Key result
A Gradient Boosting model predicted 30-day hospital readmission with an AUC of 97.3%, identifying previous admissions, hyperosmolar imbalance, and comorbidity index as top risk factors.
Why the study?
Unplanned hospital readmissions within 30 days are frequent and costly, prompting a descriptive and predictive analysis to understand risk factors and predict readmissions.
Can machine learning models using EHR data predict unplanned hospital readmission within 30 days?
Observational
Can machine learning models using EHR data predict unplanned hospital readmission within 30 days?
Answers are grounded in this paper and cite their sources.
A machine learning model using EHR data can highly accurately predict 30-day unplanned hospital readmissions.
Hospital readmission involves the unplanned emergency admission of patients within 30 days from discharge after the previous admission. According to the Canadian Health Institute (CIHI), 1 in 11 patients were readmitted within 30 days of leaving the hospital in 2021. In the USA, nearly 20% of Medicare patients were readmitted after discharge, where the average cost of readmission was approximately USD 15,000, as reported by the Agency for Healthcare Research and Quality (AHQR) in 2018. To tackle this issue, we first conducted a descriptive analysis study to understand the risk factors associated with hospital readmission, and then we applied machine learning approaches to predict hospital readmission by using patients’ demographic and clinical data extracted from the Electronic Health Record of the MIMIC-III clinical database. The results showed that the number of previous admissions during the last 12 months, hyperosmolar imbalance and comorbidity index were the top three significant factors for hospital readmission. The predictive model achieved a performance of 95.6% AP and an AUC = 97.3% using the Gradient Boosting algorithm trained on all features.
No takes yet. Share an insight, caveat, or question.
Dafrallah et al. (2024) conducted an observational in Unplanned hospital readmission. Risk factors (previous admissions, hyperosmolar imbalance, comorbidity index) was evaluated on Hospital readmission prediction (AUC 97.3%, AP 95.6%). A Gradient Boosting model predicted 30-day hospital readmission with an AUC of 97.3%, identifying previous admissions, hyperosmolar imbalance, and comorbidity index as top risk factors.
Synapse has enriched 10 related papers. Consider them for comparative context: