Evaluating inter‐hospital variability in mortality rates over time, allowing for time‐varying effects

Abstract

In outcome studies, quality of care in various institutions is typically assessed by comparing observed to expected outcome rates, after adjusting for patients' case-mix factors in logistic regression models. However, differences in patterns of outcome rates over time, especially when there is a distinction between the determinants affecting early and later events, are rarely studied. We use six-month mortality after coronary artery bypass graft operation (CABG) as an example. We present a statistically valid approach to estimate expected survival curves for different subgroups, based on a Cox survival model with time-varying effects. Bootstrap confidence intervals around the expected survival curves are constructed. This approach is applied for examining the pattern of deviation of high-mortality hospitals after CABG. Implications for quality assessment in comparative outcome studies are discussed.

Survival of critically ill patients hospitalized in and out of intensive care units under paucity of intensive care unit beds*

“…Variables chosen for entry into the model were those depicted by Knaus (21) as the most important predictors of mortality in ICU patients. Departments were introduced into the model as time-varying covariates, allowing the hazard ratio of ICU and SCU relative to regular department to change over time, for example, for early period (0 -3 days) and late period (4 -30 days) (25)(26). The choice of the early period as 3 days corresponded to the 72 hrs after deterioration, which is regarded as a critical period for the assessment of success in stabilizing critically ill patients in ICU (27).…”

Evaluating survival model performance: a graphical approach

“…The ÿrst example is taken from a study of mortality after coronary artery bypass grafting (CABG) operation where time varying e ects of covariates are explored [19]. Understanding whether risk factors for postoperative mortality have constant or varying e ect over time may have implications for the management policy of patients.…”