Chance, continuity, and change in hospital mortality rates. Coronary artery bypass graft patients in California hospitals, 1983 to 1989

Abstract

Objective: To assess whether risk-adjusted mortality rates for hospitals reflect primarily chance variation.

Design: Observation over time.

Setting: All 115 California hospitals with five or more coronary artery bypass graft (CABG) patients in any year 1983 to 1989.

Patients: All CABG patients aged 18 years and older excluding those with other open heart procedures and percutaneous transluminal coronary angioplasty on the day of CABG surgery (n = 132,750).

Outcome measure: Inpatient mortality adjusted for age, gender, chronic comorbidities, timing of surgery, and presence of additional procedures.

Methods: Data were derived from routinely collected hospital discharge abstracts. Observed and expected semiannual mortality rates were examined for each hospital to identify consistent patterns over time. Using the quartile of patients with the highest predicted risk (average mortality, 10%), high- and low-outlier hospitals were identified from 2 consecutive years of pooled data and outcomes 2 years later were examined. Each hospital-year observation was also examined individually to identify outliers and to assess differences for observed and expected death rates, patient volume, and other characteristics.

Results: Some hospitals showed consistently lower-than-expected inpatient mortality. Some hospitals had periods of significantly higher-than-expected mortality followed by a correction. High-outlier hospitals that were selected based on 2 years of data had mortality rates 2 years later that averaged 31% above expected, while low-outlier hospitals had rates 2 years later that averaged 28% below expected. On a contemporaneous basis, high outliers had proportionately more transfers to other acute care hospitals and longer postoperative stays among survivors.

Conclusions: Risk-adjusted outcomes for CABG patients derived from administrative data exhibit substantial patterns of consistency. Such patterns cannot be detected for low-risk patients but are evident for the top quartile of patients stratified by risk. Even with reporting lags and changes in hospital outcomes over time, a policy of channeling high-risk patients away from high-outlier hospitals and toward low-outlier hospitals could lower their overall risk-adjusted mortality rate by 45% [corrected].