Patient population characteristics across the five trials
The five studies in this meta-analysis enrolled a total of 4,824 patients — 2,410 in the experimental group receiving either Lecanemab or Donanemab, and 2,414 in the placebo control group. Across all trials, the mean participant age clustered tightly between 71 and 75 years, reflecting the characteristic demographic of early Alzheimer's disease research cohorts. Female participants constituted a substantial and often majority share of enrolment in the studies that reported sex data, consistent with the higher population prevalence of Alzheimer's disease in women.
APOE ε4 carrier status was reported in the majority of trials and represented a substantial proportion of participants in each. In van Dyck 2023 (Lecanemab), 592 of 859 experimental participants and 600 of 875 control participants were APOE ε4 carriers. In Honig 2024 (Lecanemab), the figures were 620 of 898 (experimental) and 611 of 897 (control). In Mintun 2021 (Donanemab), 95 of 131 experimental and 92 of 126 control participants were carriers. This high carrier frequency is clinically significant: according to research published by the Nature portfolio, APOE ε4 status is associated both with elevated Alzheimer's disease risk and with increased incidence of amyloid-related imaging abnormalities (ARIA), a key safety endpoint in these trials.
The five-trial Lecanemab/Donanemab meta-analysis enrolled 4,824 patients in total — 2,410 in the experimental arm and 2,414 in the placebo arm — with mean participant ages ranging from 71.0 to 75.4 years across studies, all focused on adults with early Alzheimer's disease.
The McDade 2022 dataset was split into two sub-cohorts within the meta-analysis (McDade-1: n=70; McDade-2: n=484). Notably, McDade-1 did not report female sex ratio or APOE ε4 carrier frequency, introducing a small gap in the characterisation data. The Swanson 2021 cohort reported age as a range (experimental: 71, range 53–90; control: 72, range 50–89) rather than a standard deviation, reflecting a slightly different reporting convention from the other trials.
Sample size variation and its effect on pooled weight
Sample size variation across the five studies is substantial, and this directly determines how much each trial drives the pooled estimate. The Honig 2024 trial contributed the largest cohort at 1,795 patients, closely followed by van Dyck 2023 at 1,734 patients. Together, these two Lecanemab trials account for approximately 73% of the total enrolled population. At the other extreme, McDade-1 enrolled just 70 patients, and Mintun 2021 enrolled 257 — the only Donanemab study in the pool.
This disparity matters for interpreting the pooled results. In a meta-analysis, larger trials exert greater influence on the combined effect estimate. Because four of the five trials tested Lecanemab and only one (Mintun 2021, n=257) tested Donanemab, the pooled conclusions more accurately characterise Lecanemab's profile. Donanemab targets deposited amyloid plaques with high-affinity characteristics, while Lecanemab targets soluble protofibrils — a mechanistic distinction that may underlie some of the observed variability, particularly in amyloid PET outcomes.
Donanemab targets deposited mature amyloid plaques with high-affinity clearance characteristics, whereas Lecanemab targets soluble amyloid protofibrils. This difference in molecular target may contribute to the high heterogeneity (I²=95%) observed specifically for amyloid PET outcomes in the pooled analysis.
How I² heterogeneity (0%–95%) shapes reliability by endpoint
Heterogeneity statistics varied dramatically across endpoints in this meta-analysis, and the choice of statistical model — fixed-effects versus random-effects — followed accordingly. For ADCOMS, CDR-SB, ARIA-E, ARIA-H, and the majority of adverse events (including death, serious adverse events, falls, dizziness, arthralgia, and anxiety), I² values were 0% or very low. These outcomes used a fixed-effects model and are considered highly reliable: the studies were measuring the same underlying effect with minimal true between-study variation.
In the five-trial Lecanemab/Donanemab meta-analysis, ADCOMS, CDR-SB, ARIA-E, ARIA-H, and most adverse event endpoints showed I²=0%, supporting use of a fixed-effects model and indicating high reliability and generalisability of pooled results for these outcomes in early Alzheimer's disease patients.
ADAS-Cog 14 presented moderate heterogeneity at I²=47%. A fixed-effects model was still applied, but the presence of some between-study variability — potentially attributable to differences in patient characteristics, the specific drug used, or measurement nuances — slightly reduces the precision of the pooled estimate compared to the zero-heterogeneity endpoints. The overall conclusion of significant cognitive improvement remained robust given the strong associated P-value.
Need to search and analyse clinical trial data across anti-amyloid therapies? PatSnap Eureka gives you AI-powered access to the full patent and literature landscape.
Explore Clinical Trial Data in PatSnap Eureka →The I²=95% problem: amyloid PET and the limits of pooled precision
The amyloid PET endpoint produced an I² of 95% (P < 0.00001) — the highest heterogeneity value in the entire meta-analysis — necessitating a random-effects model. The pooled effect size was HR = −72.99 SUVr (95% CI: −88.58 to −57.41). An I² of 95% means that 95% of the observed variability in amyloid reduction across the five trials reflects true differences between studies, not random sampling error. The wide confidence interval is a direct consequence of this variability.
"An I² of 95% means that 95% of the observed variability in amyloid clearance effects is due to true differences between studies — not chance. The pooled effect size of −72.99 SUVr must therefore be interpreted as an average across substantially different treatment contexts, not a universal figure."
Several factors plausibly drive this extreme heterogeneity. First, the mechanistic difference between Donanemab (which targets deposited mature plaques with high-affinity clearance characteristics) and Lecanemab (which targets soluble protofibrils) means these drugs operate on different amyloid species and may produce quantitatively different PET signal reductions. Second, PET imaging protocols — including the choice of amyloid tracer, acquisition timing, and quantification method — may differ across study sites and trial designs, introducing measurement-level variability. Third, differences in baseline amyloid burden across the enrolled populations could produce different absolute magnitudes of reduction even if the relative treatment effect were identical.
In the Lecanemab/Donanemab meta-analysis, amyloid PET data showed I²=95% (P<0.00001), requiring a random-effects model. The pooled effect was HR=−72.99 SUVr (95% CI: −88.58 to −57.41). This high heterogeneity means the precise magnitude of amyloid clearance may vary considerably across different clinical settings, patient subgroups, and PET imaging protocols.
The practical implication for clinicians and researchers is important: while the overall direction of effect — that Lecanemab and Donanemab significantly reduce amyloid burden on PET — is strongly supported, the precise magnitude of −72.99 SUVr should not be applied uniformly across clinical contexts without accounting for these sources of variability. As noted by regulatory guidance from the EMA and reflected in WIPO-indexed patent literature on amyloid imaging biomarkers, the relationship between PET-measured amyloid reduction and downstream clinical benefit remains an area of active investigation.
The random-effects model correctly accounts for between-study variability in amyloid PET outcomes, but the resulting wide confidence interval (−88.58 to −57.41 SUVr) reflects genuine uncertainty about the magnitude of amyloid clearance in any specific clinical setting. The conclusion that these drugs reduce amyloid is robust; the exact degree of reduction is not.
Generalisability, inclusion criteria, and clinical applicability
The meta-analysis inclusion criteria — published RCTs or clinical trials in adults with early Alzheimer's disease, comparing Lecanemab or Donanemab against placebo, with a minimum treatment duration of 72 weeks — were designed to ensure a coherent and clinically relevant evidence base. The 72-week minimum is particularly important: disease-modifying effects in early Alzheimer's disease require extended observation to manifest as detectable changes in cognitive scales and biomarkers. Studies of shorter duration would risk underestimating treatment effects and were appropriately excluded.
The consistent age range (71–75 years), early-stage disease focus, and high APOE ε4 carrier representation across trials enhance the internal coherence of the pooled population. However, two considerations limit broader generalisability. First, the variation in what constitutes "early Alzheimer's disease" across individual trial protocols — whether defined by amyloid PET positivity, CSF biomarker thresholds, or clinical staging criteria — introduces subtle definitional heterogeneity that is not fully captured by the I² statistic. Second, the predominance of Lecanemab trials (four of five) means the pooled safety and efficacy conclusions are more directly applicable to Lecanemab than to Donanemab, despite both drugs being included under the broader anti-amyloid antibody category.
The Lecanemab/Donanemab meta-analysis required a minimum treatment duration of 72 weeks for trial inclusion, targeting adults with early Alzheimer's disease. Four of the five included trials tested Lecanemab and one tested Donanemab, meaning pooled conclusions are most directly generalisable to Lecanemab in early Alzheimer's disease populations with high APOE ε4 carrier representation.
For clinical applicability, the low-heterogeneity outcomes (ADCOMS, CDR-SB, ARIA-E, ARIA-H, most adverse events) provide the most actionable pooled estimates. These conclusions — that anti-amyloid antibodies produce measurable cognitive benefits and carry defined ARIA risks — are reliable across the range of patient populations represented in the meta-analysis. The ADAS-Cog 14 finding, with moderate heterogeneity (I²=47%), remains clinically meaningful but warrants awareness that individual study estimates vary more than the pooled figure suggests.
Tracking the evolving patent and clinical evidence landscape around anti-amyloid therapies? PatSnap Eureka surfaces the connections between trial data, biomarker patents, and regulatory filings.
Search Anti-Amyloid Patent Intelligence in PatSnap Eureka →The high APOE ε4 carrier frequency across trials — particularly in van Dyck 2023 and Honig 2024, where carriers represented the majority of both arms — means that the safety profile for ARIA, which is known to be elevated in APOE ε4 carriers, is well-characterised for this subgroup. However, the generalisability of these ARIA findings to APOE ε4 non-carriers, who are under-represented in the pooled population, is less certain. Clinicians treating non-carrier patients should interpret the pooled ARIA rates with this caveat in mind.
From a research and drug development perspective, the high amyloid PET heterogeneity signals an unresolved question: does the degree of PET-measured amyloid clearance predict clinical benefit proportionally, and does this relationship hold equally for Lecanemab and Donanemab given their distinct mechanisms? Addressing this through pre-specified subgroup analyses in future trials, or through individual patient data meta-analysis, would substantially improve the clinical applicability of amyloid PET as a surrogate endpoint. Resources such as PatSnap's life sciences intelligence platform and databases maintained by PatSnap can support systematic mapping of the biomarker patent landscape as this field evolves.