Problems and suggested solutions in creating an archive of clinical trials data to permit later meta-analysis: An example of methotrexate trials in rheumatoid arthritis

Because data archives contain patient-based rather than study-based data, they can address meta-analytic questions on uncommon outcomes and on predefined patient subsets, questions that are difficult to address using the traditional meta-analytic approach based on grouped data. We report the tasks involved in establishing the first data archive of rheumatoid arthritis trials. In general, problems stem from the heterogeneity of trials in the archive and we suggest some solutions. In the initial phases, difficulties include recruitment and incomplete participation of trial investigators, whereas later on, other issues arise, such as quality control, coping with different dataset designs, and incomplete documentation. Other issues include heterogeneous measures, missing variables, and comparing data across different visit intervals and trial lengths. Suggested solutions include requesting trial data in predefined archive-wide structures and asking for all possible documentation for each dataset. Data cleaning is necessary, as is rescaling of variables or developing unit-free outcomes, and estimating data for missing variables. Archive design should allow for referencing a patient's data among various datasets. Although one goal is to reduce the quantity of data in the archive while retaining information content, data from early stages of archive building must be accessible for developing new analysis datasets. Documentation of archive building and software choices are discussed. Our experience suggests data archiving for meta-analysis is time consuming and expensive, yet it provides a useful method for analyzing data from multiple trials.