Properties of the cumulative risk-adjusted mortality (CRAM) chart, including the number of deaths before a doubling of the death rate is detected.

摘要

The cumulative risk-adjusted mortality (CRAM) chart is intended to detect changes in the failure rate in a sequence of trials when the relative risk but not the absolute risk of failure at each trial is known. For example, surgical mortality can be monitored taking into account the heterogeneity of patients and procedures. We simulate abrupt change to twice or half the previous failure rate and series with no change. As with any control procedure, changes are eventually signaled whether or not real changes have occurred. With initial mean failure rates ranging from 1% to 25%, the median number of failures before a doubling is signaled is about 20 (i.e., about 10 extra failures) and is nearly independent of the extent of heterogeneity in the series. CRAM chart calculations give an estimate of the risk of failure at the next trial, and we document the performance of this estimate.