Hospital rankings changed markedly after accounting for stroke severity, highlighting the limitations of administrative data for comparing the quality of stroke care.

Fair comparisons of hospitals‘ performance require reliable methods to account for differences in patients‘ baseline characteristics and predicted risk for adverse events, because otherwise hospitals would be penalized for treating sicker patients. The Centers for Medicare and Medicaid Services (CMS) has proposed a risk-adjustment model for comparing hospitals’ mortality rates after stroke. However, this model relies on administrative data and does not account for stroke severity, prompting concern that it would imperfectly capture baseline differences. To address this concern, investigators have compared the performance of the proposed CMS model with and without an added measure of baseline stroke severity.

By linking CMS claims data to the Get With The Guidelines–Stroke registry, the researchers could add patients’ baseline NIH Stroke Scale (NIHSS) scores to the administrative data in the CMS model. The addition of NIHSS scores substantially improved the model’s prediction of 30-day mortality: Its C statistic increased significantly, from 0.772 to 0.864. Strikingly, when hospitals were divided into the top and bottom 20% and middle 60%, 26.3% of hospitals changed category after the addition of NIHSS scores. Of hospitals with “worse than expected” mortality using the standard CMS model, 57.7% were reclassified as having “as expected” mortality after the addition of NIHSS scores.

Comment: Efforts to improve hospital accountability and institute bundled payments to hospitals and providers will rely on risk-adjustment models to account for case mix. The details of such models can be arcane, yet the stakes are enormous, because flawed models will unfairly penalize those caring for sicker patients and enable gaming of publicly-reported outcomes and reimbursement rates. This timely study is a reminder that clinicians must actively engage with policymakers to develop reliable risk-adjustment models. Measures of stroke severity clearly must be accounted for, although doing so efficiently and reliably will take some creative thinking.

Source: Journal Watch Neurology