Hospital-Wide Quality Assessment Using All-Condition Versus Summed-Condition Risk Adjustment

To compare risk-adjusted 30-day mortality using two hospital-wide methods: one across all admissions ("all-condition") versus a model with combined results from 15 condition/procedure cohorts ("summed-condition").

We developed models on 2019 and 2020 (validated prospectively on 2020 and 2021) using Medicare admissions for patients aged 65 to 94 years. We measured hospitals' 30-day risk standardized mortality rate (RSMR) using both approaches, as well as two covariate selection methods: Risk Stratification Index (RSI) and Hierarchical Condition Categories (HCC). We reported model performance statistics, descriptive statistics comparing hospitals' RSMR between methods, and an exploration of hospital characteristics associated with different RSMR estimates from the two approaches using random forest regression.

100% sample of Medicare fee-for-service claims. We included 3638 general US hospitals with a secondary analysis on 1355 large hospitals.

There was high correlation between RSMRs from all- and summed-condition methods (0.89 correlation using RSI; 0.87 HCC). RSMR correlation between RSI and HCC was higher when using the all-condition versus summed-condition method (0.87 vs. 0.85). Model performance was higher using RSI (C-statistic 0.88 for all-condition; 0.84 to 0.95 summed-condition) than for HCC (C-statistic 0.87 for all-condition; 0.80 to 0.93 summed-condition). Small hospitals and those with a high proportion of non-surgical pulmonary admissions had lower estimated RSMRs under the summed-condition method compared to all-condition. For large hospitals, all- and summed-condition RSMRs were highly correlated using either RSI or HCC (correlation 0.97 RSI; 0.95 HCC).

All- and summed-condition RSMRs using RSI were highly correlated. Some hospital characteristics were associated with comparative RSMR results between all- versus summed-condition. For large hospitals, the choice of RSI versus HCC impacts RSMR more than the choice of all- versus summed-condition method.