Challenges in conducting a systematic review of predictors and moderators of outcome: an example in psychological therapies for OCD

Article type
Authors
Knopp J1, Knowles S1, Bee P2, Lovell K2, Bower P1
1NIHR School for Primary Care Research, Centre for Primary Care, Manchester Academic Health Science Centre, The University of Manchester, UK
2School of Nursing, Midwifery, and Social Work, The University of Manchester, UK
Abstract
Background: Psychological interventions play an increasingly significant role in treating mental and physical health problems. However, treatment effectiveness varies across conditions and individuals. The systematic assessment of predictors and moderators of outcome in trials has the potential to help understand this variability and provide treatments focussed on the needs of individual patients.

Objectives: We sought to synthesize current evidence on predictors and moderators in psychological therapies for adults with Obsessive-Compulsive Disorder (OCD).

Methods: Relevant trials of any language were identified through electronic database searches (CENTRAL, MEDLINE, PsycINFO and EMBASE), key author contact and searches of systematic reviews. We assessed quality of predictor and moderator analyses using published criteria - predictors measured via a validated pre-randomization measure, < 5 predictors assessed, using a test of interaction, and a-priori hypotheses of anticipated predictor effects.

Results: 55% (38/69) of trials reported baseline factors associated with outcome. Predictors were commonly assessed via a validated pre-randomization measure, though few trials adopted existing guidelines by stating a-priori hypotheses or conducting the appropriate test of interaction. Inadequate data reporting prevented the use of meta-analytic procedures; the analysis was restricted to a variant of the box-score. None of the common predictors showed a consistent association with outcome. Commonly reported variables, showing a lack of association comprised medication use, age of onset, OCD-related beliefs, and educational level.

Conclusions: Whilst analyses of predictors and moderators are common, their utility is limited by methodological weaknesses. Advances in this field require the implementation of existing guidelines and full reporting of the planning and conduct of predictor analyses; explicitly, the use of gold standard statistical procedures in assessing moderators, the reporting of a-priori hypotheses of anticipated predictor effects, and increased consideration of statistical power in predictor analyses.