A genetic risk score improves risk stratification for anticoagulation-related intracerebral hemorrhage

Background Intracerebral hemorrhage (ICH) is the most devastating adverse outcome for patients on anticoagulants. Clinical risk scores (CRS) that quantify bleeding risk can guide decision making in situations when indication or duration for anticoagulation is uncertain. We investigated whether integration of a genetic risk score (GRS) into an existing risk factor-based CRS could improve risk stratification for anticoagulation-related ICH. Methods We constructed 153 GRS from genome-wide association data of 1,545 ICH cases and 1,481 controls and validated them in 431 ICH cases and 431 matched controls from the population-based UK Biobank. The score that explained the largest variance in ICH risk was selected and tested for prediction of incident ICH in an independent cohort of 5,530 anticoagulant users. A CRS for major anticoagulation-related hemorrhage, based on 8/9 components of the HAS-BLED score, was compared with an enhanced score incorporating an additional point for high genetic risk for ICH (CRS+G). Results Among anticoagulated individuals, 94 ICH occurred over a mean follow-up of 11.9 years. Compared to the lowest GRS tertile, being in the highest tertile was associated with a two-fold increased risk for incident ICH (HR: 2.08 [95% CI 1.22, 3.56]). While the CRS predicted incident ICH with a HR of 1.24 per one point increase (95% CI [1.01, 1.53]), adding a point for high genetic ICH risk led to a stronger association (HR of 1.33 per one point increase, 95% CI [1.11, 1.59]) with improved risk stratification (C-index 0.58 vs. 0.53) and maintained calibration (integrated calibration index 0.001 for both). The new score (CRS+G) showed 20% improvement of high-risk classification among individuals with ICH and a net reclassification improvement of 0.10. Conclusions Among anticoagulant users, a prediction score incorporating genomic information is superior to a clinical risk score alone for ICH risk stratification and could serve in clinicaldecision making.