PREDICT vs Framingham: why New Zealand's CVD risk tool catches Pacific patients Framingham misses

Framingham is the most widely-used CVD risk equation in the world. It is also wrong, by a substantial margin, for entire populations the Framingham cohort didn't represent. New Zealand's PREDICT cohort study — 401,752 patients, eight years of follow-up — quantified the gap and gave the country a domestically-derived alternative that has been the official primary-care tool since 2018.

What Framingham gets wrong

Framingham's original cohort was 5,209 residents of one Massachusetts town, almost entirely white, recruited in 1948. The equation works well for the population it was derived from. For Māori, Pacific, Indian South Asian, and Chinese patients in New Zealand, it consistently underestimates 5-year CVD risk by 30–60% — meaning patients at high risk are classified as low risk, missing the threshold for statin or BP intervention.

The clinical consequence is not hypothetical. Māori cardiovascular mortality is 1.6× the rate of non-Māori non-Pacific peers. Pacific cardiovascular mortality is similar. Some of that gap is structural; some of it is access; some of it is the equation telling clinicians to do nothing in patients who should have been on a statin three years ago.

What PREDICT does differently

• Derived from a 401,752-patient New Zealand cohort with full ethnicity stratification.
• Eight predictors plus ethnicity, deprivation index (NZDep), and family history.
• Calibrated separately for Māori, Pacific, Indian, Chinese, and European populations.
• Recommends earlier risk assessment age: 30 for Māori/Pacific/Indian men, 40 for European men; 40 for Māori/Pacific/Indian women, 55 for European women.
• Outputs both 5-year total CVD risk and absolute benefit from treatment.

What this looks like in a consult

A 42-year-old Māori male non-smoker with BP 138/86, total cholesterol 5.4, HDL 1.1, no diabetes, no family history. Framingham 5-year CVD risk: 3.8% — well below the 5% threshold for statin discussion. PREDICT 5-year CVD risk: 6.9% — above the 5% threshold. Different conversation, different prescription, different outcome.

“It changed how I trained my registrars. Don't use Framingham. Ever. We have a better tool — use it.”

GP, Tauranga

What other countries can learn

The US, UK, and Australia all face the same problem with different populations. US Pooled Cohort Equations underestimate risk for African American and South Asian Americans. QRISK3 in the UK partially corrects for South Asian ancestry but doesn't have full ethnicity stratification. Australia's AusCVD risk equation is newer and improving. The general principle holds: a risk calculator derived from one population is wrong, by clinically-significant margins, for populations not in its derivation cohort.

What good AI tooling should do

MedMETs's CVD-RA workflow defaults to PREDICT for New Zealand patients, AusCVD for Australian patients, ASCVD Risk Estimator Plus for US patients, and QRISK3 for UK patients. The ethnicity field is mandatory, not optional. The output is the absolute risk plus the absolute benefit from treatment. No clinician should be calculating CVD risk by hand on a population they were never trained for.