Statistical analysis

Baseline clinical features were compared stratified by different diabetes diagnosis and functional outcome. All continuous variables were presented as mean ± standard deviation for normal distribution and median (Quartile 1 to 3) for skewed distribution. Kolmogorov-Smirnov test was used for normal distribution test. Categorical variables were presented as frequencies and/or prevalence/incidence rate. For the continuous variables, when the group numbers were more than two, if the variable was normal distribution, one-way ANOVA test was used to perform the comparison (LSD-test for variables with homogeneity of variance and Games-Howell test for variables with heterogeneity of variance), otherwise Kruskal-Wallis H test was performed instead; when the group numbers were just two, if the variable was normal distribution, T-test was used to perform the comparison, otherwise Mann–Whitney U test was used instead. For categorical variables, the comparisons were used Chi-square test and Fisher exact test. The adjustment of the correlation between different variables was done by multivariate logistic regression. Variables with a P value < 0.1 in univariate analysis were included in the multivariate regression test. Variables, which had shown potential association to the outcome of stroke during the literature search, were also included in the multivariate regression test. The independent factors included in the multivariate analysis were age, gender, education degree, tobacco use, alcohol consumption, BMI, high density lipoprotein, low density lipoprotein, creatinine, systolic and diastolic blood pressure, HOMA2-IR, homocysteine, a history of coronary heart disease, a history of hypertension, a history of atrial fibrillation, stroke types and stroke severity at admission. The use of Aspirin, Diuretics and Coumadin [28], which could influence SUA concentration, were also analyzed. SUA was compared between “use” and “non-use” of them respectively and stratified by different glycometabolism first, then it will be recognized as a confounding factor if P < 0.05 and entered into the according multivariate analysis. The fitness of the models was evaluated by using the Hosmer and Lemeshow goodness-of-fit test and a P value > 0.2 was considered as a good fit. All the statistical analyses were done with SPSS ver. 19.0 software (SPSS Inc., Chicago, IL). A P <0.05 was considered statistically significant.