[P4–013]: ESTIMATING CLINICALLY RELEVANT CHANGE AND DIFFERENCE THRESHOLDS FOR THE CLINICAL DEMENTIA RATING: SUM OF BOXES (CDR‐SB) FOR EARLY ALZHEIMER's DISEASE

Minimally Clinically Important Difference (MCID) is a key concept for interpreting clinical trial data and responder criteria provide critical information regarding patient level benefit. Methodology in this area has evolved in recent years. There is increasing recognition of the need for separate estimates of inter-group difference and intra-individual change, and consideration of demographic and clinical characteristics of the target population. Furthermore, estimates for improvement may differ from those for worsening–important in the context of progressive disease and treatment that may moderate course. The Clinical Dementia Rating (CDR) has inherent clinical relevance and a single box score increment of 0.5 or 1.0 was proposed as a mean change reflecting efficacy, and as a responder criterion for early Alzheimer's disease (AD). There is a need to explore this using statistical approaches and clinical consensus, to generate estimates of improvement and worsening, at different stages of disease severity. Calculation of anchor and distribution-based estimates used data from ADNI, and the SCarlet RoAD (SR) (NCT01224106) and ABBY (NCT01343966) clinical trials. Separate estimates were generated for biomarker confirmed prodromal AD (pAD) and mild AD (mAD) populations. Anchor-based estimates used change in CDR-Global Score (GS) and change in MMSE-based severity as anchors, for overall improvement and worsening of disease. Distribution-based estimates included standard deviation and standard error of measurement, as effect size metrics. For pAD, distribution-based estimates at baseline and for change at 12 months showed consistency between ADNI and SR (0.5xSD: ADNI 0.46 and 0.57; SR 0.49 and 0.55). Anchor-based estimates for ‘minimally worsened’ and ‘minimally improved’ (smallest change in CDR-GS) were also consistent between populations (CDR-GS change at 12 months: ADNI 2.32 and -1.03; SR 2.15 and -1.42). The number of patients in the ‘minimally improved’ group was too low to constitute a robust estimate. Selection of suitable anchors is crucial to generating robust estimates. These results showed that, depending on the approach used, estimates vary markedly. Due to nature of the data collected, the selection of anchors was limited. Further analyses investigating additional approaches are planned, including exploration of relationships between anchor and target measures and optimal scoring.