Before this cohort, the sex gap in dementia had been measured mostly in people 65 and older, and Rachel Whitmer, the senior author, says nobody knew whether it held after 90 [12]. Filling that gap needed a particular kind of data, and the Kaiser Permanente charts are it: for some participants the records reach back to the 1960s [4]. Work the arithmetic on a participant at the cohort's median age of 92 when enrolment began in 2018, and you get a birth year around 1926 [14], which puts a 1960s record at about age 34 [16] and their sixties in the late 1980s [15]. So the hypertension and high cholesterol carried by people who arrived at 90 cognitively intact [10] were written down at the time by a clinician with no dementia hypothesis in mind, not reconstructed from a nonagenarian's memory. That clean, contemporaneous record is the study's real asset, one no other design could assemble quickly.
What the entry criterion buys in cleanliness, it costs in interpretation. Everyone here survived to 90 without dementia before the first assessment [2]. The APOE4 result is where that bites: null across the cohort as a whole, higher risk among men, roughly doubled among Black participants [9]. A variant that is flat overall and strong inside subgroups can mean the subgroups genuinely differ. It can also mean the carriers in whom it would have shown up never reached the enrolment desk. Hilary Colbeth, the first author, says the finding has pushed the group to examine how APOE genotype affects mortality among those with and without dementia by 90 [11], which is the analysis that separates those two readings.
What these figures leave out is the baseline: how often dementia actually occurs after 90. The doubling for women, the 75% gap between Black and Asian participants and the 60% protection from APOE2 are all relative [19]. Without the incidence rate in the reference group, a factor of two is not yet something to plan care around. Cell sizes deserve the same caution. Split 800 people by two sexes, four race and ethnicity groups and three APOE categories and even an even split averages about 33 people per stratum [20], and real cohorts are never even. The direction of these subgroup effects is more trustworthy than their magnitude.
The resilient group is the finding I would spend money on. It is a natural experiment sitting inside a cohort of more than 800 with measured midlife exposures [3], assessed every six months [2], in a population projected to number roughly 230 million by 2100 [13]. It is descriptive today, but it could become sampleable tomorrow if someone characterises it.
