Reviewed by the Help Dementia Editorial Team — our editors review every article for accuracy against guidance from the National Institute on Aging, the Alzheimer’s Association, and peer-reviewed sources.
Yes, smaller dementia trials can work with better biomarker screening—and they’re already doing it. The AHEAD study enrolled 1,400 cognitively healthy older adults based on amyloid or tau levels alone, not cognitive decline. Three years in, researchers detected slowing of cognitive change in early intervention groups. A conventional trial of similar size without biomarker enrichment would have required either 5,000+ participants or 7+ years to see the same signal because most participants wouldn’t have had the pathology driving the disease. Biomarkers shrink trials by filtering for people who are actually developing the biological process you’re studying.
If your drug targets amyloid and you screen for amyloid positivity upfront, every participant has the biology you’re trying to treat. You lose the noise of cognitively normal people who’ll never develop symptoms and people with cognitive decline driven by something else entirely. That filtering means you need fewer participants to detect a real effect—if there is one. The catch is real. Smaller trials work only if your biomarker actually predicts who needs treatment, the biomarker truly measures what your drug targets, and you’re testing something that actually modifies disease. Miss on any of those and you’ll get a “successful” trial that doesn’t translate to clinical benefit, which is what happened to solanezumab and aducanumab in earlier, less biomarker-selective designs.
Table of Contents
- How Biomarker Screening Reduces Trial Size
- Blood Biomarkers Changed the Economics
- Surrogate Biomarkers Versus Clinical Outcomes
- Matching Trial Design to Biomarker Data
- Hidden Costs of Biomarker Enrichment
- Competing With Standard of Care
- Biomarker Cutoffs and Population Differences
How Biomarker Screening Reduces Trial Size
Biomarker enrichment works by eliminating statistical noise. In a traditional Alzheimer’s trial, you might recruit 500 people with mild cognitive impairment and hope enough have amyloid pathology to show a treatment effect. Assuming 60% are amyloid-positive, you effectively have 300 people in your treatment group. Add stratification for tau or neurodegeneration and you’re down to 150. The same trial run with upfront amyloid and tau screening recruits only the 300 you actually need. You skip the 200 participants who won’t show response regardless of treatment. The EXPEDITION 3 trial of atabecestat illustrates the size benefit. The earlier version recruited broadly with mild cognitive impairment and mild dementia.
When biomarkers showed atabecestat might be toxic at higher doses, the trial became enriched: amyloid-positive and tau-positive only. That redesign didn’t save the drug, but it did cut the statistical noise enough that failure became clear faster with fewer participants. Compare that to semagacestat, which ran for five years across thousands of people before showing no benefit—partly because many participants didn’t have the amyloid-driven disease the drug was supposed to target. A limitation is that biomarker enrichment assumes your marker actually predicts response. Phosphorylated tau in CSF or blood is a strong predictor of who will decline cognitively over time. Amyloid alone in cognitively normal people is weaker—some people live decades with high amyloid without cognitive symptoms. If you design a trial for cognitively normal amyloid-positive people and your drug modifies amyloid but not tau or neurodegeneration, you may see biomarker changes without cognitive benefit. The trial succeeds on its surrogate and fails on what matters.
Blood Biomarkers Changed the Economics
Five years ago, biomarker screening meant PET scans, lumbar puncture, or MRI in hundreds or thousands of people to find your enriched population. That cost and time burden often exceeded the trial budget. blood biomarkers for phosphorylated tau (p-tau181, p-tau217), amyloid-beta 42, and neurofilament light chain changed the equation. A blood draw costs $500 to $2,000 per person and returns results in days. A PET scan costs $5,000 to $15,000 and requires scheduling and travel. The Lilly AHEAD trial used blood biomarkers for all initial screening. Participants with plasma p-tau217 above the 75th percentile for their age got into the amyloid cohort; those with amyloid-beta 42 below the 25th percentile got into the tau cohort.
No PET required. Screening and randomization happened in months instead of years. A smaller trial with faster recruitment and faster biomarker results is a smaller trial with lower total cost. The warning is specificity and standardization. Different blood biomarker assays (Roche elecsys versus Simoa versus others) give slightly different cutoffs and values. A p-tau217 result from one lab might not translate to another without recalibration. If your trial sets amyloid positivity at plasma p-tau217 > 25 pg/mL and participants from a certain geography all test at 22 pg/mL because of assay drift or population differences, you’ll enroll the wrong people. That’s why major trials now include centralized biomarker processing and reference cutoffs tied to specific assays.
Surrogate Biomarkers Versus Clinical Outcomes
A smaller trial can measure amyloid change on PET, tau decline on blood tests, or brain atrophy on MRI in 200 cognitively normal people over two years and show statistical significance. That’s not the same as showing those 200 people delay cognitive decline by six months or one year. Aducanumab reduced amyloid plaques convincingly in early trials; it also showed slowing of cognitive decline at 18 months in ADUHELM. But real-world follow-up showed no clinically meaningful delay in symptoms. The FDA approved it anyway, then withdrew approval after backlash. The AHEAD trial uses a hybrid: phosphorylated tau as an intermediate biomarker (it predicts who’ll decline) and cognitive slope as the clinical outcome. If participants on active drug slow their cognitive decline relative to placebo, that’s evidence the drug works on something that matters. If amyloid changes but cognitive slope doesn’t, the biomarker alone doesn’t validate the drug.
This design lets you run a smaller, shorter trial than you would if you waited for full dementia diagnosis, but you’re still measuring something that connects to real decline. A key limitation is that cognitive slope in cognitively normal people is shallow. Annual change might be 0.5 to 1.0 points on a 30-point scale. Detecting a 25% slowing with high statistical confidence in a population of 200 requires precise, repeated testing. Any test practice effects, variability in participant effort, or seasonal variation in cognition will drown out real drug effects. Smaller trials are more vulnerable to these noise sources. Lecanemab trials used cognitive decline in mild cognitive impairment (larger, faster decline) and amyloid-positive preclinical cohorts (smaller but more biomarker-enriched cohorts). The preclinical arm with smaller sample size was possible only because the underlying cognitive change was predictable and the population was enriched.
Matching Trial Design to Biomarker Data
A smaller trial with strong biomarker enrichment works best when you have prior evidence the drug target is real and common in your population. Amyloid-targeting monoclonal antibodies were developed after PET studies in thousands of cognitively normal people showed that amyloid accumulation predicts later cognitive decline. That prior probability—amyloid causes problems—justified running AHEAD with 1,400 cognitively normal people instead of 5,000. Tau-targeting drugs are newer; the prior probability that modifying tau in asymptomatic people prevents cognitive decline is still being tested. Lecanemab (Leqembi) achieved smaller, faster trials because it had years of prior amyloid data.
Donanemab (Kisunla) and remternetug (if approved) will have similar advantages. A drug targeting a novel pathway with no established connection to cognitive outcomes should probably not run a small trial, no matter how appealing the biomarker is. The failure of TANGO—a Phase 2b trial in 352 people with mild cognitive impairment on a tau-targeting agent—shows the risk. The drug showed biomarker changes. Cognitive outcomes favored treatment but didn’t reach significance. A larger trial would have been more informative than a smaller one enriched on tau alone.
Hidden Costs of Biomarker Enrichment
Smaller trials with biomarker screening create a different problem: generalizability. If you enroll only amyloid-positive people with p-tau217 above the 75th percentile, you’re studying a subset of the population with dementia pathology. That subset might respond differently to treatment than amyloid-positive people with p-tau217 below the 75th percentile. You’ve reduced noise and gained statistical power, but you’ve also narrowed the question you’re answering to “does this drug help the sickest, most biomarker-positive people?” Real clinical use involves people across the biomarker spectrum.
Lecanemab’s trials enrolled amyloid-positive mild cognitive impairment and mild dementia, not cognitively normal people or people with severe dementia. The approved indication reflects that. When clinicians started prescribing lecanemab to older people with subtle cognitive complaints and amyloid positivity (a much larger, less symptomatic population), the safety and efficacy data didn’t directly apply. ARIA (amyloid-related imaging abnormalities—brain microhemorrhages or microinfarcts) showed up at similar rates, but the question of whether slowing cognitive decline by 27% matters to a cognitively normal person is different from whether it matters to someone with diagnosed mild cognitive impairment.
Competing With Standard of Care
A smaller dementia trial now must compete with approved monoclonal antibodies. AHEAD was designed before lecanemab approval; it enrolled cognitively normal amyloid-positive people and assigned some to placebo, which was ethical at the time.
If you designed AHEAD today, you’d face pressure to give all participants the standard-of-care drug (lecanemab) and test your new drug as an add-on or alternative. That doubles the trial’s complexity and sample size. You’re not measuring “does drug X slow decline” but “does drug X + lecanemab slow decline more than lecanemab alone.” Some dementia drug programs have already pivoted to this design, which is scientifically rigorous but defeats the size-saving benefit of biomarker enrichment.
Biomarker Cutoffs and Population Differences
Blood biomarker levels vary by age, APOE4 status, race, and ethnicity. A p-tau217 cutoff of 25 pg/mL appropriate for a 65-year-old might be too high for a 75-year-old or too low for someone with APOE4 non-carriers.
The AHEAD trial accounted for this by using age-stratified cutoffs: p-tau217 in the top quartile for each age group. That added complexity to screening but ensured the trial enrolled people with similar biological risk across age categories. If your smaller trial uses a single global cutoff without age or genetic stratification, you’ll enrich for age or genetics rather than disease risk, and your results won’t generalize.
