Study population
We identified 211,809 participates (54.82% men and 45.18% women) who met our inclusion criteria (Fig. 1) of whom 29,377 (13.87%) with hypertension and 182,432 (86.13%) without hypertension. The mean age of the population was 42.10±12.65 years. A total of 4173 participants developed diabetes during the median follow-up of 3.12±0.94 years. The number of participants with missing value of TC, TG, HDL-C and LDL-C were 4854, 4887, 94556 and 93415, respectively. Besides, the missing value of Scr, ALT and AST were 11173, 1782 and 123279, respectively. In addition, the missing value of smoking and drinking status were 151,583 and 151583. Before propensity-score matching, there were differences between the two groups in several of the baseline variables (Table 1). We found that participants with hypertension generally had higher age, BMI, FPG, Scr, ALT, TC and TG. Participants with hypertension also had a higher percentage of male and higher rates of current smoker and drinker. With the use of one to one propensity-score matching, 28,946 hypertensive patients matched with 28,946 non-hypertensive subjects. After matching, the standardized differences were less than 10.0% for almost all variables, indicating that the propensity scores were well matched. Namely, there were only small differences in baseline characteristics between the two groups.
Incidence rate of diabetes
Table 2 showed the incidence of diabetes by hypertension exposure before and after propensity-score matching. Before propensity-score matching, a total of 4173 participants developed incident diabetes during follow-up. The morbidity rate in the overall population was 630.947 per 100,000 person-years, specifically, 1693.144 per 100,000 person-years in the hypertensive group and 460.303 per 100,000 person-years in the non- hypertensive group, respectively. The corresponding cumulative incidence of diabetes in the hypertension and non-hypertension group were 5.276(5.021-5.532) and 1.438(1.383-1.492), respectively. This crude difference in the morbidity rate between the two groups changed significantly after the PS-matching procedure (1521.335 per 100,000 person-years among the overall population, 1627.690 per 100,000 person-years among the hypertensive subjects and 1414.422 per 100,000 person-years among the non-hypertensive subjects). The corresponding cumulative incidence in the hypertension and non-hypertension group were 5.072(4.819-5.324) and 4.384(4.148-4.620), respectively. Besides, we assigned participants into subgroup based on propensity score tertile. In comparison with those in a low propensity score level, participants with an increased propensity score level had a higher cumulative incidence in the original cohort (p for trend<0.00001 ) . The correlation still exists in the propensity-score matching cohort (p for trend<0.00001 ).
Kaplan–Meier analysis demonstrated that participants with hypertension had a higher incidence of diabetes than those without hypertension in the original cohort. (log-rank test; P < 0.0001; Fig. 2). After propensity-score matching, the difference of morbidity rate between the two groups reduced significantly. In addition, there was a significant higher incidence of diabetes in the population with high propensity score, especially in the propensity-score matching cohort(Fig. 3).
Association between hypertension and incident diabetes
We used cox proportional hazard regression model to evaluate the associations between hypertension and incident diabetes in the propensity-score–matched cohort. We simultaneously showed the results from unadjusted, minimally adjusted analysis, fully adjusted analysis and propensity-score adjusted analysis. (Table 3) In crude model, hypertension had a significant correlation with incident diabetes (HR=1.140, 95% confidence interval (CI): 1.058-1.229, P=0.00063). That is, the risk of developing diabetes increased by 14.0% among hypertensive participants than those without hypertension. In minimally adjusted model (adjusted age, gender, BMI, family history of diabetes, smoking and drinking status), the result did not have obvious changes (HR: 1.147, 95%CI: 1.064-1.237, P=0.00034). After adjusting for the full covariates (age, gender, BMI, FPG, TC, TG, HDL-C, LDL-C, ALT, AST, Scr, family history of diabetes, smoking and drinking status), we could also detect the connection (HR=1.131, 95%CI: 1.049-1.220,P= 0.00143). In other words, compared to the non-hypertensive group in the full model, the risk of incident diabetes increased by 13.1% in the hypertensive group. In the propensity-score adjusted model, this correlation still exists, and the risk of developing diabetes increased by 15.4% in the population with hypertension (HR=1.154, 95%CI:1.070-1.244, P=0.00019).
Subgroup analysis
We used a subgroup analysis to detect the effect of potential confounders which may affect the relationship between hypertension and incident diabetes. We treated age, BMI, FPG, Scr, ALT, TC, TG as the stratification variables to evaluate the trend of effect sizes in these variables. Table 4 showed that none of the interactions were observed based on our priori specification. The analysis revealed that the variables listed above will not affect the association between hypertension and incident diabetes after propensity-score matching. However, we detected the interaction based on propensity score tertile (Fig.4). Specifically, with reference to the non-hypertensive population with the low propensity score level, the hazard ratios of low, medium and high propensity score level in the hypertensive population were 1.231 (0.870, 1.742), 3.399 (2.529, 4.568) and 3.610 (2.604, 5.005), respectively. Thus, there was a stronger association between hypertension and incident diabetes in the population with high propensity score level.
Sensitivity analysis
We used inverse probability of treatment weights(IPTW)to generate a weighted cohort. To ensure the robustness of the results, we performed the cox proportional hazard regression model to assess the relationship between hypertension and incident diabetes in the original cohort and the weighted cohort, respectively. Table 5 simultaneously showed the unadjusted, minimally and fully adjusted models in these two cohorts. We found that hypertension was associated with the likelihood of developing diabetes in both the original cohort and the weighted cohort. Compared with the non-hypertensive group in the full model, the risk of diabetes in the hypertensive group increased by 11.7% in the original cohort (HR=1.117,95%CI: 1.044-1.196,P= 0.00134)and 19.9% in the weighted cohort(HR=1.199,95%CI: 1.149-1.250,P <0.00001), respectively.