Classifications

• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
  • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
  • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
  • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
  • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders

Definitions

• Various aspects of this disclosure generally relate to a clinical decision support system, and more particularly, to a predictive analytics solution for personalized clinical decision support.
• a core challenge in the delivery of modern healthcare is to provide highest quality care with minimal utilization of increasingly scarce resources (e.g., healthcare budget, physician's time, hospital beds, etc.).
• resources e.g., healthcare budget, physician's time, hospital beds, etc.
• provider organizations such as hospitals, clinics, and service centres are in dire need of methods to help plan care in ways that balance trade-offs between resource cost and care quality.
• Prevailing clinical practice guidelines tend to be tailored to the 'average' patient, and not the individual patient for whom decisions need to be made. Further, prevailing guidelines may not always be relevant to the policies and care provision systems of individual provider organizations, and thus may or may not be effective. Finally, prevailing guidelines tend to be largely 'static', and revisions require large clinical trials or controlled patient cohort-based studies. With increasing amounts of medical data accumulated within provider organizations, predictive analytics methodologies present compelling opportunities to overcome the limitations of prevailing clinical care guidelines. Specifically, predictive analytics methodologies can leverage rich clinical care datasets comprising physiological and administrative information, extract valuable insights and apply these insights to derive personalized and dynamically evolving care delivery guidelines.
• a method, a computer-readable medium, and an apparatus for personalized clinical decision support employs machine learning algorithms on multimodal patient datasets associated with a group of past patients to learn a plurality of models for predicting a plurality of care objectives (e.g., cost, quality of outcomes, recovery levels, complication rates). .
• the method optimizes the plurality of predicted care objectives to determine an ideal treatment workflow.
• the optimization could be subject to a set of variable constraints defined to ensure the resulting treatment workflows are feasible.
• variable constraints could be derived based on prior clinical knowledge or based on a subset of multimodal patient data associated with the group of past patients. For the latter case, the subset of multimodal patient data may correspond to patient data for one or more patients identified to be most similar to a new patient presenting for care.
• FIG. 1 is a diagram illustrating an example of multimodal patient data.
• FIG. 2 is a diagram illustrating an example of a predictive analytics solution that leverages multimodal patient data for personalized clinical decision-support.
• FIG. 3 is a diagram conceptually illustrating an example of using the predictive analytics solution described in FIG. 2 to determine a treatment workflow.
• FIG. 4 is a table illustrating examples of non-apparent insights revealed by the predictive analytics solution (described in FIG. 2 above), alongside the recommended changes to the treatment workflow to improve the care objectives of interest (e.g., cost of care, risk of readmission).
• FIG. 5 is a flowchart of a method of personalized clinical decision support.
• FIG. 6 is a conceptual data flow diagram illustrating the data flow between different means/components in an exemplary apparatus.
• FIG. 7 is a diagram illustrating an example of a hardware implementation for an apparatus employing a processing system.
• processors include microprocessors, microcontrollers, graphics processing units (GPUs), central processing units (CPUs), application processors, digital signal processors (DSPs), reduced instruction set computing (RISC) processors, systems on a chip (SoC), baseband processors, field programmable gate arrays (FPGAs), programmable logic devices (PLDs), state machines, gated logic, discrete hardware circuits, and other suitable hardware configured to perform the various functionality described throughout this disclosure.
• processors in the processing system may execute software.
• Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software components, applications, software applications, software packages, routines, subroutines, objects, executables, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.
• the functions described may be implemented in hardware, software, or any combination thereof. If implemented in software, the functions may be stored on or encoded as one or more instructions or code on a computer-readable medium.
• Computer-readable media includes computer storage media. Storage media may be any available media that can be accessed by a computer.
• such computer-readable media may include a random-access memory (RAM), a read-only memory (ROM), an electrically erasable programmable ROM (EEPROM), optical disk storage, magnetic disk storage, other magnetic storage devices, combinations of the aforementioned types of computer- readable media, or any other medium that can be used to store computer executable code in the form of instructions or data structures that can be accessed by a computer.
• RAM random-access memory
• ROM read-only memory
• EEPROM electrically erasable programmable ROM
• optical disk storage magnetic disk storage
• magnetic disk storage other magnetic storage devices
• combinations of the aforementioned types of computer- readable media or any other medium that can be used to store computer executable code in the form of instructions or data structures that can be accessed by a computer.
• FIG. 1 is a diagram 100 illustrating an example of multimodal patient data 102.
• the multimodal patient data 102 may be stored in a single database system.
• the multimodal patient data 102 may be stored in multiple database systems.
• the multimodal patient data may include physiological data 110, administrative data 112, billing data 114, medical history 116, admitting characteristics 1 18, inpatient care data 120, etc.
• the physiological data 110 may include measurements related to a range of physiological parameters.
• the physiological data 110 may include measurements that predominantly focus on accessing the function of major organ systems (e.g., the central and peripheral nervous system).
• the physiological data 110 may include vital signs measurements such as blood pressure, body temperature, body weight.
• the physiological data 110 may include the levels of chemicals in the blood and/or other diagnostic test results.
• the administrative data 112 may include information collected primarily for care delivery purposes.
• the administrative data 112 may be collected by medical service provider organizations for the purpose of registration, transaction, and record keeping, usually during the delivery of a service.
• the administrative data 112 may include patient identifier (ID), date of birth, date of death, residential address, etc.
• ID patient identifier
• the administrative data 112 may include the duration for which a patient stays in a hospital, details on the scheduling of diagnostic tests, the class of ward and/or any transfers between wards or units within a hospital, or the care delivery departments which have had influence on the patient's care. In one embodiment, the administrative data 112 may include the reason for delay in the patient's discharge from the hospital, and/or the destination that the patient is discharged to following inpatient stay (e.g., community nursing home or home).
• inpatient stay e.g., community nursing home or home.
• the billing data 114 may include service provider organization's claims with insurance companies in order to receive payment for services rendered, and well as bills to patients for payment due.
• the billing data 114 may include the billing codes and descriptions for the patient's clinical condition, diagnosis, and/or procedures performed.
• the billing data 114 may be part of the administrative data 112.
• the medical history 116 may include information gained by a physician by asking specific questions, either of the patient or of other people who know the person and can give suitable information, with the aim of obtaining information useful in formulating a diagnosis and providing medical care to the patient.
• the medical history 116 may include symptoms, clinical signs, past medical history, family diseases, social history (e.g., living arrangements, occupation, marital status, drug use, recent foreign travel, exposure to environmental pathogens through recreational activities or pets), medications, allergies, etc.
• the admitting characteristics 118 may include patient characteristics when admitted into inpatient care.
• the admitting characteristics 118 may include demographic characteristics such as patient age, gender, and race.
• the admitting characteristics 118 may include the patient's state of health at admission (e.g., restricted mobility due to fracture, or decompensating heart failure).
• the inpatient care data 120 may include data related to inpatient treatments received by patients.
• the inpatient care data 120 may include data on the mix and dose of medications prescribed for the patient during the hospital stay.
• the inpatient care data 120 may include data from surgical or other interventional procedures performed on the patient during the hospital stay.
• Predictive analytics deals with extracting information from data and using it to predict trends, categories, and future events that can be expected based on the statistical patterns within the data.
• Predictive models are models of the relation between one or more known attributes or features of a given data sample, and the unknown label (numerical or categorical prediction) that characterizes the data sample. The objective of the model is to predict the unknown label based on the attributes or features of a new data sample.
• Clinical decision support assists clinicians or associated clinical care staff by presenting specific intelligently filtered and timely information to enhance health care delivery and/or patient outcomes.
• a clinical decision support system may encompasses a variety of tools and interventions such as computerized alerts and reminders, clinical guidelines, order sets, patient data reports and interfaces, documentation templates, diagnostic support tools, and/or clinical workflow improvement tools.
• predictive analytics may, in principle, be personalized to specific patients/groups, individual provider organizations, and evolve as medical records datasets evolve. Furthermore, unlike existing decision support tools based on semantic query matching and/or visualization, predictive analytics methodologies would enable the extraction of valuable insights from the rich multimodal patient data streams possessed by healthcare providing organizations, as well as the application of these insights to derive personalized and dynamically evolving care delivery guidelines for clinical decision support.
• FIG. 2 is a diagram illustrating an example of a predictive analytics solution 200 that leverages multimodal patient data for personalized clinical decision-support.
• the predictive analytics solution integrates machine learning and optimization techniques in a unique manner, as detailed below.
• machine learning algorithms e.g., random forests, support vector regression, gradient-based methods, neural networks, or deep learning-based methods
• the predictive models 204 may be used to elicit insightful drivers of clinical outcomes (e.g., cost and quality) based on multimodal patient data 208.
• X may represent important attributes (within the multimodal patient data 208) driving cost and quality
• the predicted cost may be a first function of X and the predicted quality may be a second function of X.
• the quality may be defined by several quality of care metrics, e.g., risk of readmission, functional recovery score, chance of complication, etc.
• the multimodal patient data 208 may be the multimodal patient data 102 described above with reference to FIG. 1.
• the implementation at 202 may include two degrees of cross validation, both in the choice of test and training datasets, as well as in the choice of the machine learning algorithms/learning parameters. This enables superior accuracy and flexibility.
• the best predictive model resulting from the training may be used to relate the key attributes in the data to the outcome, simulate scenarios for different outcomes (e.g., cost and length of stay) and compute consequences of various possible treatment workflow decisions.
• feature importance metrics may be computed and data visualizations may be developed to provide clinical intuition underlying the features that drive the objectives of interest. This may be particularly relevant for doctors who wish to gain intuitive insights on what features matter most for desired outcomes.
• feature engineering may be employed to reduce the dimensionality and correlations within the feature space.
• machine learning and optimization methodologies may be adapted to appropriately handle sparse and missing data, imbalanced classes, and uncertainty in predicted labels.
• a model-based multi-objective optimization may be performed to determine possible decisions 212 that would lead to the best trade-offs amongst outcomes of interest (e.g., optimal trade-off between cost and the risk for readmission).
• the optimization at 210 may be based on the black box predictive models 204 learned at 202. While the machine learning models employ both intrinsic features (fixed features that describe the patient at admission) and extrinsic features (representing decisions/quantities that are related to clinical actions and/or treatment workflow), the optimizer is designed to only tune the extrinsic features/decisions conditional on the patient's intrinsic features/attributes.
• the goal at 210 is to optimize multiple quantities simultaneously (multi-objective optimization), for example to minimize cost and maximize quality of care.
• the problem may be formulated as the optimization of an aggregated weighted-sum of multiple single objectives.
• the covariance matrix adaptation (CMA) algorithm may be employed at 210 as it is (a) highly suited for problems involving unknown, multivariate and complex (even ill-conditioned) objective functions, and (b) based on an evolutionary strategy that is informed primarily by the dependencies within the dataset.
• CMA requires very few assumptions on the forms and numerical features of the objective function, and thus is highly suitable for models derived using machine learning algorithms.
• the implementation at 210 may ensure that features implicitly correlated with the decision variables (secondary decision variables) are updated together with the decision variables during optimization.
• the optimizer at 210 may have three outputs: (1) the optimal decisions 212 for the patient under evaluation, (2) the importance ranking of these decisions so that a clinical practitioner can prioritize accordingly, and (3) the Pareto front 214 logging the entire space of possible optimal combinations of outcomes (constructed by sweeping the weights of the objectives with varying degrees of relative importance assigned to the multiple objectives (outcomes) of interest).
• the Pareto front 214 gives an intuitive way to assess the baseline outcomes (e.g., outcome 216) before optimization and the improved outcomes (e.g., outcome 218) after optimization for any given patient.
• the integrative analytics may be packaged into a decision support dashboard that can facilitate personalized decision making on treatment workflows for a newly admitting patient.
• One aspect of this disclosure focuses on personalizing the decisions for an individual patient and their condition/history, and incorporates constraints to ensure that the recommended interventions are practically feasible in the clinical setting (i.e., actionable optimal decisions).
• This step involves (a) multivariate clustering (at 220) to identify past patients 224 most similar to the new patient 222, and (b) derivation of personalized constraints (e.g., personalized constraint function h(X)) on optimization decision variables using these past patient examples 224.
• personalized constraints e.g., personalized constraint function h(X)
• the optimizer 210 constrains the decision variable search to within [A, B], and penalizes decisions outside of this range.
• the decision support dashboard may display alerts that could guide clinicians during the care for this new patient, or recommend treatment workflow decisions that could optimize cost-quality trade-off for this new patient.
• the predictive analytics solution 200 may (a) learn a predictive model 204 for outcomes of interest 206 based on multimodal patient data 208 from past patients, (b) employ the predictive model 204 to predict the outcomes and jointly optimize the trade-offs amongst outcomes for a new patient, and (c) constrain the optimization result to reasonable decisions possible given the clinical condition of the new patient.
• predictive modelling may be employed based on past patient data to help clinicians determine a treatment workflow that would maximize the quality of care and at the same time, minimize the cost of care for a new patient.
• FIG. 3 is a diagram 300 conceptually illustrating an example of using the predictive analytics solution 200 described in FIG. 2 to determine a treatment workflow. Specifically, this example illustrates the different use cases of business intelligence, alerts generation and target setting, and how different aspects of the predictive analytics solution 200 contribute to the use cases.
• multimodal patient data 302 may be utilized to develop predictive models 304.
• the predictive models 304 may provide the ability to perform scenario analysis for resource-efficient case management, be employed to assess improvement relative to the patient's condition at admission, enable assessment of risk for complication through the inpatient care and/or after discharge, and/or assess readiness for discharge.
• the use of the predictive model for scenario analysis applications may provide business intelligence for resource-efficient case management decisions.
• the use of predictive models for assessing improvement relative to admission, assessing risk for complication, and assessing readiness for discharge may provide decision support to providers and enable improved personalized clinical decisions.
• the multimodal patient data 302 may be the multimodal patient data 208 described above with reference to FIG. 2, and the predictive models 304 may be the predictive models 204 described above with reference to FIG. 2.
• the predictive models 304 could be used as the basis of a multi- objective optimization algorithm to determine the best treatment workflow for a set of care delivery objectives of interest. Derivation of an optimal treatment workflow may enable clinicians to better set care management targets, optimize cost-quality trade-offs, optimize trade-offs between multiple clinical objectives in care delivery, and/or better plan for post- discharge management. Specifically, the optimization of the cost-quality trade-offs in care delivery may provide business intelligence for cost-efficient delivery of quality care. The use of optimization to set care management targets for better recovery, to optimize clinical trade-offs in care delivery, and to better plan post-discharge management may provide decision support to providers and enable improved personalized clinical decisions.
• the multi-objective optimization performed at 306 may include the operations performed at 210 in FIG. 2.
• the treatment workflow may be the optimal decisions 212 described above with reference to FIG. 2.
• the predictive analytics solution 200 described above in FIG. 2 may suggest personalized optimal decisions spanning the care continuum. For example, the suggested personalized optimal decisions may impact decisions across the care continuum including diagnosis, inpatient management, discharge, or post-discharge care.
• One key limitation in translating the predictive analytics solution 200 to a living tool for health provision is an information systems challenge.
• the challenge lies in collating disparate and diversified data sources into a machine learning friendly format.
• this process may be automated within a series of automated pre-processing pipelines for feature engineering and dimensionality reduction.
• Such pipelines could be expanded in collaboration with hospital information systems providers to become more customized and/or generalizable as needed, and could also help deploy best practices for data collection and storage.
• the aforementioned integrative predictive analytics approaches have been developed and successfully tested on hospital datasets for hip fracture and heart failure patients.
• the machine learning models have achieved predictive accuracy in the 70-90% range for several cost and quality metrics, and revealed non- apparent insights into the drivers of cost and quality of care.
• FIG. 4 is a table 400 illustrating examples of non-apparent insights revealed by the predictive analytics solution 200 described in FIG. 2 above alongside the recommended treatment workflow decisions for improving cost and quality of care.
• the predictive analytics solution 200 may reveal that document transfer efficiency between hospital A and community hospital X increases costs by $3,000 on average. Therefore, document transfer efficiency between hospital A and community hospital X may need to be improved.
• the predictive analytics solution 200 may reveal that heart failure patient clinical indicators at admission are highly predictive for whether a patient is likely to be readmitted for heart failure related reasons or for other causes. As a result, the patient may be flagged for focused heart failure related care or for comorbidities management by non-cardiac clinical specialties.
• the optimization may drop cost and improve quality of care by at least 25-30%.
• the optimization may be expanded to a variety of relevant cost and quality of care metrics such as cost of hospital stay, length of hospital stay, occurrence of inpatient complications, recovery metrics such as postoperative weight bearing ability, risk of readmission, and recovery of pre-operative functional mobility levels.
• an intelligent clinical decision support tool may be developed to facilitate and improve patient care management.
• a framework is provided to learn a predictive model based on multimodal patient data. This predictive model may map the patient data to care outcomes of interest, and have the ability to learn from every new patient just like a clinician accumulates knowledge from clinical experience.
• a framework is provided to use the predictive model learned to determine optimal care decisions. This involves integrating the predictive model developed with multivariate optimization and clustering techniques to derive optimal decisions that can facilitate improved patient care. Together, the predictive model and the optimization approach may be packaged into a clinical decision support tool which goes beyond mere patient profiling or semantic querying.
• machine learning based on patient/clinical data may be employed to develop a predictive model, which then forms the basis of decision support guidelines offered.
• multimodal data spanning the spectrum of patient care e.g., physiological data, medical history, admitting characteristics, inpatient care, administrative, and billing details
• the predictive analytics solution 200 may help distil the features driving outcomes of interest, drill down why they are relevant to the outcomes, and thus could help clinicians understand and intuit the driving features, as well as actions derived therefrom.
• the predictive analytics solution 200 may generate alerts to stratify/flag patients who are "high resource utilization" or "high clinical risk", so as to enable focused care management actions.
• the predictive analytics solution 200 may be able to go beyond triggering alerts to planning decisions.
• the predictive analytics solution 200 may build in optimization abilities to inform clinicians on personalized decisions that may be optimal for care outcomes - across the spectrum of treatment workflow decisions through the inpatient, discharge, and post discharge continuum.
• the predictive analytics solution 200 described above may enable health care providers to optimize trade-offs in clinical care, as well as improve delivery of care in a personalized manner.
• the predictive analytics solution 200 may generate possible decisions that could optimize trade-offs for clinical care - for instance cost or resource utilization targets and quality of care metrics.
• the predictive analytics solution 200 may be applicable to any combination of outcomes/objectives that trade off each other.
• the predictive analytics solution 200 may set care management targets and incentivize clinicians to achieve these targets. In one embodiment, the predictive analytics solution 200 may serve as a business intelligence tool that can enable healthcare providers to minimize the overall costs/resource utilization while maximizing the quality of care.
• FIG. 5 is a flowchart 500 of a method of personalized clinical decision support.
• the method may be performed by the predictive analytics solution 200 described above with reference to FIG. 2.
• the method may be performed by a computing device or system (e.g., the apparatus 602/602').
• the method may generate, using machine learning algorithms, a plurality of predictive models (e.g., the predictive models 204) based on multimodal patient data (e.g., the multimodal patient data 208) associated with a group of patients.
• multimodal patient data e.g., the multimodal patient data 208 associated with a group of patients.
• the multimodal patient data associated with the group of patients may include two or more of physiological data, administrative data, billing data, medical history, admitting characteristics, or inpatient care data.
• the method may optionally identify a plurality of patients (e.g., the past patient 224) within the group of patients that are most similar to a particular patient (e.g., the patient 222).
• the plurality of patients may be identified using multivariate clustering.
• the method may optionally derive a set of variable constraints based on multimodal patient data associated with the plurality of patients or based on prior clinical knowledge. For instance, if the plurality of patients has a range [A, B] for some clinical indicator, the method may derive a set of variable constraints that constrains the decision variable search to within [A, B], and penalizes decisions outside of this range.
• the method may determine a treatment workflow (e.g., the treatment workflow 300) by performing optimization on a plurality of objectives based on the plurality of predictive models. In one embodiment, the optimization may be performed subject to a set of variable constraints. In one embodiment, the set of variable constraints may be derived at 506. In one embodiment, the set of variable constraints may be derived based on prior clinical knowledge.
• the plurality of objectives may include cost and quality.
• the optimization may maximize the quality while minimizing the cost.
• the plurality of objectives may tradeoff against each other.
• the optimization may be performed using covariance matrix adaptation algorithm.
• the treatment workflow may be personalized to an individual patient who presents for care, e.g., by performing the optimization subject to variable constraints derived from patients that are similar to the individual patient.
• the treatment workflow may be personalized to individual provider organization, e.g., by generating predictive models based on patient data associated with the individual provider organization.
• the treatment workflow and the plurality of predictive models may evolve as the multimodal patient data evolves. For example, as the multimodal patient data evolves, the predictive models and variable constraints may be updated accordingly. As a result, the treatment workflow may be updated as well.
• the treatment workflow may include treatment decisions for two or more of diagnosis, inpatient, discharge, or post discharge care.
• each of the plurality of predictive models may relate a set of attributes of the multimodal patient data to one or more treatment decisions.
• the method may compute an importance metrics for the set of attributes based on an impact of each of the set of attributes on the plurality of objectives. The method may provide a visualization of the importance metrics.
• FIG. 6 is a conceptual data flow diagram 600 illustrating the data flow between different means/components in an exemplary apparatus 602.
• the apparatus 602 may be a computing device or a system including multiple computing devices.
• the apparatus 602 may include a predictive model generator 604 that generates predictive models based on multimodal patient data.
• the predictive model generator 604 may include a machine learning component 612 that employs machine learning algorithms on multimodal patient data to learn predictive models.
• the predictive model generator 604 may perform the operations described above with reference to 502 in FIG. 5.
• the apparatus 602 may include a constraint generation component 610 that identifies patients that are most similar to a particular patient and derives variable constraints based on multimodal patient data associated with the identified patients. Alternatively or additionally, the constraint generation component 610 may derive variable constraints based on prior clinical knowledge. In one embodiment, the constraint generation component 610 may perform the operations described above with reference to 504 or 506 in FIG. 5.
• the apparatus 602 may include a multi-objective optimization component 608 that determines an ideal treatment workflow by performing optimization on multiple objectives based on the predictive models provided by the predictive model generator 604. The optimization may be performed subject to the variable constraints received from the constraint generation component 610. In one embodiment, the multi-objective optimization component 608 may output recommendations on a treatment workflow. In one embodiment, the multi-objective optimization component 608 may perform the operations described above with reference to 508 in FIG. 5.
• the apparatus 602 may include additional components that perform each of the blocks of the algorithm in the aforementioned flowchart of FIG. 5. As such, each block in the aforementioned flowchart of FIG. 5 may be performed by a component and the apparatus may include one or more of those components.
• the components may be one or more hardware components specifically configured to carry out the stated processes/algorithm, implemented by a processor configured to perform the stated processes/algorithm, stored within a computer-readable medium for implementation by a processor, or some combination thereof.
• FIG. 7 is a diagram 700 illustrating an example of a hardware implementation for an apparatus 602' employing a processing system 714.
• the processing system 714 may be implemented with a bus architecture, represented generally by the bus 724.
• the bus 724 may include any number of interconnecting buses and bridges depending on the specific application of the processing system 714 and the overall design constraints.
• the bus 724 links together various circuits including one or more processors and/or hardware components, represented by the processor 704, the components 604, 608, 610, and the computer-readable medium / memory 706.
• the bus 724 may also link various other circuits such as timing sources, peripherals, voltage regulators, and power management circuits, which are well known in the art, and therefore, will not be described any further.
• the processing system 714 includes a processor 704 coupled to a computer- readable medium / memory 706.
• the processor 704 is responsible for general processing, including the execution of software stored on the computer-readable medium / memory 706.
• the software when executed by the processor 704, causes the processing system 714 to perform the various functions described supra for any particular apparatus.
• the computer-readable medium / memory 706 may also be used for storing data that is manipulated by the processor 704 when executing software.
• the processing system 714 further includes at least one of the components 604, 608, 610.
• the components may be software components running in the processor 704, resident/stored in the computer readable medium / memory 706, one or more hardware components coupled to the processor 704, or some combination thereof.
• the processing system 714 may include a data storage 708 for storing data (e.g., multimodal patient data).
• Example 1 is a method or apparatus for personalized clinical decision support.
• the method or apparatus may generate, using machine learning algorithms, a plurality of predictive models based on patient data associated with a group of patients.
• the method or apparatus may determine a treatment workflow by performing optimization on a plurality of objectives based on the plurality of predictive models.
• Example 2 the subject matter of Example 1 may optionally include that the patient data associated with the group of patients may be multimodal - i.e., include two or more of physiological data, administrative data, billing data, medical history, admitting characteristics, or inpatient care data.
• Example 3 the subject matter of any one of Examples 1 to 2 may optionally include that the treatment workflow may be personalized to an individual patient who presents for care.
• Example 4 the subject matter of any one of Examples 1 to 3 may optionally include that the treatment workflow may be personalized to individual provider organization.
• Example 5 the subject matter of any one of Examples 1 to 4 may optionally include that the treatment workflow and the plurality of predictive models may evolve as the multimodal patient data evolves.
• Example 6 the subject matter of any one of Examples 1 to 5 may optionally include that the treatment workflow may cut across one or more aspects of the spectrum of diagnosis, inpatient, discharge, or post discharge care.
• Example 7 the subject matter of any one of Examples 1 to 6 may optionally include that the plurality of objectives may include cost and quality.
• the optimization may maximize the quality while minimizing the cost.
• Example 8 the subject matter of any one of Examples 1 to 7 may optionally include that each of the plurality of predictive models may relate a set of attributes of the multimodal patient data to one or more care objectives or outcomes.
• Example 9 the subject matter of Examples 8 may optionally include that the method or apparatus may compute feature importance metrics for the set of attributes based on the impact that each attribute has on the plurality of objectives, and provide a visualization of the feature importance metrics.
• Example 10 the subject matter of any one of Examples 1 to 9 may optionally include that the optimization may be performed using covariance matrix adaptation algorithm.
• Example 11 the subject matter of any one of Examples 1 to 10 may optionally include that the method or apparatus may identify a plurality of patients within the group of patients that are most similar to a particular patient using multivariate clustering approaches, and derive a set of variable constraints based on multimodal patient data associated with the plurality of patients. The optimization may be performed subjective to the set of variable constraints.
• Example 12 the subject matter of any one of Examples 1 to 11 may optionally include that the method or apparatus may derive a set of variable constraints based on prior clinical knowledge. The optimization may be performed subjective to the set of variable constraints.
• Example 13 the subject matter of any one of Examples 1 to 12 may optionally include that the plurality of objectives may trade-off against each other.
• Combinations such as "at least one of A, B, or C,” “one or more of A, B, or C,” “at least one of A, B, and C,” “one or more of A, B, and C,” and "A, B, C, or any combination thereof include any combination of A, B, and/or C, and may include multiples of A, multiples of B, or multiples of C.
• combinations such as “at least one of A, B, or C,” “one or more of A, B, or C,” “at least one of A, B, and C,” “one or more of A, B, and C,” and “A, B, C, or any combination thereof may be A only, B only, C only, A and B, A and C, B and C, or A and B and C, where any such combinations may contain one or more member or members of A, B, or C.

Landscapes

• Engineering & Computer Science (AREA)
• Health & Medical Sciences (AREA)
• Medical Informatics (AREA)
• Public Health (AREA)
• Data Mining & Analysis (AREA)
• Biomedical Technology (AREA)
• Databases & Information Systems (AREA)
• Pathology (AREA)
• Epidemiology (AREA)
• General Health & Medical Sciences (AREA)
• Primary Health Care (AREA)
• Medical Treatment And Welfare Office Work (AREA)

Priority Applications (1)

Applications Claiming Priority (1)

Publications (1)

Family

ID=65525017

Family Applications (1)

Country Status (1)

Cited By (8)

• 2017
  • 2017-08-28 WO PCT/SG2017/050422 patent/WO2019045637A2/fr not_active Ceased

Similar Documents

Legal Events