Development and external validation of the SMA²SH²ERS risk prediction model for aneurysmal subarachnoid haemorrhage in the general population: a population-based prospective cohort study.

Abstract

Aneurysmal subarachnoid haemorrhage (ASAH) is a severe stroke type, preventable by screening for intracranial aneurysms followed by treatment in high-risk individuals. We aimed to develop and validate a risk prediction model for ASAH in the general population to identify high-risk individuals. We used the population-based prospective cohort studies of the United Kingdom (UK) Biobank for model development and the Trøndelag Health (HUNT) Study for model validation. Participants missing data were excluded. A total of 456 856 individuals from the UK Biobank and 46 483 individuals from the HUNT Study were included. Incident ASAH identified using the International Classification of Diseases codes, ICD-9 430 and ICD-10 I600 to I609 codes. In the development cohort, ASAH occurred in 738 (0.2%) during 5 407 909 person-years of follow-up. We developed a multivariable Cox regression model to identify predictors for ASAH. Predictive performance was assessed using discrimination and calibration, and we corrected for overfitting using bootstrapping techniques. Predictors for ASAH were sex (S), diabetes mellitus (M), age and alcohol consumption (A²), smoking (S), hypertension and hypercholesterolaemia (H²), educational attainment (E), regular physical activity (R) and family history of stroke (S; SMA²SH²ERS), and multiple interactions between these predictors. The concordance statistic (c-statistic) of the model in the development cohort was 0.62 (95% CI 0.60 to 0.64). Predicted absolute 10-year ASAH risk varied from 0.042% to 0.52%. In the validation cohort, 220 individuals developed ASAH, and the c-statistic of this model was 0.64 (95% CI 0.58 to 0.69). Both models showed reasonable calibration. Our SMA²SH²ERS model provides ASAH risk estimates between 0.042% and 0.52% for the general population. While overall ASAH risk is low, the model identifies individuals with up to 12 times increased risk compared with those at the lowest risk.