Correlating Predicted Admissions with Unscheduled Return Visits: Insights from a Machine Learning Model in Emergency
Wayne A Martini1, Rachelle Beste2, Andrej Urumov1
1Department of Emergency Medicine, Mayo Clinic, Phoenix, Arizona.
Background:
Emergency departments (EDs) are adopting machine learning (ML) models to predict hospital admissions and improve patient flow. The association between admission predictions and unscheduled return visits (URVs) remains underexplored.
Objectives:
This study evaluated the relationship between real-time ML-predicted admission likelihood and 72-hour URVs leading to admission across three tertiary care centers.
Methods:
We analyzed 169,288 ED visits from January 1 to December 31, 2023, using an internally developed ML model with 47 clinical features, including demographics, vital signs, and protocol activations. Data were extracted from electronic health records. Expected return visits were excluded.
Results:
The ML model demonstrated strong predictive performance, with an area under the curve (AUC) of 0.88 for hospital admission prediction across the pooled test set. Of 1,996 URVs within 72 hours (1.18%), 6.61% involved multiple returns. Patients with higher admission scores had increased URV rates and hospital admissions: artificial intelligence (AI) Score ≤ 25%: 90,860 patients; 0.97% URVs; 31.81% admitted on return. AI Score > 25% to ≤ 50%: 42,809 patients; 1.94% URVs; 23.88% admitted on return. AI Score > 50% to ≤ 75%: 21,668 patients; 3.80% URVs; 38.38% admitted on return. AI Score > 75%: 13,951 patients; 4.86% URVs; 76.54% admitted on return CONCLUSION: Higher predicted admission scores were associated with increased URVs and hospital admissions. This suggests ML models can improve ED operations by identifying high-risk patients for early intervention and better resource allocation.
Related Concept Videos
01:11Residuals and Least-Squares Property
If the observed data point lies above the line, the residual is positive, and the line underestimates the actual data value for y. If the observed data point lies below the line, the residual is negative, and the line overestimates the actual data value for y.
The process of fitting the best-fit...
01:18Steps in Outbreak Investigation
02:20Correlations
01:03Prediction Intervals
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.
01:15Receiver Operating Characteristic Plot
01:19Variation
When independent and dependent variables are plotted on a scatter plot, the slope of a line is a value that describes the rate of change between the two...