Skip to main content

Table 1 A focus on difference in PGS classification accuracies between groups can mask potential utility when baseline risks differ

From: Recent advances in polygenic scores: translation, equitability, methods and FAIR tools

Many critiques of PGS focus on their modest effect size (e.g., odds or hazard ratio) and the risk stratification between the top and bottom quantiles of genetic predisposition, which is related to the proportion of variance explained (r 2) and classification accuracy (AUROC or C-index) [41, 42]. As noted, the risk stratification capacity of PGS decreases proportional to the genetic distance from the training population, leading to attenuated but non-null effect sizes in non-European ancestry groups [43] (specific analyses of PGS for coronary artery disease or CAD [44], breast [45] and prostate cancer [46]). Given that these effect sizes (and thus classification accuracies) are non-null indicates that they may still be useful for stratification. This may be particularly true when the baseline risk is higher in the groups with lower effect sizes, and it is the case that many non-European ancestry groups have a significantly higher incidence for some common diseases [47, 48]. In cases where demographic groups have different average/baseline risks, it can be difficult to infer clinical utility of a PGS when looking at PGS alone [49] (Fig. 1).

Existing European-biased PGS integrated into clinical risk tools have been shown to improve the reclassification of cases in non-European ancestries in multiple studies of cardiometabolic diseases [33, 37, 50,51,52]. In these circumstances, higher baseline risk can compensate (partially or completely) for attenuated PGS performance with respect to metrics of disease prevention in each group (e.g. number of events prevented, number needed to treat/screen to prevent one event, etc.).

While analysis of risk stratification and utility is more interpretable using absolute rather than relative risk differences, it of course does not address the underlying representational bias in the data. Global efforts to collect genomic data in more diverse cohorts should certainly continue and form the foundation for greater equity in the future.