In 4,392 cognitively healthy adults in Taiwan, APOE4 carriers pulled away from non-carriers on a standard thinking test after about age 70. Two copies pulled away hardest. A genome-wide Alzheimer risk score that excluded APOE did not. [1]
That is a quieter finding than a new drug, and more useful than it sounds. Most of the ε4 cognitive-aging literature is European-ancestry. East Asian cohorts are thinner, even though some work suggests the ε4 multiplier may run higher there, not lower. See APOE4 risk by ancestry. A community study from Taiwan that can separate APOE from the rest of the genome fills a real gap in that picture.
What this page answers
- What changed on a 30-point thinking screen, and when
- Why “the other genes” did not show up in six years
- What a 70-year divergence means if you are 55
What the Taiwan study measured
Yu-Chu Chung, Wei Chen, and colleagues used the Healthy Aging Longitudinal Study in Taiwan, a community cohort started in 2009. They analyzed 4,392 people aged 55 and older who had both genetic data and Mini-Mental State Examination (MMSE) scores, a 30-point clinic screen of orientation, memory, attention, language, and spatial skill (higher is better; it is coarse and better at catching obvious impairment than subtle midlife change), across two waves from 2009 to 2019. Mean age was 68.2 years, 54% were women, 723 (16.5%) carried one ε4 copy, and 33 (0.8%) carried two. Mean follow-up was 6.3 years. [1]
Carrier frequency itself is a finding. About 17% carried at least one ε4 copy, below the 20 to 30% often quoted for European-ancestry samples, which matches older cross-population work. [3] A rarer allele in a population can still pack a punch in the people who have it.
Average MMSE decline was 0.2 points per year. In the 3,259 people with complete two-wave scores, the mean drop was 1.3 points over the interval. Cross-sectional scores looked similar by genotype. Change over time did not. The drop steepened with age more in carriers than in non-carriers, and most of all in people with two copies. Plotted by 5-year age bands, the lines started to split between 70 and 75. [1]
The genetic-score piece is the twist. The team built a polygenic risk score, adding up many common DNA variants each with a tiny effect, from the rest of the genome after taking APOE and nearby related variants out of the model. That score, whether treated as a continuous number or split into thirds, did not track MMSE decline over this follow-up. [1]
That does not mean “only APOE matters.” It means that in six years of MMSE in this cohort, APOE4 was the genetic signal that showed up, and the remaining common-variant load did not. A longer follow-up, a harder cognitive battery, or a dementia endpoint could still let the rest of the genome speak.
Why the timing matters more than the test score
A split after 70 is easy to misread as “nothing to do until then.” The biology argues the opposite. Amyloid and tau changes in people with two copies are already common in midlife. [2] MMSE is a blunt instrument. It will not see a decade of silent protein change. It will see the moment that change starts to cost points on a 30-item screen. If the test only diverges at 70, the work of blood pressure, exercise, and metabolic health still belongs in the 50s and 60s, when the slope is being set. That is the same logic as APOE4 and Alzheimer’s risk: relative risk is not destiny, and the levers are earlier than the symptoms.
Two copies still look more serious, which fits the two-copy explainer. The catch is sample size. Thirty-three people with two copies is enough to see a direction, not enough to quote a precise annual point loss as if it were a personal forecast.
How steep the MMSE split was
| What | Figure | Context |
|---|---|---|
| Sample | 4,392 community-dwelling adults, Taiwan | mean age 68; two MMSE waves |
| One copy / two copies | 723 (16.5%) / 33 (0.8%) | ~17% any ε4, lower than many European cohorts |
| Follow-up | 6.3 years mean | 2009–2019 |
| Mean annual MMSE change | −0.2 points (SD 0.5) | 30-point scale |
| Two-wave mean drop | 1.3 points (SD 2.9) | 3,259 people with complete scores |
| Carrier vs non-carrier decline | steeper with age in carriers (P = 0.001) | mixed-effects model |
| Two copies | steeper still (P = 0.03) | 33 people |
| Divergence | after ~age 70 | 5-year age bands |
| Gene score without APOE | no link with MMSE decline | this follow-up window |
What an observational aging study cannot do
This is observational. People who return for a second wave differ from people who do not: excluded and incomplete-follow-up participants were older, less educated, less active, and sicker. About a quarter of wave-2 MMSE data were entirely missing. The authors used models that try to correct for who came back, which is the right instinct, not a magic eraser. [1]
MMSE is also a poor tool for high-functioning people and a noisy one in cohorts with limited schooling. The HALST team has used an MMSE cut-off of 16 to separate frank impairment from low scores driven by literacy, which tells you the instrument is under strain. There were no dementia diagnoses, no p-tau217, and no PET. Faster MMSE decline is not the same as Alzheimer’s, vascular dementia, or “your genes have started.” For how to keep an observational aging study in its lane, see reading a study like a skeptic.
What this means if you are 55, or 75
- If you are in midlife, treat 70 as a reminder that the slope is long, not as a date when worry starts. The Start Here levers are the same ones this paper cannot see on an MMSE.
- If you are East Asian or have East Asian ancestry, do not inflate a European 3- to 4-fold ε4 figure, and do not shrink it either. This study is one more reason to take ε4 seriously in this ancestry group, with still-thin local data. Pair it with ancestry and the risk gradient.
- Do not buy a polygenic Alzheimer score because this paper found the non-APOE score quiet. Quiet over six years of MMSE is not “the rest of the genome is irrelevant.”
- If thinking has already changed, skip the internet interpretation of a 1.3-point drop and get a clinical workup. Mild cognitive impairment and early detection is the practical next page.
Common questions
Does this mean decline only starts at 70? It means a 30-point screen started to show a genotype gap then. Biology can run earlier. Midlife prevention still has the better timing argument.
I carry two copies. Is that steeper number my personal slope? No. It is a 33-person subgroup on a coarse test. Directionally it matches a lot of other two-copy data. It is not a calendar.
Should I get a polygenic risk score if I already know my APOE? Not on the strength of this paper. APOE was the genetic signal that moved MMSE here. A consumer score would add cost and anxiety without a clear action.
In this Taiwanese cohort, APOE4, not the rest of a common-variant score, tracked faster late-life decline on a simple thinking test. Use it as a reason to work the midlife levers, not as a countdown.