Simultaneous derivation, validation, and comparison of predictor hazard ratios for cardiovascular risk prediction equations in patients with diabetes from high versus non-high income countries: cohort study.
Liang J, Choi Y, et al. • BMJ (Clinical research ed.) • 2026
Applying a high income country derived CVD risk prediction equation to a population from a non-high income country with diabetes may introduce systematic errors, primarily driven by differences in the effect of age, and updating key predictor coefficients, particularly age, should be considered for more accurate risk stratification.
Key Findings
Methods
Two equivalent population-based cohorts of people with diabetes and no prior CVD were derived from New Zealand and China for comparison of cardiovascular risk prediction equations.
New Zealand cohort: 47,958 participants aged 30-74 years with diabetes and no CVD, based on electronic health records, between 2004 and 2018.
Chinese cohort: 46,558 participants aged 30-74 years with diabetes and no CVD, based on electronic health records, between 2010 and 2018.
New Zealand participants had 5,622 (11.7%) first CVD events; Chinese participants had 3,650 (7.8%) first CVD events.
Country-specific five-year equations were derived using standardised methods to predict first cardiovascular events, with sex-specific equations including 16 predictors.
Results
Hazard ratios for most predictors were similar between New Zealand and Chinese equations but differed markedly for age.
For women, the age hazard ratio per 10 years was 1.61 (95% CI 1.51 to 1.71) in New Zealand versus 2.51 (95% CI 2.32 to 2.72) in China.
The age effect was substantially larger in the Chinese cohort compared to the New Zealand cohort for both sexes.
Most other predictor hazard ratios were similar between the two country-specific equations.
Sex-specific equations included 16 predictors in both countries.
Results
Standard recalibration of New Zealand equations applied to the Chinese cohort did not adequately improve calibration.
After standard recalibration, the expected to observed (E/O) ratio was 0.938 in men and 0.809 in women in the Chinese cohort.
An E/O ratio of 1.0 represents perfect calibration; values below 1.0 indicate overestimation of risk.
Standard recalibration adjusts baseline risk and overall scale but does not update individual predictor coefficients.
The poor calibration after standard recalibration was primarily attributed to the marked difference in the age hazard ratio between the two populations.
Results
Replacing the New Zealand age coefficients with Chinese age coefficients substantially improved calibration of the New Zealand equations in the Chinese cohort.
After replacing only the age coefficients, the E/O ratio improved to 1.027 in men and 1.025 in women.
This approach represents updating a single key predictor coefficient (age) rather than full re-derivation or simple recalibration.
The improvement from updating age coefficients was described as 'noticeable' compared to standard recalibration.
This finding suggests that the age-CVD risk relationship differs systematically between high income and non-high income country populations with diabetes.
Discussion
The study identified that the stronger age effect on CVD risk in the Chinese (non-high income) cohort is the primary driver of poor performance when high income country equations are applied directly.
The age hazard ratio in China was approximately 56% larger per 10 years than in New Zealand for women (2.51 vs 1.61).
This differential age effect represents a systematic error source when transferring equations across income-level country settings.
The authors propose that this may reflect differences in cardiovascular risk factor burden by age between high income and non-high income country populations.
The study highlights 'potential limitations of high income country equation updating practices' when applied to non-high income country populations.
Conclusions
The study proposes that moving beyond simple recalibration to updating key predictor coefficients, particularly age, is a feasible approach to improve CVD risk equation accuracy in non-high income country populations.
Simple recalibration (adjusting baseline hazard and mean risk) was insufficient to correct systematic errors introduced by differential age effects.
Selectively updating statistically significantly different predictor hazard ratios, particularly age, achieved near-perfect calibration (E/O ~1.03) in the Chinese cohort.
The authors characterize this as a feasible intermediate strategy between full re-derivation and simple recalibration.
The approach requires identification of which predictor coefficients differ statistically significantly between populations.
What This Means
This research studied whether cardiovascular disease (CVD) risk prediction tools developed in wealthy countries work accurately when applied to populations in lower-income countries. The researchers compared two large groups of people with diabetes — nearly 48,000 from New Zealand and nearly 47,000 from China — and developed separate risk prediction equations for each country using identical methods. They then tested what happened when the New Zealand equation was applied to the Chinese population, both with and without standard adjustments.
The study found that most factors in the risk equations (such as blood pressure, cholesterol, and smoking) had similar predictive effects in both countries. However, age had a much stronger effect on CVD risk in China than in New Zealand — for women, the risk increase per decade of age was about 56% larger in China. When the New Zealand equation was applied to the Chinese group using standard recalibration (a common adjustment technique), it still substantially overestimated risk, particularly in women. However, simply swapping in the Chinese age coefficients while keeping all other parts of the New Zealand equation nearly perfectly corrected the predictions.
This research suggests that borrowing CVD risk equations from high-income countries and applying them in lower-income countries can introduce meaningful errors, largely because the relationship between age and cardiovascular risk differs between these settings — possibly due to differences in how risk factors accumulate with age. The practical implication is that rather than full redevelopment of risk tools (which requires large datasets), or simple recalibration (which is insufficient), selectively updating the age component of existing equations may be a practical and effective way to improve CVD risk prediction accuracy in non-high-income country populations.
Liang J, Choi Y, Shen P, Wells S, Poppe K, Fu Z, et al.. (2026). Simultaneous derivation, validation, and comparison of predictor hazard ratios for cardiovascular risk prediction equations in patients with diabetes from high versus non-high income countries: cohort study.. BMJ (Clinical research ed.). https://doi.org/10.1136/bmj-2026-100535