Background Combining data from different clinical trials (observational studies) is getting more frequent for the analysis of DMARD efficacy (effectiveness) and safety. The underlying objective is plausible: the generation of data with sufficient power for the investigation of rare adverse events or different treatment strategies. However, the approach of simply pooling data from different studies within an analysis has been shown to be prone to bias which applies for both, observational and clinical trial data [1,2].
Objectives To describe the agreement of recommended and applied methods in the analyses of data originating from different studies.
Methods A systematic literature review with PubMed and Web of Science was independently conducted by two researchers (RA, CJ) using a common strategy. We included full text publications if: ≥2 different data bases (clinical trial/observational data) with individual patient data were used, efficacy/effectiveness or safety were investigated and if the studies were conducted in patients with rheumatoid arthritis. We excluded studies using claims data, genetic/proteomic association studies and studies of prognostic markers. Outcomes of interest were the number of data bases, the analysis-type (pooled/separated/both) and the illustration and discussion of study heterogeneity.
Results After screening 2002 search results, initial consent of the eligibility of each publication for this analyses was found in 58 publications (safety: n=34 (59%), effectiveness: n=20 (34%), both: n=4 (7%)). From these 28 publications (48%) were based on observational data, 24 (42%) on clinical trials and 6 (10%) based on a mixture of study types. The No. of databases in one publication ranged from 2–42 for clinical trial data, 2–10 in observational data and 4–13 if a mixture of databases was investigated. The analysis of pooled data (n=28, 48%) was more frequent in analyses with trials (n=18) than with data from observational studies (n=8). However, the latter studies were primarily conducted in recent years (Table). In only 1/8 publications (13%) from pooled observational data the baseline characteristics were shown separately for each database. Heterogeneity of study populations was discussed in only 2/8 (25%) publications based on pooled analysis, in 8/15 publications (53%) using a separate analysis of observational studies the heterogeneity of patients was discussed.
Conclusions Collaborative analyses of several sources of observational data or multiple clinical trials may deliver superior results than individual studies but only if an appropriate analysis is applied. The approach of simply pooling the data neglects existing heterogeneity of patient cohorts, assessment of risk factors/events, missing data and dropouts. The use of individual patient data from multiple clinical trials could be superior to meta-analysis of published results but loses this advantage if the data are simply pooled. Regarding the considerable differences between observational studies, e.g. European register , the utility and the recent increase of pooled analyses from observational studies is disputable.
Hammad et al. Clinical Trials 8.5 (2011): 559–570.
Van der Steen et al. Palliative medicine 2008;22:750–759.
Kearsley-Fleet et al. Rheumatology 54.6 (2015): 1074–1079.
Disclosure of Interest None declared