A Cochrane review published this April looked at seventeen trials of the drugs meant to clear amyloid from the brain in Alzheimer's disease, and it concluded that they probably make little to no difference to a person's memory or thinking eighteen months in. That sentence is true. It is also, standing alone, the wrong thing to hand a family deciding whether to start lecanemab or donanemab, because fifteen of those seventeen trials were not testing lecanemab or donanemab.

The review, led by Francesco Nonino with Edo Richard as senior author, pooled 20,342 people across nine different antibodies: aducanumab, bapineuzumab, crenezumab, donanemab, gantenerumab, lecanemab, ponezumab, remternetug, and solanezumab. Twelve of the seventeen trials tested drugs that missed their own primary endpoint. Three of those twelve tested aducanumab, later pulled from the market outright. Only two trials in the whole review, one for lecanemab and one for donanemab, tested the antibodies now actually prescribed. On the pooled measure of cognitive decline, the effect came out as a standardized mean difference of -0.11, moderate certainty. Dementia got a shade slower too, about the same size of effect but on weaker evidence (a standardized mean difference of -0.12, low certainty, against -0.11 for cognition). Nonino put it plainly: "Unfortunately, the evidence suggests that these drugs make no meaningful difference to patients." Richard, who still sees patients weekly, added that existing drugs "offer some benefit for some patients, but there remains a high unmet need for more effective treatments. Sadly, anti-amyloid drugs do not offer this and bring additional risks."

The pushback arrived within days, and some of it is hard to dismiss. Bart De Strooper of the UK Dementia Research Institute called the pooling a category error: "By mixing failed drugs with the only antibodies that have actually changed clinical practice, it turns therapeutic progress into statistical noise." Jonathan Schott, at the same institute, made the statistical point plainly: pool enough failures into one average and "it is almost inevitable that the conclusion will be that as a group they are clinically ineffective." Richard's defense, when asked, was that the nine drugs share one target even if they work by different mechanisms: "they all have the same target: amyloid-beta proteins." That is true, and it does not change that a class average can honestly report a mostly failed class while obscuring what its two survivors did. The most useful voice here may be David Knopman of the Mayo Clinic, who backed the methodological objection without letting it become an endorsement: "I have some serious doubts about the clinical benefits of lecanemab and donanemab, but the Cochrane review didn't address that."

So look at the two drugs alone. In Clarity AD, lecanemab slowed decline on an eighteen-point dementia severity scale by 0.45 of a point over eighteen months, against an expected untreated decline of a bit over a point a year. That is roughly six months of decline deferred over a year and a half of infusions, every two weeks, with brain swelling on imaging in about 126 of every 1,000 people treated and microbleeds in about 173 of every 1,000. In TRAILBLAZER-ALZ 2, donanemab slowed decline by 3.25 points on a 144-point scale, about a third slower than placebo, over roughly the same period, with three deaths in the trial adjudicated as related to the drug's brain-swelling-and-bleeding side effect. Across the whole nine-drug pool, that same swelling turned up in 119 of every 1,000 treated people against 12 of every 1,000 on placebo. None of this is a rounding error, and none of it is the transformation the drug names have acquired in casual conversation.

It is also not free, and not necessarily available. Britain's regulator licensed both drugs in 2024; donanemab is restricted to people with zero or one copy of the APOE4 gene, since two copies carries materially higher swelling risk. NICE, England's cost-effectiveness body, has separately and repeatedly declined to fund either drug on the National Health Service, putting the price at five to six times what it normally accepts per quality-adjusted life year, the standard measure regulators use to weigh what a year of good health is worth against what it costs to buy. A patient there can be legally prescribed a drug her own health service has judged not worth its price.

Meanwhile the field is quietly hedging its bets. Jeffrey Cummings' 2026 pipeline count found 158 drugs in 192 trials, a field grown by nearly 40 percent in a decade, and amyloid's share of it has fallen from about a third to about a fifth, with tau and inflammation each climbing to roughly the same fifth. "Alzheimer's is a complex disease with many contributing elements," Cummings said of the shift toward inflammation. Amyloid did not disappear from the research. It just stopped being the only bet.

Nobody in this argument is lying to you. The review is not wrong that most of the class failed. The critics are not wrong that two drugs, examined on their own, did something small and real. Half a point on an eighteen-point scale is not nothing, and it is not much either. It buys months, not years, and it comes with a scanning schedule, a gene test, and, for some patients, a bill their own health system has already decided not to pay.