Predicting heart failure in cancer survivors: use of polygenic risk vs clinical score

Abstract Background The risk of heart failure (HF) is increased among cancer survivors, but predicting individual HF risk is difficult. Polygenic risk scores (PRS) for HF prediction summarize the combined effect of multiple genetic variants, specific to the individual. We sought to compare clinical HF prediction models with PRS in both cancer- and non-cancer populations. Methods Cancer and HF diagnoses were identified in the UK-Biobank from International Classification of Diseases (ICD)-10 codes. HF risk was calculated using the Atherosclerosis Risk in Communities score (ARIC-HF). PRS-HF was calculated using the Global Biobank Meta-analysis Initiative. The predictive performance of ARIC-HF and PRS-HF were compared using the area under the curve (AUC) in both cancer- and non-cancer populations. Results After excluding 2,644 participants with HF prior to consent, we identified 440,813 non-cancer participants (mean age 57 years, 53% female) and 43,720 cancer survivors (mean age 60 years, 65% female). Both ARIC-HF and PRS-HF were significant predictors for HF occurrence after adjustment for chronic kidney disease, overall health rating and total cholesterol level. The PRS-HF performed poorly in predicting HF among cancer (AUC: 0.552) and non-cancer populations (AUC: 0.571). However, the ARIC-HF predicted HF incidence in the non-cancer population (AUC: 0.806) and a reasonable performance among the cancer population (AUC: 0.748). Conclusions Prediction of HF occurrence in cancer survivors based on conventional risk factors is superior to PRS. However, both models demonstrate suboptimal discriminative capabilities in forecasting HF, emphasizing the imperative for the development of a cancer-specific HF predictive model.