Applying Deep Learning in Heart Failure: Hospital Readmission is Not Like Other Health Quality Metrics

Background Early identification of heart failure patients at increased risk for near-term adverse outcomes would assist clinicians in efficient resource allocation and improved care. Deep learning can improve identification of these patients. Methods This retrospective study examined adult heart failure patients admitted to a tertiary care institution between January 2009 and December 2018. A deep learning model was constructed with a dense input layer, three long short-term memory (LSTM) layers, and a dense hidden layer to cohesively extract features from time-series and non-time-series EHR data. Primary outcomes were all-cause hospital readmission or death within 30 days after hospital discharge. Results Among a final subset of 49,675 heart failure patients, we identified 171,563 hospital admissions described by 330 million EHR data points. There were 22,111 (13%) admissions followed by adverse 30-day outcomes, including 19,122 readmissions (87%) and mortality in 3,330 patients (15%). Our final deep learning model achieved an area under the receiver-operator characteristic curve (AUC) of 0.613 and precision-recall (PR) AUC of 0.38. Conclusions This EHR-based deep learning model developed from a decade of heart failure care achieved marginal clinical accuracy in predicting very early hospital readmission or death despite previous accurate prediction of 1-year mortality in this large study cohort. These findings suggest that factors unavailable in standard EHR data play pivotal roles in influencing early hospital readmission. ### Competing Interest Statement The authors have declared no competing interest. ### Funding Statement No external funding received. ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Not Applicable The details of the IRB/oversight body that provided approval or exemption for the research described are given below: The human studies Institutional Review Board at Washington University School of Medicine approved this study. I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Not Applicable I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Not Applicable I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Not Applicable Data is available upon request. Provision of patient-protected data is not possible to protect patient identity.