External Validation And Expanding Application Of SOAR And GWTG-stroke Mortality Predictive Model

ObjectiveTo external validate the Get With the Guidelines–Stroke(GWTG-Stroke)predictive model and the Stroke subtype, Oxfordshire Community Stroke ProjectClassification, Age, and prestroke Rankin stroke(SOAR) score in China NationalStroke Register (CNSR). Meanwhile to explore their ability to predict long-termmortality.Materials and MethodsAccording to the original article’s admittance and exclusion criteria, we selectedthe subdata base from CNSR. The primary endpoint is in-hospital death, and thesecondary endpoint was30-day,3-month,6-month and1-year death. The diffenencebetween the CNSR and the original data was analysed by using single factor analysis.For continuous variables, we used t-test; and for classified variable, we used χ2test.The P<0.01indicated the difference with statistic significance. Logistic regressionwas used to determine the discrimination ability and calibration ability. The areaunder the receiver operating curve(C value) and95%confidence interval(CI) wasused to evaluate the discrimination ability, and the C value was more closer to1, thediscrimination ability was higher. Calibration was assessed by comparing predictedand observed probability using Pearson correlation coefficient. The Pearsonstatistic>0.9indicates good fitness. Result1. For GWTG-Stroke modelsThe patients in CNSR were quite different from those in GWTG in demographiccharacteristics and risk factors. The Chinese patients were younger and the proportionof male was higher with significant difference (P<0.0001). Chinese patients weremore likely to be transferred to hospital by private transport. Compared with theGWTG data, our patients had a significantly lower prevalence of atrial fibrillation,prosthetic heart valve, previous stroke/TIA, coronary artery disease, diabetes mellitus,peripheral vascular disease, hypertension and dyslipidemia.The in-hospital mortalityof CNSR was6.3%, which was significant lower than the in-hospital mortality ofGWTG. The C value(95%CI) of the modle without NIHSS to predict in-hospitalmortality was0.76(0.75-0.78), and of the model with NIHSS was0.86(0.84-0.88).The C value(95%CI) of the modle with NIHSS to predict30-day,3-month,6-monthand1-year mortality was0.86（0.84-0.88），0.84（0.83-0.86），0.83（0.81-0.84），0.82（0.80-0.83） respectively. The C value(95%CI) of the modle with NIHSS topredict long-term mortality was between0.71-0.76. There was a relatively lowcorrelation between observed and expected probability of in-hospital death for thetwo models (Pearson correlation coefficient was0.213and0.689, respectively).2. For SOAR modelThe Chinese patients were younger and the proportion of male was higher withsignificant difference (P<0.0001). Compared with the SOAR data, our patients had asignificantly different distribution of pre-stroke mRS and OCSP classification. Thein-hosptial mortality of CNSR was4.7%, much lower then the SOAR’s, whichreached to19.8%. The C value of SOAR model to predict in-hospital,30-day,3-month,6-month and1-year mortality was0.73（0.71-0.75），0.71（0.69-0.73），0.71（0.69-0.72），0.71（0.69-0.72），0.71（0.69-0.72） respectively. There was a high correlation between observed and expected probability (Pearson correlationcoefficient>0.9).Conclusion1. GWTG-Stroke models can predict in-hospital death in Chinese population.2. GWTG-Stroke models can predict30-day,3-month,6-month and1-year death inChinese population, but the modle with NIHSS score had better predictive ability.3. SOAR model can predict in-hospital,30-day,3-month,6-month and1-year deathin Chinese population.