The Impact of a Novel Transfer Process on Patient Bed Days and Length of Stay: A Five-Year Comparative Study at the Mayo Clinic in Rochester and Mankato Quaternary and Tertiary Care Centers

Abstract

Introduction: This study evaluated the impact of parallel-level patient transfers on bed utilization efficiency within the Mayo Clinic Health System in Southern Minnesota, focusing on optimizing resources across tertiary and critical access hospitals. Methods: A retrospective analysis of 179,066 Emergency Department visits (2018-2022) was conducted, with ~2% involving parallel-level transfers for observation or admission. Machine learning was utilized to identify patients suitable for parallel transfers based on demographics, comorbidities, and clinical factors. A Random Forest model with an AUROC of 0.87 guided transfer decisions. Saved patient days were calculated as the difference between the actual LOS and the benchmark LOS based on Diagnosis-Related Groups (DRGs). Generalized estimating equations analyzed length of stay (LOS) differences, adjusted for confounders, with 95% confidence intervals (CI). Statistical analyses were conducted using SPSS (v.26). Results: The mean patient age was 56 years (SD = 17.2), with 51.4% being female. Saved patient days increased from ~600 to 5200 days over the study period. Transferred patients had a 5.7% longer unadjusted LOS compared to non-transferred patients (95% CI: 2.9-8.6%, p < 0.001). After adjustment for demographics and comorbidities, the LOS difference was not significant (adjusted mean difference: 0.4%, 95% CI: -1.7-2.5%, p = 0.51). Conclusions: Parallel-level transfers increased saved patient days, reflecting enhanced resource utilization. However, the adjusted LOS differences were not significant, highlighting the need for robust transfer protocols and controlled studies to confirm these findings.

Keywords: bed utilization efficiency; parallel transfer; saved patient days; tertiary care centers.

Figure 1

ROC curve for length of stay (LOS) prediction: This shows the model’s performance in distinguishing between transferred and non-transferred patients, with the AUC score illustrating the balance between sensitivity and specificity.

Figure 2

Confusion matrix for transfer prediction: This matrix provides a breakdown of the true positives, true negatives, false positives, and false negatives, indicating the model’s accuracy in predicting whether a patient would be transferred.

Figure 3

Kaplan–Meier survival curves displaying the probability of discharge over time for the transferred and non-transferred patient groups.

Figure 4

A chart demonstrating progressive bed day savings across the years. The patient bed day savings for each year were calculated as the difference between the actual length of stay (LOS) and the expected LOS based on Diagnosis-Related Group (DRG) benchmarks. The adjusted LOS estimates were derived using multivariable regression models accounting for age, gender, and comorbidities. The cumulative difference was summed for all eligible patients annually to quantify the total savings.

Figure 5

Gender analysis of patients over the years from 2018–2022.