2012 HSR&D/QUERI National Conference Abstract
3010 — A Less Bad Alternative to Diagnosis-Based Denominators for Health Care Quality Measures
Sox-Harris A, VA Palo Alto HCS;
Process quality measures are usually formulated as a ratio of the number of patients who receive some type of care divided by the number of patients likely to benefit, usually defined by particular diagnoses. Because diagnosing patterns vary widely, many process measures of treatment quality are sensitive to case finding/identification effort. Some health care facilities only diagnose patients who are interested or involved in treatment. Other facilities have active programs of screening that identify broader groups of patients with particular disorders, only some of whom are interested in treatment. Such differences in case finding and identification effort dramatically affect the calculation of quality measures and seriously impact the subsequent validity of cross-facility comparisons.
To demonstrate the impact of case identification in quality measure performance, two versions of several addiction treatment quality measures were calculated: 1) the usual method, including a denominator with all substance use disorder (SUD)-diagnosed patients, 2) an alternative “population-based” denominator, including the entire facility census (all patients with and without SUD). Differences in facility percentile rank under the specifications were calculated.
The percentile rank of many facilities shifted dramatically. For example, for a quality measure of medication treatment for alcohol dependence, the mean difference in percentile rank was zero and half of the facilities shifted percentile rank less than 6 percent. However, the other half of facilities changed percentile rank between 7 and 33%.
These results imply that much of the observed between-facility differences in performance on denominator-based metrics may be sensitive to diagnosing patterns or case-finding efforts. However, some of the observed between-facility differences in performance between the alternative metrics may be an artifact of real differences in prevalence and need. Choosing between these imperfect choices is not simple and may be of greater consequence in clinical areas where diagnosing patterns are more discretionary, stigmatized, and variable.
Quality managers and other stakeholders must decide if the validity threats introduced by diagnosing patterns are greater or less than the validity threats introduced with the population-based denominator approach described here.