Introduction: Most stroke risk prediction equations were developed in Western populations. These risk scores perform less well in Asian populations, who have different background risk.
Hypothesis: This study aimed to develop and validate a new stroke risk algorithm for estimating 5-year risk of developing stroke in a Thai general population using large electronic health records.
Methods: This retrospective cohort was derived from the linkage of health checks data with diagnostic information from electronic health records of 483,285 men and women residing in Ubon Ratchathani. It was randomly and evenly divided into the derivation and validation cohorts. An outcome of interest was first recorded diagnosis of stroke over a period of 6 years between January 2006 and December 2012. A Cox proportional hazards model was used to estimate effects of risk factors on stroke risk and to derive a risk equation in the derivation cohort. Measures of discrimination, global model fits and calibration were calculated in the validation cohort.
Results: The derivation cohort comprised of 241,643 individuals aged ≥30 years, who contributed 1,393,990 person-years of follow-up and 1,884 incident cases of stroke, while the validation cohort comprised of 241,642 individuals (1,393,420 person-years), with 1,915 incident cases of stroke. The risk equation was 0.0584 x Age (years) + 0.4538 x Sex (Male=1) + 0.0040 x Systolic BP (mmHg) + 0.2886 x Hypertension (present=1) + 0.5214 x Diabetes (present=1) + 0.0266 x Body mass index (kg/m2) - 0.0039 (if exercise 1-2 days/week) or -0.2827 (if exercise 3-5 days/week) or -0.3179 (if exercise >5 days/week). The stroke risk equation had a reasonably good discriminatory ability in the validation cohort, with the area under the receiver operating characteristic curve of 0.756 (95%CI 0.750-0.772). The risk equation had good global model fit as measured by Bayesian information criteria. The Gronnesby and Borgan test showed good calibration, with chi-square statistic of 701.43 (p<0.001).
Conclusions: This simple stroke risk score is the first risk algorithm to estimate the 5-year risk of stroke in a Thai general population. The risk score does not need laboratory tests and can therefore be used in clinical settings and also by the public.