A risk-adjustment technique for comparing prematurity rates among clinic populations.

Abstract

RISK characteristics often vary appreciably from one clinic population to another so that comparison of the effectiveness of a health program among these populations is difficult. Failure to take such differences into consideration can lead to serious errors in the interpretation of patients' performance. Although a health program would be expected to affect the measures of outcome which reflect patients' performance, there are usually other factors asscc1ated with populations of patients which also alter outcome. We shall refer to these concomitant sources of outcome variability as risk factors. A statistical treatment of outcome data which will account for the risk factors is highly desirable. Several techniques may be used to adjust for the risk factors. A sophisticated statistical procedure, such as analysis of covariance, is sometimes used for this purpose. By analysis of covariance, one can isolate and measure the effect of each possible source of outcome variability which is identified. Although this method is a powerful means of controlling for the concomitant variables, it has the disadvantage of being computationally complex (usually requiring a computer) and is dependent on the specification of an appropriate mathematical model. A rigorous discussion of analysis of covariance is given by Cochran (I) . Alternatively, an intuitive method is often applied, in which the groups to be compared are stratified according to the concomitant variables and comparisons are made only within similar strata. Although this approach is straightforward computationally, interpretation of the results is complicated because a separate set of comparisons The authors are in the department of preventive medicine and rehabilitation, University of Maryland School of Med:cine. Dr. Hebel is an associate professor of biostatistics, Dr. Tayback is a pro fessor of biostatihlics, and Dr. Entwisle is chairman of the department. The work described was supported by Public Health Service Grant No. Ph700. Tearsheet requests to Dr. Richard Hebel, Department of Preventive Medicine and Rehabilitation, University of Maryland School of Medicine, Baltimore, Md. 21201.