A comparison of models for predicting early hospital readmissions

Abstract

Risk sharing arrangements between hospitals and payers together with penalties imposed by the Centers for Medicare and Medicaid (CMS) are driving an interest in decreasing early readmissions. There are a number of published risk models predicting 30day readmissions for particular patient populations, however they often exhibit poor predictive performance and would be unsuitable for use in a clinical setting. In this work we describe and compare several predictive models, some of which have never been applied to this task and which outperform the regression methods that are typically applied in the healthcare literature. In addition, we apply methods from deep learning to the five conditions CMS is using to penalize hospitals, and offer a simple framework for determining which conditions are most cost effective to target.

Predicting all-cause risk of 30-day hospital readmission using artificial neural networks

“…Numerous attempts to build such predictive models have been made [6–12]. However, the majority of them suffer from at least one of the following shortcomings: (1) the model is not predictive enough compared to LACE [11], the industry-standard scoring model [13], (2) the model uses insurance claim data, which would not be available in a real-time clinical setting [6,7], (3) the model does not consider social determinants of health (SDoH) [13,8], which have proven to be predictive [14], (4) the model is limited to a particular medical condition, and thus, limited in scope [9,10]. …”

Machine learning‐based prediction of heart failure readmission or death: implications of choosing the right model and the right metrics

“…9,11,13 However, our MLP-based approach produces a comparable AUC of 0.63 with only 47 variables (vs. hundreds in other studies); signifying more variables may not necessarily lead to better prediction, and basic variables from the core hospitalization and death datasets are sufficient. 9,11,13 However, our MLP-based approach produces a comparable AUC of 0.63 with only 47 variables (vs. hundreds in other studies); signifying more variables may not necessarily lead to better prediction, and basic variables from the core hospitalization and death datasets are sufficient.…”

“…Koulaouzidis et al 23 used the naive Bayes classifier on telemonitored data, such as left ventricular systolic dysfunction, New York Health Association score, co-morbidities, blood pressure, and medications, to predict HF readmissions. 11,16 Recently, Frizzell et al 10 compared the effectiveness of ML algorithms and standard regression methods to predict 30 day all-cause readmissions in HF patients. 11,16 Recently, Frizzell et al 10 compared the effectiveness of ML algorithms and standard regression methods to predict 30 day all-cause readmissions in HF patients.…”

“…7 This is because the availability of large amounts of medical data has the potential to be used in new ways to gain insights on how to improve healthcare outcomes. [9][10][11][12][13][14][15] Lastly, most ML studies have used area under the receiver operating characteristic curve (AUC) as the only performance metric to assess the model performance. 8 Previous studies have complemented administrative data with clinical information to develop predictive models.…”

Predictive Modeling of Hospital Readmission Rates Using Electronic Medical Record-Wide Machine Learning: A Case-Study Using Mount Sinai Heart Failure Cohort

“…Using a diabetes cohort from a hospital in India, Duggal et.al 17 showed that Naïve Bayes (0.67) showed higher readmission associated savings compared to logistic regression (0.67), Random Forests (0.68), Adaboost (0.67) and Neural Networks (0.62). Futoma et.al 30 showed that Random Forests (0.68) and deep learning using neural networks (0.67) have similar accuracy rate with >1 million patients and > 3 million admission. However, Penalized Logistic Regression had similar accuracy rates as we have shown in our orthogonal validation methods.…”