Comparison of the statistical efficiency of case-crossover and case-control designs: application to severe cutaneous adverse reactions.

Although case-crossover analyses have lately emerged as an alternative to case-control analyses in epidemiological studies, it is not yet known in which situations they give reliable conclusions. In this work, the case-crossover and the case-control designs were first compared on the basis of a dataset from a published study of severe cutaneous adverse reactions resulting from drug exposures of various durations and prevalences of use (245 cases, 1147 controls, and exposures to 23 drug classes). Next, the statistical efficiency of each design was compared via Monte Carlo simulations. Eight of the 13 risk factors identified by case-control analysis of the published data were also identified by the case-crossover analysis, with fairly good agreement on ranks of risk estimates (Spearman's correlation coefficient = 0.71, P < 0.001 ). Simulation studies showed that for relative risks below 8, the case-crossover design (250 cases, 4 control periods/case) had a higher power than the case-control design (250 cases, 4 controls/case), and that the case-crossover design was more conservative than the case-control design for prevalences of drug use below 10%. We conclude that the case-crossover design is not suitable for long-term exposures, but is an appropriate alternative for assessing rare risks associated with transient to short-term exposures.