Health Economics and Outcomes Research in Alzheimer’s: RCT vs. RWE Gap

Lecanemab's trial enrolled the patient we want. A database of 6M+ neurological records shows the patient we have. That contrast is the working problem in Alzheimer's health economics and outcomes research (HEOR) right now. The pivotal trials behind anti-amyloid therapies enrolled a deliberately narrow population; the patients arriving in real-world neurology practice look different on nearly every axis. When HEOR and market-access teams model uptake, budget impact, or effectiveness off trial baselines, they are modeling a patient who is uncommon in the clinic. Across 62,746 patients first diagnosed with Alzheimer's disease (AD) or mild cognitive impairment (MCI) since 2021, only 10.3% met the core trial-aligned treatment-readiness criteria that can be assessed in routine care and just 4.0% when trial-grade cognitive staging is required. This piece quantifies that real-world evidence gap and why it matters for evidence generation.

Real Life vs. Clinical Trials: Who Really Gets Treated?

A randomized controlled trial (RCT) enrolls patients who meet strict inclusion and exclusion criteria, creating a defined population to isolate a drug's effect. Real-world evidence (RWE) refers to clinical evidence about a therapy's use and outcomes, drawn from sources such as EHRs, claims, and registries that reflect the patients who actually present in routine care. The “population gap” is the difference between these groups, which is particularly pronounced for anti-amyloid Alzheimer's drugs.

CLARITY-AD, the phase 3 trial of lecanemab, randomized 1,795 participants with a mean age of approximately 71, early symptomatic disease (MMSE 22-30, CDR-Global 0.5-1.0), confirmed amyloid pathology, and a baseline MRI without protocol-defined exclusionary findings [1]. Stable anticoagulant use was permitted in CLARITY-AD when anticoagulation was optimized and the dose was stable; however, subsequent lecanemab appropriate-use recommendations advise against treatment in patients requiring anticoagulants because of the risk of intracerebral hemorrhage [4]. Only a minority of real-world AD/MCI patients match the overall trial profile.

Why This Difference Matters for HEOR, Patients, and Care

To ensure a fair comparison, the analysis includes only patients first diagnosed in 2021 or later. This focuses on a contemporary, therapy-relevant population and avoids diluting the estimate with older patients diagnosed years earlier and lost to follow-up.

Amyloid confirmation is the primary bottleneck. Only 21.5% of the post-2021 AD/MCI population had any amyloid test on record, and 12.0% were amyloid-positive. Since amyloid positivity is required for treatment, the eligible population is a small subset of those diagnosed. Low documented testing rates, rather than low disease prevalence, account for most of this reduction. Tests performed outside the available clinical network may not always be captured, so absence of a recorded test should be interpreted as a documentation finding rather than proof that testing never occurred.

Age: Real-world patients are older than those in the trial, with a mean age of 74.6 compared with approximately 71 in CLARITY-AD, and 58.7% are aged 75 or older. About 8.4% fall outside the trial's 50-90 age range. Effectiveness and safety estimates based on a younger cohort may not apply directly to this population.

Concomitant medications: Anticoagulant use was present in 8.6% of amyloid-positive patients. Although stable anticoagulation was permitted in CLARITY-AD, current appropriate-use recommendations advise against lecanemab treatment in patients requiring anticoagulants because of hemorrhage risk [4]. Anticoagulant avoidance is therefore included here as a real-world treatment-readiness criterion rather than a direct CLARITY-AD exclusion criterion.

Stacking the assessable criteria produces a stark funnel: 62,746 diagnosed → 57,167 within the age window → 7,227 amyloid-positive → 6,604 without anticoagulants → 6,444 without a baseline ARIA finding = 10.3%.

Roughly nine of every ten real-world AD/MCI patients would therefore not enter this trial-aligned treatment-readiness cohort before cognitive staging is even considered. Because anticoagulant avoidance reflects current appropriate-use guidance rather than an original CLARITY-AD exclusion, the 10.3% estimate should be viewed as a pragmatic treatment-readiness estimate rather than an exact recreation of CLARITY-AD eligibility.

Real-World vs. Treatment-Ready vs. CLARITY-AD (published)

Measure	CLARITY-AD (published) [1]	Real-world treatment-ready	Real-world AD/MCI overall (dx ≥2021)
Patients (N)	1,795	6,444	62,746
Mean age, years	~71	75.4	74.6
Female	~52%	57.7%	58.5%
White (of patients with known race)	~77%	47.3%	42.5%
On anticoagulant	Permitted if optimized/stable	0% by analysis definition	7.4%
Amyloid-positive	100%	100%	12.0%

The Challenges: Staging Documentation And Safety Readiness

Two data challenges complicate this analysis beyond a simple criteria mismatch. Both arise because the underlying variables were extracted from clinical notes using large language models.

Cognitive staging is inconsistently documented. After note-level extraction, MMSE scores were available for 49.6% of amyloid-positive patients, and 39.2% fell within the trial's 22-30 range. Requiring a documented MMSE of 22-30, as an enrolling site would, reduces the treatment-readiness estimate from 10.3% to 4.0%. This drop largely reflects incomplete documentation of formal cognitive staging, which is recorded for only about half of patients, although some patients with documented scores also fall outside the trial range. A missing MMSE should therefore not automatically be interpreted as patient ineligibility.

CDR-Global, a key CLARITY-AD criterion, could not be reliably operationalized because routine documentation did not consistently distinguish the global score from CDR Sum of Boxes or capture the global score at the level required to apply the trial criterion.

Baseline safety workup is often incomplete. Anti-amyloid therapy requires MRI monitoring for ARIA [4], yet 42.6% of amyloid-positive patients had no MRI documented in their records. This represents a significant documented readiness gap for those most likely to receive treatment. Imaging performed outside the available clinical network may not always be captured, which should be considered when interpreting this finding.

The Current Industry Approach

HEOR and pharmaceutical teams already use RWE to address this gap through external control arms, label-expansion support, and budget-impact and cost-effectiveness models based on real-world utilization. The FDA's Real-World Evidence Framework and subsequent guidance on EHR and claims data formalize how real-world data can support regulatory decision-making [2][3]. Separately, HEOR and market-access teams use RWE to inform payer, access, utilization, budget-impact, and cost-effectiveness questions.

The main limitation is not interest, but whether the underlying data can reconstruct trial-defining variables such as amyloid status, cognitive stage, and ARIA/MRI findings, which are often found in unstructured text rather than coded fields.

How AI Closes The Gap

This capability is demonstrated in each figure above. Amyloid results, MMSE and CDR staging, and ARIA findings were extracted from clinical notes using large language models; the same NeuroDiscovery AI applies across a base of 6M+ patient records and 3M+ active patients in the US.

That extraction is what converts an otherwise difficult-to-answer question into a quantified one: it shows that amyloid testing is documented in only about a fifth of diagnosed patients, that a documented MMSE exists for about half of amyloid-positive patients, and that 42.6% lack a documented MRI. For HEOR teams, reading the notes at scale means modeling the treatment-eligible population as it actually exists and locating precisely where real-world care diverges from the trial-defined ideal.

Benefits

• A truer eligible-population denominator for forecasting, budget impact, and access strategy: Built on the contemporary, treatment-relevant population, not trial proportions.
• Better trial design and feasibility: Inclusion/exclusion criteria informed by the real, older population without sacrificing rigor.
• Stronger HEOR modeling inputs: Real-world age, amyloid-testing, and concomitant-medication distributions feeding effectiveness and cost-effectiveness models.
• Safety and readiness insight: Quantifying gaps like the 42.6% no-documented-MRI finding that bear directly on ARIA monitoring and rollout.

Conclusion

The Alzheimer's RWE-RCT gap is not a rounding error; it is the difference between a ~71-year-old, amyloid-confirmed, cognitively staged and MRI-assessed trial participant and a broader real-world population that is older, largely untested for amyloid, and frequently without a documented baseline MRI. Health economics and outcomes research that models the former without accounting for the latter risks misjudging the population encountered in practice.

Closing the gap begins with real-world data at scale and with the ability to read what clinicians actually wrote. Explore how large-scale, note-level real-world data can reframe your Alzheimer's HEOR and access strategy.

References

1. van Dyck CH, Swanson CJ, Aisen P, et al. Lecanemab in Early Alzheimer's Disease. N Engl J Med. 2023;388(1):9-21. doi:10.1056/NEJMoa2212948.
2. U.S. Food and Drug Administration. Framework for FDA's Real-World Evidence Program. December 2018.
3. U.S. Food and Drug Administration. Real-World Data: Assessing Electronic Health Records and Medical Claims Data To Support Regulatory Decision-Making for Drug and Biological Products Guidance for Industry. 2024.
4. Cummings J, Apostolova L, Rabinovici GD, et al. Lecanemab: Appropriate Use Recommendations. J Prev Alzheimers Dis. 2023;10(3):362-377.
5. Alzheimer's Association. 2024 Alzheimer's Disease Facts and Figures. Alzheimers Dement. 2024;20(5):3708-3821.