Statistical analysis

Baseline characteristics of the study population are described as the mean (standard deviation; SD) values for continuous variables, and as frequencies (%) for categorical variables. Comparison of the baseline characteristics between male and female participants, and also respondents (entered in the study) and non‐respondent (including those with missing data at the baseline or with no follow‐up data) individuals was carried out using the Student’s t‐test for normally distributed continuous variables, the χ²‐test for categorical variables, and the Mann–Whitney U‐statistic for the skewed and ordered variables. The cumulative incidence rate of retinopathy was calculated by dividing the number of event cases by the total number of participants. The crude incidence rate (95% confidence interval [CI]) of retinopathy was calculated by dividing the number of new cases of retinopathy by person‐years at risk for each sex and the whole population.

Cox proportional hazards models were applied to evaluate the association of the potential risk factors with incident retinopathy. Univariable analysis was carried out for each potential retinopathy risk factor including sex (reference: women), age (reference: 20–39 years), BMI (reference: normal), smoking status (reference: never), education levels (reference: <6 years), low physical activity, BP level (reference: normal), low HDL‐C, high TC, high TG and FPG level categories (reference: <7.22 mmol/L), as well as for central obesity, prevalent cardiovascular disease, CKD and aspirin medication. Those covariates with a P‐value <0.2 were entered in the multivariable analysis.

The hazard ratios (HRs) and 95% CIs were reported for adjusted risk factors. The proportionality in the Cox model was evaluated with the Schoenfeld residual test and, generally, all proportionality assumptions were appropriate. The event date was defined as the date of the incident PDR. Those who met the following criteria were considered to be censored: leaving the residential area, loss to follow up or end of follow up. For individuals with incident PDR, survival time was considered as the time between the entered date and the severe‐NPDR/PDR date. Additionally, for the censored participants, the survival time was considered as the difference between the entered date and the last available follow‐up date.

As a sensitivity analysis, a multivariable analysis was carried out among those who had not been on diabetes medications (newly diagnosed patients with type 2 diabetes mellitus) at enrollment.

As another sensitivity analysis, to minimize selection bias, due to missing data at the baseline, the propensity score was calculated and adjusted in the analysis. The propensity score calculated the estimated probability of non‐responders based on individual characteristics at baseline. This measure was computed using maximum likelihood logistic regression analysis ²⁵ .

All tests were carried out using Stata version 14 SE (StataCorp LP, College Station, TX, USA), which was considered to be significant with a two‐tailed P‐value of <0.05.