Using a distribution-based approach and systematic review methods to derive minimum clinically important differences

BACKGROUND: Clinical interpretation of changes measured on a scale is dependent on knowing the minimum clinically important difference (MCID) for that scale: the threshold above which clinicians, patients, and researchers perceive an outcome difference. Until now, approaches to determining MCIDs were based upon individual studies or surveys of experts. However, the comparison of meta-analytic treatment effects to a MCID derived from a distribution of standard deviations (SDs) associated with all trial-specific outcomes in a meta-analysis could improve our clinical understanding of meta-analytic treatment effects. METHODS: We approximated MCIDs using a distribution-based approach that pooled SDs associated with baseline mean or mean change values for two scales (i.e. Mini-Mental State Exam [MMSE] and Alzheimer Disease Assessment Scale – Cognitive Subscale [ADAS-Cog]), as reported in parallel randomized trials (RCTs) that were included in a systematic review of cognitive enhancing medications for dementia (i.e. cholinesterase inhibitors and memantine). We excluded RCTs that did not report baseline or mean change SD values. We derived MCIDs at 0.4 and 0.5 SDs of the pooled SD and compared our derived MCIDs to previously published MCIDs for the MMSE and ADAS-Cog. RESULTS: We showed that MCIDs derived from a distribution-based approach approximated published MCIDs for the MMSE and ADAS-Cog. For the MMSE (51 RCTs, 12,449 patients), we derived a MCID of 1.6 at 0.4 SDs and 2 at 0.5 SDs using baseline SDs and we derived a MCID of 1.4 at 0.4 SDs and 1.8 at 0.5 SDs using mean change SDs. For the ADAS-Cog (37 RCTs, 10,006 patients), we derived a MCID of 4 at 0.4 SDs and 5 at 0.5 SDs using baseline SDs and we derived a MCID of 2.6 at 0.4 SDs and 3.2 at 0.5 SDs using mean change SDs. CONCLUSION: A distribution-based approach using data included in a systematic review approximated known MCIDs. Our approach performed better when we derived MCIDs from baseline as opposed to mean change SDs. This approach could facilitate clinical interpretation of outcome measures reported in RCTs and systematic reviews of interventions. Future research should focus on the generalizability of this method to other clinical scenarios.