An exploratory analysis of lumping and splitting in system­atic reviews

Article type
Authors
Weir M, Mayhew A, Fergusson D, Grimshaw J
Abstract
Background: A key issue for systematic reviewers is how broad or narrow their research question should be in terms of populations, interventions, study designs and outcomes. The scope may have implications on the findings and generalizeability of the review. Health professional behaviour change interventions can be lumped or split according to population groups (physicians, nurses, pharmacists); interventions (educational materials, reminders); study designs (randomized controlled trials, interrupted time series), and outcomes (physician prescribing, test ordering). Differing opinions exist regarding the best approach. Whether professional behaviour change interventions are lumped or split according to these groups likely has an effect on how the results can be utilized. Furthermore, it is unclear if authors of new systematic reviews consider the findings of previously published reviews. Objectives: To explore the effects of lumping and splitting of reviews of professional behaviour change interventions, and the possible impact of splitting on the final results. To explore how authors cited other systematic reviews in the field. Methods: A descriptive and exploratory methodology was taken to examine how reviews were lumped or split across all professional behaviour change interventions and how authors chose to cite other reviews for the same interventions. References and included studies were examined in detail to address these issues. Results: Authors generally split their reviews across at least one category, which results in a proliferation of systematic reviews in the same topic area. In addition, review authors tended not to cite other similar reviews. Conclusions: The results of this study provide advice about the optimal organization and methodology of systematic reviews of professional behaviour change interventions. Further empirical research is needed to assess the impact of lumping and splitting on results, and whether splitting leads to biased conclusions.