Anti-amyloid antibodies: what the trials showed

Lecanemab, donanemab and aducanumab are monoclonal antibodies, given by infusion into a vein, that bind amyloid-beta, the protein that collects into plaques in the brain in Alzheimer's. They bind different forms of it. Lecanemab binds soluble protofibrils, the clumps that form before plaque[1]. Donanemab targets a modified form of amyloid already deposited in plaque[6]. Aducanumab binds aggregated forms, both soluble and insoluble[5]. The idea is that removing amyloid slows the disease. The trials agree that the drugs remove amyloid and that people on them worsened a little less, on rating scales, over about 18 months. The argument is about whether that difference is big enough to matter. The FDA called the lecanemab result "statistically significant and clinically meaningful"[FDA]. Two independent reviews that pooled the trials reached a different reading: one found that none of the changes, including those for lecanemab, aducanumab and donanemab, exceeded the smallest difference usually counted as clinically important[14], and a 2026 Cochrane review concluded the antibodies probably make little to no difference to cognition at 18 months[13]. The first of the three, aducanumab, was approved after the FDA's own advisory committee had voted against it[7][12]. The harm side is not in dispute. All three cause ARIA, amyloid-related imaging abnormalities: swelling (ARIA-E) or small bleeds (ARIA-H) in the brain, seen on MRI, usually without symptoms but occasionally serious[FDA][FDA]. The two drugs whose approval stands carry a boxed warning for it[FDA][FDA].

The FDA decision

The FDA granted accelerated approval on 6 January 2023[FDA]. That decision rested on reduction of amyloid plaque on PET in a phase 2 trial, with full approval made to depend on the phase 3 trial[8]. On 6 July 2023 the FDA converted it to traditional approval, saying a confirmatory trial had verified clinical benefit[FDA][9]. It was the first antibody of this kind to reach full approval[9].

The pivotal trial

Clarity AD randomised 1,795 people with early Alzheimer's and confirmed amyloid (898 lecanemab, 897 placebo) to 10 mg per kg into a vein every 2 weeks, or placebo, for 18 months[1]. The primary outcome was the change on the CDR-SB, a dementia-severity scale that runs from 0 to 18. Scores started at about 3.2 in both groups[1]. They rose by 1.21 on lecanemab and 1.66 on placebo: 0.45 points less worsening (95% CI 0.23 to 0.67), or 27% less worsening[1][FDA]. The trial authors' own phrase for this was "moderately less decline", and they wrote that longer trials were warranted[1]. Infusion-related reactions occurred in 26.4% of participants[2].

ARIA

From the label: ARIA of any kind occurred in 21% vs 9% (lecanemab vs placebo); swelling, ARIA-E, in 13% vs 2% (113 of 898 vs 15 of 897); bleeding, ARIA-H, in 17% vs 9%[FDA]. ARIA caused symptoms in 3% (29 of 898) of people on the drug, and serious symptoms in 0.7% (6 of 898)[FDA]. The risk depends heavily on the APOE4 gene. Any ARIA occurred in 45% vs 22% in people with two copies, 19% vs 9% with one copy, and 13% vs 4% with none[FDA]. In the two-copy group, which was 15% of the trial, an exploratory analysis found no treatment effect on the primary scale[FDA]. The label tells prescribers to test for APOE4 before starting[FDA]. In a specialty memory clinic that treated 234 people outside a trial, 9.8% stopped treatment, 4.3% because of ARIA, and symptomatic ARIA was far more common in mild dementia than in MCI (27% in mild dementia vs 1.8% in MCI or very mild dementia)[20].

Who it is for

The label's indication is Alzheimer's disease; treatment should be initiated in mild cognitive impairment or mild dementia, the population in which treatment was started in the trials[FDA]. Amyloid must be confirmed before treatment[FDA]. The trial population was 62% mild cognitive impairment, 38% mild dementia; 77% White, 17% Asian, 3% Black; median age 72[FDA]. Canada's drug agency recommends reimbursing it for APOE4 non-carriers or heterozygotes only, with timely MRI access, and only if the price is reduced[28].

The FDA decision

The FDA approved donanemab on 2 July 2024[FDA][FDA][10]. Unlike the other two it went straight to full approval, on a trial that measured clinical decline rather than amyloid alone[FDA][10].

The pivotal trial

TRAILBLAZER-ALZ 2 randomised 1,736 people with early symptomatic Alzheimer's and both amyloid and tau on PET (860 donanemab, 876 placebo)[3]. The drug was given into a vein every 4 weeks for up to 72 weeks, switched to placebo once amyloid fell below a set level[FDA]. 1,320 (76%) completed the trial[3]. The primary outcome was the iADRS, a combined cognition-and-function score from 0 to 144 where lower is worse, at 76 weeks[3]. In people with low or medium tau it fell −6.02 vs −9.27 (difference 3.25, 95% CI 1.88 to 4.62); across everyone, −10.2 vs −13.1 (difference 2.92, 95% CI 1.51 to 4.33)[3]. The trial report puts that as 35.1% and 22.3% less decline[4], and 23 of 24 of its pre-set outcomes reached statistical significance[3]. On the CDR-SB, the label records worsening of 1.72 on donanemab and 2.42 on placebo across all participants: 0.70 points less worsening (29%)[FDA]. Dosing stopped once scans showed amyloid had fallen below a set level; 17%, 47% and 69% of treated people were eligible to switch to placebo at weeks 24, 52 and 76[FDA]. The label is explicit that there is, in its words, "no data beyond the 76-week duration of Study 1"[FDA].

ARIA

From the label: any ARIA in 36% vs 14%; ARIA-E in 24% vs 2% (201 of 853 vs 17 of 874); ARIA-H in 31% vs 13%; ARIA with symptoms in 6% (52 of 853)[FDA]. A brain haemorrhage larger than 1 cm occurred in 0.5% vs 0.2%, and fatal haemorrhages have been observed[FDA]. The trial report counts 3 in the donanemab group, 1 in the placebo group[4]. By APOE4 status, any ARIA occurred in 55% vs 22% with two copies, 36% vs 13% with one, and 25% vs 12% with none[FDA]. Across both donanemab trials, the risk of ARIA-E rose with the number of APOE4 copies, with microbleeds already on the baseline scan, with amyloid load and with blood pressure[17]. A later trial of 843 people found that building the dose up more slowly cut ARIA-E at 24 weeks to 13.7% vs 23.7%, with the same amyloid reduction[18].

Who it is for

The label's indication is Alzheimer's disease; treatment should be initiated in mild cognitive impairment or mild dementia[FDA]. Amyloid must be confirmed, and the label says APOE4 testing should be done before treatment to inform the risk of ARIA[FDA]. The trial population: mean age 73 (59 to 86); 71% APOE4 carriers; 91% White, 6% Asian, 4% Hispanic or Latino, 2% Black[FDA].

The FDA decision

The FDA granted accelerated approval on 7 June 2021, the first new Alzheimer's drug since 2003[7][12]. The label states the basis: approved under accelerated approval based on reduction in amyloid beta plaques; continued approval may be contingent upon verification of clinical benefit in confirmatory trial(s)[FDA]. The plaque reduction cited was 71% in EMERGE, 59% in ENGAGE[7]. The FDA's outside advisory committee had voted against approval in November 2020[7], and a 2021 commentary in the library raised concerns that safeguards were insufficient to mitigate regulatory capture within the FDA[12]. The maker discontinued the drug on 31 January 2024, and the required confirmatory trial (ENVISION) was ended; its stated reason was "not related to any safety or efficacy concerns"[11].

The pivotal trials

There were two identical trials, EMERGE (Study 1) and ENGAGE (Study 2), both stopped before planned completion[FDA], run at 348 sites in 20 countries[5]. EMERGE randomised 1,638 people and ENGAGE 1,647, each to a low dose, a high dose or placebo, for 18 months, with the primary endpoint at week 78; the primary endpoint was the CDR-SB[FDA]. In EMERGE the high dose worsened by 1.35 against 1.74 on placebo: 0.39 points less worsening (22%), p = 0.012[FDA]. In ENGAGE there was no statistically significant difference from placebo on CDR-SB at 78 weeks[FDA]. The approval rested on the amyloid reduction, not on these clinical results[7][FDA].

ARIA

At the 10 mg/kg dose, ARIA of any kind occurred in 41% vs 10% (454 of 1,105 vs 111 of 1,087); ARIA-E in 35% vs 3%, and 42% vs 20% in APOE4 carriers against non-carriers[FDA]. ARIA-H alongside swelling occurred in 21% vs 1%[FDA]. Among people with ARIA, 24% vs 5% had symptoms, and serious symptoms were reported in 0.3% of those treated[FDA].

Who it was for

The 2021 label's indication was Alzheimer's disease, under accelerated approval[FDA], studied in people with mild cognitive impairment or mild dementia with confirmed amyloid; mean age 71 (50 to 85)[FDA]. Since 31 January 2024 it is no longer available[11].

Benefit beyond the trial horizon, or on outcomes people feel

Every primary result above is a score on a rating scale at 18 months or 76 weeks. None of the trials followed people until they lost independence, moved into care, or died, and the FDA label for donanemab says there is "no data beyond the 76-week duration of Study 1"[FDA]. Lecanemab's extension study reports benefit continuing to 36 months, but after the first 18 months everyone received the drug, so there is no placebo group to compare against[19]. The estimates of extra months of independence that circulate (10 months for lecanemab (95% CI 4 to 18) and 13 months for donanemab in the low or medium tau group (95% CI 6 to 24)) come from a model that maps CDR-SB scores onto when 282 people followed for an average of 2.9 years lost independence, not from the trials observing it[24]. The analysis reporting that donanemab cut the risk of progressing from MCI to mild dementia (HR 0.67 for MCI to mild dementia; HR 0.50 for mild to moderate dementia) carries caveats its own record notes: it was authored by manufacturer employees and consultants, with several post hoc analyses and a 76-week horizon[23].

Pooled across the class, the Cochrane review of 17 studies and 20,342 participants rated the cognition result as "probably little to no difference" (moderate certainty) and the CDR-SB result as "may make little to no difference" (low certainty)[13]. It noted that there was only one trial each of lecanemab and donanemab, that side effects may unblind participants, and that all the trials were industry-funded[13]. The 2024 meta-analysis of 19 publications, 23,202 participants and 8 antibodies put the pooled CDR-SB difference at 0.18 points (95% CI 0.03 to 0.34) and found that none of the changes, including those for lecanemab, aducanumab and donanemab, exceeded the minimal clinically important difference[14]. An independent value review in 2023 asked whether the evidence showed a net health benefit over supportive care; its committee voted 12 to 3 that it did not, and it priced the benefit at $8,900 to $21,500 a year vs $26,500[21].

Long-term brain safety

A meta-analysis of brain-volume data found that ARIA-causing antibodies accelerated ventricular enlargement by 2.1 mL (38.7% more than placebo; 95% CI 1.5 to 2.8), and projected that treated people with MCI would reach brain volumes typical of Alzheimer's dementia about 8 months earlier than untreated[22]. What that means for a person is not known. The Alzheimer's Association's ARIA workgroup estimates that roughly 1 to 2% of ARIA cases end in permanent disability or death, and warns that ARIA-E can mimic a stroke and must be ruled out before clot-busting treatment[16]. Across nine phase 3 cohorts with 6,315 patients, ARIA-E occurred in 25.5% after removing outlier cohorts, symptomatic ARIA-E in 6.7% and ARIA-H in 17.8%, and the odds of ARIA-E were 5.6 times for two copies, 1.9 times for carriers, vs non-carriers[15]. The Cochrane figures are 107 more per 1,000 treated (95% CI 77 to 148) for any ARIA-E and 29 more per 1,000 treated (95% CI 22 to 38) with symptoms[13]; the 2024 meta-analysis gives a number needed to harm of 9 for ARIA-E, 13 for ARIA-H, 86 for symptomatic ARIA-E[14].

Whether the results carry over to people outside the trials

The trial populations were narrow. Clarity AD was 77% White, 17% Asian, 3% Black[FDA]; TRAILBLAZER-ALZ 2 was 91% White, 6% Asian, 4% Hispanic or Latino, 2% Black[FDA]. The donanemab ARIA analysis itself records that the trial populations exclude many real-world patients, for example those on anticoagulants[17]. When one memory clinic applied the lecanemab criteria to 1,309 new patients, 8% (21% of those with MCI or Alzheimer's) were eligible, falling to 6% (15% of those with MCI or Alzheimer's) once APOE4 homozygotes and people on anticoagulants were excluded[25]. Among 10,804 people screened for the lecanemab and related trials, the odds of failing the amyloid criterion, compared with non-Hispanic White participants, were 3.20 (Hispanic Black), 4.15 (Hispanic White), 2.35 (non-Hispanic Asian), 3.75 (non-Hispanic Black)[26]. Early use follows the same pattern: the 2,512 Medicare patients who received lecanemab had a mean age of 75.4 years, against 80.8 for untreated patients with the same diagnoses; 91.1% vs 86% were White and 1.1% vs 6.7% were Black, treated against untreated[27]. Whether the drugs work the same way in older, sicker, less White and less tightly monitored people than the trials enrolled is not something the trials can answer.