Suitability of dementia risk scores as surrogate outcomes for lifestyle‐based multidomain prevention trials: Epidemiology / Innovative methods in epidemiology (i.e., assessment methods, design, recruitment strategies, statistical methods, etc.).

Background: Although not designed as such, dementia risk scores, such as the CAIDE or LIBRA, might be useful as surrogate outcome measures for dementia prevention trials conducted from midlife onwards. In the risk scores' original categorical scoring systems, changes in risk factors are only taken into account when crossing cut‐off values, which may limit responsiveness and ability to detect intervention effects. We hypothesized that scoring systems based on continuous z‐score measures of risk factors may improve the performance of risk scores in this respect. Method: In three large multidomain dementia prevention trials (MAPT, preDIVA and HATICE) with 1.5‐2 years of follow up, we assessed of the original and z‐score versions of the CAIDE and LIBRA scores: 1) responsiveness and 2) actual and simulated intervention effects. Result: Across the three trials, 48 to 58% of intervention group participants underwent change on the CAIDE score between baseline and follow‐up. The LIBRA index was more responsive: 61 to 79% of participants changed over time. The z‐score versions registered changes in risk factors that were not detected by the original scores: subjects with no change over time in the original score showed great variability in their change in z‐scores, ranging from as much as ‐1.25 to 1.25 for the CAIDE. Conversely, for very small changes in z‐score, the change in original CAIDE or LIBRA scores was as much as ‐5 to 5 points, indicating multiple small changes in risk factors just across the cut‐off values for the original scoring system. Original scores and z‐scores were equally able to detect modest but significant intervention effects. When simulating very large intervention effects, mean between‐group differences of ‐0.50 points (23% of potential for improvement) could be attained for original CAIDE score and ‐0.52 points for original LIBRA index (15% of potential for improvement). Their z‐score versions could attain a maximum of ‐0.24 standard deviations for CAIDE and ‐0.15 standard deviations for the LIBRA index. Conclusion: Dementia risk scores as well as their z‐score versions show potential as surrogate outcomes. How changes in risk scores affect long‐term dementia incidence rates remains to be determined. [ABSTRACT FROM AUTHOR]