Study population
All the participants (≥65 y) were recruited from local communities who have taken healthy check-up at Health Management Center, Ren Ji Hospital from January 1, 2014 to May 31, 2019. A total number of 9,902 Chinese aged population were eligible for the study. BMI was measured at baseline (2014). Fasting blood glucose (FBG) and glycated hemoglobinA1c (HbA1c) were measured annually during follow-up (2014-2019). We first excluded the extremist values (>99th percentile or <1st percentile) (n=1,930) at baseline. Then, we excluded participants whose FBG ≥ 7.0mmol/L, or HbA1c ≥ 6.5mmol/L, or with self-reporting DM at baseline (n=1,471). Finally, we excluded those who were lost during follow up (n=12), a total number of 6,489 Chinese aged population [3,828 men and 2,661 women, aged 69 (interquartile range: 67, 74) years] were included in the study (Supplementary Figure 1). Compared with those out of the study, the participants included in the study were with similar level of BMI and with lower level of FBG and HbA1c (Supplementary Table 1). The study protocol was approved by the Ethical Committee of Ren Ji Hospital, School of Medicine, Shanghai Jiao Tong University (Reference Number: KY-2019-112). As a re-identified study, the signed consent was waived by the Ethical Committee.
Exposures (BMI)
Body weight and height were measured in light clothes with no shoes at baseline, and BMI was calculated by body weight in kilogram divided by square of height in meter. All the participants were further classified into eight groups based on baseline BMI with 2.0 kg/m2 interval: ≤17.9 kg/m2, 18.0-19.9 kg/m2, 20.0-21.9 kg/m2, 22.0-23.9 kg/m2, 24.0-25.9 kg/m2, 26.0-27.9 kg/m2, 28.0-29.9 kg/m2, and ≥30.0 kg/m2. The BMI was also used as a continuous variable with interval of 1.0 kg/m2.
Outcomes (incident DM)
Venous blood samples were drawn and transfused into vacuum tubes containing EDTA in the morning after participants were fasted overnight for eight hours. FBG was measured by enzyme linked immunosorbent assay (Roche 701 Bioanalyzer, Roche, UK). HbA1c were measured by a high-performance liquid chromatography method (Variant II automatic glycosylated hemoglobin analyzer, Bio-Rad, America). DM was confirmed if either FBG ≥ 7.0 mmol/L or HbA1c ≥ 6.5% (15).
Assessment of other confounders
Blood pressure was measured twice using an automatic blood-pressure meter [HBP-9020, OMRON (China) Co., Ltd.] after participants were seated for at least 10 mins. The average of two measurements was recorded for further analysis. Total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, were measured as well. The estimating glomerular filtration (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration 2-level race equation(16). All the biochemical measurements were completed in the Clinical Laboratory of our hospital.
Statistical analysis
We completed all statistical analysis by SAS version 9.4 (SAS Institute, Inc, Cary, NC). Formal hypothesis testing will be Wilcoxon test for rank sum with a significant level of 0.05.
In the current study, we used the Cox proportional hazards regression model to evaluate the association between BMI and incident DM in whole group. The person-time of follow-up for each participant was determined from the baseline to (January 1, 2014) to either the onset date of DM, loss to follow up, or the end of follow-up (May 31, 2019), whichever came first.
With the analysis of dose-response trend, more specifically, the continuous variable of the change in BMI was used to fit into a restricted cubic spline model (17) and to obtain a smooth representation of the hazard ratio as a function of the change in BMI adjusted by potential confounders. We used 5 knots defined at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles to divide continuous change in BMI into 5 intervals.
We adjusted for potential confounders in different models: model 1, adjusting for age (y) and sex; model 2, adjusting for variables in model 1, and systolic blood pressure (mmHg), diastolic blood pressure (mmHg), total cholesterol (mmol/L), triglyceride (mmol/L), low density lipoprotein cholesterol (mmol/L), high density lipoprotein cholesterol (mmol/L), eGFR (mL/min per 1.73 m2); model 3, adjusting age (y), systolic blood pressure (mmHg), diastolic blood pressure (mmHg), total cholesterol (mmol/L), triglyceride (mmol/L), low density lipoprotein cholesterol (mmol/L), high density lipoprotein cholesterol (mmol/L), eGFR (mL/min per 1.73 m2); model 4, adjusting sex, systolic blood pressure (mmHg), diastolic blood pressure (mmHg), total cholesterol (mmol/L), triglyceride (mmol/L), low density lipoprotein cholesterol (mmol/L), high density lipoprotein cholesterol (mmol/L), eGFR (mL/min per 1.73 m2).
The interaction between continuous BMI and sex, age groups was tested by adding the cross-product terms in the multivariable model. To test the robustness of the results obtained from the main analysis, we conducted three sensitivity analyses: excluding participants with high blood pressure (systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg (18)) , with abnormal lipid metabolism ( total cholesterol ≥ 5.7mmol/L or triglyceride ≥ 1.7mmol/L or low density lipoprotein cholesterol ≥ 3.4mmol/L or high density lipoprotein cholesterol < 1.0 mmol/L for man or high density lipoprotein cholesterol < 0.9 mmol/L for female) (19), or with decreased eGFR ( ≤ 60 mL/min per 1.73 m2) (16).