Predicting 30-Day Hospital Readmission in Medicare Patients: Insights from an LSTM Deep Learning Model

BackgroundReadmissions among Medicare beneficiaries are a major problem for the US healthcare system from a perspective of both healthcare operations and patient caregiving outcomes. Our study analyzes Medicare hospital readmissions using LSTM networks with feature engineering to assess feature contributions. DesignThe 21002 senior patient admission data from MIMIC-III clinical database at Beth Israel Deaconess Medical Center.is utilized in the study We selected variables from admission-level data, inpatient medical history and patient demography. The baseline model is a logistic-regression model based on the LACE index, and the LSTM model is designed to capture temporal dynamic in the data from admission-level and patient-level data. We leveraged Area Under the Curve metric, precision and recall to evaluate the models performance. ResultsThe LSTM model outperformed the logistic regression baseline, accurately leveraging temporal features to predict readmission. The major features were the Charlson Comorbidity Index, hospital length of stay, the hospital admissions over the past 6 months or the number of medications before discharge, while demographic variables were less impactful LimitationsThe use of a single-center database from the MIMIC-III database limits the generalizability of the findings. Additionally, the exclusion for specific chronic conditions and external factors limit the models ability to capture the complexities of chronic diseases. ConclusionsThis work suggests that LSTM networks offers a more promising approach to improve Medicare patient readmission prediction. It captures temporal interactions in patient databases, enhancing current prediction models for healthcare providers. ImplicationsAdoption of predictive models into clinical practice may be more effective in identifying Medicare patients to provide early and targeted interventions to improve patient outcomes. HighlightsO_LIImproved Prediction: Our LSTM model outperforms the logistic regression model with LACE index in predicting Medicare patient readmissions. C_LIO_LIFeature Contribution: Feature engineering ranks variables base on the impact, deprioritizing the importance of patient demographic variables, highlighting the importance of patients chronic diseases in leading hospitalizations and guiding targeted interventions to prevent senior hospital readmissions for healthcare providers. C_LIO_LIEffective Use of Data: Our LSTM model incorporates with time-series data from MIMIC-III database to enhance the accuracy of all-cause hospital readmission predictions, especially for the high-risk patients. C_LIO_LIActionable Insights: The result demonstrates the utilization of deep learning model in healthcare decision-making to reduce hospital readmissions for seniors. C_LI