All the participants were recruited from Health Management Center from January 1, 2013 to October 31, 2018. A total number of 55,155 adults with measurements of FBG, HbA1c and hs-CRP was eligible for the study. The concentration of hs-CRP was measured at 2013 (baseline), 2014, and 2015 and the trajectory was identified based on the three measurements. HbA1c and FBG were repeatedly annually measured during five years of follow up. We excluded participants with history of diabetes, high HbA1c (³ 6.5%), high fasting blood glucose (³7.0 mmol/L) or extremely high concentration of hs-CRP (³10 mg/L) during 2013-2015, and those lost to follow up. Because hs-CRP status is strongly influenced by presence of cardiovascular disease, cancer and major metabolic disorders (hypertension, dyslipidemia and hyperuricemia), we further excluded participants with these conditions. Finally, included were 6,349 adults (4,111 men and 2,238 women; 18-89 years old) in the analysis (Supplemental Figure 1). Participants included in the study were younger, with higher proportion of women, and lower concentration of hs-CRP, FBG, and HbA1c at baseline, compared with those were not included (Supplemental Table 1). The study protocol was approved by the Ethical Committee of Ren Ji Hospital, School of Medicine, Shanghai Jiao Tong University. As a de-identified secondary data analysis, patients’ consent was waived by the Ethical Committee.
2.2 Assessment of incident diabetes
Venous blood samples were drawn and transfused into vacuum tubes containing EDTA in the morning after participants were fasted for six hours. The whole blood was stored at 4℃ for further analysis. The level of HbA1c was measured by high performance liquid chromatography, using the fully automated VARIANT™ II Hemoglobin Testing System (Bio-Rad, U.S). The measurement range was between 2.0% and 18.0%. An automatic analyzer (Roche 701 Bioanalyzer, Roche, UK) was used to measure FBG with the hexokinase/glucose-6-phosphate dehydrogenase method. The coefficient of variation using blind quality control specimens was 2.0%. Diabetes was confirmed if FBG (³7.0 mmol/L=126mg/dl) or HbA1c (³6.5%) (17, 18).
Measurement of hs-CRP and other biochemical parameters
The concentration of hs-CRP was measured by immunotubidimetric method (CardioPhase hsCRP kit, Siemens Healthcare Diagnostics Products GmbH, German). The lower limit of detection was 0.01 mg/L. The intraassay CV was 7.6% and the interassay CV was 4.0%. Hs-CRP concentration during 2013 -2015 was classified into 3 categories: low (<1.0 mg/L), moderate (1.0–3.0 mg/L), and high (³3.0 mg/L) based on a statement by American Heart Association (19).
The following variables were measured at baseline. Total cholesterol, triglycerides, high-density-lipoprotein cholesterol, low-density- lipoprotein cholesterol, and blood creatinine were measured by enzyme linked immunosorbent assay (Roche 701 Bioanalyzer, Roche, UK). White blood cell (WBC) were also measured. All the measurements were completed in the Clinical Laboratory of Ren Ji Hospital. The eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration 2-level race equation (20).
Assessment of other confounders
Body weight and height were annually measured, and BMI was calculated by body weight (kg) divided by height square (m2). The latest BMI was calculated as the latest BMI measured before the onset of the diabetes (e.g., if incident diabetes was determined in 2016, then BMI measured in 2015 was used). Blood pressure and history of the diseases were obtained at baseline. 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 min. The average of two measurements was recorded for further analysis. The history of hypertension, diabetes/impaired fasting glucose, dyslipidemia, hyperuricemia, stroke and hemorrhage, and coronary heart diseases (coronary atherosclerosis, coronary artery bypass grafting, stent surgery, and ischemic infarction) collected via a self-report questionnaire if the participants were diagnosed with these diseases by a physician or they were taking any drugs for these diseases (21).
Data were presented as mean±standard deviation. Since hs-CRP was in abnormal distribution, it was square transformed. We completed all statistical analyses by SAS version 9.4 (SAS Institute, Inc, Cary, NC). The correlation between FBG, HbA1c, BMI, blood pressure, lipid profile, and eGFR was evaluated by Spearman correlation analysis. Formal hypothesis testing will be two-sided with a significant level of 0.05. The person-time of follow-up for each participant was determined from January 1, 2016 to either the onset date of diabetes, or the end of follow-up (December 31, 2018), whichever came first. The time unit for the analysis was a month.
The hs-CRP trajectory was identified by PROC TRAJ procedure based on three measurement obtained in 2013, 2014, and 2015 (12, 22). A basic one group model was fitted with all groups set to a quadratic equation. Then, we fitted a two-group, three-group, three-group, four-group, and five-group model as well. The model with four groups was identified as the best, as suggested by the lowest value of Bayesian information criterion. We then compared the model with different functional forms. In the final model, we had one pattern with a quadratic order term and three patterns with a cubic order term. Further, we calculated the posterior predicted probability for each participant of being a member of each trajectory (15). The average posterior probability for each trajectory was 0.84, 0.87, 0.85, and 0.83.
In secondary analyses, we used baseline hs-CRP, hs-CRP in 2015, and the average of three measurements of CRP as exposures. Participants were further classified into following groups based on either single assessment of hs-CRP in 2013, in 2015, or the average of hs-CRP during 2013-2015: low‐risk (<1.0 mg/L), intermediate‐risk (1.0–3.0 mg/L), and high‐risk (³3.0 mg/L) (19).
We used the proportional hazards Cox model to evaluate the association between the hs-CRP trajectory and incident diabetes. We adjusted for potential confounders in two different models: model 1, adjusting for age (y) and sex; and model 2 further adjusting baseline BMI (kg/m2), systolic blood pressure (mmHg), diastolic blood pressure (mmHg), total cholesterol (mmol/L), triglycerides (mmol/L), low-density-lipoprotein cholesterol (mmol/L), high-density-lipoprotein cholesterol (mmol/L), eGFR (ml/min/1.73m2), FBG (mmol/L), and HbA1c (%) because these variables were closely associated with both FBG and HbA1c (Supplemental Table 2). In a secondary analysis, we further adjusted for hs-CRP (2013) to understand whether the potential association between hs-CRP trajectory and diabetes risk was driven by the baseline CRP status, although we were aware of the risk of over-adjustment and collinearity.
We tested the interaction between hs-CRP trajectory and sex, age (<65y vs. ³65y) (12), and BMI (<28.0 vs. ³28.0 kg/m2) (23), in relation to incident diabetes.