Locally tuned combinations of genetic risk scores better distinguished coronary disease and two cancers in biobank tests. A peer-reviewed study trained and evaluated models within two datasets, addressing how inherited-risk tools fit different patient populations. It did not test whether using the revised scores to guide care improves health outcomes.
Combining inherited risk scores using local patient data improved disease ranking in a peer-reviewed analysis of two biobanks. The approach covered breast and prostate cancers alongside coronary artery disease. It addresses a practical problem: a score developed in one population can perform differently when used in another clinical setting.[1]
A polygenic score brings together estimated effects of many inherited genetic variants. Instead of discovering new variants, the researchers reweighted combinations of published scores using data from UK Biobank, a British research resource, and the US Mass General Brigham Biobank. Age and available clinical covariates were also considered.[1]
Each dataset was divided into 70% for training and 30% for held-out testing. Model selection and tuning stayed within the training portion. Evaluation on the separate test group therefore assessed predictions for participants who had not been used to fit the model, while retaining the characteristics of the same cohort.[1]
The tuned models improved discrimination and concentration of cases in high-risk groups in the tested datasets. Disease outcomes came from biobank and health data rather than a trial of care guided by the scores. Whether the same frozen model transfers to another clinic or ancestry group requires further measurement. The authors make analysis code available, while access to individual-level biobank data remains controlled.[1]
