Improving the reliability of polygenic risk score-based prediction for cardiovascular and renal complications across ancestries in type 2 diabetes using Mondrian Cross-Conformal Prediction

Uncertainty quantification via Mondrian Cross-Conformal Prediction applied to logistic regression on multiple polygenic risk scores (multiPRS).

Polygenic risk scores developed in European populations lose predictive performance in non-European populations, limiting clinical utility and raising equity concerns. The authors assessed whether Mondrian Cross-Conformal Prediction (MCCP), an uncertainty quantification framework, improves prediction of nephropathy, stroke and myocardial infarction in individuals with type 2 diabetes. Two training frameworks were compared: 4,098 European-ancestry participants from the ADVANCE trial for training with testing in 17,574 White British, 1,145 South Asian and 749 African UK Biobank participants; then training on the 17,574 White British participants with testing in the South Asian and African groups. Logistic regression already gave robust performance across populations and MCCP did not improve it. What MCCP added was absent from probability-based stratification: an explicit confidence and credibility level per individual, a tolerated error rate set in advance and respected in most settings, and flagging of individuals for whom no reliable prediction could be made.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

The most honest result here is a negative one: MCCP does not improve performance, it changes what the model is allowed to say. That is precisely what polygenic scores lack in clinic — the ability to answer "I cannot conclude for this person" instead of returning a reassuring but non-transferable probability. Still, the African (749) and South Asian (1,145) test groups are too small to settle the question, and no benefit on management is shown: this improves the metrology of uncertainty, not yet the medical decision.

Analysis by Dr Thibaut Benquey