Which risk factors still matter at ninety, and which have stopped

Design

pooled population-based cohort study of three Chinese cohorts, age-stratified Cox models with population attributable fractions

Population

61,045 participants in age bands 30-49, 50-59, 60-79, 80-99 and 100+, followed for a mean of 3.79 to 13.20 years

Primary outcome

all-cause mortality by modifiable risk factor score and by individual factor, within each age band

Effect

risk score HR 1.11 (1.01-1.21) to 1.31 (1.21-1.36); metabolic HR 1.43 (1.14-1.80) middle age to 1.07 (0.99-1.17) centenarians; hypertension PAF 16.2% to 0.1%; low education PAF 6.4% to 30.8%

This pooled analysis combined three Chinese cohorts — the Kadoorie Biobank, the Hainan Centenarian Cohort and the Guangzhou Biobank — to ask whether the same modifiable risks carry the same weight at every age. It covered 61,045 participants across bands from 30-49 up to 100 and over, with follow-up ranging from 3.79 to 13.20 years, and scored 11 modifiable factors.

The overall score predicted mortality at every age (hazard ratios from 1.11, 95% CI 1.01 to 1.21, up to 1.31, 95% CI 1.21 to 1.36). But its composition shifted. Metabolic risk weakened steadily: hazard ratio 1.43 (95% CI 1.14 to 1.80) in middle age against 1.07 (95% CI 0.99 to 1.17) in centenarians, with diabetes and hypertension driving the cluster. Hypertension-attributable mortality peaked at 16.2% (95% CI 8.3 to 22.6) in the presenium and fell to 0.6% and 0.1% in the longest-lived groups. Running the other way, low educational attainment mattered more with age (HR 1.55, 95% CI 1.13 to 2.14, rising to 1.74, 95% CI 1.01 to 3.01; attributable fraction 6.4% to 30.8%), and physical inactivity contributed in both middle age and among centenarians.

The survivor effect is the honest explanation for at least part of this: people who reach 100 with hypertension are, by definition, those in whom it did not kill them, so the surviving cohort is enriched for people who tolerate it. That does not make the finding useless — it is exactly the population sitting in front of a geriatrician. It supports what good geriatric practice already does: relax metabolic targets in the oldest-old where the burden of treatment is real, and keep pushing on activity and social and educational disadvantage, which do not attenuate.

• Weight blood pressure and glycaemic targets against burden of treatment in the very old; the attributable mortality is small
• Do not extrapolate a deprescribing decision from population attributable fractions to an individual with symptomatic disease
• Keep physical activity on the plan at every age — it contributed in middle age and in centenarians alike
• Treat social and educational disadvantage as a clinical risk factor that grows rather than shrinks with age
• Read the attenuation of metabolic risk partly as survivor effect, not purely as changed biology

The statistics, in plain English

A population attributable fraction estimates what share of deaths might be avoided if a risk factor were removed from the population, so hypertension falling from 16.2% to 0.6% means treating blood pressure has far less to offer at the population level in the very old — it does not mean an individual's severe hypertension is harmless. Confidence intervals widen sharply in the oldest groups (the centenarian education hazard ratio runs 1.01 to 3.01) because there are few people and few deaths to work with. And because these are cross-sectional age bands rather than the same people followed from 40 to 100, differences between bands mix ageing with survivorship.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.