Machine-learning algorithms for predicting hospital re-admissions in sickle cell disease
Arisha Patel1, Kyra Gan2, Andrew A Li2
1Tepper School of Business, Carnegie Mellon University, Pittsburgh, PA, USA.
British Journal of Haematology
|November 10, 2020
Summary
Machine learning algorithms significantly outperformed standard scoring systems in predicting hospital readmissions for Sickle Cell Disease patients. This advancement offers a promising approach to reduce readmissions and improve patient care.
Area of Science:
- Health Informatics
- Computational Medicine
- Hematology
Background:
- Preventable hospital readmissions in Sickle Cell Disease (SCD) contribute to poor patient outcomes and increased healthcare expenditures.
- Existing scoring systems (LACE and HOSPITAL indices) may not accurately predict readmission risk in high-risk patient populations like those with SCD.
Purpose of the Study:
- To evaluate the efficacy of Machine Learning (ML) algorithms in predicting unplanned hospital readmissions for patients with Sickle Cell Disease.
- To compare the predictive performance of ML models against established scoring systems (LACE and HOSPITAL indices).
Main Methods:
- A retrospective study analyzed electronic health records of 446 SCD patients with at least one unplanned inpatient stay between 2013 and 2018.
- Potential predictors (n=486) were extracted using data-driven methods and clinical knowledge.
- Three ML algorithms—Logistic Regression (LR), Support-Vector Machine (SVM), and Random Forest (RF)—were applied and compared using C-statistic, sensitivity, and specificity.
Main Results:
- ML algorithms demonstrated superior performance compared to LACE (C-statistic 0.6) and HOSPITAL (C-statistic 0.69) indices.
- The Random Forest (RF) and Logistic Regression (LR) models achieved the highest predictive accuracy, with a C-statistic of 0.77.
- Key predictors for readmission were identified, providing insights for targeted interventions.
Conclusions:
- Machine Learning algorithms are effective tools for predicting hospital readmissions in high-risk SCD patient groups.
- Implementing ML-based prediction models can potentially reduce readmissions, improve patient outcomes, and decrease healthcare costs in SCD management.
