Affiliations: [a] Massachusetts General Hospital Biostatistics Center, Boston, MA, USA | [b] Harvard Medical School, Boston, MA, USA | [c] Department of Psychiatry, Massachusetts General Hospital, Boston, MA, USA | [d] Department of Epidemiology, Harvard School of Public Health, Boston, MA, USA | [e] Department of Neurology, Massachusetts General Hospital, Boston, MA, USA | [f] Department of Biostatistics, Harvard School of Public Health, Boston, MA, USA
Correspondence:
[*]
Correspondence to: Eric A. Macklin, PhD, MGH Biostatistics Center, Massachusetts General Hospital, 50 Staniford St, Suite 560, Boston, MA 02114, USA. Tel.: +1 617 724 9828; Fax: +1 617 724 9878; E-mail: [email protected].
Abstract: Alzheimer's disease (AD) trials initiated during or before the prodrome are costly and lengthy because patients are enrolled long before clinical symptoms are apparent, when disease progression is slow. We hypothesized that design of such trials could be improved by: 1) selecting individuals at moderate near-term risk of progression to AD dementia (the current clinical standard) and 2) by using short-term surrogate endpoints that predict progression to AD dementia. We used a longitudinal cohort of older, initially non-demented, community-dwelling participants (n = 358) to derive selection criteria and surrogate endpoints and tested them in an independent national data set (n = 6,243). To identify a “mid-risk” subgroup, we applied conditional tree-based survival models to Clinical Dementia Rating (CDR) scale scores and common neuropsychological tests. In the validation cohort, a time-to-AD dementia trial applying these mid-risk selection criteria to a pool of all non-demented individuals could achieve equivalent power with 47% fewer participants than enrolling at random from that pool. We evaluated surrogate endpoints measureable over two years of follow-up based on cross-validated concordance between predictions from Cox models and observed time to AD dementia. The best performing surrogate, rate of change in CDR sum-of-boxes, did not reduce the trial duration required for equivalent power using estimates from the validation cohort, but alternative surrogates with better ability to predict time to AD dementia should be able to do so. The approach tested here might improve efficiency of prodromal AD trials using other potential measures and could be generalized to other diseases with long prodromal phases.
Keywords: Alzheimer's disease, clinical trials as topic, National Alzheimer's Coordinating Center Uniform Data Set, surrogate endpoint, survival analysis
