Fasting plasma glucose variability and all-cause mortality among type 2 diabetes patients: a dynamic cohort study in Shanghai, China

The study aims to examine whether the variation of fasting plasma glucose (FPG), represented by coefficient of variation (CV), independently predicts all-cause mortality among Chinese type 2 diabetes patients. This retrospective cohort study was designed based on a standardized electronic management system of diabetes patients in Shanghai, China. 8871 type 2 diabetes patients were enrolled between 1 January 2007 and 31 December 2007 and were followed-up for all-cause mortality until 31 December 2014. All patients were grouped by the quartiles of CV of FPG. 1136 patients deceased during following-up. After adjusting for other risk factors, CV of FPG was not independently associated with all-cause mortality. Stratified analysis by mean FPG levels (<7 mmol/L and ≥7 mmol/L) observed a significant modifying effect of CV of FPG (P for interact test <0.01). CV of FPG was independently associated with all-cause mortality in patients whose glucose control was poor, with the HRs (95% CI) for the second, third, fourth vs first quartiles of CV of FPG being 1.23(0.94–1.61), 1.23(0.94–1.61), and 1.63(1.25–2.13), respectively. Our results suggest that variability of FPG may be an important predictor of mortality among type 2 diabetes in China, particularly for those with their glycemic status uncontrolled.

Scientific RepoRts | 6:39633 | DOI: 10.1038/srep39633 (n = 226). Table 1 shows the comparisons of baseline socio-demographic and clinical factors of survivors and the deceased after an average of 6.43 years of following-up. Compared with the survival patients, the deceased patients were more likely to be male and older, had a longer duration of diabetes, lower mean of BMI, higher mean of SBP, higher mean FPG and CV of FPG, and more frequently used insulin. Table 2 presents baseline socio-demographic and clinical factors in subgroups of patients by quartiles of CV of FPG. Along with the increasing quartiles of CV of FPG, decreasing average ages and significantly increasing baseline FPG levels and duration of diabetes were observed in our participants.
As shown in Fig. 1, patients in the top quartile of CV of FPG experienced higher mortality than patients of the other quartiles(P < 0.001) during the 8-year following-up period. No significant difference was observed in survival among other quartiles groups during the following-up. Table 3 shows HRs for death from all causes in subjects grouped by quartiles of CV of FPG. Compared to patients with the lowest quartile, age-gender adjusted HRs (95%CI) in the second, third and highest CV of FPG quartiles were 0.89 (0.75-1.05), 0.95 (0.81-1.13) and 1.24 (1.06-1.45), respectively. When further adjusting for duration of diabetes, methods of DM treatment, baseline smoking status, physical activity, SBP, DBP, family history and BMI categories (model 2), the effect of CV of FPG attenuated, but still remained statistically significant (HR 1.18, 95%CI 1.01-1.39). Further adjusting for baseline FPG showed that it was baseline FPG, but not CV of FPG, was an independent predictor of mortality. Other predictors of mortality included male sex (all three models), age (all three models), BMI < 18.5 (model 2 and model 3), BMI ≥ 30.0 (model 2 and model 3) and insulin treatment (model 2 and model 3). Sensitivity analysis by excluding participants who died during the first 2 years of follow-up did not find that the results changed substantially. We also excluded patients who died due to external (injury-related) causes (n = 42). There was little difference seen for the effect of fasting plasma glucose variability on all-cause mortality. Then we investigated associations between fasting plasma glucose variability and cause-specific mortality (Supplementary Table 1).
We found significant interaction effects of FPG-CV and mean FPG (P for interact test < 0.01). In order to rule out the effect of glucose status of control on variability of FPG, we performed cox regression models stratified by mean FPG levels of subjects (< 7 mmol/L vs. ≥ 7 mmol/L). As shown in Table 4, in the group with mean FPG < 7 mmol/L, CV of FPG was not associated with the risk of death, while in the group with FPG ≥ 7 mmol/L, CV of FPG was an independent predictor of mortality in all three models, with HRs (95%CI) for all-cause mortality in model 3 being 1.23(0.94-1.61), 1.23(0.94-1.61) and 1.63(1.25-2.13), respectively, across increasing quartile groups comparing with the lowest quartile group.

Discussion
In this first large-scale study to investigate the association of glycemic variability with all-cause mortality among type 2 diabetes patients in Mainland China, we found that variability of FPG was independently associated with all-cause mortality in type 2 diabetes patients whose glucose level was poorly controlled, but not a predictor of all-cause mortality in patients with their glucose well controlled. In previous studies 10,13,14 , the CV of FPG was observed a significant predictor for all-cause mortality in type 2 diabetes patients. The Taichung Diabetes Study 10 measured glycemic variability by computing annual CV of all FPG measurements within each year, and showed that annual CV of FPG was independently associated with all-cause mortality in patients with type 2 diabetes aged 30 years and over. The Verona Diabetes Study 13 found that FPG variability, as assessed by CV of FPG over a period of three years, was an independent predictor of all-cause in patients with type 2 diabetes aged 56-74 years. In our analysis, we examined the relationship between variability of FPG and all-cause mortality in subgroups stratified by glucose status of control. The present results showed that variability of FPG was only independently associated with a higher risk of subsequent all-cause death in those with their glycemic status uncontrolled. A large national survey in China showed that the mean BMI was 25.2 for men with previously diagnosed diabetes and the mean BMI was 24.6 for women with previously diagnosed diabetes 15 . The mean BMI was 23.9 for type 2 diabetes patients in our study, which was lower than that of previously diagnosed diabetes patients in the national survey 15 . However, the duration of diabetes was not indicated in the national survey, and the mean duration of diabetes in our study was 13.0 years.
Several potential mechanisms may explainthe association between glycemic variability and all-cause mortality. First, glycemic variability might be an indicator of irregular compliance to therapy due to a variety of reasons (poor healtheducation, insufficient awareness of the severity of the disease) 13 . Our study had considered the effect of adherence to therapy or diet guideline on all-cause mortality. The proportion of adherence to therapy or diet guideline for different groups of CV of FPG was similar. And adherence to therapy or diet guideline wasn't in the final regression model. Second, glycemic variability might be an indication of poor health, comorbidity, or complication that results in the increase of mortality. A previous study 10 had considered the associations of baseline comorbidity or complication with mortality and they can explain only a small amount of these associations. Third, glucose fluctuation has been shown to cause over production of superoxide that is a key risk factor in the pathogenesis of diabetes complications 16,17 . The increases in diabetes complications further result in the increase of mortality.
Our study has some strengths, including retrospective cohort study design, a large number of diabetes patients, long-term of following-up and standardized procedure for data collection according to the Diabetes Prevention Guide. In addition, to our best knowledge, it is the first study to examine the relationship between variability of FPG and all-cause mortality in subgroups stratified by glycemic status of control. However, this study has several limitations. First, unlike RCTs, FPG measurements in this study were derived from clinical follow-ups, thus the frequency of FPG measurements and the intervals between measurements varied across patients. Although we adjusted for the effect of the frequency of FPG measurements on variability, the difference of intervals between FPG measurements had not been fully addressed. Second, we didn't adjust for baseline comorbidity and presence of diabetic complications due to lacking of the information, which may have biased our results. Finally, baseline Variables (% or mean ± SD) Baseline quartiles of FPG-CV glycated hemoglobin determinations were not available for all participants and they have been observed as one of the independent risk factors of macrovascular events which may result in the increase of mortality 18 .
In conclusion, glycemic variability was a potent independent predictor of all-cause mortality in type 2 diabetes patients whose glucose was poorly controlled. Tight control of FPG variability may provide further protection against subsequent death in type 2 diabetes. Further randomized, controlled trials investigating the favorable effectsof an intervention of maintaining the stable glycemia are needed to confirm our results and elucidate the direct causality.     20 , and were followed up regularly once per month or every 3 months, depending on the glycemic status of control. At the following-up visits, the status of adherence to therapy or diet guideline had been evaluated by General Practitioners. All following-up data was recorded in electronic health records (eHR) database 21 .
Study design. We included type 2 diabetes patients enrolled in the registry between 1 January 2007 and 31 December 2007. 122 subjects with type 1 diabetes or malnutrition-related diabetes were excluded, and 10,712 (4763 men, 5949 women) type 2 diabetes patients were eligible (Fig. 2). And 1841 subjects without ≥ 3 records fasting plasma glucose (FPG) in the first one-year of following-up to analyze variability were excluded. Then 8871 (3909 men, 4962 women) subjects were finally included in the analysis. Written informed consent was obtained for all study participants when they were enrolled in the registry. The study was approved by the Institutional Review Board of Minhang Center for Disease Control and Prevention. The methods were carried out in accordance with the approved guidelines. Date of diabetes patients registered into the standardized managementwas defined as the index date. The survival status of patients up to 31 December 2014 was ascertained according to the recordsin the management system. We also identified the survival status through record-linkage of the eHR system with the Vital Statistics system in Minhang district, which is part of national Disease Surveillance Point System 22 . We used the underlying cause of death ICD-10 code on the Vital Statistics system to group causes of death. The categories were as follows: diabetes (E10-E14), cardiovascular disease (I00-99), cancer (C00-97), external (injury-related) causes (V00-Y89) and all other codes. The end points of follow-up were defined as the time when he/she was dead or censored (due to lost to follow up). 1136 deaths were identified in this cohort.
Measurements. The demographic characteristics (birth date and sex), measurements of body height, weight, BMI, systolic blood pressure (SBP), diastolic blood pressure (DBP), FPG levels, family history, diagnosis date of diabetes, methods of DM treatment (oral hypoglycemic drug, insulin injection, both, or diet or exercise) and adherence to therapy or diet guideline were obtained from the eHR database for all diabetes patients. Duration of diabetes was calculated as the calendar date when he/she was dead or censored minus the calendar date of diagnosis with type 2 diabetes. A number of studies suggested a protective effect of overweight in patients with type 2 diabetes (obesity paradox) 23,24 . Adjusting for BMI as a continuous may result in some bias, and we adjusted for BMI on mortality using BMI categories. BMI categories were defined as follows: < 18.5, 18.5 to 24.9 (reference), 25.0 to 29.9 and ≥ 30.0, respectively 15 . Statistical analysis. The variability (coefficient of variation, CV) of FPG measurements from diabetes patients visits within the first year of follow-up for each patientwas calculated. Considering that the frequency of visits may affect the evaluation of variability, the CV of FPG was adjusted by dividing by the square root of the ratio of total visits divided by total visits minus 1 25,26 . All patients were classified into four groups by quartiles of CV of FPG.