OBJECTIVES: Analyses of real world data are widely used in pharmaceutical research. Retrospective analysis was used to support the design of the 5-years Insights to Model Alzheimer’s Progression in real life study (iMAP) that studies the clinical meaningfulness of early changes in cognitive test scores (Alzheimer's Prevention Initiative Cognitive Composite (APCC), Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) and Clinical Dementia Rating Sum of Boxes (CDR-SB)) through all stages of Alzheimer disease (AD).

METHODS: To determine sample sizes, we sampled from observational cohort studies. The inclusion/exclusion criteria of iMAP study were applied to mimic the study population. A set of simulations was conducted for each of the iMAP cohorts, respectively (unimpaired participants, mild cognitive impairment (MCI), and dementia due to AD). The whole records of subjects were sampled from databases to preserve the correlation among the variables. Required sample size was determined when the regression coefficient for change from baseline to year 1 in APCC or CDR-SB was significantly different from zero (corresponding 99% confidence intervals (CI) do not include 0).

RESULTS: For the cohort of unimpaired participants, a Cox model was used. The sample size of 300 participants was sufficient to achieve non-zero 99% CI of APCC change. For the MCI cohort, the linear regression model was implemented. A sample size of 120 participants was sufficient to achieve a non-zero 99% CI of APCC change. For the dementia cohort, a Cox model was used. A sample size of 140 participants was sufficient to achieve a non-zero 99% CI of CDR-SB change. To account for additional uncertainty and potential higher variability, additional discounts were introduced.

CONCLUSIONS: Complex prospective studies require careful planning with estimation of optimal number of participants. Patient-level simulations of study populations by sampling from existing real world datasets were successfully used to inform the study design.