ABSTRACT
We tested the ability of a physiologically driven minimally invasive closed-loop algorithm, called Resuscitation based on Functional Hemodynamic Monitoring (ReFit), to stabilize for up to 3 h a porcine model of noncompressible hemorrhage induced by severe liver injury and do so during both ground and air transport. Twelve animals were resuscitated using ReFit to drive fluid and vasopressor infusion to a mean arterial pressure (MAP) > 60 mmHg and heart rate < 110 min-1 30 min after MAP < 40 mmHg following liver injury. ReFit was initially validated in 8 animals in the laboratory, then in 4 animals during air (23nm and 35nm) and ground (9 mi) to air (9.5nm and 83m) transport returning to the laboratory. The ReFit algorithm kept all animals stable for ~ 3 h. Thus, ReFit algorithm can diagnose and treat ongoing hemorrhagic shock independent to the site of care or during transport. These results have implications for treatment of critically ill patients in remote, austere and contested environments and during transport to a higher level of care.
ABSTRACT
Forecasting healthcare time series is crucial for early detection of adverse outcomes and for patient monitoring. Forecasting, however, can be difficult in practice due to noisy and intermittent data. The challenges are often exacerbated by change points induced via extrinsic factors, such as the administration of medication. To address these challenges, we propose a novel hybrid global-local architecture and a pharmacokinetic encoder that informs deep learning models of patient-specific treatment effects. We showcase the efficacy of our approach in achieving significant accuracy gains for a blood glucose forecasting task using both realistically simulated and real-world data. Our global-local architecture improves over patient-specific models by 9.2-14.6%. Additionally, our pharmacokinetic encoder improves over alternative encoding techniques by 4.4% on simulated data and 2.1% on real-world data. The proposed approach can have multiple beneficial applications in clinical practice, such as issuing early warnings about unexpected treatment responses, or helping to characterize patient-specific treatment effects in terms of drug absorption and elimination characteristics.
ABSTRACT
BACKGROUND: Despite the morbidity associated with acute atrial fibrillation (AF), no models currently exist to forecast its imminent onset. We sought to evaluate the ability of deep learning to forecast the imminent onset of AF with sufficient lead time, which has important implications for inpatient care. METHODS: We utilized the Physiobank Long-Term AF Database, which contains 24-h, labeled ECG recordings from patients with a history of AF. AF episodes were defined as ≥5 min of sustained AF. Three deep learning models incorporating convolutional and transformer layers were created for forecasting, with two models focusing on the predictive nature of sinus rhythm segments and AF epochs separately preceding an AF episode, and one model utilizing all preceding waveform as input. Cross-validated performance was evaluated using area under time-dependent receiver operating characteristic curves (AUC(t)) at 7.5-, 15-, 30-, and 60-min lead times, precision-recall curves, and imminent AF risk trajectories. RESULTS: There were 367 AF episodes from 84 ECG recordings. All models showed average risk trajectory divergence of those with an AF episode from those without â¼15 min before the episode. Highest AUC was associated with the sinus rhythm model [AUC = 0.74; 7.5-min lead time], though the model using all preceding waveform data had similar performance and higher AUCs at longer lead times. CONCLUSIONS: In this proof-of-concept study, we demonstrated the potential utility of neural networks to forecast the onset of AF in long-term ECG recordings with a clinically relevant lead time. External validation in larger cohorts is required before deploying these models clinically.
Subject(s)
Atrial Fibrillation , Humans , Atrial Fibrillation/diagnosis , Electrocardiography , Neural Networks, Computer , ROC Curve , Time FactorsABSTRACT
BACKGROUND: Healthcare-associated bacterial pathogens frequently carry plasmids that contribute to antibiotic resistance and virulence. The horizontal transfer of plasmids in healthcare settings has been previously documented, but genomic and epidemiologic methods to study this phenomenon remain underdeveloped. The objectives of this study were to apply whole-genome sequencing to systematically resolve and track plasmids carried by nosocomial pathogens in a single hospital, and to identify epidemiologic links that indicated likely horizontal plasmid transfer. METHODS: We performed an observational study of plasmids circulating among bacterial isolates infecting patients at a large hospital. We first examined plasmids carried by isolates sampled from the same patient over time and isolates that caused clonal outbreaks in the same hospital to develop thresholds with which horizontal plasmid transfer within a tertiary hospital could be inferred. We then applied those sequence similarity thresholds to perform a systematic screen of 3074 genomes of nosocomial bacterial isolates from a single hospital for the presence of 89 plasmids. We also collected and reviewed data from electronic health records for evidence of geotemporal links between patients infected with bacteria encoding plasmids of interest. FINDINGS: Our analyses determined that 95% of analyzed genomes maintained roughly 95% of their plasmid genetic content and accumulated fewer than 15 SNPs per 100 kb of plasmid sequence. Applying these similarity thresholds to identify horizontal plasmid transfer identified 45 plasmids that potentially circulated among clinical isolates. Ten highly preserved plasmids met criteria for geotemporal links associated with horizontal transfer. Several plasmids with shared backbones also encoded different additional mobile genetic element content, and these elements were variably present among the sampled clinical isolate genomes. INTERPRETATION: Evidence suggests that the horizontal transfer of plasmids among nosocomial bacterial pathogens appears to be frequent within hospitals and can be monitored with whole genome sequencing and comparative genomics approaches. These approaches should incorporate both nucleotide identity and reference sequence coverage to study the dynamics of plasmid transfer in the hospital. FUNDING: This research was supported by the US National Institute of Allergy and Infectious Disease (NIAID) and the University of Pittsburgh School of Medicine.
Subject(s)
Anti-Bacterial Agents , Cross Infection , Humans , Plasmids/genetics , Genomics , Bacteria/genetics , Cross Infection/epidemiology , Genome, BacterialABSTRACT
The idea that we can detect subacute potentially catastrophic illness earlier by using statistical models trained on clinical data is now well-established. We review evidence that supports the role of continuous cardiorespiratory monitoring in these predictive analytics monitoring tools. In particular, we review how continuous ECG monitoring reflects the patient and not the clinician, is less likely to be biased, is unaffected by changes in practice patterns, captures signatures of illnesses that are interpretable by clinicians, and is an underappreciated and underutilized source of detailed information for new mathematical methods to reveal.
Subject(s)
Clinical Deterioration , Electrocardiography , Humans , Electrocardiography/methods , Monitoring, Physiologic , Models, Statistical , Artificial IntelligenceABSTRACT
Background: Whole-genome sequencing (WGS) has traditionally been used in infection prevention to confirm or refute the presence of an outbreak after it has occurred. Due to decreasing costs of WGS, an increasing number of institutions have been utilizing WGS-based surveillance. Additionally, machine learning or statistical modeling to supplement infection prevention practice have also been used. We systematically reviewed the use of WGS surveillance and machine learning to detect and investigate outbreaks in healthcare settings. Methods: We performed a PubMed search using separate terms for WGS surveillance and/or machine-learning technologies for infection prevention through March 15, 2021. Results: Of 767 studies returned using the WGS search terms, 42 articles were included for review. Only 2 studies (4.8%) were performed in real time, and 39 (92.9%) studied only 1 pathogen. Nearly all studies (n = 41, 97.6%) found genetic relatedness between some isolates collected. Across all studies, 525 outbreaks were detected among 2,837 related isolates (average, 5.4 isolates per outbreak). Also, 35 studies (83.3%) only utilized geotemporal clustering to identify outbreak transmission routes. Of 21 studies identified using the machine-learning search terms, 4 were included for review. In each study, machine learning aided outbreak investigations by complementing methods to gather epidemiologic data and automating identification of transmission pathways. Conclusions: WGS surveillance is an emerging method that can enhance outbreak detection. Machine learning has the potential to identify novel routes of pathogen transmission. Broader incorporation of WGS surveillance into infection prevention practice has the potential to transform the detection and control of healthcare outbreaks.
ABSTRACT
For fluid resuscitation of critically ill individuals to be effective, it must be well calibrated in terms of timing and dosages of treatments. In current practice, the cardiovascular sufficiency of patients during fluid resuscitation is determined using primarily invasively measured vital signs, including Arterial Pressure and Mixed Venous Oxygen Saturation (SvO2), which may not be available in outside-of-hospital settings, particularly in the field when treating subjects injured in traffic accidents or wounded in combat where only non-invasive monitoring is available to drive care. In this paper, we propose (1) a Machine Learning (ML) approach to estimate the sufficiency utilizing features extracted from non-invasive vital signs and (2) a novel framework to address the detrimental impact of inter-patient diversity on the ability of ML models to generalize well to unseen subjects. Through comprehensive evaluation on the physiological data collected in laboratory animal experiments, we demonstrate that the proposed approaches can achieve competitive performance on new patients using only non-invasive measurements. These characteristics enable effective monitoring of fluid resuscitation in real-world acute settings with limited monitoring resources and can help facilitate broader adoption of ML in this important subfield of healthcare.
Subject(s)
Cardiovascular System , Fluid Therapy , Animals , Critical Illness , Heart , Humans , OximetryABSTRACT
Early recognition of pathologic cardiorespiratory stress and forecasting cardiorespiratory decompensation in the critically ill is difficult even in highly monitored patients in the Intensive Care Unit (ICU). Instability can be intuitively defined as the overt manifestation of the failure of the host to adequately respond to cardiorespiratory stress. The enormous volume of patient data available in ICU environments, both of high-frequency numeric and waveform data accessible from bedside monitors, plus Electronic Health Record (EHR) data, presents a platform ripe for Artificial Intelligence (AI) approaches for the detection and forecasting of instability, and data-driven intelligent clinical decision support (CDS). Building unbiased, reliable, and usable AI-based systems across health care sites is rapidly becoming a high priority, specifically as these systems relate to diagnostics, forecasting, and bedside clinical decision support. The ICU environment is particularly well-positioned to demonstrate the value of AI in saving lives. The goal is to create AI models embedded in a real-time CDS for forecasting and mitigation of critical instability in ICU patients of sufficient readiness to be deployed at the bedside. Such a system must leverage multi-source patient data, machine learning, systems engineering, and human action expertise, the latter being key to successful CDS implementation in the clinical workflow and evaluation of bias. We present one approach to create an operationally relevant AI-based forecasting CDS system.
Subject(s)
Decision Support Systems, Clinical , Artificial Intelligence , Critical Care , Humans , Intensive Care Units , Machine LearningABSTRACT
We consider whether one can forecast the emergence of variants of concern in the SARS-CoV-2 outbreak and similar pandemics. We explore methods of population genetics and identify key relevant principles in both deterministic and stochastic models of spread of infectious disease. Finally, we demonstrate that fitness variation, defined as a trait for which an increase in its value is associated with an increase in net Darwinian fitness if the value of other traits are held constant, is a strong indicator of imminent transition in the viral population.
Subject(s)
COVID-19/epidemiology , Forecasting/methods , SARS-CoV-2/genetics , COVID-19/transmission , Epidemiological Models , Genetic Fitness/genetics , Genetics, Population/methods , Humans , Pandemics , SARS-CoV-2/pathogenicityABSTRACT
BACKGROUND: Artificial Intelligence (AI) is increasingly used to support bedside clinical decisions, but information must be presented in usable ways within workflow. Graphical User Interfaces (GUI) are front-facing presentations for communicating AI outputs, but clinicians are not routinely invited to participate in their design, hindering AI solution potential. PURPOSE: To inform early user-engaged design of a GUI prototype aimed at predicting future Cardiorespiratory Insufficiency (CRI) by exploring clinician methods for identifying at-risk patients, previous experience with implementing new technologies into clinical workflow, and user perspectives on GUI screen changes. METHODS: We conducted a qualitative focus group study to elicit iterative design feedback from clinical end-users on an early GUI prototype display. Five online focus group sessions were held, each moderated by an expert focus group methodologist. Iterative design changes were made sequentially, and the updated GUI display was presented to the next group of participants. RESULTS: 23 clinicians were recruited (14 nurses, 4 nurse practitioners, 5 physicians; median participant age â¼35 years; 60% female; median clinical experience 8 years). Five themes emerged from thematic content analysis: trend evolution, context (risk evolution relative to vital signs and interventions), evaluation/interpretation/explanation (sub theme: continuity of evaluation), clinician intuition, and clinical operations. Based on these themes, GUI display changes were made. For example, color and scale adjustments, integration of clinical information, and threshold personalization. CONCLUSIONS: Early user-engaged design was useful in adjusting GUI presentation of AI output. Next steps involve clinical testing and further design modification of the AI output to optimally facilitate clinician surveillance and decisions. Clinicians should be involved early and often in clinical decision support design to optimize efficacy of AI tools.
Subject(s)
Decision Support Systems, Clinical , Physicians , Adult , Artificial Intelligence , Delivery of Health Care , Female , Humans , Male , WorkflowABSTRACT
A significant proportion of clinical physiologic monitoring alarms are false. This often leads to alarm fatigue in clinical personnel, inevitably compromising patient safety. To combat this issue, researchers have attempted to build Machine Learning (ML) models capable of accurately adjudicating Vital Sign (VS) alerts raised at the bedside of hemodynamically monitored patients as real or artifact. Previous studies have utilized supervised ML techniques that require substantial amounts of hand-labeled data. However, manually harvesting such data can be costly, time-consuming, and mundane, and is a key factor limiting the widespread adoption of ML in healthcare (HC). Instead, we explore the use of multiple, individually imperfect heuristics to automatically assign probabilistic labels to unlabeled training data using weak supervision. Our weakly supervised models perform competitively with traditional supervised techniques and require less involvement from domain experts, demonstrating their use as efficient and practical alternatives to supervised learning in HC applications of ML.
Subject(s)
Artifacts , Monitoring, Physiologic , Supervised Machine Learning , Vital Signs , Humans , Monitoring, Physiologic/methods , Monitoring, Physiologic/standards , Heuristics , AutomationABSTRACT
BACKGROUND: Most hospitals use traditional infection prevention (IP) methods for outbreak detection. We developed the Enhanced Detection System for Healthcare-Associated Transmission (EDS-HAT), which combines whole-genome sequencing (WGS) surveillance and machine learning (ML) of the electronic health record (EHR) to identify undetected outbreaks and the responsible transmission routes, respectively. METHODS: We performed WGS surveillance of healthcare-associated bacterial pathogens from November 2016 to November 2018. EHR ML was used to identify the transmission routes for WGS-detected outbreaks, which were investigated by an IP expert. Potential infections prevented were estimated and compared with traditional IP practice during the same period. RESULTS: Of 3165 isolates, there were 2752 unique patient isolates in 99 clusters involving 297 (10.8%) patient isolates identified by WGS; clusters ranged from 2-14 patients. At least 1 transmission route was detected for 65.7% of clusters. During the same time, traditional IP investigation prompted WGS for 15 suspected outbreaks involving 133 patients, for which transmission events were identified for 5 (3.8%). If EDS-HAT had been running in real time, 25-63 transmissions could have been prevented. EDS-HAT was found to be cost-saving and more effective than traditional IP practice, with overall savings of $192â 408-$692â 532. CONCLUSIONS: EDS-HAT detected multiple outbreaks not identified using traditional IP methods, correctly identified the transmission routes for most outbreaks, and would save the hospital substantial costs. Traditional IP practice misidentified outbreaks for which transmission did not occur. WGS surveillance combined with EHR ML has the potential to save costs and enhance patient safety.
Subject(s)
Cross Infection , Electronic Health Records , Cross Infection/epidemiology , Cross Infection/microbiology , Cross Infection/prevention & control , Delivery of Health Care , Disease Outbreaks , Genome, Bacterial , Humans , Machine Learning , Whole Genome Sequencing/methodsABSTRACT
Big data analytics research using heterogeneous electronic health record (EHR) data requires accurate identification of disease phenotype cases and controls. Overreliance on ground truth determination based on administrative data can lead to biased and inaccurate findings. Hospital-acquired venous thromboembolism (HA-VTE) is challenging to identify due to its temporal evolution and variable EHR documentation. To establish ground truth for machine learning modeling, we compared accuracy of HA-VTE diagnoses made by administrative coding to manual review of gold standard diagnostic test results. We performed retrospective analysis of EHR data on 3680 adult stepdown unit patients identifying HA-VTE. International Classification of Diseases, Ninth Revision (ICD-9-CM) codes for VTE were identified. 4544 radiology reports associated with VTE diagnostic tests were screened using terminology extraction and then manually reviewed by a clinical expert to confirm diagnosis. Of 415 cases with ICD-9-CM codes for VTE, 219 were identified with acute onset type codes. Test report review identified 158 new-onset HA-VTE cases. Only 40% of ICD-9-CM coded cases (n = 87) were confirmed by a positive diagnostic test report, leaving the majority of administratively coded cases unsubstantiated by confirmatory diagnostic test. Additionally, 45% of diagnostic test confirmed HA-VTE cases lacked corresponding ICD codes. ICD-9-CM coding missed diagnostic test-confirmed HA-VTE cases and inaccurately assigned cases without confirmed VTE, suggesting dependence on administrative coding leads to inaccurate HA-VTE phenotyping. Alternative methods to develop more sensitive and specific VTE phenotype solutions portable across EHR vendor data are needed to support case-finding in big-data analytics.
Subject(s)
Venous Thromboembolism , Big Data , Hospitals , Humans , Machine Learning , Retrospective Studies , Venous Thromboembolism/diagnosisABSTRACT
BACKGROUND: We explored sex-based differences in discharge location after resuscitation from cardiac arrest. METHODS: We performed a single-center retrospective cohort study including patients hospitalized after resuscitation from cardiac arrest from January 2010 to May 2020. We identified patients from a prospective registry, from which we extracted standard demographic and clinical variables. We explored favorable discharge location, defined as discharge to home or acute rehabilitation for survivors to hospital discharge. We tested the association of sex with the residuals of a multivariable logistic regression built using bidirectional selection to control for clinically relevant covariates. RESULTS: We included 2,278 patients. Mean age was 59 (SD 16), 40% were women, and 77% were admitted after out-of-hospital cardiac arrest. A total of 970 patients (43%) survived to discharge; of those, 607 (63% of survivors) had a favorable discharge location. Female sex showed a weak independent association with unfavorable discharge location (adjusted OR 0.94 (95%CI 0.89-0.99)). CONCLUSIONS: Our results suggest a possible sex-based disparity in discharge location after cardiac arrest.
ABSTRACT
Background This study evaluated the role of supplementing Society of Thoracic Surgeons (STS) risk models for surgical aortic valve replacement with machine learning (ML). Methods and Results Adults undergoing isolated surgical aortic valve replacement in the STS National Database between 2007 and 2017 were included. ML models for operative mortality and major morbidity were previously developed using extreme gradient boosting. Concordance and discordance in predicted risk between ML and STS models were defined using equal-size tertile-based thresholds of risk. Calibration metrics and discriminatory capability were compared between concordant and discordant patients. A total of 243 142 patients were included. Nearly all calibration metrics were improved in cases of concordance. Similarly, concordance indices improved substantially in cases of concordance for all models with the exception of deep sternal wound infection. The greatest improvements in concordant versus discordant cases were in renal failure: ML model (concordance index, 0.660 [95% CI, 0.632-0.687] discordant versus 0.808 [95% CI, 0.794-0.822] concordant) and STS model (concordance index, 0.573 [95% CI, 0.549-0.576] discordant versus 0.797 [95% CI, 0.782-0.811] concordant) (each P<0.001). Excluding deep sternal wound infection, the concordance indices ranged from 0.549 to 0.660 for discordant cases and 0.674 to 0.808 for concordant cases. Conclusions Supplementing ML models with existing STS models for surgical aortic valve replacement may have an important role in risk prediction and should be explored further. In particular, for the roughly 25% to 50% of patients demonstrating discordance in estimated risk between ML and STS, there appears to be a substantial decline in predictive performance suggesting vulnerability of the existing models in these patient subsets.
Subject(s)
Aortic Valve Stenosis , Heart Valve Prosthesis Implantation , Heart Valve Prosthesis , Wound Infection , Adult , Aortic Valve/surgery , Aortic Valve Stenosis/epidemiology , Aortic Valve Stenosis/surgery , Heart Valve Prosthesis Implantation/adverse effects , Humans , Machine Learning , Risk Assessment , Risk Factors , Treatment OutcomeABSTRACT
Objective.To develop a standardized format for exchanging clinical and physiologic data generated in the intensive care unit. Our goal was to develop a format that would accommodate the data collection pipelines of various sites but would not require dataset-specific schemas or ad-hoc tools for decoding and analysis.Approach.A number of centers had independently developed solutions for storing clinical and physiologic data using Hierarchical Data Format-Version 5 (HDF5), a well-supported standard already in use in multiple other fields. These individual solutions involved design choices that made the data difficult to share despite the underlying common framework. A collaborative process was used to form the basis of a proposed standard that would allow for interoperability and data sharing with common analysis tools.Main Results.We developed the HDF5-based critical care data exchange format which stores multiparametric data in an efficient, self-describing, hierarchical structure and supports real-time streaming and compression. In addition to cardiorespiratory and laboratory data, the format can, in future, accommodate other large datasets such as imaging and genomics. We demonstated the feasibility of a standardized format by converting data from three sites as well as the MIMIC III dataset.Significance.Individual approaches to the storage of multiparametric clinical data are proliferating, representing both a duplication of effort and a missed opportunity for collaboration. Adoption of a standardized format for clinical data exchange will enable the development of a digital biobank, facilitate the external validation of machine learning models and be a powerful tool for sharing multiparametric, high frequency patient level data in multisite clinical trials. Our proposed solution focuses on supporting standardized ontologies such as LOINC allowing easy reading of data regardless of the source and in so doing provides a useful method to integrate large amounts of existing data.
Subject(s)
Critical Care , Genomics , Humans , Intensive Care UnitsABSTRACT
Risk models have historically displayed only moderate predictive performance in estimating mortality risk in left ventricular assist device therapy. This study evaluated whether machine learning can improve risk prediction for left ventricular assist devices. Primary durable left ventricular assist devices reported in the Interagency Registry for Mechanically Assisted Circulatory Support between March 1, 2006 and December 31, 2016 were included. The study cohort was randomly divided 3:1 into training and testing sets. Logistic regression and machine learning models (extreme gradient boosting) were created in the training set for 90-day and 1-year mortality and their performance was evaluated after bootstrapping with 1000 replications in the testing set. Differences in model performance were also evaluated in cases of concordance versus discordance in predicted risk between logistic regression and extreme gradient boosting as defined by equal size patient tertiles. A total of 16,120 patients were included. Calibration metrics were comparable between logistic regression and extreme gradient boosting. C-index was improved with extreme gradient boosting (90-day: 0.707 [0.683-0.730] versus 0.740 [0.717-0.762] and 1-year: 0.691 [0.673-0.710] versus 0.714 [0.695-0.734]; each p<0.001). Net reclassification index analysis similarly demonstrated an improvement of 48.8% and 36.9% for 90-day and 1-year mortality, respectively, with extreme gradient boosting (each p<0.001). Concordance in predicted risk between logistic regression and extreme gradient boosting resulted in substantially improved c-index for both logistic regression and extreme gradient boosting (90-day logistic regression 0.536 versus 0.752, 1-year logistic regression 0.555 versus 0.726, 90-day extreme gradient boosting 0.623 versus 0.772, 1-year extreme gradient boosting 0.613 versus 0.742, each p<0.001). These results demonstrate that machine learning can improve risk model performance for durable left ventricular assist devices, both independently and as an adjunct to logistic regression.
Subject(s)
Forecasting/methods , Heart-Assist Devices/trends , Ventricular Dysfunction, Left/surgery , Cohort Studies , Decision Support Systems, Clinical/trends , Humans , Logistic Models , Machine Learning , Models, Statistical , Risk FactorsABSTRACT
We describe a new approach to estimating relative risks in time-to-event prediction problems with censored data in a fully parametric manner. Our approach does not require making strong assumptions of constant proportional hazards of the underlying survival distribution, as required by the Cox-proportional hazard model. By jointly learning deep nonlinear representations of the input covariates, we demonstrate the benefits of our approach when used to estimate survival risks through extensive experimentation on multiple real world datasets with different levels of censoring. We further demonstrate advantages of our model in the competing risks scenario. To the best of our knowledge, this is the first work involving fully parametric estimation of survival times with competing risks in the presence of censoring.
Subject(s)
Models, Statistical , Humans , Proportional Hazards Models , Risk , Survival AnalysisABSTRACT
The adoption of best practices has been shown to increase performance in healthcare institutions and is consistently demanded by both patients, payers, and external overseers. Nevertheless, transferring practices between healthcare organizations is a challenging and underexplored task. In this paper, we take a step towards enabling the transfer of best practices by identifying the likely beneficial opportunities for such transfer. Specifically, we analyze the output of machine learning models trained at different organizations with the aims of (i) detecting the opportunity for the transfer of best practices, and (ii) providing a stop-gap solution while the actual transfer process takes place. We show the benefits ofthis methodology on a dataset ofmedical inpatient claims, demonstrating our abilityto identify practice gaps and to support the transfer processes that address these gaps.
Subject(s)
Delivery of Health Care , Machine Learning , Health Facilities , HumansABSTRACT
Analysing electrocardiograms (ECGs) is an inexpensive and non-invasive, yet powerful way to diagnose heart disease. ECG studies using Machine Learning to automatically detect abnormal heartbeats so far depend on large, manually annotated datasets. While collecting vast amounts of unlabeled data can be straightforward, the point-by-point annotation of abnormal heartbeats is tedious and expensive. We explore the use of multiple weak supervision sources to learn diagnostic models of abnormal heartbeats via human designed heuristics, without using ground truth labels on individual data points. Our work is among the first to define weak supervision sources directly on time series data. Results show that with as few as six intuitive time series heuristics, we are able to infer high quality probabilistic label estimates for over 100,000 heartbeats with little human effort, and use the estimated labels to train competitive classifiers evaluated on held out test data.