Systems, apparatuses, and methods for monitoring organ health

Abstract

A system for training at least one graph neural network (GNN) for monitoring organ health includes at least one memory configured to store instructions and at least one processor configured to execute the instructions to cause the system to obtain a plurality of parameters from healthy individuals and diseased individuals, perform causal discovery to determine one or more causal relationships between the plurality of organ parameters, train the at least one GNN with the one or more causal relationships to create a trained GNN configured to output a multi-organ health status based on patient parameters. The system is also configured to train at least one acute event prediction model to predict at least one acute or adverse event for at least one organ

Description

Attorney Docket No.: 32860U-004101-WO-POASYSTEMS, APPARATUSES, AND METHODS FOR MONITORING ORGAN HEALTHCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is a PCT International Application of U.S. Provisional Application No. 63 / 753,745, filed on February 4, 2025, the entire contents of which are incorporated herein by reference.TECHNICAL FIELD

[0002] The present disclosure relates to systems, apparatuses, and / or methods for monitoring the health of multiple organs and predicting acute adverse events.BACKGROUND

[0003] Organ health may be interconnected and involve multidirectional interactions among various organs of an individual.SUMMARY

[0004] At least one example embodiment relates to a system for training at least one graph neural network (GNN) for monitoring organ health. The system may include at least one memory configured to store instructions and at least one processor configured to execute the instructions to cause the system to obtain a plurality of first parameter values of parameters from healthy individuals and diseased individuals, perform causal discovery to determine one or more causal relationships between the plurality of parameters, and train the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi-organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.Attorney Docket No.: 32860U-004101-WO-POA

[0005] In at least one example embodiment, the performing the causal discovery may include creating a causal structure between the plurality of parameters from the healthy individuals and the diseased individuals based on the first parameter values. In at least one example embodiment, the causal structure may be a directed acyclic graph. In at least one example embodiment, the training the at least one GNN may include inputting the causal structure into the at least one GNN.

[0006] In at least one example embodiment, the at least one GNN may include a plurality of GNNs each trained with an organ specific model.

[0007] In at least one example embodiment, the at least one GNN may include a plurality of task-specific branches of linear layers for multiple tasks specific to each organ of a plurality of organs. In at least one example embodiment, the at least one GNN may be trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.

[0008] In at least one example embodiment, the multi-organ health status may include a health score or an end-organ risk score for at least one organ for the patient. In at least one example embodiment, the system may be further configured to output an alert based on a threshold and at least one of the health score or the end-organ risk score.

[0009] In at least one example embodiment, the plurality of parameters may include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

[0010] In at least one example embodiment, the system may be further configured to train at least one acute event prediction model to predict at least one acute or adverse event for at least one organ. In at least one example embodiment, the at least one acute event prediction model may be trained with the plurality of parameters and multi-organ health status data.Attorney Docket No.: 32860U-004101-WO-POA

[0011] Also described herein is a system for monitoring multi-organ health. The system may include at least one memory configured to store instructions and at least one processor configured to execute the instructions to cause the system to obtain a plurality of first parameter values of a patient, form a causal structure from the plurality of parameter values, the causal structure being based on a causal discovery indicative of one or more causal relationships between the parameters, and input the causal structure into a graph neural network (GNN) to obtain a multi -organ health status of the patient from the GNN.

[0012] In at least one example embodiment, the GNN may be trained based on the causal discovery.

[0013] In at least one example embodiment, the system may be further configured to input the multi-organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ. In at least one example embodiment, the system may be further configured to input the plurality of parameters into the acute event prediction model.

[0014] In at least one example embodiment, the plurality of parameters may include one or more of lab parameters, vitals, or demographic information for the individual.

[0015] Also described herein is a method for training at least one graph neural network (GNN) for monitoring organ health. The method may include obtaining a plurality of first parameter values of parameters from healthy individuals and diseased individuals, performing causal discovery to determine one or more causal relationships between the plurality of parameters, and training the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi -organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.

[0016] In at least one example embodiment, the performing the causal discovery may include creating a causal structure between the plurality of parameters from the healthyAttorney Docket No.: 32860U-004101-WO-POA individuals and the diseased individuals based on the first parameter values. In at least one example embodiment, the causal structure may be a directed acyclic graph. In at least one example embodiment, the training the at least one GNN may include inputting the causal structure into the at least one GNN.

[0017] In at least one example embodiment, the at least one GNN may include a plurality of GNNs each trained with an organ specific model.

[0018] In at least one example embodiment, the at least one GNN may include a plurality of task-specific branches of linear layers for multiple tasks specific to each organ of a plurality of organs. In at least one example embodiment, the at least one GNN may be trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.

[0019] In at least one example embodiment, the multi-organ health status may include a health score or an end-organ risk score for at least one organ for the patient.

[0020] In at least one example embodiment, the method may further include outputting an alert based on a threshold and at least one of the health score or the end-organ risk score.

[0021] In at least one example embodiment, the plurality of parameters may include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

[0022] In at least one example embodiment, the method may further include training at least one acute event prediction model to predict at least one acute or adverse event for at least one organ. In at least one example embodiment, the at least one acute event prediction model is trained with the plurality of parameters and multi-organ health status data.

[0023] Also described herein is a method for monitoring multi-organ health. The method may include obtaining a plurality of first parameter values of parameters from a patient, forming a causal structure from the plurality of parameter values, the causal structure beingAttorney Docket No.: 32860U-004101-WO-POA based on a causal discovery indicative of one or more causal relationships between the parameters, and inputting the causal structure into a graph neural network (GNN) to obtain a multi-organ health status of the patient from the GNN.

[0024] In at least one example embodiment, the GNN may be trained based on the causal discovery.

[0025] In at least one example embodiment, the method may further include inputting the multi-organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ. In at least one example embodiment, the method may further include inputting the plurality of parameters into the acute event prediction model.

[0026] In at least one example embodiment, the plurality of parameters may include one or more of lab parameters, vitals, or demographic information for the individual.

[0027] Also described herein is a non-transitory computer-readable storage medium storing computer-executable instruction that, when executed by at least one processor of a system, cause the system to perform a method for training at least one graph neural network (GNN) for monitoring organ health. The method may include obtaining a plurality of first parameter values of parameters from healthy individuals and diseased individuals, performing causal discovery to determine one or more causal relationships between the plurality of parameters, and training the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi-organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.

[0028] In at least one example embodiment, the performing the causal discovery may include creating a causal structure between the plurality of parameters from the healthy individuals and the diseased individuals based on the first parameter values. In at least one example embodiment, the causal structure may be a directed acyclic graph. In at least oneAttorney Docket No.: 32860U-004101-WO-POA example embodiment, the training the at least one GNN may include inputting the causal structure into the at least one GNN.

[0029] In at least one example embodiment, the at least one GNN may include a plurality of GNNs each trained with an organ specific model.

[0030] In at least one example embodiment, the at least one GNN may include a plurality of task-specific branches of linear layers for multiple tasks specific to each organ of a plurality of organs. In at least one example embodiment, the at least one GNN may be trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.

[0031] In at least one example embodiment, the multi-organ health status may include a health score or an end-organ risk score for at least one organ for the patient.

[0032] In at least one example embodiment, the method may further include outputting an alert based on a threshold and at least one of the health score or the end-organ risk score.

[0033] In at least one example embodiment, the plurality of parameters may include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

[0034] In at least one example embodiment, the method may further include training at least one acute event prediction model to predict at least one acute or adverse event for at least one organ. In at least one example embodiment, the at least one acute event prediction model is trained with the plurality of parameters and multi-organ health status data.

[0035] Also described herein is a non-transitory computer-readable storage medium storing computer-executable instruction that, when executed by at least one processor of a system, cause the system to perform a method for monitoring multi-organ health. The method may include obtaining a plurality of first parameter values of parameters from a patient, forming a causal structure from the plurality of parameter values, the causal structure beingAttorney Docket No.: 32860U-004101-WO-POA based on a causal discovery indicative of one or more causal relationships between the parameters, and inputting the causal structure into a graph neural network (GNN) to obtain a multi-organ health status of the patient from the GNN.

[0036] In at least one example embodiment, the GNN may be trained based on the causal discovery.

[0037] In at least one example embodiment, the method may further include inputting the multi-organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ. In at least one example embodiment, the method may further include inputting the plurality of parameters into the acute event prediction model.

[0038] In at least one example embodiment, the plurality of parameters may include one or more of lab parameters, vitals, or demographic information for the individual.BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The various features and advantages of the non-limiting embodiments herein may become more apparent upon review of the detailed description in conjunction with the accompanying drawings. The accompanying drawings are merely provided for illustrative purposes and should not be interpreted to limit the scope of the claims. The accompanying drawings are not to be considered as drawn to scale unless explicitly noted. For purposes of clarity, various dimensions of the drawings may have been exaggerated.

[0040] FIG. 1 is a block diagram of a system that may implement the various methods described herein according to at least one example embodiment.

[0041] FIG. 2 illustrates a method of training a graph neural network and deploying the graph neural network to monitor organ health according to at least one example embodiment.

[0042] FIG. 3 illustrates a directed acyclic graph according to at least one example embodiment.Attorney Docket No.: 32860U-004101-WO-POA

[0043] FIG. 4 illustrates a graph neural network according to at least one example embodiment.

[0044] FIG. 5 illustrates a flow chart of a method of training a graph neural network for monitoring organ health according to at least one example embodiment.

[0045] FIG. 6 illustrates a flow chart of a method of implementing the graph neural network of FIG. 5 according to at least one example embodiment.DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

[0046] Independent of the grammatical term usage, individuals with male, female or other gender identities are included within the term.

[0047] Some detailed example embodiments are disclosed herein. However, specific structural and functional details disclosed herein are merely representative for purposes of describing some example embodiments. Example embodiments may, however, be embodied in many alternate forms and should not be construed as limited to only example embodiments set forth herein.

[0048] Accordingly, while example embodiments are capable of various modifications and alternative forms, example embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that there is no intent to limit an example embodiment to the particular forms disclosed, but to the contrary, example embodiments are to cover all modifications, combinations, equivalents, and alternatives falling within the scope of an example embodiment. Like numbers refer to like elements throughout the description of the figures.

[0049] It should be understood that when an element or layer is referred to as being “on,” “connected to,” “coupled to,” or “covering” another element or layer, it may be directly on, connected to, coupled to, or covering the other element or layer or intervening elements orAttorney Docket No.: 32860U-004101-WO-POA layers may be present. In contrast, when an element is referred to as being “directly on,” “directly connected to,” or “directly coupled to” another element or layer, there are no intervening elements or layers present. Like numbers refer to like elements throughout the specification. As used herein, the term “and / or” includes any and all combinations of one or more of the associated listed items.

[0050] It should be understood that, although the terms first, second, third, etc. may be used herein to describe various elements, regions, layers and / or sections, these elements, regions, layers, and / or sections should not be limited by these terms. These terms are only used to distinguish one element, region, layer, or section from another region, layer, or section. Thus, a first element, region, layer, or section discussed below could be termed a second element, region, layer, or section without departing from the teachings of example embodiment.

[0051] The terminology used herein is for the purpose of describing various example embodiment only and is not intended to be limiting of example embodiment. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “includes,” “including,” “comprises,” and / or “comprising,” when used in this specification, specify the presence of stated features, integers, steps, operations, and / or elements, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements and / or groups thereof.

[0052] When the words “about” and “substantially” are used in this specification in connection with a numerical value, it is intended that the associated numerical value include a tolerance of ±10% around the stated numerical value, unless otherwise explicitly defined. Moreover, when the terms “generally” or “substantially” are used in connection with geometric shapes, it is intended that precision of the geometric shape is not required but that latitude for the shape is within the scope of the disclosure. Furthermore, regardless of whether numericalAttorney Docket No.: 32860U-004101-WO-POA values or shapes are modified as “about,” “generally,” or “substantially,” it will be understood that these values and shapes should be construed as including a manufacturing or operational tolerance (e.g., ±10%) around the stated numerical values or shapes.

[0053] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which example embodiment belong. It will be further understood that terms, including those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0054] FIG. 1 illustrates a system 100 that may implement the methods and algorithms described herein. For example, the system 100 may be used for causal discovery, generating a causal structure, generating and / or implementing an organ health prediction model, and / or training and / or implementing an adverse event prediction model. In at least one example embodiment, the system 100 includes at least one processor 102, at least one memory 104, and at least one communication interface 106. The at least one memory 104 may be configured to store instructions that may be executed by the at least one processor 102 to cause the system 100 to perform one or more functions.

[0055] As will be appreciated, depending on the implementation of the system 100, the system 100 may include additional components. However, it is not necessary that all of these generally conventional components be shown in order to disclose the illustrative example embodiment. For example purposes, the system 100 will be discussed with regard to the at least one processor 102. However, it should be understood that the system 100 may include one or more processors or other processing circuitry, such as one or more Application Specific Integrated Circuits (ASICs).Attorney Docket No.: 32860U-004101-WO-POA

[0056] The at least one processor 102 may include, but is not limited to, a central processing unit (CPU), an arithmetic logic unit (ALU), a graphics processing unit (GPU), an application processor (AP), a digital signal processor (DSP), a microcomputer, a field programmable gate array (FPGA), and programmable logic unit, application-specific integrated circuit (ASIC), a neural network processing unit (NPU), an Electronic Control Unit (ECU), a quantum computer, and the like. In some example embodiments, the processing circuitry may include a non-transitory computer readable storage medium or device (e.g., memory), for example a solid state drive (SSD), storing a program of instructions, and a processor (e.g., CPU) configured to execute the program of instructions to implement the functionality and / or methods performed by some or all of the systems according to any of the example embodiments.

[0057] The at least one memory 104 may be a computer readable storage medium that generally includes a random access memory (RAM), read only memory (ROM), and / or a permanent mass storage device, such as a disk drive. The at least one may also store an operating system and any other routines / modules / applications for providing the functionalities of the system 100 to be executed by the at least one processor 102. These software components may also be loaded from a separate computer readable storage medium into the at least one using a drive mechanism (not shown). Such separate computer readable storage medium may include a disc, tape, DVD / CD-ROM drive, memory card, or other like computer readable storage medium (not shown). In some example embodiments, software components may be loaded into the at least one memory 104 via one of the at least one communication interface 106, rather than via a computer readable storage medium.

[0058] The at least one processor 102 or other processing circuitry may be configured to carry out instructions of a computer program by performing the arithmetical, logical, and input / output operations of the system. Instructions may be provided to the at least one processor 102 by the at least one memory 104.Attorney Docket No.: 32860U-004101-WO-POA

[0059] The at least one communication interface 106 may be wired and may include components that interface the at least one processor 102 with the other input / output components. As will be understood, the at least one communication interface 106 and programs stored in the at least one memory 104 to set forth the special purpose functionalities of the system 100 will vary depending on the implementation of the system 100.

[0060] The at least one communication interface 106 may also include one or more user input devices (e.g., a keyboard, a keypad, a mouse, or the like) and user output devices (e.g., a display, a speaker, or the like).

[0061] As disclosed herein, the term "storage medium," "computer readable storage medium" or "non-transitory computer readable storage medium” may represent one or more devices for storing data, including read only memory (ROM), random access memory (RAM), magnetic RAM, core memory, magnetic disk storage mediums, optical storage mediums, flash memory devices and / or other tangible machine-readable mediums for storing information. The term "computer-readable medium" may include, but is not limited to, portable or fixed storage devices, optical storage devices, and various other mediums capable of storing, containing or carrying instruction(s) and / or data.

[0062] Furthermore, example embodiments may be implemented by hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware or microcode, the program code or code segments to perform the necessary tasks may be stored in a machine or computer readable medium such as a computer readable storage medium. When implemented in software, a processor or processors will perform the necessary tasks. For example, as mentioned above, according to one or more example embodiments, at least one memory may include or store a computer program or computer program code, and the at least one memory and the computer program code may be configured to, with at least one processor, the methodsAttorney Docket No.: 32860U-004101-WO-POA described herein. Additionally, the processor, memory and example algorithms, encoded as computer program code, serve as means for providing or causing performance of operations discussed herein. At least one other example embodiment may include a computer program including program segments or instructions that, when executed by at least one processor of a system, cause the system to perform the functions and methods described herein.

[0063] A code segment of a computer program may represent a procedure, function, subprogram, program, routine, subroutine, module, software package, class, or any combination of instructions, data structures or program statements. A code segment may be coupled to another code segment or a hardware circuit by passing and / or receiving information, data, arguments, parameters or memory contents. Information, arguments, parameters, data, etc. may be passed, forwarded, or transmitted via any suitable technique including memory sharing, message passing, token passing, network transmission, etc.

[0064] The systems and methods described herein provide improved disease prediction methods. In particular, current machine learning and deep learning algorithms lack generalizability and explainability which poses challenges in clinical settings. The systems and methods described herein utilize causal machine learning, focused on cause-and-effect relationships to provide improved disease prediction models. As described herein, linear causal relationships may be defined among clinical parameters using causal discovery with a linear non-Gaussian acyclic model. These linear causal relationships may be used to define a causal structure that may be used to train a simplicial graph neural network which may capture nonlinear relationships between the clinical parameters. The systems and methods described herein have been tested and validated with different population datasets which illustrates generalizability and explainability for early sepsis prediction as a case study.

[0065] FIG. 2 illustrates a method for monitoring multi-organ health in accordance with at least one example embodiment. In at least one example embodiment, the method may begin atAttorney Docket No.: 32860U-004101-WO-POA S202 where parameters and values of the parameters are obtained from healthy individuals and from diseased individuals. The parameters and the values of the parameters may be obtained from lab data related to one or more organs of each individual, vitals of each individual, and demographic information of each individual. In at least one example embodiment, the parameters may include routine lab parameters such as one or more of: lactate, alanine aminotransferase (“ALT”), partial thromboplastin time (“PTT”), international normalized ratio (“INR”), globulin, monocyte count, hematocrit, magnesium, pH, sodium, urea nitrogen, triglycerides, creatinine, bilirubin total, cholesterol total, aspartate aminotransferase (“AST”), mean corpuscular hemoglobin concentration (“MCHC”), percent eosinophils, phosphate, C-reactive protein, prothrombin time (“PT”), neutrophil count, basophil count, glucose, D-dimer, albumin, percent neutrophils, percent lymphocytes, calcium total, chloride, % hemoglobin Ale, high density lipoprotein (“HDL”), red blood cell (“RBC”) count, red blood cell distribution width (“RDW”), white blood cell (“WBC”) count, troponin T, mean corpuscular hemoglobin (“MCH”), percent monocytes, eosinophil count, alkaline phosphatase (“ALP”), fibrinogen functional, hemoglobin, low density lipoprotein (“LDL”), mean corpuscular volume (“MCV”), percent basophils, lymphocyte count, platelet count, protein total, potassium, and bicarbonate. The parameters may also include vitals such as diastolic blood pressure (“DBP”), systolic blood pressure (“SBP”), oxygen (O2) saturation, respiratory rate, heart rate, and temperature as well as demographic parameters such as age and gender. The parameters are not limited herein. In at least one example embodiment, the parameters may be obtained from data sources that are publicly accessible. For example, the parameters may be obtained from a MIMIC-IV dataset in at least one example embodiment.

[0066] After the parameters are obtained, at S204 causal discovery may be performed which may output a causal structure 205. In at least one example embodiment, causal discovery may be performed using a Linear Non-Gaussian Acyclic Model (“LiNGAM”). AlthoughAttorney Docket No.: 32860U-004101-WO-POA embodiments herein are described with reference to causal discovery performed using LiNGAM, other causal discovery algorithms may be used. For example, one or more of Peter-Clark (“PC”), Fast Causal Inference (“FQ”), Greedy Equivalence Search (“GES”), or Max-Min Hill Climbing (“MMHC”) may be used to perform causal discovery. Causal discovery may establish a cause and effect relationship between one or more variables. In particular, inputting the obtained parameters and the values of the obtained parameters into the LiNGAM may output the causal structure 205 such as a directed acyclic graph. The LiNGAM may perform causal discovery by leveraging an assumption that the input data generated by a linear structure with non-Gaussian noise. The model may use statistical techniques like Independent Component Analysis (ICA) to identify underlying independent sources driving the observed data. By examining the non-Gaussianity of the residuals after modeling linear dependencies, LiNGAM can infer the causal directionality between variables, uncovering the acyclic structure of the input data. This method may distinguish causal relationships from mere correlations.

[0067] The directed acyclic graph may include one or more nodes connected via one or more edges. The one or more nodes may represent the parameters obtained at S202. When a first node is connected to a second node via an edge, the edge represents a causal influence between the nodes. Thus, one of the first node or the second node may impact the other of the first node or the second node. This may indicate a cause-and-effect relationship between connected nodes of the directed acyclic graph. Each node may be connected to as many nodes as are determined to be in a cause-and-effect relationship with that node.

[0068] At S206, the causal structure, such as the directed acyclic graph, is input into a Graph Neural Network (GNN). In at least one example embodiment, the GNN may be an organ health prediction model. The GNN may include multiple layers. Each layer may incorporate activation functions to introduce non-linearity. The GNN may be trained using a messagepassing mechanism, where each node (representing a clinical parameter) is configured toAttorney Docket No.: 32860U-004101-WO-POA aggregate information from neighboring nodes to update its own representation. These updated representations are then passed to the nodes in the next layer. During classification, the final node representations, after multiple layers of message passing, are used as input to a classification head (often a fully connected layer) that produces an output. The GNN may be trained iteratively by optimizing a loss function that compares the predicted classification with the ground truth, allowing the GNN to adjust its parameters and improve its predictions over time. In particular, the causal structure 205 with values of the parameters for patients from the data obtained at S202 may be input with the ground truth for each patient into the GNN. In at least one example embodiment, data for each patient may be input into the GNN individually and independently from additional data for other patients from the healthy individual and the diseased individual data sets. The ground truth for each patient may include an organ status for one or more organs of the individual. For example, if the patient has a chronic condition or other disease or condition indicating lack of health for an organ, the ground truth may be indicated as a “1” while if the patient does not have a chronic condition or other disease or condition indicating poor health for an organ, the ground truth may be indicated as a “0”. In at least one example embodiment, when a causal structure / graph is fed into the GNN for classification, the GNN may guide the information flow across the connections / edges that encode causal relationships, ensuring that only relevant relationships are used in the prediction process.

[0069] In at least one example embodiment, the GNN may be at least one simplicial GNN which may be configured to generalize operations and concepts of GNNs to simplicial concepts. The GNN may be trained for multiple disease conditions, including early disease stages, for one or more organs. Thus, the GNN may be utilized for determining organ health in general rather than only providing information about a particular disease. The GNN may utilize the directed acyclic graph to inform its learning process. For example, using a causal structureAttorney Docket No.: 32860U-004101-WO-POA such as a directed acyclic graph in its learning process may allow the GNN to account for confounding variables and ignore spurious relationships that occur without a causal basis. The GNN may use the edges of the directed acyclic graph to understand and propagate information through the directed acyclic graph which may capture both linear and non-linear relationships between the parameters. Linear relationships may be determined directly from edges connecting two nodes. Non-linear relationships in a directed acyclic graph may be captured through transformations applied across layers during message passing in the GNN. The layers may include learnable weights and non-linear activation functions which may enable the GNN to combine and propagate information from connected nodes which may reflect complex dependencies beyond direct edges. The GNN may also be configured to extract features from the directed acyclic graph for classification.

[0070] In at least one example embodiment, the GNN trained with the directed acyclic graph may be trained with either organ-specific models or with a multi-task classification approach.

[0071] In at least one example embodiment, a plurality of organ-specific models may be developed using GNNs trained with the directed acyclic graph. Each of the organ-specific models may be developed for a specific organ and may be configured to output organ health scores for the particular organ. The outputs of the organ-specific models may then be combined for a multi-organ output. Each organ-specific model may be trained using data from one or more patients related to that particular organ. For example, patient data for any patient with chronic kidney disease, including early and late stage disease, and any other kidney disease at any stage of the disease is used to train a kidney-specific model. The organ health scores that are output by the organ-specific models may be used for predicting an adverse event as described in further detail herein. The final output may be scored based on one or more of accuracy, sensitivity, specificity, positive predictive value (“PPV”), negative predictive value (“NPV”), area under the receiver operating characteristic (“AUROC”), and area under theAttorney Docket No.: 32860U-004101-WO-POA precision-recall curve (“AUPRC”). In at least one example embodiment, the ensemble of organ-specific models returned the following scores when applied to a test data set for kidney health: accuracy: 80.2%, sensitivity: 91.27%, specificity: 72.97%, PPV: 77.05%, NPV: 89.36%, AUROC: 91.35%, and AUPRC: 90.59%.

[0072] In at least one example embodiment, a GNN is augmented by task-specific branches including one or more of linear layers, convolutional layers, or additional GNN layers for multiple tasks specific to one or more organs. Each branch may include task-specific layers that process the shared features to produce task-specific outputs. The at least one GNN is trained simultaneously for multiple tasks by jointly optimizing the loss for each task. The loss for each task may be measured using a task specific loss function such as one or more of a cross-entropy loss function, a Kullback-Leibler (“KL”) loss function, or a Focal Loss function. By training the GNN for multiple tasks, the GNN may be able to generalize better and improve the performance on individual tasks. This is achieved by leveraging shared representations and learnings between tasks. In at least one example embodiment, the ensemble of organ-specific models returned the following scores when applied to a test data set for kidney health and liver health: Kidney accuracy: 74.23%, Kidney sensitivity: 75.91%, Kidney specificity: 73.02%, Kidney PPV: 67.10%, Kidney NPV: 80.70%, Kidney AUROC: 74.46%, and Kidney AUPRC: 61.06%; Liver accuracy: 78.83%, Liver sensitivity: 83.09%, Liver specificity: 71.43%, Liver PPV: 83.50%, Liver NPV: 70.83%, Liver AUROC: 77.26%, and Liver AUPRC: 80.11%.

[0073] In at least one example embodiment, a training process for the GNN may include the steps S202 to S206. Upon completion of the training process, the GNN may be deployed to output multi-organ health scores for input patient data. The multi-organ health scores may be referred to herein as multi-organ health status.

[0074] After the GNN is trained, parameters 207 and values of the parameters 207 from an individual may be obtained. The parameters 207 may include one or more of the parametersAttorney Docket No.: 32860U-004101-WO-POA described above at step S202 while the values of the parameters 207 may be individual to the patient. The values of the parameters 207 may be input into the causal structure 205 determined from the causal discovery performed at S204.

[0075] The causal structure 205 including the values of the parameters 207 may then be input into the trained GNN. The trained GNN may be configured to output multi-organ health scores 209 for the patient. The multi-organ health scores may be scores for each of a plurality of organs that may be determined from the parameters 207, the values of the parameters 207, and the causal structure 205. For example, the GNN may output a health score for a liver equal to 0.4, a kidney equal to 0.7, and a heart equal to 0.8. A lower health score may indicate that an organ is in poor health which a higher health score may indicate more optimal functioning of an organ. Thus, based on the example health scores given above, the liver may be in poor health while the kidney and the heart may not be in poor health.

[0076] At S208, multi-organ health scores derived from the GNN for the patient may be used for predicting one or more acute and / or adverse events 211, such as end-organ failure including heart failure, stroke, acute kidney injury, acute liver failure, etc. with an acute event prediction model. In at least one example embodiment, the values of the parameters 207 for the individual may also be input into the acute event prediction model to predict one or more acute and / or adverse events.

[0077] In at least one example embodiment, the acute event prediction model may be a machine learning model such as a linear regressor model, a neural network, or a supervised machine learning algorithm which may be configured to output end-organ damages from the health score input. The adverse event prediction model may be trained using organ health scores derived from the GNN, with or without incorporating patient data as additional input. The training process may include a binary classification task, with patients identified as having adverse events labeled as positive. Depending on the approach, the adverse event predictionAttorney Docket No.: 32860U-004101-WO-POA model may be trained either as a single unified model to predict all adverse events or as separate models, trained for each possible acute or adverse event.

[0078] Using the example health scores above, the adverse event prediction model may output a risk of acute liver failure equal to 0.8, a risk of stroke equal to 0.2, a risk of acute kidney injury equal to 0.3, and a risk of heart failure equal to 0.4. In at least one example embodiment, a threshold may be identified for either the organ health scores and / or the endorgan damage scores. Scores over the threshold score may be identified as organs or events of greater risk. For example, if the threshold for end-organ damage is defined as greater than 0.7, the machine learning model may output acute liver failure as being greater than the threshold.

[0079] FIG. 3 is a causal structure 300 obtained using LiNGAM as described above. The causal structure 300 is an example embodiment of the causal structure 205 described above in FIG. 2. The causal structure 300 may be a directed acyclic graph. The causal structure 300 includes a plurality of nodes 302 and edges 304. Each of the plurality of nodes 302 may be coupled to one or more nodes 302 via the edges 304. Each node of the plurality of nodes 302 may represent a parameter. Example parameters include hemoglobin, temperature, respiratory rate, gender, Basophils percentage, Eosinophils percentage, white blood cell count, and heartrate. The parameters are not limited herein and may include any of the parameters listed above with respect to FIG. 2. Each of the edges 304 may include an associated number. The number may represent a causal influence or direct effect of one node of the plurality of nodes 302 on a connected node of the plurality of nodes 302. In at least one example embodiment, the numbers indicative of the causal influence between nodes may be determined by analyzing a non-Gaussian nature of the data to identify a directionality of the causal relationships. The causal structure 300 may define cause-and-effect relationships between the nodes which may enable a GNN to determine organ health and / or end-organ damages as described above.Attorney Docket No.: 32860U-004101-WO-POA

[0080] In at least one example embodiment, the directionality of the edges, shown by arrows in FIG. 3 and the associated numbers for the edges may be omitted when the causal structure 300 is input into the GNN. The GNN may only receive the plurality of nodes 302 and the edges 304. Further, then patient data such as the parameters 207 is input into the causal structure 300, each node of the plurality of nodes 302 that includes a corresponding patient parameter of the parameters 207 may be replaced by the patient data for that parameter. The causal structure 300 with the patient data may then be input into the GNN to determine the multi-organ health scores 209 for the patient.

[0081] FIG. 4 is an GNN 400 according to at least one example embodiment. The GNN 400 may correspond to the GNN described above at S206 described above with respect to FIG. 2. The GNN may include one or more layers. The one or more layers 402 may be interconnected and may be configured to receive an input and output a final health score. The one or more layers may include hidden layers and may be one or more of linear layers, convolutional layers, or additional GNN layers for multiple tasks specific to one or more organs. The GNN may receive an input 404 which may be fed into a first layer of the one or more layers 402. The GNN may include one or more nodes 406 that may be connected by one or more edges 408. The one or more edges 408 may be configured to connect the one or more nodes 406 within the one or more layers 402 of the GNN. The GNN may be configured to generate an output 410. The output 410 may be a multi-organ health score as described herein.

[0082] FIG. 5 is a flow chart of a method 500 for training a GNN to predict a multi-organ health status. The method 500 may begin at S502 where parameters and values of the parameters are obtained. The parameters and the values of the parameters may be obtained from one or more data sources that may be publicly accessible. The parameters and the values of the parameters may be from one or more healthy and diseased individuals as described above.Attorney Docket No.: 32860U-004101-WO-POA

[0083] At step S504. causal discovery may be performed. As described above, causal discovery may be performed by a model or algorithm that takes the parameters and the values of the parameters as inputs and outputs causal relationships of the parameters. For example, as described above, causal discovery may be performed using LiNGAM which may output a causal structure such as a directed acyclic graph. The directed acyclic graph may include one or more nodes connected via one or more edges.

[0084] At step S506, at least one GNN is trained with the causal relationships determined from the causal discovery. The at least one GNN may be trained as described above with reference to FIG. 2 such that the GNN is configured to output a multi -organ health status when the GNN is deployed after is it trained.

[0085] FIG. 6 is a flow chart of a method 600 for utilizing a trained GNN to predict a multiorgan health status. The method 600 may begin at S602 where patient data such as the parameters 207 and the values of the parameters 207 are obtained for a patient.

[0086] At step S604, the values of the parameters 207 are input into a causal structure such as the causal structure 205 determined during training of the GNN.

[0087] At S606, the GNN is used to generate a multi-organ health status. In at least one example embodiment, the multi -organ health status may be determined for a particular patient after the GNN is trained. The input may include the causal structure determined at S604. Thus, the GNN may be configured to receive the input data of the patient in the form of the causal structure 205 and determine a multi-organ health status such as the multi-organ health scores 209 for that individual.

[0088] At optional step S608, the multi-organ health scores 209 may be input to a machine learning model to predict at least one acute or adverse event for at least one organ. The machine learning model may be the adverse event prediction model described above with respect to FIG. 2. In at least one example embodiment, the values of the parameters 207 may also be inputAttorney Docket No.: 32860U-004101-WO-POA into the machine learning model. As described above, one or more organs may be given a score indicative of their health. The machine learning model may be trained to output acute or adverse events, such as the acute or adverse event predictions 211 described above, that may occur to one or more organs based on the scores for the one or more organs. The output from the machine learning model may be used by a user such as a health provider to intervene and provide one or more of preventive or critical care to treat one or more organs.

[0089] The methods described herein may be implemented by the system 100 described with respect to FIG. 1 in at least one example embodiment. Additionally, in at least one example embodiment, a means for processing may be configured to perform the methods described herein.

[0090] The above-described systems and methods provide improved monitoring of organ health. The organ health scores and end-organ damage scores provide critical insights into organs and cause and effect relationships between organs. The GNN described herein may be applicable in a variety of healthcare applications as it may be able to identify effective combinations of routine parameters or integrate routine parameters with biomarkers for enhanced risk prediction of various organ conditions. For example, in at least one example embodiment, the progression in CKM (Cardio-Kidney-Metabolic) syndrome stages may be predicted based on an output of the GNN. The GNN may not only highlight contributing factors for various organ conditions but may also enable actionable recommendations based on risk profiles.

[0091] For instance, when analyzing the progression of CKM in a preventative health setting, the health scores output from the GNN may be input into an adverse event prediction model to output an acute or adverse event prediction which may indicate a progression of the CKM syndrome. For example, the combination of the GNN and the adverse event prediction model may be configured to identify patients at Stage 2 CKM with a high risk of advancing toAttorney Docket No.: 32860U-004101-WO-POA Stage 3 CKM. This output may be used to flag those patients requiring early interventions. Moreover, the combination of the GNN and the adverse event prediction model may leverage annual health check-up data to monitor organ health which may allow a practitioner to recommend targeted tests which may improve early detection and care. In critical care or in-hospital scenarios, the combination of the GNN and the adverse event prediction model may aid in identifying patients at a higher risk of in-hospital or short-term adverse events, such as cardiac events or end-organ failure (e.g., kidney or liver failure). For example, the combination of the GNN and the adverse event prediction model may flag patients at an elevated risk of progressing to Stage 4 of CKM syndrome, allowing healthcare providers to prioritize care.

[0092] The combination of the GNN and the adverse event prediction model may improve health care including early disease detection and prevention by outputting health scores and / or acute or adverse event predictions which may be used to recommend unmeasured but informative tests that are personalized for an individual. For example, individuals with a higher risk of cardiac events may be advised to undergo biomarker tests such as NT-proBNP or hsTroponin, offering a deeper understanding of their cardiovascular health and aiding in precise intervention planning. Thus, the combination of the GNN and the adverse event prediction model may enable prediction and personalized recommendations which may empower healthcare professionals to intervene early, improve outcomes, and optimize resource use in both preventive and critical care environments.

[0093] Appendix A, incorporated by reference in U.S. Provisional Application No.63 / 753,745, filed on February 4, 2025, is incorporated herein by reference in its entirety.

[0094] Example embodiments have been disclosed herein, it should be understood that other variations may be possible. Such variations are not to be regarded as a departure from the spirit and scope of the present disclosure, and all such modifications as would be obvious to one skilled in the art are intended to be included within the scope of the following claims.Attorney Docket No.: 32860U-004101-WO-POA NON-LIMITING ILLUSTRATIVE EMBODIMENTS

[0095] The following is a list of non-limiting illustrative embodiments disclosed herein:

[0096] Illustrative embodiment 1 includes a system for training at least one graph neural network (GNN) for monitoring organ health, the system comprises: at least one memory configured to store instructions; and at least one processor configured to execute the instructions to cause the system to obtain a plurality of first parameters values of parameters from healthy individuals and diseased individuals, perform causal discovery to determine one or more causal relationships between the plurality of parameters, and train the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi -organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.

[0097] Illustrative embodiment 2 includes the system of illustrative embodiment 1, wherein the performing the causal discovery includes creating a causal structure between the plurality of parameters from the healthy individuals and the diseased individuals based on the first parameter values.

[0098] Illustrative embodiment 3 includes the system of illustrative embodiment 2, wherein the causal structure is a directed acyclic graph.

[0099] Illustrative embodiment 4 includes the system of any one of illustrative embodiments 2 and 3, wherein the training the at least one GNN includes inputting the causal structure into the at least one GNN.

[0100] Illustrative embodiment 5 includes the system of any one of illustrative embodiments 1, 2, 3, and 4, wherein the at least one GNN includes a plurality of GNNs each trained with an organ specific model.Attorney Docket No.: 32860U-004101-WO-POA

[0101] Illustrative embodiment 6 includes the system of any one of illustrative embodiments 1, 2, 3, 4, and 5, wherein the at least one GNN includes a plurality of task-specific branches of linear layers for multiple tasks specific to each organ of a plurality of organs.

[0102] Illustrative embodiment 7 includes the system of illustrative embodiment 6, wherein the at least one GNN is trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.

[0103] Illustrative embodiment 8 includes the system of any one of illustrative embodiments 1, 2, 3, 4, 5, 6, and 7, wherein the multi-organ health status includes a health score or an end-organ risk score for at least one organ for a patient.

[0104] Illustrative embodiment 9 includes the system of illustrative embodiment 8, wherein the system is further configured to output an alert based on a threshold and at least one of the health score or the end-organ risk score.

[0105] Illustrative embodiment 10 includes the system of any one of illustrative embodiments 1 , 2, 3, 4, 5, 6, 7, 8, and 9, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

[0106] Illustrative embodiment 11 includes the system of any one of illustrative embodiments 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10, wherein the system is further configured to train at least one acute event prediction model to predict at least one acute or adverse event for at least one organ.

[0107] Illustrative embodiment 12 includes the system of illustrative embodiment 11, wherein the at least one acute event prediction model is trained with the plurality of parameters and multi-organ health status data.

[0108] Illustrative embodiment 13 includes a system for monitoring multi -organ health, the system comprises: at least one memory configured to store instructions; and at least oneAttorney Docket No.: 32860U-004101-WO-POA processor configured to execute the instructions to cause the system to obtain a plurality of first parameter values of parameters of a patient, form a causal structure from the plurality of parameter values, the causal structure being based on a causal discovery indicative of one or more causal relationships between the parameters, and input the causal structure into a graph neural network (GNN) to obtain a multi-organ health status of the patient from the GNN.

[0109] Illustrative embodiment 14 includes the system of illustrative embodiment 13, wherein the GNN is trained based on the causal discovery.

[0110] Illustrative embodiment 15 includes the system of any one of illustrative embodiments 13 and 14, wherein the system is further configured to input the multi-organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ.

[0111] Illustrative embodiment 16 includes the system of illustrative embodiment 15, wherein the system is further configured to input the plurality of parameters into the acute event prediction model.

[0112] Illustrative embodiment 17 includes the system of any one of illustrative embodiments 13, 14, 15, and 16, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the individual.

[0113] Illustrative embodiment 18 includes a method for training at least one graph neural network (GNN) for monitoring organ health, the method comprising: obtaining a plurality of first parameter values of parameters from healthy individuals and diseased individuals, performing causal discovery to determine one or more causal relationships between the plurality of parameters, training the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi-organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.Attorney Docket No.: 32860U-004101-WO-POA

[0114] Illustrative embodiment 19 includes the method of illustrative embodiment 18, wherein the performing the causal discovery includes creating a causal structure between the plurality of parameters from the healthy individuals and the diseased individuals based on the first parameter values.

[0115] Illustrative embodiment 20 includes the method of illustrative embodiment 19, wherein the causal structure is a directed acyclic graph.

[0116] Illustrative embodiment 21 includes the method of any one of illustrative embodiments 19 and 20, wherein the training the at least one GNN includes inputting the causal structure into the at least one GNN.

[0117] Illustrative embodiment 22 includes the method of any one of illustrative embodiments 18, 19, 20, and 21, wherein the at least one GNN includes a plurality of GNNs each trained with an organ specific model.

[0118] Illustrative embodiment 23 includes the method of any one of illustrative embodiments 18, 19, 20, 21, and 22, wherein the at least one GNN includes a plurality of taskspecific branches of linear layers for multiple tasks specific to each organ of a plurality of organs.

[0119] Illustrative embodiment 24 includes the method of illustrative embodiment 23, wherein the at least one GNN is trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.

[0120] Illustrative embodiment 25 includes the method of any one of illustrative embodiments 18, 19, 20, 21, 22, 23, and 24, wherein the multi-organ health status includes a health score or an end-organ risk score for at least one organ for a patient.

[0121] Illustrative embodiment 26 includes the method of illustrative embodiment 25, wherein the method further includes outputting an alert based on a threshold and at least one of the health score or the end-organ risk score.Attorney Docket No.: 32860U-004101-WO-POA

[0122] Illustrative embodiment 27 includes the method of any one of illustrative embodiments 18, 19, 20, 21, 22, 23, 24, 25, and 26, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

[0123] Illustrative embodiment 28 includes the method of any one of illustrative embodiments 18, 19, 20, 21, 22, 23, 24, 25, 26, and 27, the method further comprising: training at least one acute event prediction model to predict at least one acute or adverse event for at least one organ.

[0124] Illustrative embodiment 29 includes the method of illustrative embodiment 28, wherein the at least one acute event prediction model is trained with the plurality of parameters and multi-organ health status data.

[0125] Illustrative embodiment 30 includes a method for monitoring multi-organ health, the method comprising: obtaining a plurality of first parameter values of parameters of a patient, forming a causal structure from the plurality of parameter values, the causal structure being based on a causal discovery indicative of one or more causal relationships between the parameters, and inputting the causal structure into a graph neural network (GNN) to obtain a multi-organ health status of the patient from the GNN.

[0126] Illustrative embodiment 31 includes the method of illustrative embodiment 30, wherein the GNN is trained based on the causal discovery.

[0127] Illustrative embodiment 32 includes the method of any one of illustrative embodiments 30 and 31, further comprising: inputting the multi-organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ.

[0128] Illustrative embodiment 33 includes the method of illustrative embodiment 32, further comprising: inputting the plurality of parameters into the acute event prediction model.Attorney Docket No.: 32860U-004101-WO-POA

[0129] Illustrative embodiment 34 includes the method of any one of illustrative embodiments 30, 31, 32, and 33, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the individual.

[0130] Illustrative embodiment 35 includes a non-transitory computer-readable storage medium storing computer-executable instruction that, when executed by at least one processor of a system, cause the system to perform a method for training at least one graph neural network (GNN) for monitoring organ health, the method comprising: obtaining a plurality of first parameter values of parameters from healthy individuals and diseased individuals, performing causal discovery to determine one or more causal relationships between the plurality of parameters, training the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi-organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.

[0131] Illustrative embodiment 36 includes the non-transitory computer-readable storage medium of illustrative embodiment 35, wherein the performing the causal discovery includes creating a causal structure between the plurality of parameters from the healthy individuals and the diseased individuals based on the first parameter values.

[0132] Illustrative embodiment 37 includes the non-transitory computer-readable storage medium of illustrative embodiment 36, wherein the causal structure is a directed acyclic graph.

[0133] Illustrative embodiment 38 includes the non-transitory computer-readable storage medium of any one of illustrative embodiments 36 and 37, wherein the training the at least one GNN includes inputting the causal structure into the at least one GNN.

[0134] Illustrative embodiment 39 includes the non-transitory computer-readable storage medium of any one of illustrative embodiments 35, 36, 37, and 38, wherein the at least one GNN includes a plurality of GNNs each trained with an organ specific model.Attorney Docket No.: 32860U-004101-WO-POA

[0135] Illustrative embodiment 40 includes the non-transitory computer-readable storage medium of any one of illustrative embodiments 35, 36, 37, 38, and 39, wherein the at least one GNN includes a plurality of task-specific branches of linear layers for multiple tasks specific to each organ of a plurality of organs.

[0136] Illustrative embodiment 41 includes the non-transitory computer-readable storage medium of illustrative embodiment 40, wherein the at least one GNN is trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.

[0137] Illustrative embodiment 42 includes the non-transitory computer-readable storage medium of any one of illustrative embodiments 35, 36, 37, 38, 39, 40, and 41, wherein the multi-organ health status includes a health score or an end-organ risk score for at least one organ for a patient.

[0138] Illustrative embodiment 43 includes the non-transitory computer-readable storage medium of illustrative embodiment 42, wherein the method further includes outputting an alert based on a threshold and at least one of the health score or the end-organ risk score.

[0139] Illustrative embodiment 44 includes the non-transitory computer-readable storage medium of any one of illustrative embodiments 35, 36, 37, 38, 39, 40, 41, 42, and 43, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

[0140] Illustrative embodiment 45 includes the non-transitory computer-readable storage medium of any one of illustrative embodiments 35, 36, 37, 38, 39, 40, 41, 42, 43, and 44, the method further comprising: training at least one acute event prediction model to predict at least one acute or adverse event for at least one organ.

[0141] Illustrative embodiment 46 includes the non-transitory computer-readable storage medium of illustrative embodiment 45, wherein the at least one acute event prediction model is trained with the plurality of parameters and multi-organ health status data.Attorney Docket No.: 32860U-004101-WO-POA

[0142] Illustrative embodiment 47 includes a non-transitory computer-readable storage medium storing computer-executable instruction that, when executed by at least one processor of a system, cause the system to perform a method for monitoring multi-organ health, the method comprising: obtaining a plurality of first parameter values of parameters of a patient, forming a causal structure from the plurality of parameter values, the causal structure being based on a causal discovery indicative of one or more causal relationships between the parameters, and inputting the causal structure into a graph neural network (GNN) to obtain a multi-organ health status of the patient from the GNN.

[0143] Illustrative embodiment 48 includes the non-transitory computer-readable storage medium of illustrative embodiment 47, wherein the GNN is trained based on the causal discovery.

[0144] Illustrative embodiment 49 includes the non-transitory computer-readable storage medium of any one of illustrative embodiments 47 and 48, further comprising: inputting the multi-organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ.

[0145] Illustrative embodiment 50 includes the non-transitory computer-readable storage medium of illustrative embodiment 49, further comprising: inputting the plurality of parameters into the acute event prediction model.

[0146] Illustrative embodiment 51 includes the non-transitory computer-readable storage medium of any one of illustrative embodiments 47, 48, 49, and 50, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the individual.

Claims

Attorney Docket No.: 32860U-004101-WO-POAWE CLAIM:

1. A system for training at least one graph neural network (GNN) for monitoring organ health, the system comprising:at least one memory configured to store instructions; andat least one processor configured to execute the instructions to cause the system to obtain a plurality of first parameter values of parameters from healthy individuals and diseased individuals,perform causal discovery to determine one or more causal relationships between the plurality of parameters, andtrain the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi-organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.

2. The system of claim 1, wherein the performing the causal discovery includes creating a causal structure between the plurality of parameters from the healthy individuals and the diseased individuals based on the first parameter values.

3. The system of claim 2, wherein the causal structure is a directed acyclic graph.

4. The system of claim 2, wherein the training the at least one GNN includes inputting the causal structure into the at least one GNN.

5. The system of claim 1, wherein the at least one GNN includes a plurality of GNNs each trained with an organ specific model.

6. The system of claim 1, wherein the at least one GNN includes a plurality of taskspecific branches of linear layers for multiple tasks specific to each organ of a plurality of organs.

7. The system of claim 6, wherein the at least one GNN is trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.Attorney Docket No.: 32860U-004101-WO-POA8. The system of claim 1, wherein the multi -organ health status includes a health score or an end-organ risk score for at least one organ for the patient.

9. The system of claim 8, wherein the system is further configured to output an alert based on a threshold and at least one of the health score or the end-organ risk score.

10. The system of claim 1 , wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

11. The system of claim 1, wherein the system is further configured totrain at least one acute event prediction model to predict at least one acute or adverse event for at least one organ.

12. The system of claim 11 , wherein the at least one acute event prediction model is trained with the plurality of first parameter values of the parameters from the healthy individuals and the diseased individuals and multi-organ health status data.

13. A system for monitoring multi-organ health, the system comprising:at least one memory configured to store instructions; andat least one processor configured to execute the instructions to cause the system to obtain a plurality of first parameter values of parameters of a patient, form a causal structure from the plurality of parameter values, the causal structure being based on a causal discovery indicative of one or more causal relationships between the parameters, andinput the causal structure into a graph neural network (GNN) to obtain a multiorgan health status of the patient from the GNN.

14. The system of clam 13, wherein the GNN is trained based on the causal discovery.

15. The system of claim 13, wherein the system is further configured toinput the multi-organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ.Attorney Docket No.: 32860U-004101-WO-POA16. The system of claim 15, wherein the system is further configured toinput the plurality of parameters into the acute event prediction model.

17. The system of claim 13, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the individual.

18. A method fortraining at least one graph neural network (GNN) for monitoring organ health, the method comprising:obtaining a plurality of first parameter values of parameters from healthy individuals and diseased individuals,performing causal discovery to determine one or more causal relationships between the plurality of parameters, andtraining the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi-organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.

19. The method of claim 18, wherein the performing the causal discovery includes creating a causal structure between the plurality of parameters from the healthy individuals and the diseased individuals based on the first parameter values.

20. The method of claim 19, wherein the causal structure is a directed acyclic graph.

21. The method of claim 19, wherein the training the at least one GNN includes inputting the causal structure into the at least one GNN.

22. The method of claim 18, wherein the at least one GNN includes a plurality of GNNs each trained with an organ specific model.

23. The method of claim 18, wherein the at least one GNN includes a plurality of task-specific branches of linear layers for multiple tasks specific to each organ of a plurality of organs.Attorney Docket No.: 32860U-004101-WO-POA24. The method of claim 23, wherein the at least one GNN is trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.

25. The method of claim 18, wherein the multi-organ health status includes a health score or an end-organ risk score for at least one organ for a patient.

26. The method of claim 25, further comprising:outputting an alert based on a threshold and at least one of the health score or the endorgan risk score.

27. The method of claim 18, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

28. The method of claim 18, further comprising:training at least one acute event prediction model to predict at least one acute or adverse event for at least one organ.

29. The method of claim 28, wherein the at least one acute event prediction model is trained with the plurality of parameters and multi-organ health status data.

30. A method for monitoring multi-organ health, the method comprising: obtaining a plurality of first parameter values of parameters of a patient, forming a causal structure from the plurality of parameter values, the causal structure being based on a causal discovery indicative of one or more causal relationships between the parameters, andinputting the causal structure into a graph neural network (GNN) to obtain a multiorgan health status of the patient from the GNN.

31. The method of clam 30, wherein the GNN is trained based on the causal discovery.

32. The method of claim 30, further comprising:Attorney Docket No.: 32860U-004101-WO-POAinputting the multi -organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ.

33. The method of claim 32, further comprising:inputting the plurality of parameters into the acute event prediction model.

34. The method of claim 30, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the individual.

35. A non-transitory computer-readable storage medium storing computer-executable instruction that, when executed by at least one processor of a system, cause the system to perform a method for training at least one graph neural network (GNN) for monitoring organ health, the method including:obtaining a plurality of first parameter values of parameters from healthy individuals and diseased individuals,performing causal discovery to determine one or more causal relationships between the plurality of parameters, andtraining the at least one GNN with the one or more causal relationships to create a trained GNN, the trained GNN configured to output a multi-organ health status based on second parameter values, the second parameter values being parameter values of parameters from a patient.

36. The non-transitory computer-readable storage medium of claim 35, wherein the performing the causal discovery includes creating a causal structure between the plurality of parameters from the healthy individuals and the diseased individuals based on the first parameter values.

37. The non-transitory computer-readable storage medium of claim 36, wherein the causal structure is a directed acyclic graph.

38. The non-transitory computer-readable storage medium of claim 36, wherein the training the at least one GNN includes inputting the causal structure into the at least one GNN.Attorney Docket No.: 32860U-004101-WO-POA39. The non-transitory computer-readable storage medium of claim 35, wherein the at least one GNN includes a plurality of GNNs each trained with an organ specific model.

40. The non-transitory computer-readable storage medium of claim 35, wherein the at least one GNN includes a plurality of task-specific branches of linear layers for multiple tasks specific to each organ of a plurality of organs.

41. The non-transitory computer-readable storage medium of claim 40, wherein the at least one GNN is trained simultaneously for the multiple tasks by jointly optimizing a loss for each task of the multiple tasks.

42. The non-transitory computer-readable storage medium of claim 35, wherein the multi-organ health status includes a health score or an end-organ risk score for at least one organ for a patient.

43. The non-transitory computer-readable storage medium of claim 42, wherein the method further comprises:outputting an alert based on a threshold and at least one of the health score or the endorgan risk score.

44. The non-transitory computer-readable storage medium of claim 35, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the healthy individuals and the diseased individuals.

45. The non-transitory computer-readable storage medium of claim 35, wherein the method further comprises:training at least one acute event prediction model to predict at least one acute or adverse event for at least one organ.

46. The non-transitory computer-readable storage medium of claim 45, wherein the at least one acute event prediction model is trained with the plurality of parameters and multiorgan health status data.Attorney Docket No.: 32860U-004101-WO-POA47. A non-transitory computer-readable storage medium storing computer-executable instruction that, when executed by at least one processor of a system, cause the system to perform a method for monitoring multi-organ health, the method comprising:obtaining a plurality of first parameter values of parameters of a patient, forming a causal structure from the plurality of parameter values, the causal structure being based on a causal discovery indicative of one or more causal relationships between the parameters, andinputting the causal structure into a graph neural network (GNN) to obtain a multiorgan health status of the patient from the GNN.

48. The non-transitory computer-readable storage medium of clam 47, wherein the GNN is trained based on the causal discovery.

49. The non-transitory computer-readable storage medium of claim 47, wherein the method further comprises:inputting the multi -organ health status into an acute event prediction model to predict at least one acute or adverse event for at least one organ.

50. The non-transitory computer-readable storage medium of claim 49, wherein the method further comprises:inputting the plurality of parameters into the acute event prediction model.

51. The method of claim 47, wherein the plurality of parameters include one or more of lab parameters, vitals, or demographic information for the individual.

Images