Background: Length of stay (LOS) is a key performance indicator that drives continuous process improvement initiatives in the hospital setting. Case-mix index-adjusted resource length of stay (CARLOS), which adjusts hospital length of stay for case-mix index (CMI), is used at our institution as a measure of hospital efficiency and has been the focus of many tactics to optimize value and create bed capacity.[1] However, growing evidence suggests that administrative case-mix adjustments incompletely account for patient complexity, comorbid burden, social determinants of health, and other clinical processes that influence LOS.[2] The current risk adjustment models for LOS lack calibration for institutional and service-level variation [2] or outliers [3], and ICD-10-based case-mix algorithms demonstrate better performance.[4,5]
Purpose: This project aimed to identify the drivers of CARLOS variation among Hospital Medicine services at Vanderbilt University Medical Center and to inform the design of a “Hospital Medicine Report Card” enabling targeted, actionable, disease-specific performance improvement.
Description: Using fiscal year 2024-25 data, CARLOS (total LOS/total CMI) was analyzed across diagnosis-related groups (DRGs), individual providers, and interdisciplinary teams. Unadjusted LOS and CARLOS values were compared with geometric mean LOS (GMLOS) and arithmetic mean LOS (AMLOS) to correct for “mega-outliers” who were defined as LOS ≥5 SD above mean. Correlations between CARLOS, CMI, and DRG composition were fit tested as a linear regression model to assess whether current adjustments adequately reflected resource intensity and case complexity.
Conclusions: Average CARLOS increased by 0.4% from 2024 to 2025, coinciding with a 2% rise in LOS among low-reimbursing DRGs. CARLOS distribution was bimodal and right skewed, with higher CMI correlating to a higher CARLOS, suggesting incomplete adjustment for complexity. This was best demonstrated with three of Hospital Medicine’s most common DRGs: Esophagitis, Sepsis w/MCC, and UTI w/wo MCC. For esophagitis, CARLOS overestimated inefficiency with GMLOS aligning with predicted LOS. Sepsis w/ MCC patients where CARLOS was above national average; both GMLOS and AMLOS were > 1 day above expected (~ 20% increased LOS), indicating systemic efficiency gaps rather than outlier effect. UTI w/wo MCC showed CARLOS elevation, but GMLOS fell within the expected range.CARLOS is a useful but imperfect measure of operational efficiency in the hospital setting. Static CARLOS targets ignore its sensitivity to DRG mix, outlier frequency, and reimbursement heterogeneity, obscuring meaningful performance trends and opportunities. Developing a dynamic, DRG aware report card that integrates CARLOS with GMLOS/AMLOS benchmarking, and service-line context could better guide value-based performance evaluation and refine process improvement in hospital medicine.