A daughter sits in a neurology office with a printout in her lap. Her mother, seventy-four, has mild cognitive impairment, an amyloid PET scan that came back positive, and the offer of an infusion every two weeks. On the printout is a number: 0.45. The clinician explains that this is the treatment effect. The daughter asks the only question that matters to her. Forty-five hundredths of what?

Of a point, on a scale that runs from 0 to 18.

That scale is the Clinical Dementia Rating Sum of Boxes, and it is built out of conversation. A clinician interviews the person and the care partner about memory, orientation, judgment, community affairs, home life and hobbies, and personal care. Six domains, each scored in steps, summed. In the phase 3 Clarity AD trial of lecanemab, published in the New England Journal of Medicine, participants started at an average score of about 3.2, which is the territory of someone still running her own household while losing the thread of appointments and names.

Over eighteen months, the placebo group's score worsened by 1.66. The lecanemab group's worsened by 1.21. The gap, 0.45 points, with a confidence interval from 0.23 to 0.67, is the entire clinical finding. Both groups declined. One declined slightly less steeply.

The trial was large and properly done: 1,795 people aged 50 to 90, randomly assigned 1:1 to intravenous lecanemab at 10 mg per kilogram every two weeks or to placebo, double-blind, eighteen months. Everyone enrolled had mild cognitive impairment or mild dementia due to Alzheimer's, with amyloid confirmed by PET scan or spinal fluid. That enrollment list is the answer to the first question a family usually asks, which is whether this is for them. The evidence covers people at the early end, with a positive amyloid test, and it stops there. People with moderate or advanced dementia were not in this trial.

The drug did what it was designed to do inside the skull. In an imaging substudy of 698 participants, amyloid burden fell by 59.1 centiloids more than placebo, enough to take many scans from clearly positive to below the threshold for positivity. Plaque is removable. The removal buys a fraction of a point.

Then there is the part that brings people back for MRIs.

ARIA, amyloid-related imaging abnormalities, is the signature risk of this entire drug class: swelling in the brain tissue (ARIA-E) and small bleeds or iron deposits (ARIA-H). Most of it shows up on a scan in someone who feels fine. Some of it produces headache, confusion, visual changes or seizures, and a small number of cases have been severe and have included deaths. The US prescribing information for lecanemab carries a boxed warning for ARIA, calls for MRI before dosing and at set points during treatment, and flags APOE4 genotype, because carriers of the gene variant, particularly those with two copies, have shown higher ARIA rates. That is why a genetic test now arrives early in these conversations, attached to a risk estimate rather than to a diagnosis.

The regulatory record on donanemab makes the same point from another direction. The phase 3 TRAILBLAZER-ALZ 2 trial was published in JAMA in July 2023. Two years later, in July 2025, Lilly announced that the FDA had approved an updated label with a modified titration schedule, a slower ramp to the full dose, adopted specifically to lower the rate of ARIA-E. When a manufacturer redesigns the dosing of an approved drug to reduce one adverse event, that event is common enough to engineer around.

Now the logistics, which no trial result captures and every family feels. Eighteen months of infusions every two weeks is roughly thirty-nine appointments. Add the MRI scans on the monitoring schedule, the drive, the parking, the hour in the chair, the day someone takes off work. For a son who lives ninety minutes away, the real currency is not the copay. It is Thursdays.

And the arithmetic of the benefit stays modest at the level of one person. A 0.45-point difference is smaller than a single half-point step in one of the six boxes. Across a trial population it is a statistically solid signal that the disease course bent. At a kitchen table it may look like a few extra months of managing the checkbook, or it may be invisible. Researchers continue to follow participants in open-label extension studies, with three-year data presented at the Alzheimer's Association International Conference in 2024, though open-label follow-up has no placebo group left to compare against, which limits what can be read from it.

So the decision in front of that daughter is not really about Alzheimer's biology. It is a trade between a measurable but small slowing, a monitored risk that is mostly silent and occasionally serious, and eighteen months of a family's calendar. Trials give the first two numbers. The third one belongs to the people doing the driving, and it is theirs to weigh with the clinician who knows the scans, the genotype and the medical history in front of them.

The amyloid comes off the scan. The clock keeps its own time. Families deserve to see both of those facts written down before they sign anything.