I-4 RESOURCES FOR FUTURE GUIDELINES & RESEARCH: AN OARSI ‘TOOLBOX’ AND AN ‘EVIDENCE-BASED OA RESEARCH DATABASE’ (EBOARD)

Abstract

to, (b) the result of hard quantitative data on the other side. If we use the perfect consensus model, then we are back to where evidencebased medicine (EBM) started twenty years ago. EBM proponents then demanded that clinical experts should step down as review authors because of their notoriously unsystematic evaluations of the literature. In this perspective, a high level of consensus may rather be a measure of the opinions among stakeholders, than synonymous with best evidence. On the other hand, we still struggle with the handling of quantitative data to make trustworthy comparisons across interventions. A systematic review of non-steroidal anti-inflammatory drugs (NSAID) in knee osteoarthritis which we performed, can serve as an example of the difficulties associated with quantitative data. We found that some trials only recruited known responders to NSAID who had 49.4% higher effect size compared to patients in trials which did not. The inflated effect size in the subgroup of biased trials, led to an overall inflated effect size by 24.3%. This fact hampers valid comparisons between NSAID and other interventions. Because of the strict exclusion of cointerventions in most NSAID-trials, there is also a lack of data for the effect of NSAID in combination with potentially effective exercise therapy. There are examples of trials with other interventions which do the opposite, and recruit known non-responders in addition to allowing effective cointerventions. Comparisons across interventions may then be flawed. Conclusions: There is still a way to go before we can be satisfied with our methods for synthesizing best evidence. When a consensus-oriented approach is selected, it seems important to balance guidelines developer groups with involved stakeholders. If best evidence is sought from hard quantitative data, more attention should probably be paid to differences in patient selection criteria, intervention characteristics and allowed cointerventions.