Comparison of Machine Learning Algorithms for Predicting Hospital Readmissions and Worsening Heart Failure Events in Patients With Heart Failure With Reduced Ejection Fraction: Modeling Study

BACKGROUND

Heart failure (HF) is highly prevalent in the United States. Approximately one-third to one-half of HF cases are categorized as HF with reduced ejection fraction (HFrEF). Patients with HFrEF are at risk of worsening HF, have a high risk of adverse outcomes, and experience higher health care use and costs. Therefore, it is crucial to identify patients with HFrEF who are at high risk of subsequent events after HF hospitalization.

OBJECTIVE

Machine learning (ML) has been used to predict HF-related outcomes. The objective of this study was to compare different ML prediction models and feature construction methods to predict 30-, 90-, and 365-day hospital readmissions and worsening HF events (WHFEs).

METHODS

We used the Veradigm PINNACLE outpatient registry linked to Symphony Health's Integrated Dataverse data from July 1, 2013, to September 30, 2017. Adults with a confirmed diagnosis of HFrEF and HF-related hospitalization were included. WHFEs were defined as HF-related hospitalizations or outpatient intravenous diuretic use within 1 year of the first HF hospitalization. We used different approaches to construct ML features from clinical codes, including frequencies of clinical classification software (CCS) categories, Bidirectional Encoder Representations From Transformers (BERT) trained with CCS sequences (BERT + CCS), BERT trained on raw clinical codes (BERT + raw), and prespecified features based on clinical knowledge. A multilayer perceptron neural network, extreme gradient boosting (XGBoost), random forest, and logistic regression prediction models were applied and compared.

RESULTS

A total of 30,687 adult patients with HFrEF were included in the analysis; 11.41% (3184/27,917) of adults experienced a hospital readmission within 30 days of their first HF hospitalization, and nearly half (9231/21,562, 42.81%) of the patients experienced at least 1 WHFE within 1 year after HF hospitalization. The prediction models and feature combinations with the best area under the receiver operating characteristic curve (AUC) for each outcome were XGBoost with CCS frequency (AUC=0.595) for 30-day readmission, random forest with CCS frequency (AUC=0.630) for 90-day readmission, XGBoost with CCS frequency (AUC=0.649) for 365-day readmission, and XGBoost with CCS frequency (AUC=0.640) for WHFEs. Our ML models could discriminate between readmission and WHFE among patients with HFrEF. Our model performance was mediocre, especially for the 30-day readmission events, most likely owing to limitations of the data, including an imbalance between positive and negative cases and high missing rates of many clinical variables and outcome definitions.

CONCLUSIONS

We predicted readmissions and WHFEs after HF hospitalizations in patients with HFrEF. Features identified by data-driven approaches may be comparable with those identified by clinical domain knowledge. Future work may be warranted to validate and improve the models using more longitudinal electronic health records that are complete, are comprehensive, and have a longer follow-up time.

Calibrating a Transformer-Based Model's Confidence on Community-Engaged Research Studies: Decision Support Evaluation Study

BACKGROUND

Deep learning offers great benefits in classification tasks such as medical imaging diagnostics or stock trading, especially when compared with human-level performances, and can be a viable option for classifying distinct levels within community-engaged research (CEnR). CEnR is a collaborative approach between academics and community partners with the aim of conducting research that is relevant to community needs while incorporating diverse forms of expertise. In the field of deep learning and artificial intelligence (AI), training multiple models to obtain the highest validation accuracy is common practice; however, it can overfit toward that specific data set and not generalize well to a real-world population, which creates issues of bias and potentially dangerous algorithmic decisions. Consequently, if we plan on automating human decision-making, there is a need for creating techniques and exhaustive evaluative processes for these powerful unexplainable models to ensure that we do not incorporate and blindly trust poor AI models to make real-world decisions.

OBJECTIVE

We aimed to conduct an evaluation study to see whether our most accurate transformer-based models derived from previous studies could emulate our own classification spectrum for tracking CEnR studies as well as whether the use of calibrated confidence scores was meaningful.

METHODS

We compared the results from 3 domain experts, who classified a sample of 45 studies derived from our university's institutional review board database, with those from 3 previously trained transformer-based models, as well as investigated whether calibrated confidence scores can be a viable technique for using AI in a support role for complex decision-making systems.

RESULTS

Our findings reveal that certain models exhibit an overestimation of their performance through high confidence scores, despite not achieving the highest validation accuracy.

CONCLUSIONS

Future studies should be conducted with larger sample sizes to generalize the results more effectively. Although our study addresses the concerns of bias and overfitting in deep learning models, there is a need to further explore methods that allow domain experts to trust our models more. The use of a calibrated confidence score can be a misleading metric when determining our AI model's level of competency.