How to Read an Alzheimer Drug Trial: CDR-SB and What Counts as Meaningful

A statistically firm result and a life-changing result are not the same thing, and nowhere is that gap more visible than in Alzheimer drug trials. Most of these studies come down to a single figure: how far apart the treated group and the placebo group end up on a scale called the CDR-SB. When lecanemab landed a 0.45-point separation over 18 months, the number was solid enough to survive any statistical objection. Whether it changes a person's daily life is a different question, and the honest answer is that experts still disagree. Reading these trials well means keeping those two questions apart on purpose.

Key points#

• The CDR-SB is a 0-to-18 clinician-rated scale of thinking and daily function; higher scores are worse.
• In CLARITY-AD, lecanemab slowed 18-month decline by 0.45 points versus placebo, a real and statistically strong effect.
• "Statistically real" is not the same as "large enough to notice." A big trial can pin down a tiny true difference.
• There is no single agreed threshold for a meaningful CDR-SB change, which is why reasonable analysts reach opposite verdicts on the same data.
• Group averages hide wide individual variation, so any one patient's benefit cannot be read off the mean.

Start with the number, not the percentage#

Headlines love a percentage. Lecanemab was widely reported as slowing decline by 27 percent, and translated into an estimated five to six months of delayed progression across the trial. Both figures come from the same underlying data, but the percentage is the more flattering way to say it. A 27 percent slowing of a small amount of change is still a small amount of change. The first move any careful reader makes is to convert the story back into the raw metric: how many points, on what scale, over how long. For lecanemab the answer is 0.45 points on the CDR-SB across 18 months. Everything else is interpretation layered on top of that one figure.

What the CDR-SB is actually counting#

The Clinical Dementia Rating Sum of Boxes is not a memory quiz. A trained rater sits down with the patient and someone who knows them well, then scores six areas of life: memory, orientation, judgment and problem-solving, community affairs, home and hobbies, and personal care. Each area is rated from 0 (normal) to 3 (severe), and the six ratings are summed. The total therefore runs from 0 to 18, and a higher score means more impairment.

What makes this scale useful is that it fuses cognition with function. It does not care about a biomarker in the blood; it cares whether someone can still balance a checkbook, follow a recipe, or get dressed without help. That is exactly what patients and families are trying to protect. As a rough anchor, people with early Alzheimer disease left untreated tend to worsen by about one to two points a year on the CDR-SB. That natural drift is the yardstick every drug is measured against.

Reading the 0.45-point result honestly#

In the CLARITY-AD trial, reported in the New England Journal of Medicine, the placebo group declined by an adjusted mean of 1.66 points over 18 months while the lecanemab group declined by 1.21 points. Subtract them and you get the 0.45-point difference, with a P value below 0.001.

Two things are simultaneously true, and skipping either one produces a bad take. First, the result is statistically robust. The confidence interval sits comfortably away from zero, so this is very unlikely to be noise. Second, statistical robustness only certifies that the effect is real, not that it is big. With enough participants, a trial can detect a genuine but minuscule difference with near-total confidence. Firmness answers "did something happen?" It does not answer "is what happened enough to matter?"

Why "meaningful" has no fixed line#

The tool the field reaches for is the minimal clinically important difference, or MCID: the smallest change a patient or clinician would actually notice and care about. The problem is that no agreed MCID for the CDR-SB exists. One commonly cited benchmark treats roughly a 1-point change over a year as the floor for being clinically noticeable. Other work has floated thresholds near 0.98 points at the mild-cognitive-impairment stage and higher for mild dementia. A review in Brain Communications is direct about the underlying issue: there is no standardized way to define meaningfulness in dementia trials, and published estimates for related scales scatter widely.

Now hold 0.45 points up against those benchmarks, and the whole debate becomes legible. An analysis in Scientific Reports pooled anti-amyloid trials and found lecanemab's CDR-SB effect worked out to a standardized mean difference of about -0.19, a small effect size that landed below common cutoffs for clinical importance. A separate analysis in Alzheimer's Research and Therapy argued the group-level difference was still plausibly tangible, while granting that validated thresholds have not been established. Neither side is fumbling the math. They are pressing different, defensible rulers against the same 0.45.

The average is not your patient#

A mean difference describes a gap between two groups, not a promise made to any individual. It does not mean every treated person did 0.45 points better than they otherwise would have. Inside any trial, some people decline quickly and others barely budge, in both arms. A modest average slowing can sit on top of many patients who felt nothing and a smaller number who may have benefited more. This is the reason a result below a proposed MCID does not prove "no one is helped," and a result above one does not guarantee a specific person will be.

The same logic governs cross-drug comparisons. The Brain Communications review notes that donanemab reported a larger relative slowing on the CDR-SB within one baseline subgroup, and that modeling exercises have projected multi-year delays in reaching later disease stages. Those projections are built from trial data rather than observed over time, and they carry their own assumptions. Treat them as hypotheses about the future, not as measurements of it.

A short checklist for the next headline#

When a fresh Alzheimer readout crosses your feed, four questions strip away most of the hype. What was the absolute difference on the CDR-SB in points, not just the percentage slowing? How does that difference compare to any proposed MCID, and does the paper admit the threshold is contested? Is the benefit set against harms such as imaging abnormalities and the monitoring they demand? And are the long-term claims measured outcomes or model projections? Run those four questions and a headline number turns back into something you can actually weigh.