- Design
- population-based cohort study with parallel derivation, validation and coefficient comparison
- Population
- 47,958 New Zealand and 46,558 Chinese adults aged 30-74 with diabetes and no cardiovascular disease
- Primary outcome
- first cardiovascular event within five years; comparison of predictor hazard ratios
- Effect
- age HR per decade in women 1.61 (95% CI 1.51-1.71) New Zealand vs 2.51 (2.32-2.72) China; expected-to-observed 0.809 improving to 1.025 after replacing the age coefficient
Two matched primary prevention cohorts of adults aged 30 to 74 with diabetes and no cardiovascular disease — 47,958 people in New Zealand with 5,622 first events, and 46,558 in China with 3,650 — were used to derive five-year risk equations by identical methods, then compare the predictor effects directly. Sixteen predictors went into sex-specific equations.
Most coefficients agreed across the two countries. Age did not. In women, the hazard ratio per decade was 1.61 (95% CI 1.51 to 1.71) in New Zealand against 2.51 (2.32 to 2.72) in China. That is not a small discrepancy: risk climbs far more steeply with age in the Chinese cohort, so an equation built elsewhere will systematically underestimate older patients and overestimate younger ones.
The practical finding is what failed to work. Standard recalibration — the usual remedy, rescaling an imported equation to local event rates — did not correct calibration in the Chinese cohort, with expected-to-observed ratios of 0.938 in men and 0.809 in women. Replacing the New Zealand age coefficient with the Chinese one brought both close to 1.0 (1.027 and 1.025). The fix was not adjusting the overall level but repairing the one coefficient that genuinely differed.
For Indian practice this is directly relevant. Risk calculators in routine use here are derived elsewhere and, at best, recalibrated. This study suggests that is not enough, and that the error will fall hardest on exactly the patients in whom the treat-or-not decision is closest.
- Treat a risk score from an imported equation as an ordering, not an absolute percentage, when counselling.
- Be alert that the error is age-dependent — younger patients may be over-treated and older ones under-treated.
- Where a locally derived equation exists for your population, prefer it to a recalibrated foreign one.
- Do not let a borderline calculated risk override strong individual risk factors.
- If your institution is validating a risk tool, compare predictor coefficients and not only overall calibration.
Why it matters
The risk percentage on the screen is treated as a measurement when it is an estimate built in another population.
Don't overread it
China is not India — this demonstrates that transportability fails, not what the Indian coefficients are.
The statistics, in plain English
Hazard ratios of 1.61 and 2.51 per decade have non-overlapping confidence intervals, so the difference in how age drives risk between these populations is real rather than chance. Expected-to-observed ratios measure calibration: 0.809 means the equation predicted about 19% fewer events than actually occurred in Chinese women, while 1.025 after fixing the age term is close to correct. This is a cohort study of two populations, so it shows that equations do not transport — it does not tell you what an Indian equation's coefficients would be.
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.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for top clinical updates, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free