Fasting Plasma Glucose Variability in Midlife and the Risk of Parkinson’s Disease: A Nationwide Population-Based Study

Background — Athough an association exists between type 2 diabetes and Parkinson’s disease (PD), the implications of glycemic variability on PD are unknown. We assessed the future risk of incident PD according to visit-to-visit fasting plasma glucose (FPG) variability; this was calculated using standard deviation (FPG-SD), coefficient variance (FPG-CV), and variability independent of the mean (FPG-VIM). Methods — Using the Korean National Health Insurance Service–Health Screening Cohort, we followed 131,625 Korean adults without diabetes. This study population was divided into a midlife group (<65 years) and an elderly group (≥65 years), during a median follow-up of 8.4 years. Results — The adjusted hazard ratios (HRs) were calculated using a multivariable Cox proportional hazard analysis. In the midlife group, the HRs for incident PD in the highest quartile of FPG variability, as measured using SD, CV, and VIM, were 1.35 (95% confidence interval (CI), 1.07–1.70), 1.31 (95% CI, 1.04–1.65), and 1.33 (95% CI, 1.06–1.67), respectively, when compared to the lowest quartile group. However, the incident PD was not different depending on FPG variability in the elderly group. Kaplan–Meier curves of PD probability showed a progressively increasing risk of PD according to the higher FPG variability in the midlife group. According to a multivariable adjusted model, a 1-SD unit increment in glycemic variability was associated with a 9% higher risk for incident PD in the midlife group. Conclusions Increased long-term glycemic variability the to demonstrate the impact of long-term glycemic variability on the development of PD. We used a standardized and validated database created by the Korean government which has a huge sample size and extensive information, including medical diagnoses and medications. In this study, we included only the non-diabetic population to eliminate the possible confounding effects of anti-diabetic medication or diabetic complications, including

Therefore, we hypothesized that long-term glycemic variability and PD are intimately related due to their shared mechanistic pathophysiology. Nevertheless, to the best of our knowledge, no previous studies have showed the relationship between long-term glucose variability and incident PD. Our study was based on the longitudinal National Health Insurance Service-National Health Screening Cohort (NHIS-HEALS) database in Korea. Using this data, we examined the association between visitto-visit variability of FPG and the risk of developing PD in the general population without diabetes.

Study participants
The NHIS of South Korea is a government-managed mandatory public health insurance program that covers approximately 50 million Korean residents (almost 98% of the entire population). Under this program, it is recommended to have a general health examination at least biannually [18]. The NHIS includes sociodemographic information (e.g., age, sex, and income), a health examination database (e.g., standardized self-reported questionnaires on physical activity, smoking status, and alcohol consumption; anthropometric measurements; blood pressure; and laboratory tests), and claim databases of outpatient and inpatient records (e.g., diagnoses, pharmacy, and death).
Anthropometric measurements (e.g., height and weight) and laboratory tests were done after an overnight fast, and quality control procedures were confirmed by the Korean Association of Laboratory Quality Control. The NHIS-HEALS database was randomly selected to include almost 10% of the entire population between 40  Association. Informed consent was waived because anonymous and de-identified information was used for the analysis.

Definition of FPG variability
The FPG variability was calculated using three or more FPG values measured during the participants' health examinations that were completed between January 1, 2002 to December 31, 2007. The following quantities of FPG value measurements per participant were included: 3 measurements (n = 67120; 51.0%), 4 measurements (n = 14075; 10.7%), 5 measurements (n = 18794; 14.3%), and 6 measurements (n = 31636; 24.0%). We used three indices for prescriptive FPG variability: standard deviation (FPG-SD), coefficient of variation (FPG-CV), and variability independent of the mean (FPG-VIM). The CV was defined as SD/mean × 100% and VIM was defined as 100xSD/Meanβ, where β is the regression coefficient, based on the natural logarithm of the SD divided by the natural logarithm of the mean.

Measurements and definitions
We examined newly occurring PD as primary outcomes from January 1, 2008 to December 31, 2015.
The diagnosis of PD was defined using the International Classification of Diseases, Tenth Revision (ICD-10) code G20. During hospitalisation, patients with one or more diagnoses of G20 codes and during outpatient clinic visits, those with G20 codes having been recorded at least twice were considered to have PD [19,20]. The presence of diabetes was defined based on the criteria of a fasting glucose level ≥ 7.0 mmol/L or by having at least one prescription claim per year for antidiabetic medication under the ICD-10 codes E10-E14. The presence of hypertension was defined based on the criteria of systolic/diastolic blood pressure ≥ 140/90 mmHg or by having at least one prescription claim per year for an antihypertensive agent under ICD-10 codes I10-I15. The presence of dyslipidemia was defined based on the criteria of total cholesterol ≥ 6.2 mmol/L or by having at least one prescription claim per year for an anti-hyperlipidemic agent under ICD-10 code E78. The diagnosis of stroke was defined as ICD-10 codes I60-I64 on the admission record with computerized tomography or magnetic resonance imaging claim data. The diagnosis of chronic kidney disease (CKD) was defined as ICD-10 codes N18 or N19. Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters (kg/m 2 ). Smoking status and information concerning alcohol consumption were obtained from a questionnaire undertaken during the health examination. Regular exercise was defined as strenuous physical activity for at least 20 minutes, ≥ 5 times/week. Income levels were dichotomized at the lower 10%.

Statistical analysis
The baseline characteristics are presented as the mean ± standard deviation for continuous variables or percentages for categorical variables. The participants were divided by age 65 years at the index year (2007). Then, they were classified into quartile groups according to the FPG variability value. Any differences between the groups were identified using the analysis of variance (ANOVA) for continuous variables or the χ 2 -test for categorical variables. Kaplan-Meier curves for the probability of PD were obtained for the four groups which were classified and expressed as quartiles of FPG variability.
Hazard ratios (HRs) and 95% confidence interval (CI) values for PD incidence were analyzed using Cox proportional hazards models for quartile groups of FPG variability. This was adjusted for age, sex, BMI, hypertension, dyslipidemia, smoking status, alcohol consumption, regular exercise, income, history of stroke, history of CKD, and mean FPG. We tested the assumption of proportionality of hazards using the numerical method proposed by Lin et al., derived from the cumulative sums of martingale-based residuals [21]. We found no evidence that violates the proportional hazards assumption. We used multivariable linear regression for the risk of PD to evaluate the association of a 1-SD increment for continuous FPG-SD, FPG-CV, and FPG-VIM after adjusting compounding factors. Especially, to estimate whether the FPG variability to PD occurrence association is independent for other risk factors, we analyzed with stepwise models adjusting for: 1) age and sex; 2) age, sex, BMI, hypertension, dyslipidemia, smoking status, alcohol intake, regular exercise, and income; 3) age, sex, body mass index, hypertension, dyslipidemia, smoking status, alcohol intake, regular exercise, income, history of stroke, and history of CKD; 4) age, sex, body mass index, hypertension, dyslipidemia, smoking status, alcohol intake, regular exercise, income, history of stroke, history of CKD, and mean FPG. All of the statistical results were analyzed using SAS 9.4 (SAS Institute Inc., Cary, NC, USA). A P-value < 0.05 was assumed to indicate statistical significance. All the statistical analyses were performed by an experienced professional statistician, who was also one of authors (J.S.L.).

Results
Baseline characteristics of the study population Table 1 presented the general characteristics of the study participants according to the quartiles of SD for FPG variability in individuals < 65 years (midlife group) and those ≥ 65 years (elderly group).
The participants in the higher quartile of FPG-SD were older and heavier, more likely to be male, and had less healthy behaviors including smoking, alcohol consumption, and lack of exercise than the lower quartile group. In the midlife group, higher FPG variability was also incrementally associated with blood pressure, total cholesterol level, and the prevalence of hypertension and dyslipidemia.
However, there were no significant increases in systolic blood pressure, total cholesterol level, or the prevalence of dyslipidemia according to the quartiles of FPG variability in the elderly group. In addition, FPG variability was not associated with the history of stroke or the history of CKD in either age groups. Similar patterns of characteristics among the study population were shown according to the quartiles of FPG-CV (Supplemental Table 1) and FPG-VIM (Supplemental Table 2).  However, in the elderly group, there was no significant difference in incident PD according to the quartiles of FPG variability before and after adjusting for multi-variables and mean FPG.  Sensitivity analyses excluding participants who were diagnosed dementia (Table 3) or stroke (Table 4), also showed that increased risk for incident PD in the highest quartile group are maintained compared to the lowest quartile group in the midlife group.  Within the midlife group, a 1 SD-increment of PFG variability was associated with an increased risk of PD in all models (Table 5). Moreover, these associations persisted after stepwise adjustment for exposure to PD risk factors in all models and after adjusting for diverse confounding factors and mean FPG levels. We observed a significant association with a 1-SD unit increment in FPG variability in   Controlled Evaluation (ADVANCE) trial, increased Hemoglobin A1C (HbA1c)-SD and FPG-SD were related to a high risk of microvascular and macrovascular events and mortality [14]. Moreover, it is showed that not only severe hypoglycemic events [22] but also peak glucose levels [23] are strongly related to the risk of dementia, the most common neurodegenerative disease, in subjects with diabetes. In the Taiwan Diabetes Study cohort (n = 16,709), an increased glycemic variability, determined by FPG-CV and HbA1c-CV, were independently related to an elevated risk of developing Alzheimer's disease (AD) during a median follow-up of period of 8.9 years [24]. Additionally, in the same cohort, greater glycemic variability was associated with higher risk of ischemic stroke [25] and end-stage renal disease [13] In a meta-analysis study, Gorst et al. reported that a high long-term HbA1c variability was significantly associated with micro-and macrovascular complications and increased incident mortality in patients with type 1 and type 2 diabetes [12].
However, there have been limited studies evaluating the effect of long-term glycemic variability in subjects without diabetes. In a study of the Chinese population (including diabatic patients, 2.5%), visit-to-visit FPG variability was positively associated with the risk of CVD and all-cause mortality, even after adjusting for confounding factors such as mean FPG level [15]. Recently, Bancks et al.
reported an association between long-term FPG variability and decreased cognitive function, after adjusting for mean FPG level in the subjects without diabetes, among the Coronary Artery Risk Development in Young Adults (CARDIA) Study cohort [26]. In the present study, we found that longterm FPG variability is independently associated with the risk of developing PD in the middle-aged population without diabetes. In particular, the subjects with highest quartile of FPG variability consistently showed an increased the risk of developing PD by over 30% compared to those with the lowest quartile in all three different indicators such as FPG-SD, FPG-CV, and FPG-VIM. Moreover, multivariable linear regression showed a significantly increased risk of future PD based on a 1-SD change of glycemic variability, regardless of the calculating method of variability (e.g, SD, CV, and VIM), in the midlife group. Even with a stepwise adjustment for clinical and behavior risk factors, the degree of increased risk for PD did not change meaningfully. Therefore, long-term glycemic variability may act as a predictor for the future risk of PD as we showed that this relationship persisted despite adjusting for various confounding factors including mean FPG level.
In this study, the association between glycemic variability and the future risk of PD was only evident in the midlife group. Although we could not provide the exact reason for this age-dependent effect, growing evidence has shown that glucose metabolic disorders have a greater effect on neurodegenerative diseases in midlife populations than elderly populations. In Taiwan's population-based cohort, stratified by sex and age, diabetes was significantly linked to an elevated risk for incident PD; particularly, the association was stronger in the midlife population (< 60 years) [4].
Similar to Taiwan's study, a Danish study found the risk of developing PD was significantly higher among subjects with diabetes if early-onset (< 60 years) rather than late-onset PD (≥ 60 years) [6].
Likewise, Arvanitakis et al. demonstrated that the association between diabetes and parkinsonian signs was more pronounced in the midlife group than in the elderly group [27]. Furthermore, in 3,307 black and Caucasian subjects without diabetes, a higher FPG variability during young adulthood was related to poorer cognitive functioning in midlife, independent of FPG levels [26].
There are some potential mechanisms that may explain the effect of glycemic variability on the development PD. Reactive oxygen species (ROS) production coupled with insufficient antioxidant defense can lead to oxidative stress; this causes various pathophysiological conditions such as aging, type 2 diabetes, and neurodegenerative diseases (e.g., PD and AD) [28]. DNA damage originating from ROS is a conspicuous feature of PD [29]. Previously, it is suggested the vicious cycle among glycemic variability, oxidative stress, increased pro-inflammatory cytokine, and insulin resistance.
Some studies showed that increased glycemic variability caused more extreme oxidative stress than chronic hyperglycemia in the subjects with or without type 2 diabetes [10,11]. Abrupt hyperglycemia stimulates inflammatory cytokines including interleukin-6 and tumor necrosis factor-alpha via oxidative stress [30]. Concurrently, hyperglycemia leads to insulin over-secretion by compensatory mechanisms, which may lead to hypoglycemic events, increasing peripheral or central insulin resistance [31]. Further, this can cause vascular damage through the aggravation of hypertension, dyslipidemia and the stimulation of the proliferation and migration of arterial smooth muscle cells [32]. Moreover, accumulating evidence has revealed that the insulin signaling pathway may be central to the pathogenesis of PD [33]. It is known that insulin can pass through the blood-brainbarrier; insulin receptors exist in the substantia nigra and basal ganglia [34]. Much less insulin receptor mRNA expression was found in the substantia nigra from brain tissue affected by PD compared with tissue from normal controls [35]. Insulin exhibits neuroprotective effects by modulating neuronal growth and survival, maintenance of synapses, and dopaminergic transmission.
Conversely, defective insulin signaling pathways can lead to the downregulation of the AKT pathway, mitochondrial dysfunction, oxidative stress, inflammation, decreased expression of peroxisome proliferator-activated receptor-gamma coactivator 1α, and the increase of α-synuclein in the brain [33,36]. Additionally, both peripheral and neuronal insulin resistance, induced by metabolic stress (e.g., glucose fluctuation and inflammation), contribute to reduced dopaminergic signaling and increased nigrostriatal neurodegeneration, which results in the progression of PD [33,34,37,38].
Several antidiabetic medications have demonstrated beneficial effects on the outcomes of PD in animal models and in early-phase clinical trials. In in vitro and in vivo animal models of PD, both dipeptidyl peptidase-4 (DPP-4) inhibitors and glucagon-like peptide-1 receptor agonists showed neuroprotective effects [39]. Interestingly, DPP-4 inhibitors were reported to be more effective in reducing glycemic variability compared to sulfonylurea in patients with type 2 diabetes [40]. In a nationwide case-control study, Svenningsson et al. found a significantly decreased incidence of PD in patients taking DPP-4 inhibitors [41].
This study has several limitations to be considered. First, although we attempted to adjust for multiple confounding variables that influence the development of PD, using multivariate analyses, the possibility of unmeasured or residual confounding factors remain. Second, methods of measuring blood glucose levels, including HbA1c or oral glucose tolerance tests other than FPG, were not included in the health checkup data of the NHIS-HEALS. Third, the present study adopted the definition of PD if patients had at least two recorded PD diagnoses during outpatient clinic visits or hospitalization with PD diagnosis to reduce the over-estimation of PD [19,20]. However, like previous studies using definition of PD based on medical claim codes [19,20,42,43], the possibility of the under-or over-estimation of PD could not be ruled out. Nonetheless, the current study has noteworthy strengths. Particularly, this study was the first to demonstrate the impact of long-term glycemic variability on the development of PD. We used a standardized and validated database created by the Korean government which has a huge sample size and extensive information, including medical diagnoses and medications. In this study, we included only the non-diabetic population to eliminate the possible confounding effects of anti-diabetic medication or diabetic complications, including severe hypoglycemia. In addition, because this study investigated a sample taken from the general population, it could be assumed that the findings reflect the results that exist in the real world.
Moreover, we utilized three different indicators, FPG-SD, FPG-CV, and FPG-VIM, to express the variability of glucose and the results were similar and consistent, despite the adjustment for various confounding factors.

Conclusions
This study demonstrated that visit-to-visit FPG variability in midlife is independently associated with the long-term risk of PD, beyond mean FPG levels and other risk factors. Further research is needed to confirm these findings in other ethnic groups, to discover the pathophysiological mechanisms of glycemic variability leading to the development of PD, and to evaluate the possibility of glycemic variability as a therapeutic target to prevent PD in people with and without diabetes.

Acknowledgements.
This study used NHIS data (NHIS-2018-2-107) made by National Health Insurance Service (NHIS). We wish to thank the National Health Insurance Sharing Service (NHISS).
Funding. This study was supported in part by the Korea University Research Fund (K.M.C.) and by the