THE TrME REVERSED CROSSOVER

SOLLECITO, WILLIAM A. The Time Reversed Crossover Design in Health Servi ces Research. "(Under the di rect i on of DENNIS B. GILLINGS.) Hea1th servi ces pol icy is often determi ned based on natu ra lly occurring experiments. Because true experimental designs cannot be used, quasi-experimental designs must be Bpplied to such experiments in order to draw valid conclusions. Among the wide variety of designs which are applicable in these situations, there are important differences in regard to threats to internal and external validity to which each is susceptible. The time reversed crossover design is a quasi-experimental design which can be applied to natural experiments in which the t~eatment of interest represents a-change in social or health policy, and in which data are available before and after the change has occurred. This design makes use of separate sampling and analysis strategies to improve the validity of conclusions drawn from such an experiment. The properti~s of the time reversed crossover design are presented including the use of stratification on outcome in the sampling stage which is intended to improve external validity. It is demonstrated that, although this feature of the design introduces internal validity threats due to regression toward the mean in extreme outcome strata, these effects can be measured and eliminated from the test of significance of treatment effects. Within and across stratum estimation and hypothesis testing methods are derived based on those which have been developed for the traditional two period crossover design which is widely used in clinical trials. The procedures are illustrated using data derived from a study conducted by the United Mine Workers of America Health and Retirement Funds to measure the impact of cost sharing on health care utilization among members of its health plan. A detailed analysis of these results provides an illustration of the advantages and disadvantages of the time reversed crossover design and indicates that this design is a powerful evaluation tool which can be used to make conclusions regarding the effects of health policy changes based on naturally occurring experiments.