227 Measuring Emergency Care Survival: The Implications of Risk-Adjusting for Race and Poverty

Abstract

We sought to assess the impact of including race and poverty on risk adjustment models of hospital quality used by the Center for Medicare and Medicaid Services (CMS) for public reporting and benchmarking, as well possible future reimbursement penalties and incentives. We limited our study specifically to in-hospital mortality due to five emergency care sensitive conditions (ECSCs), trauma, sepsis, stroke, cardiac arrest, and ST-elevation myocardial infarction, with evidence of sociodemographic disparities, in accordance with National Quality Forum (NQF) guidelines. We analyzed statewide, all-payer inpatient administrative discharge data covering all adult admissions to 168 hospitals in Pennsylvania in 2011. We developed risk-adjustment logistic regression models, both omitting and using patient race (white vs. non-white) and poverty status (non-poor vs. poor, if uninsured, on Medicaid, or living a zip code in the bottom quartile of median household income) to predict in-hospital mortality. We compared the goodness of fit for each model with a C-statistic, and then computed and ranked standardized mortality ratios (SMRs) for each hospital using both models. In order to identify hospitals likely to benefit or to be penalized with sociodemographics included in modeling, we examined characteristics of hospitals which, between the two models, moved into or out of the bottom and top deciles of SMR. Sociodemographics were significantly associated with higher odds of mortality: non-white adjusted odds ratio [aOR] 1.23 (95% CI 1.16-1.30), and poverty aOR 1.15, (95% CI 1.08-1.22). Inclusion of race and poverty had a significantly higher predictive accuracy than excluding these variables (p=0.0001), but absolute differences in C-statistic values were small (0.8260 without; 0.8264 with). No hospitals moved into or out of the top decile of SMR when race and poverty were included. However, the 3 hospitals which moved out of the bottom decile each had significantly larger young, non-white and poor patient populations compared to all other hospitals (Table). Inclusion of race and poverty status minimally changed the goodness of fit of risk-adjustment models, consistent with NQF’s prior analysis. However, inclusion of the socioeconomic factors did improve hospital mortality rankings for hospitals treating a large number of non-white and poor patients, traditionally known as “safety net” hospitals, which might enable these hospitals to avoid penalties in current value-based purchasing models.TableDescriptive characteristics of hospitals moving into and out of the bottom decile of standardized mortality ratio (SMR) when adding race and poverty to risk-adjustment modelHospitals that Moved in or out of Bottom Decile of Hospital Mortality Rankings (Mean (SD) with P-value by t-test against all other hospitals)All HospitalsMoved Out OfMoved IntoNumber of Hospitals33157ECSC admissions1,375 (1,605) P=0.70296 (268) P=0.301,088 (1,319)Mean Age62 (9) P=0.00574 (1) P=0.2471 (6)Median Age62 (11) P=0.00279 (1) P=0.1974 (7)Mean Number of Comorbidities2.5 (0.6) P=0.882.7 (0.3) P=0.562.6 (0.5)Median Number of Comorbidities2.3 (1.2) P=0.842.7 (0.6) P=0.432.4 (0.6)Non-white patients68% (28%) P<0.000055% (3%) P=0.5311% (17%)Poor patients (Medicaid, uninsured, or zip code in bottom quartile for median household income)56% (29%) P<0.0000510% (3%) P=0.5810% (7%) Open table in a new tab