Identification of latent classes of randomized controlled trials based on integrity and reporting characteristics

Objective: Reliable, well conducted and adequately reported research is essential for decision-making. This study uses an exploratory clustering approach to identify patterns (subgroups) of trials based on research conduct and reporting characteristics to better understand heterogeneity. Describing features of these subgroups may help elucidate mechanisms of poor planning and reporting that can be acted upon by the research community to improve the research practices. Estimating the impact of trials from specific subgroup classes on pooled treatment effects and overall certainty of evidence may inform conduct and interpretation of systematic reviews in the future.

Study design and setting: Our study used data from 456 randomized controlled trials included in our recent update of a Cochrane review, Exercise treatment for chronic low back pain, to explore groups of trials that have various characteristic patterns of research planning, conduct, reporting and publication.

Results: Using latent class analysis, we explored the patterns that exist in 43 characteristics of trial planning, conduct, reporting and publication characteristics. We identified a four-class model as the best fit for the data; classes were labelled based on the patterns of characteristics that emerged: 1. Well Resourced & Thorough, 155 trials (34%), 2. Dated, 92 trials (20%), 3. Fundamental Deficiencies, 102 trials (22%), and 4. Research Waste, 107 trials (24%). The characteristics that best differentiated the classes were: trial registration status, institution/ethics board approval, conflict of interest reported, and reporting of adverse events. There were no significant differences for estimates of treatment effect in the four classes for all comparisons of treatments with pain intensity outcomes; two classes overestimated functional limitations outcomes for exercise compared to no trial treatment and compared to other conservative treatments.

Conclusions: Using a single characteristic to define quality or trustworthiness is no longer sufficient; individual 'problematic' characteristics could be found in all four classes. This novel research could be used as a framework for further exploration of research integrity issues in the health sciences, and the development of interventions to improve our evidence-base, taking into consideration the research motivations and means of trialists.