-based algorithms may include, but are not limited to, Random Forest, XGBoost, LightGBM, CatBoost, and AdaBoost.

[0018] In one aspect, the merging of the learnings is performed by early fusion, mid fusion, or late fusion. The fusion techniques may include, but are not limited to, concatenation, late fusion, mid fusion, tensor decomposition, multi-view learning, and other attention mechanisms. The learnings include the intermediary learning and the output of the one or more data models and the multi-parameter data model. The health parameters used for training the multi-parameter data model may also be merged with the learnings.

[0019] In one aspect, the one or more data models may be executed only upon availability of the one or more health parameters related to one or more organ systems.Other aspects and advantages of the invention will become apparent from the following description, taken in conjunction with the accompanying drawings, illustrating by way of example the principles of the invention.BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings constitute a part of the description and are used to provide further understanding of the present invention. Such accompanying drawings illustrate the embodiments of the present invention and are used to describe the principles of the present invention. The embodiments are illustrated by way of example and not by way of limitation in the figures of the accompanying drawings in which like references indicate similar elements. It should be noted that references to “an” or “one” embodiment in this invention are not necessarily to the same embodiment, and such references provide at least one. In the drawings:

[0021] Figure 1 illustrates a network diagram of a system for developing a data model for predicting risk of sepsis in an individual, in accordance with an embodiment;

[0022] Figure 2 illustrates a block diagram showing internal components of a server used for developing a data model for predicting risk of sepsis in an individual, in accordance with an embodiment; and

[0023] Figure 3 illustrates a flow diagram of a method of developing a data model for predicting the risk of sepsis in an individual, in accordance with an embodiment.DETAILED DESCRIPTION

[0024] The detailed description set forth below in connection with the appended drawings is intended as a description of various embodiments of the present invention and is not intended to represent the only embodiments in which the present invention may be practiced. Each embodiment described in this disclosure is provided merely as an example or illustration of the present invention, and should not necessarily be construed as preferred or advantageous over other embodiments. The detailed description includes specific details for the purpose of providing a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without these specific details.

[0025] The present invention relates to a system and a method for developing a data model for medical use. The data model is utilized for predicting risk of sepsis in an individual. Figure 1 illustrates a network diagram of a system 100 for developing a data model for predicting risk of sepsis, in accordance with an embodiment of the present invention. The system 100 includes one or more data sources 102-1 to 102-n connected with a server 104.The one or more data sources 102-1 to 102-n may be databases storing health parameters of a plurality of individuals. The health parameters may include health parameters related to one or more organ systems of a plurality of individuals, health parameters related to one or more of multiple organ systems of the plurality of individuals, and parameters not specific to any particular organ system such as routine blood parameters and sepsis-specific parameters. For example, a first data source 102-1 may store health parameters relating to a renal system of the plurality of individuals. Similarly, a second data source 102-2 may store health parameters relating to hepatic system of the plurality of individuals. Further, an nthdata source 102-n stores both health parameters related to one or more of multiple organ systems of the plurality of individuals and parameters not specific to any particular organ system, such as sepsis-specific parameters.

[0026] The health parameters stored in the one or more data sources 102-1 to 102-n may be provided to the server 104. The server 104 may be a cloud server or a local server used for developing an aggregated data model for predicting the risk of sepsis. The server 104 may use the health parameters from the one or more data sources 102-1 to 102-n for developing the aggregated data model, which is hosted in the server 104.

[0027] Figure 2 illustrates a block diagram showing internal components of a server for developing a data model for predicting risk of sepsis in an individual, in accordance with an embodiment of the present invention. The server 104 may include an interface 202, a processor 204, and a memory 206. The health parameters related to one or more of multiple organ systems of a plurality of individuals and the health parameters related to one or more of the organ systems of the plurality of individuals, referred to as input health parameters in Figure 2, may be provided to the server 104 through the interface 202. The memory 206 may store program instructions for developing one or more data models 208, program instructions for developing a multi-parameter data model 210, program instructions for developing an aggregated data model 212, and program instructions for executing the aggregated data model to obtain an output indicative of a risk of sepsis 214. Detailed functioning of such program instructions shall become evident upon reading the details provided henceforth.

[0028] Figure 3 illustrates a flow diagram of a method of developing a data model for predicting risk of sepsis of an individual, in accordance with an embodiment of the present invention. The method includes training a multi-parameter data model 302 on one or morehealth parameters related to one or more of multiple organ systems 304, sepsis specific parameters, and routine parameters of a plurality of individuals and parameters not specific to any organ system such as routine blood parameters and sepsis-specific parameters, collectively referred to as overall health parameters in Figure 3. The multi-parameter data model 302 is a data model trained based on one or more Machine Learning (ML) techniques such as Bayesian models, probabilistic graphical models, neural networks, or tree-based algorithms including ensemble approaches using stacking, voting, and boosting. The ensemble approaches used in tree-based algorithms may include, but are not limited to, Random Forest, XGBoost, LightGBM, CatBoost, and AdaBoost. The health parameters related to one or more of multiple organ systems 304 may include, but are not limited to, heart rate, systolic blood pressure, diastolic blood pressure, mean arterial pressure, temperature, respiration rate, oxygen saturation, serum glucose, white blood cells, platelets count, blood urea nitrogen, creatinine, and total bilirubin. In addition, parameters such as the age and gender of the individual may be used for training the multi-parameter data model 302.

[0029] The method also includes training one or more data models 306 on health parameters related to one or more organ systems 308 of the plurality of individuals, referred to as organ specific health parameters in Figure 3. The one or more data models are data models trained based on one or more Machine Learning (ML) techniques such as Bayesian models, probabilistic graphical models, neural networks, or tree-based algorithms including ensemble approaches using stacking, voting, and boosting, etc. The ensemble approaches used in tree-based algorithms may include, but are not limited to, Random Forest, XGBoost, LightGBM, CatBoost, and AdaBoost. Each of the data models 308 are trained on the health parameters of an organ system 308 of the plurality of individuals, and the data model may be executed for monitoring the health status of the organ system.

[0030] The health parameters related to an organ system 308 may be collected from an organ system susceptible to sepsis or an organ system whose failure indicates sepsis in the plurality of individuals. The organ system may be a hepatic system, a renal system, a cardiovascular system, a coagulation system, or a respiratory system. The one or more data models 306 may be used for monitoring the health status of individual organ systems, and learnings of the one or more data models may be used in the prediction of risk of sepsis of an individual in the plurality of individuals. The health parameters related to an organsystem 308 used for training one of the data models 306 may include parameters related to the organ system. The data model 306 may also be trained on diseases specific to the organ system. For example, the parameters related to hepatic system may include total bilirubin, alkaline phosphatase, albumin, aspartate aminotransferase, and alanine aminotransferase, and the disease specific to hepatic system may include liver sclerosis, etc.

[0031] A first data model 306-1 may be trained on a first set of health parameters related to an organ system 308-1, a second data model 306-2 may be trained on a second set of health parameters related to an organ system 308-2, and so on. For example, the first data model 306-1 may be trained on a set of health parameters related to an organ system 308- 1, where the organ system is a hepatic system. The first set of health parameters related to an organ system 308-1 used for training the first data model 306-1 may include, but is not limited to, total bilirubin, alkaline phosphatase, albumin, aspartate aminotransferase, alanine aminotransferase, platelets count, serum glucose, FIB4 Index, and genomic biomarkers. In addition, parameters such as the age and gender of the individual may be used for training the first data model 306-1. The data model 306-1 trained using health parameters related to an organ system 308-1, where the organ system is the hepatic system, may be executed for monitoring the health of the hepatic system.

[0032] The second data model 306-2 may be trained on a set of health parameters related to an organ system 308-2, where the organ system is a renal system. The second set of health parameters related to an organ system 308-2 used for training the second data model 306-2 may include, but is not limited to, blood urea nitrogen, creatinine, sodium, potassium, chloride, calcium, inorganic phosphorus, uric acid, and genomic biomarkers. In addition, parameters such as the age and gender of the individual may be used for training the second data model 306-2. The data model 306-2 trained using health parameters related to an organ system 308-2, where the organ system is the renal system, may be executed for monitoring the health of the renal system.

[0033] Similarly, the data models 306-1 to 306-n may be trained on n sets of health parameters related to an organ system 308, where n corresponds to number of organ systems for health parameters related to an organ system 308 are available. The data models 306-1 to 306-n, cumulatively labelled as 306, may be executed for monitoring the health of the respective organ systems.

[0034] In addition to the health parameters related to one or more organ systems 308 ofthe plurality of individuals, the one or more data models 306 are also trained on either one or both of intermediary learnings of the multi-parameter data model 302 and an output of the multi-parameter data model 302. The intermediary learning of the multi-parameter data model 302 may include learning obtained at intermediary layers of the multi-parameter data model 302 and learnings obtained during the different iterations of training of the multi-parameter data model 302.

[0035] Successively, intermediary learnings and output of the one or more data models 306 may be merged with one or more of the health parameters used for training the multiparameter data model 302, intermediary learnings of the multi-parameter data model 302, and the output of the multi-parameter data model 302 to obtain a merged learning. The merging, also referred to as aggregation, may be performed by methods of fusion including, but not limited to, concatenation, late fusion, mid fusion, tensor decomposition, multi-view learning, and other attention mechanisms at act 312. The merged learning may be utilized to develop an aggregated data model 310 used for predicting the risk of sepsis in an individual of the plurality of individuals. Thus, the aggregated data model is trained on one or more of the health parameters used for training the multi-parameter data model 302, the intermediary learnings and output of the multi-parameter data model 302, and the intermediary learnings and output of one or more data models 306. The aggregated data model 310 may be a data model trained based on one or more Machine Learning (ML) techniques such as Bayesian models, probabilistic graphical models, neural networks, or tree-based algorithms including ensemble approaches using stacking, voting, and boosting, etc. The ensemble approaches used in tree-based algorithms may include, but are not limited to, Random Forest, XGBoost, LightGBM, CatBoost, and AdaBoost.

[0036] Once developed, the aggregated data model 310 may be executed to predict the risk of sepsis in an individual of the plurality of individuals. Health parameters of the individual may be used by the aggregated data model 310 for predicting the risk of sepsis in the individual. The health parameters of the individual include the health parameters related to one or more organs of the individual 304 and parameters not specific to any particular organ such as sepsis-specific parameters. Health parameters related to one or more organ systems 308 obtained from organ specific tests may also be used for predicting the risk of sepsis in the individual. The output of the aggregated data model 310 may be a metric such as a sepsis score used to indicate the risk of sepsis in the individual. The use ofthe one or more data models 306 in the execution of the aggregated data model 310 depends on availability of health parameters related to corresponding organ systems 308. The health parameters related to an organ system 308 may be obtained after one or more specific medical tests are performed as per orders of individuals such as doctors, clinicians, etc. The one or more medical tests may be conducted within a predefined time period, such as 24 hours prior to execution of the aggregated data model 310. The aggregated data model 310 may be executed after the individual gets admitted in an emergency room or hospital.

[0037] The aggregated data model 310 is able to predict the risk of sepsis in the individual using the health parameters related to one or more of multiple organ systems 304 and parameters not specific to organs such as sepsis specific parameters. Additionally, when the health parameters related to one or more organ systems 308 obtained from organ specific testing are also provided along with the health parameters related to one or more of multiple organ systems 304, the aggregated data model 310 may be able to predict the risk of sepsis with higher accuracy. The learnings of the aggregated data model 310 may provide an indication about organs of the individual present at a highest risk of being affected with sepsis or the risk to individual organ systems of being affected from sepsis. Based on the risk of sepsis determined by the aggregated data model 310, a doctor or a surgeon may design a treatment plan for the individual.

[0038] If the sepsis score obtained as output from the aggregated data model indicates the risk of sepsis, the health of organs systems at risk due to onset of sepsis may be monitored using the one or more data models 306. A plurality of detailed organ specific tests may be carried out for one or more of the organ systems at risk due to onset of sepsis, and parameters of an organ system obtained from the tests are monitored using the corresponding data model. The health parameters related to one or more organ systems 308 are monitored as snapshots. Snapshots of health parameters of a particular organ system refer to the health parameters related to the organ system 308 obtained at a specific time instance. The snapshot used for monitoring the health of the organ systems may be collected at regular or irregular intervals.

[0039] In addition to monitoring the health of organ systems based on the specific organs indicated to be at risk by the learnings of the aggregated data model 310, the monitoring of health of organ systems may also be based on user input and availability of the health parameters related to one or more organ systems 308. The user input includes requirementof monitoring health of a particular organ system raised by individuals such as doctor, clinicians, specialists, etc.

[0040] The system proposed in the present embodiments may be deployed in any hospital setting including emergency department and intensive care unit as the system monitors organ functioning and prediction of risk of sepsis in parallel, adding value in patient management. The method used in the present embodiments has many advantages. Since organ dysfunction is a major event in the continuum of sepsis, monitoring organ function in the context of sepsis may give clinicians specific information about the organs that are at highest risk and what will be its impact on the patient becoming septic. Additionally, in the case when there is already a septic event, the parallel monitoring of organs may inform clinicians which organ is at the highest risk due to the sepsis event. Hence, parallel monitoring of organs and sepsis gives clinicians actionable items by zooming into the state of organ health.

[0041] In the above detailed description, reference is made to the accompanying drawings that form a part thereof, and illustrate the best mode presently contemplated for carrying out the present embodiments. However, such description should not be considered as any limitation of scope of the present unit. The structure thus conceived in the present description is susceptible to numerous modifications and variations; all the details may be replaced with elements having technical equivalence.

[0042] As used in the present specification, the term “machine learning” refers broadly to an artificial intelligence technique in which a behavior of a computer evolves based on empirical data. In some cases, input empirical data may come from databases and yield patterns or predictions thought to be features of the mechanism that generated the data. Further, a major focus of machine learning is the design of algorithms that recognize complex patterns and makes intelligent decisions based on input data. Machine learning may incorporate a number of methods and techniques such as: supervised learning, unsupervised learning, reinforcement learning, multivariate analysis, case-based reasoning, backpropagation, and transduction. The Machine Learning (ML) techniques used in the present embodiments include Bayesian models, probabilistic graphical models, neural networks, or tree-based algorithms including ensemble approaches using stacking, voting, and boosting, etc. The ensemble approaches used in tree-based algorithms may include, but are not limited to, Random Forest, XGBoost, LightGBM, CatBoost, and AdaBoost.

[0043] Implementations described hereinabove provide a system for data model management, data model deployment, data model feedback collection, data model retraining, etc. in support of applications executable on the edge system. Data models may be selected and deployed based on characteristics shared between the edge system and other edge systems, and / or the cloud server. Information received from the edge system may be used to update / re-train instances of data models, and the data models may be tracked, documented, and stored such that data models may be specifically managed and customized for a single edge system, a group of edge systems, etc. Data models are continuously or periodically monitored for accuracy, updated based on information, and deployed to various edge systems.

[0044] The elements and features recited in the appended claims may be combined in different ways to produce new claims that likewise fall within the scope of the present invention. Thus, whereas the dependent claims appended below depend from only a single independent or dependent claim, it is to be understood that these dependent claims may, alternatively, be made to depend in the alternative from any preceding or following claim, whether independent or dependent. Such new combinations are to be understood as forming a part of the present specification.

[0045] While the present invention has been described above by reference to various embodiments, it should be understood that many changes and modifications may be made to the described embodiments. It is therefore intended that the foregoing description be regarded as illustrative rather than limiting, and that it be understood that all equivalents and / or combinations of embodiments are intended to be included in this description.

Claims

WE CLAIM:

1. A method for predicting risk of sepsis, the method comprising: training a multi-parameter data model (302) on health parameters related to one or more of multiple organ systems (304), sepsis specific parameters, and routine parameters of a plurality of individuals; training one or more data models (306), each data model of the one or more data models (306) being trained on a plurality of health parameters related to an organ system (308) of the plurality of individuals, wherein the organ system (308) is susceptible to sepsis or failure of which indicates sepsis in the plurality of individuals; merging one or more learnings of the multi-parameter data model (302) and health parameters used for training the multi-parameter data model (302) with learnings of the one or more data models (306), such that an aggregated data model (310) is obtained; and executing the aggregated data model (310) for providing an output indicative of a risk of sepsis in an individual of the plurality of individuals.

2. The method of claim 1, further comprising: training the one or more data models (306) using one or both intermediary learnings of the multi-parameter model (302); and outputting the multi-parameter data model (302).

3. The method of claim 1, wherein the merging of the one or more learnings is performed by early fusion, mid fusion, or late fusion.

4. The method of claim 1, wherein the one or more data models (306) are executed upon availability of one or more health parameters related to one or more organ systems (308) associated with the one or more data models (306).

5. The method of claim 1, wherein the organ systems include a hepatic system, a renal system, a cardiovascular system, a coagulation system, a respiratory system, or any combination thereof.

6. The method of claim 1, wherein the one or more learnings of the multi-parameter model (302) include intermediary learnings and a final learning of the multi-parameter model (302).

7. The method of claim 1, further comprising monitoring health parameters related to one or more of the organ systems (308) using the one or more data models (306) based on the risk of sepsis predicted by the aggregated data model (310).

8. The method of claim 7, wherein the monitoring of health parameters related to the one or more of the organ systems (308) is performed based on learnings of the aggregated data model (310), a user input, and availability of the health parameters related to one or more of the organ systems (308).

9. The method of claim 7, wherein the monitoring of the health parameters related to the one or more of the organ systems (308) is performed at predefined intervals.

10. The method of claim 7, wherein the monitoring of the health parameters related to one or more of the organ systems (308) is performed for an organ system determined to be present at a highest risk from sepsis.

11. A method for monitoring health of organ systems, the method comprising: training a multi-parameter data model (302) on health parameters related to one or more of multiple organ systems (304) of a plurality of individuals; training one or more data models (306), each model of the one or more data models (306) being trained on a plurality of health parameters related to an organ system (308) of the plurality of individuals, wherein the organ system (308) is susceptible to sepsis, or failure of the organ system (308) indicates sepsis in the plurality of individuals; merging one or more learnings of the multi-parameter data model (302) and the health parameters used for training the multi-parameter data model with learnings of the one or more data models (306) to obtain an aggregated data model (310); executing the aggregated data model (310) for providing an output indicative of a risk of sepsis in an individual of the plurality of individuals; andmonitoring health parameters related to one or more of the organ systems (308) using the one or more data models (306) based on the risk of sepsis predicted by the aggregated data model (310).

12. The method of claim 11, further comprising training the one or more data models using intermediary learnings of the multi-parameter model (302), an output of the multiparameter data model (302), or the intermediary learnings of the multi-parameter model (302) and the output of the multi-parameter data model (302).

13. The method of claim 11, wherein the one or more data models (306), the multiparameter data model (302), and the aggregated data model (310) are trained using one or more Machine Learning (ML) techniques including Bayesian models, probabilistic graphical models, neural networks, and tree-based algorithms.

14. The method of claim 11, wherein the merging of the learnings is performed by early fusion, mid fusion, or late fusion.

15. The method of claim 11, wherein the one or more data models are executed upon availability of the one or more health parameters related to one or more of the organ systems (308) associated with the one or more data models (306).

16. The method of claim 11, wherein the organ systems include a hepatic system, a renal system, a cardiovascular system, a coagulation system, a respiratory system, or any combination thereof.

17. The method of claim 11, wherein the learnings of the multi-parameter model (302) include intermediary learnings and a final learning of the multi-parameter model (302).

18. The method of claim 11, wherein the monitoring of health parameters related to one or more of the organ systems (308) is performed based on learnings of the aggregated data model (310), a user input, and availability of the health parameters related to one or more of the organ systems (308).

19. The method of claim 11, wherein the monitoring of the health parameters related to one or more of the organ systems (308) is performed at predefined intervals.

20. The method of claim 11, wherein the monitoring of the health parameters related to one or more of the organ systems (308) is performed for an organ system determined to be present at a highest risk from sepsis.

Images