Neuropsychiatric endpoints in dementia
Agitation, apathy and depression in dementia are heterogeneous, rater-dependent and badly measured. That is the exact problem class we have spent twenty-five years on.
Samuel Dickson, PhD · Sr. VP of Statistics
- May 4, 2026
- Auvelity approved for agitation in Alzheimer's disease, the first non-antipsychotic in the indication.
- 192 trials · 158 agents
- The 2026 Alzheimer's disease pipeline.
- ~10%
- of that pipeline now targets neuropsychiatric symptoms rather than disease modification.
The symptoms are real. The scales struggle to see them.
Disease-modification trials measure a slow, roughly monotonic decline. Neuropsychiatric symptoms do not behave that way. Agitation flares and settles. Apathy hides inside depression, and depression inside apathy. A caregiver's report on Tuesday depends on what Monday night was like.
The instruments inherit all of that. They are rated, not measured; they lean on informants; their items overlap across constructs; and their week-to-week noise can be as large as the treatment effect a sponsor is hoping to show.
None of this is new to us. Slow, heterogeneous, hard-to-measure change is the problem the methods on our methods page exist to solve. The disease-modification programs taught us how; the neuropsychiatric programs are where that transfers next.
Heterogeneous
Populations that do not decline in one direction
Composite construction and modeling of visit-to-visit variability, rather than a single change-from-baseline number.
Rater-dependent
Endpoints that depend on who is asking
Rater and site variability treated as a design and monitoring question, not a post-hoc excuse.
Badly measured
Instruments with overlap and noise
Item-level analysis to find where the signal actually lives, and to pre-specify an endpoint a reviewer will accept.
The indication is open, and the measurement question is unanswered
The first non-antipsychotic approval for agitation in Alzheimer's disease has shown that the regulatory path exists. Roughly a tenth of the Alzheimer's pipeline is now pointed at neuropsychiatric symptoms. What most of those programs share is an endpoint question nobody has yet settled: which instrument, which items, which analysis, and how to keep a rater-dependent primary from deciding the readout.
Those are the questions we want to be asked early, while the protocol is still a draft and the SAP has not locked. The left-hand side of the timeline is where this is least expensive.
Also in this indication
Thirty minutes on your endpoint
A senior statistician, your protocol, no deck. You'll leave knowing whether your measurement plan has a problem.
Book the thirty minutes