Predicting 30-Day Heart Failure Readmissions Using Machine Learning: Insights From the Kansas Health Information Network (KHIN)
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Heart failure (HF) is a major contributor to inpatient hospital utilization, with persistently high 30-day readmission rates. Existing prediction tools are frequently restricted to primary-diagnosis HF admissions, potentially excluding clinically relevant HF-related hospitalizations.
Objectives
To develop and validate risk prediction models using machine learning (ML)-based risk prediction models to predict 30-day readmissions among patients with HF using the Kansas Health Information Network, a statewide health information exchange.
Methods
This retrospective cohort study analyzed HF hospitalizations using predictors including demographics, comorbidities, laboratory results, medications, clinical quality metrics for diabetes and kidney disease management, and prior healthcare utilization. Five ML models, including regularized logistic regression, random forest, extreme gradient boosting, categorical boosting, and deep neural network, were trained using stratified 5-fold cross-validation. Model performance was evaluated on an independent test set using the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), misclassification rate (MCR), and Brier score.
Results
Among 2,734 HF patients, the 30-day readmission rate was 27%. The XGBoost model achieved the best discrimination (AUROC=0.75; AUPRC=0.58; MCR=0.21). Patients in the highest-risk decile had a positive predictive value of 76%, accounted for approximately one-third of all 30-day readmissions, and had a 3.3-fold enrichment compared with baseline risk. The key predictors included prior hospital utilization, diabetes and kidney disease management indicators, and comorbidity burden.
Conclusions
Risk stratification using routinely collected clinical data identified a subgroup at elevated risk for 30-day readmission. These findings support the potential role of data-driven risk prediction to inform targeted transitional care.
CLINICAL PERSPECTIVES
What is New?
This study is among the first to utilize health information exchange data to develop machine learning models for predicting 30-day heart failure readmissions, including patients with HF as either a primary or secondary diagnosis. Unlike prior models restricted to primary HF admissions, this approach includes hospitalizations in any diagnosis position, reflecting real-world acute-care utilization. Notably, our machine learning model correctly identified approximately three in four patients classified as high-risk in the top decile of predicted scores, highlighting its ability to identify individuals at greatest likelihood of early readmission
What are the Clinical Implications?
Accurately identifying a group of high-risk patients has meaningful implications for care delivery. By correctly capturing actual readmissions within a small, prioritized segment of the population, ML-driven risk stratification can guide clinicians and care management teams in targeting transitional care resources where they are most needed. Such focused intervention strategies may allow for efficient targeting of limited health care resources, improve post-discharge outcomes, enhance operational efficiency, and potentially reduce preventable HF readmissions.
