A quasi-experimental design based on regional variations: Discussion of a method for evaluating outcomes of medical practice

A large proportion of common medical practices are subject to substantial regional variation resulting in numerous natural experiments. Opportunities are thereby provided for outcome evaluation through quasi-experimental design. If patients treated in different regions were comparable a natural experiment involving alternative treatments could be regarded as 'pseudo randomised', but empirical investigations are needed to verify this prerequisite. This paper discusses the role of quasi-experimental designs in assessment of medical care with evaluation of outcomes after hysterectomy in Denmark as an example. The design is developed and the comparability of selected groups of patients is elucidated from administrative data, while the outcome results are not presented in this context. One indication for hysterectomy is carcinoma in situ of the cervix uteri which may be treated with either hysterectomy, or conisation. A study group of patients was selected from departments where hysterectomy was the treatment of choice for this indication while the reference group was drawn from departments in which conisation was generally preferred. The comparability of the populations, effects and information for the two groups are elicited from administrative data. We conclude that it is possible to establish a quasi-experimental design based on regional variations and that the comparability of the groups included may be assessed through registry data. The importance of technology diffusion for the prospects of performing quasi randomised studies is emphasised. In this attempt to evaluate hysterectomy, it was not possible to identify groups of patients, which were sufficiently comparable to justify a study of soft outcomes.