Provider profiling models for acute coronary syndrome mortality using administrative data

Administrative data have been used to construct risk-adjustment models for provider profiling to benchmark hospital performance for acute myocardial infarction (AMI), but much less for acute coronary syndrome (ACS). We assess the impact on risk model performance and hospital-level mortality rate ratios (SMRs) of three key issues: comorbidity measurement methods, inter-hospital transfers and post-discharge deaths.

Logistic regression models for 30-day total mortality used three years of national public hospital emergency (unplanned) admissions data for England for ACS (n=329,369) linked to death registrations. We compared using the Charlson comorbidity index with modelling previous admissions.

Prior admission for various conditions such as cancer and renal failure was associated with higher post-ACS mortality, whereas previous AMIs, PCI and unstable angina admissions were associated with lower mortality. The Charlson comorbidity index performed better than one- and five-year admission histories. Discrimination (c=0.81) was comparable with that from clinical databases. Adjusted 30-day total mortality rates ranged between hospitals from 6.3% to 13.3%. Median differences in SMRs between the comorbidity-adjustment methods were small. Although SMRs and outlier status could change, a hospital's 'qualitative' mortality rating (low, average or high) was not affected. In contrast, a sizeable minority of SMRs changed by ≥ 10 points if transfers were excluded or post-discharge deaths ignored. Model choice occasionally affected funnel plot outlier status.

Models for comparing hospitals' ACS mortality can be constructed with good discrimination using English administrative hospital data. Adjusting for transfers in and capturing post-discharge deaths are more important than the choice of comorbidity adjustment.