Prediction For Personalized Stroke Risk With Traditional And Genetic Risk Factors

Background and ObjectiveStroke is the leading cause of death in China currently.Primary prevention is the fundamental approach to reduce the tremendous health burden caused by stroke.A key step for primary prevention of stroke is to identify high-risk individuals with a robust risk assessment tool,followed by targeted preventive interventions.Currently,stroke risk prediction models based on traditional risk factors such as sex,age,blood pressure and antihypertensive treatment,smoking,diabetes,et al.play a significant role in screening high-risk individuals.The occurrence of stroke involves both genetic and environmental factors.However,genetic factors have not been considered in traditional stroke risk prediction models.Genetic risk score(GRS),which does not depend on age and other risk factors and can independently predict stroke risk,is expected to compensate for the shortcomings of traditional risk prediction models.However,the current studies on risk assessment and GRS of stroke are mainly from western populations.Due to the differences in the prevalence and genetic susceptibility of stroke between different countries and ethnic groups,whether these findings are applicable to Chinese needs to be further verified.Therefore,the first aim of this study is to develop and validate personalized 10-year and lifetime stroke risk prediction models based on traditional risk factors using data from four large prospective cohorts to provide practical tools for risk assessment of stroke in China.The second aim is to construct a stroke GRS based on a prospective Chinese cohort using effect sizes from genome-wide association study(GWAS)of stroke in East Asian population,and to evaluate its application value in improving stroke risk stratification on the basis of traditional models.Subjects and MethodsStudy Subjects:Populations of 35 to 74 years of age without cardiovascular diseases at baseline from four large prospective cohorts in the Prediction for Atherosclerotic Cardiovascular Disease Risk in China(China-PAR)project were used in this study.Two cohorts established around 2000 including 21 320 participants were used for derivation of the models.The models were externally validated by the other two cohorts established during 1992 to 1994 and 2007 to 2008,respectively,each including 14 123 and 70 838 participants.The four cohorts were recently followed up during 2012 to 2015.A total of 41 714 participants from the China-PAR project with eligible genotype data were used for developing stroke GRS and estimating the improvement of GRS on traditional risk prediction models.Baseline Survey and Follow-up:A uniform survey protocol was used to collect baseline information on demographics,lifestyle risk factors,personal and family history,and physical examinations of each subject.Samples of peripheral blood were drawn for blood tests and DNA extraction and purification.Fatal and non-fatal stroke events and all-cause deaths during follow-up were collected.Deaths from causes except for stroke were considered as competing risk events.Derivation and Validation of Traditional Models:Cox proportional hazard model was used for 10-year risk modelling,and sub-distribution hazard model considering competing risk was used for lifetime(till 85 years of age)risk modelling,for men and women separately.The predictors of the models included age,systolic blood pressure,antihypertensive treatment(yes/no),current smoking(yes/no),diabetes(yes/no),total cholesterol,high-density lipoprotein cholesterol,waist circumference,family history of stroke,urbanization,and living area(north/south).The discrimination and calibration ability of the models was assessed by C-statistics and Calibration χ2,respectively.Calibration plot was used for comparison between the predicted stroke risk and the Kaplan-Meier adjusted stroke event rates.In addition,the performance of our models(China-PAR models)among participants aged 55-74 years was compared with the new Framingham Stroke Risk Profile(FSRP).Furthermore,the usefulness of lifetime risk assessment for stroke risk stratification on the basis of 10-year risk was evaluated.Derivation and Assessment of GRS:Based on comprehensive literature review,we selected 22 SNPs which were reported to be associated with stroke or stroke-related traits of genome-wide significance.Using effect sizes from GWAS of ischemic stroke in East Asian population reported by the BioBank Japan Project(BBJ)as weights,the individual stroke GRS was calculated as the weighted sum of the number of effect alleles of the 22 SNPs.Cox proportional hazard model was used to evaluate the association between GRS and stroke risk.Changes in C-statistic and net reclassification index(NRI)were used to estimate the improvement in prediction ability after adding GRS to the China-PAR models for predicting 10-year stroke risk.Furthermore,subjects were categorized into low,intermediate,or high clinical risk according to the China-PAR score for 10-year stroke risk.Cox proportional hazard model was used to evaluate the 10-year and lifetime cumulative incidence of stroke in each combination of clinical risk categories and GRS groups(lowest decile,second to ninth deciles,and highest decile),so as to demonstrate the value of GRS on stroke risk stratification on the basis of traditional risk assessment.ResultsCharacteristics of Study Subjects:The age range of 21 320 subjects in the derivation cohort was 35-74 years,with a mean of 48.6(standard deviation[SD],9.3)years,among whom 48.5%were men.During a mean follow-up of 12.3 years,776 stroke events occurred(men,472;women,304).For validation cohorts China MUCA(1992-1994)and CIMIC,the age range was 35-59 years and 35-74 years,with a mean of 46.6(SD,7.4)years and 54.4(SD,10.2)years,and the proportion of men of 46.5%and 37.9%,respectively.During a mean follow-up of 17.1 years in China MUCA(1992-1994)and 5.9 years in CIMIC,698(men,374;women,324)and 1992(men,969;women,1023)stroke events occurred,respectively.A total of 41 714 subjects with average age of 51.4(SD,11.0)years and percentage of men being 43.1%were included for developing GRS,who were not substantially different from the remaining subjects in the China-PAR project in terms of the distributions of age,sex,and major risk factors.A total of 1282 stroke events occurred(men,676;women,606)in the 41 714 subjects during follow-up.Assessment of Traditional Models:The China-PAR models for predicting 10-year stroke risk had C statistics of 0.810(95%confidence interval[CI]:0.787-0.833)for men and 0.810(95%CI:0.783-0.837)for women,with calibration χ2 of 15.0(P=0.092)and 7.8(P=0.550),respectively.The models for predicting lifetime stroke risk also showed C statistics around 0.800 and calibration χ2 below 20 for both sexes.In the validation cohorts,similar discrimination and calibration ability was also found,with good agreement between the predicted and observed stroke risk for both 10-year and lifetime risk prediction models.Among those aged 55-74 years,the new FSRP substantially underestimated the 10-year stroke risk by 40.2%in men and 53.3%in women,whereas the China-PAR models displayed much better prediction ability,with the numbers of predicted stroke events within 10 years of 273.5 in men and 199.7 in women,which were very close to the observed numbers(272.5 and 189.0,respectively).Subjects were divided into 3 clinical risk categories as low(<3.5%),intermediate(3.5%-6.9%),or high(≥7%)risk based on the China-PAR score for 10-year stroke risk,and subjects within the top decile of lifetime risk(≥25%)were considered as high lifetime risk.Among young adults aged 35-49 years,5.7%of the subjects had a low to intermediate 10-year risk of stroke(<7.0%),but a high lifetime risk(≥25.0%),suggesting lifetime risk assessment could further identify high-risk individuals who account for 5.7%of this age group.Assessment of GRS:After multivariable adjustment,for every SD increase in GRS,the risk of stroke increased by 15%(hazard ratio[HR],1.15;95%CI:1.09-1.22,P=1.4×10-6).Compared with the lowest GRS decile,the stroke risk of subjects within the top GRS decile increased by 44%(HR,1.44;95%CI:0.86-2.38,P=0.163),118%(HR,2.18;95%CI:1.31-3.60,P=0.003),and 71%(HR,1.71;95%CI:1.18-2.49,P=0.005)among subjects with low,intermediate,and high clinical risk category,respectively.After adding GRS to the China-PAR models for predicting 10-year stroke risk,the C-statistics increased slightly by 0.002(P=0.003)in overall population,whereas increased significantly by 0.046(P=0.003)among those with intermediate clinical risk.The reclassification also improved after adding GRS,with a NRI for stroke events of 4.3%(95%CI:1.4%-7.1%).In each clinical risk category,the cumulative risk of stroke increased with the GRS level.Moreover,the individuals with intermediate clinical risk and GRS in the top decile had cumulative risk of stroke comparable to those with high clinical risk and GRS in the lowest decile(7.8%[95%CI:5.6%-9.9%]vs.8.1%[95%CI:5.5%-10.6%]for 10-year risk,and 31.9%[95%CI:23.6%-39.3%]vs.30.2%.[95%Cl:21.6%-38.0%]for lifetime risk).ConclusionsOur study developed and validated models to predict personalized 10-year and lifetime stroke risk in Chinese population aged 35 to 74 years.We found lifetime risk assessment could be used for further identification of high-risk individuals on the basis of 10-year risk assessment,especially among younger adults.Besides,the GRS based on stroke GWAS in East Asian population could effectively predict stroke risk in Chinese.Although GRS only slightly improved the traditional stroke risk prediction models,it could be used as a new factor for risk stratification,enabling further identification of high-risk individuals among those with intermediate clinical risk.Our study provided practical tools for individualized stroke risk assessment,which are expected to facilitate precise prevention programs for stroke in China.