Background: Length of stay (LOS) is a key indicator of hospitalist care quality, but its retrospective nature and care fragmentation limit its effectiveness. Efforts have been made to mitigate the disadvantages of LOS by creating individual LOS (iLOS) metrics, including the follow-up-to-discharge ratio (FDR), which correlate with case mix index-adjusted LOS. (Rothman et al. 2020) FDR has previously been investigated in single-hospital settings with limited covariate analyses. We chose to examine the discharge percentage (DP), a non-linear transformation of FDR that offers improved interpretability and modeling over FDR alone. We aimed to investigate the association between patient, hospital, and hospitalist factors and DP across a multi-hospital, single health system to determine whether it may serve as an individual LOS metric.
Methods: This retrospective cross-sectional study included adult hospitalizations (2020–2022) across four hospitals in a multi-hospital health system. The final cohort comprised 183 clinicians, 35,890 unique patients, and 56,542 hospitalizations. We focused on rounding hospitalists by excluding clinicians who submitted more admissions than follow-up billing codes during any rotating monthly period. We excluded patients with LOS of 0 or age greater than 100 years to maintain data integrity. The primary outcome variable was DP, defined as discharge billing codes divided by inpatient follow-up plus discharge billing codes. Hierarchical beta regression modeling with random effects for clinician, hospital, and patient was used to assess associations with DP.
Results: DP was unimodally distributed, with a mean of 0.18 ± 0.06. Every 0.10 unit increase in DP was associated with an approximate 16% reduction in LOS, although the overall correlation between DP and LOS was modest (Spearman ρ = −0.19, p < 0.001). DP increased with monthly patient volume and years in practice, and demonstrated a nonlinear pattern, plateauing at higher patient volumes (Figure 1). Most of the variance occurred at the clinician level after adjustment for modeled covariates (Intra-class correlation = 0.89). Higher DP was associated with weekend admissions, admission year, season, and home discharge disposition. Patient-level factors had minimal influence after correction for multiple testing.
Conclusions: DP was positively associated with hospitalist experience, as measured by monthly patient volume and years in practice, with most of the variance attributable to differences among hospitalists. High between-clinician variance suggests that unmeasured hospitalist predictors may have a significant effect on DP. These findings indicate that DP may be a feasible and interpretable clinician-level LOS metric that reflects underlying hospitalist practice patterns and suggests DP reflects clinician-associated variation beyond measured covariates, supporting its potential use as a granular performance metric and target for future quality improvement efforts.
.png)