Sex Differences In The Association Between Hemoglobin A1c And Cerebral White Matter Lesions In The General Japanese Population

The inuence of diabetes and associated sex differences on cerebral white matter lesions (WMLs) is unclear. We used data from a cross-sectional study uploaded to the DATADRYAD website by Shinkawa et al. to investigate differences in the association between hemoglobin A1c (HbA1c) levels and cerebral WMLs between men and women. The average age of all participants was56.4±11.5years old, and approximately 51.89 % of them were men. A linear relationship between HbA1c and cerebral WMLs was detected in men. Fully adjusted binary logistic regression showed no association of HbA1c with cerebral WMLs in men. A nonlinear relationship between HbA1c and cerebral WMLs was detected in women, whose cutoff point was 5.6%. The effect sizes and condence intervals of the left and right sides of the inection point were OR=0.21 (95%CI 0.06, 0.69, P=0.0098) and OR=3.5 (95%CI 1.50, 8.15, P=0.0037), respectively. In the higher HbA1c group, further subgroup analysis showed a stronger association between HbA1c and cerebral WMLs in women (OR=3.83, 95%CI 1.68, 8.72 P=0.0014) than in men (OR=1.02, 95%CI 0.76, 1.36 P=0.8986) (P for interaction with sex was 0.0004). A stronger effect of HbA1c on the risk of cerebral WMLs in women than in men was found in the higher HbA1c group. was 5.6%. a shape of the independent association between HbA1c and cerebral WMLs in women. In addition, also found that the trend of the effect sizes on the left and right sides of the inection point was not consistent (left OR = 0.21 95%CI 0.06, 0.69 P = 0.0098); right OR = 3.5 95%CI 1.50, 8.15 P = 0.0037). Interaction tests better understand the trends of HbA1c and cerebral WMLs in different populations. The results of this study found a stronger association between HbA1c levels and cerebral WMLs PS (carotid plaque score), systolic blood pressure (SBP), diastolic blood pressure (DBP), body mass index (BMI), LDL cholesterol (LDL), HDL cholesterol (HDL), LH ratio (quotient of LDL and HDL), triglyceride (TG), blood glucose level (BS), plaque number (n-plaque), age, sex, smoking habit (Smoke), metabolic syndrome (Met-syn), medication to reduce blood pressure (Med-BP), medication to reduce blood sugar or insulin injection (Med-sugar), medication to reduce the level of cholesterol (Med-cho), amount of drinking per day (Drink-V), and drinking habit.


Introduction
Rapid population aging combined with sedentary habits has made type 2 diabetes one of the largest public health problems worldwide 1,2 . Recent studies have demonstrated that in addition to diabetes, prediabetes can damage small and large blood vessels and lead to complications such as neuropathy, nephropathy and macrovascular diseases [3][4][5][6] . More recent investigations have shown considerable sex differences associated with diabetes risk factors, hormonal effects on glucose, and diabetic vascular and nonvascular outcomes [7][8][9] . It has now been well established by a variety of studies that a higher HbA1c level, as a biomarker of long-term glycemic control, is an independent risk factor for diabetes complications [10][11][12] . Cerebral WMLs are mainly chronic ischemic lesions caused by small vessel diseases, which show white matter hyperintensities (WMHs) on T2-weighted or uid-attenuated inversion recovery (FLAIR) images in magnetic resonance imaging (MRI) and are associated with cognitive impairment, gait dysfunction and focal neurological signs 13 . Many factors have been found to be related to cerebral WMLs, such as age, hypertension, dyslipidemia, smoking and various biomarkers of vascular disease 14 .
Although DM is well known as a vascular risk factor, the relationship between DM and cerebral WMLs is still controversial [15][16][17][18] . Prediabetes was also shown to be associated with brain structural abnormalities 19 . HbA1c, which re ects a measure of glycemia during the previous 2-3 months, is a biomarker for long-term glycemic control and is also indicative of prediabetes. Previous studies have shown a signi cant association between HbA1c and cerebral WMLs [20][21][22] . However, such conclusions were not con rmed by another study conducted in a larger cohort of patients 23 . In addition, the sex differences in the relationship between HbA1c and cerebral WMLs have still not been illuminated in previous studies.
In this study, a secondary data analysis was performed using existing data from a published paper 24 . In the secondary analysis, the independent variable and dependent variable were HbA1c level and cerebral WMLs, respectively. Other covariates are consistent with those in the original. This analysis sought to investigate whether sex differences exist in the association between HbA1c levels and the incident risk of cerebral WMLs in the general population.

Results
Baseline characteristics of participants A total of 1904 participants were included in the nal data analysis, with 988 men and 916 women classi ed into two groups (lower HbA1c group and higher HbA1c group) according to the clinical cutoff point of HbA1c. The baseline characteristics of these groups are reported in Table 1. In general, the average age of the 1904 participants was 56.4 ± 11.5 years old, and approximately 51.89% of them were male. No statistically signi cant differences were detected in LDL, HbA1c, or medication to reduce blood sugar or insulin injection between men and women in the lower HbA1c group (all p values > 0.05). Women had higher values in age and HDL and were more likely to exhibit the following values than men in the lower HbA1c group: metabolic syndrome (no), smoking habit (no), medication to reduce blood pressure (no), medication to reduce the level of cholesterol (yes), amount of drinking per day (< 180 ml), drinking habit (rarely), plaque number (0) and cerebral WMLs (yes). The opposite patterns were observed in LH, TG, BS, SBP, DBP, BMI, PS, metabolic syndrome (reserve and yes), smoking habit (yes), medication to reduce blood pressure (yes), medication to reduce the level of cholesterol (no), amount of drinking per day (> 180 ml), plaque number (n > = 1) and cerebral WMLs (no) in the lower HbA1c group. No statistically signi cant differences were detected in SBP or medication to reduce the level of cholesterol between men and women in the higher HbA1c group. Women had higher values of age, LDL, and HDL and were more likely to have metabolic syndrome (no), smoking habit (no), medication to reduce blood pressure (no), medication to reduce blood sugar or insulin injection (no), amount of drinking per day (< 180 ml) and drinking habit (rarely) and plaque number (0) and cerebral WMLs (yes) in the higher HbA1c group. The opposite patterns were observed in LH, TG, HbA1c, BS, DBP, BMI, PS, metabolic syndrome (reserve and yes), smoking habit (yes), medication to reduce blood pressure (yes), medication to reduce blood sugar or insulin injection (yes), amount of drinking per day (> 180 ml), drinking habit (sometimes and everyday), plaque number (n > = 1) and cerebral WMLs (no).

Univariate analysis
We listed the results of univariate analyses, adjusting for age, for men and women in Table 2. By univariate binary logistic regression adjusting for age, we found that LDL, HDL, LH, TG, HbA1c, BS, smoking habits, amount of drinking per day and plaque number (n = 1, n = 2, n > 2) were not associated with cerebral WMLs in men. We also found that PS odds ratio (OR) = 1. men. By univariate binary logistic regression adjusting for age, we found that LDL, LH, TG, HbA1c, BS, BMI, metabolic syndrome (reserve or yes), smoking habit, medication to reduce sugar or insulin injection, medication to reduce the level of cholesterol, drinking habit, plaque number (n = 1, n = 2, n > 2) and amount of drinking per day (180-360 ml) were not associated with cerebral WMLs in women. We also found that the amount of drinking per day (> 360 ml) OR = 0.14 (95%CI 0.04, 0.51 P = 0.0030) was negatively associated with cerebral WMLs in women.  The results of nonlinearity of HbA1c and cerebral white matter lesions for men and women In the present study, we analyzed the nonlinear relationship between HbA1c and cerebral WMLs for men and women ( Fig. 1a and Fig. 1b). The smooth curve and the result of the generalized additive model showed a linear association of HbA1c with cerebral WMLs in men after adjusting for age, PS, LDL, HDL, TG, BS, SBP, DBP, BMI, metabolic syndrome, medication to reduce blood pressure, medication to reduce blood sugar or insulin injection, medication to reduce the level of cholesterol, drinking habit and plaque number. In this study, we constructed two models to analyze the independent effects of HbA1c on cerebral WMLs (univariate and multivariate binary logistic regression). Binary logistic regression showed that there was no association of HbA1c with cerebral WMLs in men. The effect sizes (OR) and 95% con dence intervals are listed in Table 3. In the minimally adjusted model (model 1), the model-based effect size can be explained as a one-unit difference in HbA1c level associated with risk of WMLs. The smooth curve and the result of the generalized additive model showed that the relationship between HbA1c and cerebral WMLs was nonlinear in women after adjusting for age, PS, LDL, HDL, TG, BS, SBP, DBP, BMI, metabolic syndrome, medication to reduce blood pressure, medication to reduce blood sugar or insulin injection, medication to reduce the level of cholesterol, drinking habit and plaque number. We used both binary logistic regression and two-piecewise binary logistic regression to t the association and select the best t model based on P for the log likelihood ratio test. Fully adjusted model: we adjusted for age, HDL, LDL, TG, DBP, SBP, BMI, BS, PS, drinking habit, metabolic syndrome, medication to reduce blood pressure, medication to reduce blood sugar or insulin injection, medication to reduce cholesterol levels, and n-plaque.
Because the P for the log likelihood ratio test was less than 0.05, we chose two-piecewise binary logistic regression for tting the association between HbA1c and cerebral WMLs in women because it can accurately represent the relationship. Using a two-piecewise binary logistic regression and recursive algorithm, we calculated that the in ection point was 5.6%. On the left side of the in ection point, the effect size and 95%CI were 0.21 (0.06, 0.69), P = 0.0098. On the right side of the in ection point, the effect size and 95%CI were 3.5 (1.50, 8.15) (P = 0.0037) ( Table 4). We adjusted for age, HDL, LDL, TGs, DBP, SBP, BMI, BS, PS, drinking habits, metabolic syndrome, medication to reduce blood pressure, medication to reduce blood sugar or insulin injection, and medication to reduce cholesterol and n-plaque levels.

Interaction test
We used sex as the strati cation variable to observe the trend of effect sizes in this variable (  Our study showed that a higher level of HbA1c has a greater impact on women's risk for WMLs than on men's risk. The in ection point of the U shape on the independent association between HbA1c and cerebral WMLs was 5.6%. This indicated that more aggressive treatment should be considered in women for cerebral WML prevention and for glycemic targets in Japan. When HbA1c is lower than 5.6%, the level of HbA1c is negatively associated with cerebral WMLs, which indicates that hypoglycemic conditions may also contribute to the development of cerebral WMLs. The mechanisms that explain the sex difference in the risk of vascular disease associated with diabetes have not been identi ed. However, this excess risk among women could be due to certain underlying biological differences and health care provided for diabetes and its vascular complications between women and men 33 . Several studies supported that women underwent more pronounced exposure to hazardous metabolic risk factors than men before the onset of type 2 diabetes 34-37 . Among 500,000 individuals in the UK Biobank, the difference in waist circumference and BMI between those with and without diabetes was larger in women than men 38 . Moreover, women have similar levels of HbA1c but a remarkably higher BMI than men when rst diagnosed with diabetes 39,40 . These disadvantageous obesity-associated mechanisms in women were speculated to be partly responsible for the sex difference in the risk of vascular disease associated with diabetes. In contrast to the above conclusions, our study showed a lower BMI in women than in men in the higher HbA1c group. Previous studies showed that women who converted to diabetes showed relatively worse levels of total cholesterol, HDL cholesterol, triglycerides and DBP at baseline than men. In contrast, women with higher levels of HbA1c had relatively better levels of LH, TG and DBP than men in the Japanese population. In addition to biological differences between men and women, disparities in health care may in part explain sex differences in diabetes-related vascular complications. Previous studies showed that secondary prevention in risk factor management was generally worse in women than in men 41 . Our study showed a similar outcome. Women with higher levels of HbA1c are less likely to take medicine for BP and blood sugar than men. Sex differences in biological factors, such as both the use and provision of health care, could contribute to women's higher relative risk of diabetic vascular complications. There were still signi cant sex differences in the association between HbA1c and cerebral WMLs after adjusting for associated covariates. The clinical value of this study is as follows: (1) To the best of our knowledge, this is the rst study to observe the independent nonlinear association between HbA1c and cerebral WMLs in women; (2) to the best of our knowledge, this is the rst study to observe sex differences in the association between HbA1c and cerebral WMLs in women and men; and (3) the ndings from this study should contribute to future research on the establishment of diagnostic or predictive models of cerebral WMLs.
Our study has some strengths. (1) We performed a large population-based analysis of the general population; (2) we address the nonlinearity in the present study and further explore this; (3) as this is an observational study, it was susceptible to various confounding variables. We used strict statistical adjustments to minimize residual confounding; and (4) the effect modi er factor analysis improved the use of the data and revealed interactions in different subgroups in this study.
Several possible limitations of the present study should be considered: (1) This was a cross-sectional study. Thus, we could not exclude a causal relationship from the ndings of this study. (2) In this study, our research subjects were members of the general population attending a medical screening center in Japan. Therefore, there is a certain de ciency in the universality and extrapolation of research. (3) In this study, our research subjects were mainly prediabetic individuals. Therefore, if the scope of the population is expanded and the diabetes sample size is increased, the results obtained will be more persuasive. Despite these potential limitations, this analysis adds to the body of knowledge regarding the effect of HbA1c on the risk of WMLs by quantifying the dramatic impact of HbA1c in women after accounting for other known WML risk factors.

Perspectives And Signi cance
This study highlights sex differences in the association between cerebral WMLs and HbA1c. Women had a much higher odds ratio of cerebral WMLs associated with HbA1c than men in higher HbA1c group. These ndings suggest that more careful glycemic control may be needed in women with hyperglycemia to prevent cerebral WMLs. Sex differences should be taken into consideration in assessing the association between HbA1c and cerebral WMLs.

Data source
The secondary data were obtained from the DATADRYAD database (www.Datadryad.org). Users are permitted to download raw data freely from this website. According to the Dryad Terms of Service, we cited the Dryad data package in the present study.

Study population
Shinkawa Yuya et al. completed the entire study. The speci c details are described in the original report by Shinkawa Yuya 24 . Participant data were nonselectively and consecutively collected from subjects who underwent brain MRI and blood tests during the brain dock course of a comprehensive medical checkup some time between April 1, 2016, andOctober 31, 2017, at Shin Takeo Hospital. A total of 1904 participants, including 988 men and 916 women, were involved in this study. The data in the database were anonymous for the purpose of protecting participant privacy. Data are stored in an electronic data acquisition system. Participants' informed consent was not required in this study because of the nature of the retrospective cohort study. This study was approved by the ethical review committee of Shin Takeo Hospital.
Ethical approval. This analysis is based on summary statistics obtained from previously published analyses and therefore we have not sought additional ethical approval. All methods were performed in accordance with the relevant guidelines and regulations. Due to the retrospective nature of the study design and anonymous data collection, written informed consent was waived by the ethical review committee of Shin Takeo Hospital.

Variables
HbA1c was measured at baseline and recorded as a continuous variable. The blood and biochemical indexes were detected by the laboratory test systems C8000 (Canon Medical Systems Corporation, Tochigi, Japan) and Acute (Canon Medical Systems Corporation, Tochigi, Japan), respectively. HbA1c was measured with an automated glycohemoglobin analyzer HA8181 (Arkray Inc., Kyoto, Japan).
The outcome variable (dichotomous variable) was determined according to published guidelines and studies. Head magnetic resonance imaging (MRI) scans were acquired on MAGNETOM Symphony (Siemens Healthineers Japan, Tokyo, Japan) and MAGNETOM ESSENZA (Siemens Healthineers Japan, Tokyo, Japan) scanners. The detailed process of de nition of cerebral WMLs is described as follows: there are periventricular or deep white matter lesions on FLAIR sequence of MRI (dichotomous variable: 1 = presence of cerebral white matter lesions on MRI; 0 = absence of cerebral white matter lesions on MRI).
The variables in this study can be divided into three types: (1) demographic data; (2) variables that can affect HbA1c or cerebral WMLs reported by previous literature; and (3) variables based on our clinical experiences. We selected these covariates on the basis of their association with the outcomes or a change in effect estimate of more than 10%. Therefore, the following variables were used to construct the fully adjusted model: (1) continuous variables: HDL, LDL, TG, SBP, DBP, BMI, PS, and BS (obtained at baseline); (2) categorical variables: age, sex, metabolic syndrome, medication to reduce blood pressure, medication to reduce blood sugar or insulin injection, medication to reduce the level of cholesterol, drinking habit (every day, sometimes, or rarely drink (cannot drink)) (obtained at baseline) and plaque number. Binary variables take a value of 0 or 1 to indicate the absence or presence of some categorical effect, respectively, e.g., sex: X = 0 for men and X = 1 for women; medication to reduce blood pressure: X = 0 for "No" and X = 1 for "Yes"; medication to reduce blood sugar or insulin injection: X = 0 for "No" and X = 1 for "Yes"; medication to reduce blood pressure: X = 0 for "No" and X = 1 for "Yes". For the purpose of fully adjusting variables, we converted age from a categorical variable to a continuous variable.

Statistical analysis
Quantitative continuous variables are presented as the mean ± standard deviation (normal distribution), and categorical variables are presented as the number and percentage. We used χ2 (categorical variables) or Student's T test (normal distribution) to test for differences among men and women in different HbA1c groups (clinical cut point). The data analysis process of this study was based on three criteria: (1) what is the relationship between HbA1c and cerebral WMLs (linear or nonlinear) in men and women? (2) which factors modify or interfere with the relationship between HbA1c and cerebral WMLs in men and women? and (3) after adjustment for the interfering factors or after the strati ed analysis, what is the true relationship between HbA1c and cerebral WMLs in men and women? Therefore, data analysis can be summarized in three steps.
Step 1: Univariate and multivariate binary logistic regression were employed. We constructed two models: model 1, minimally adjusted model, adjusted only for age; model 2, fully adjusted model, adjusted for those covariates as just described.
Step 2: To address the nonlinearity of HbA1c and cerebral WMLs, a generalized additive model and smooth curve tting (penalized spline method) strati ed by sex were conducted. If there is a nonlinear relationship, a recursive algorithm is used to calculate the in ection point, and then two-piecewise binary logistic regression on both sides of the in ection point is constructed. The log likelihood ratio test was used to determine the most suitable model for tting the association between the independent variable and outcome variable.
Step 3: In view of the difference in association between HbA1c and cerebral WMLs in men and women re ected by smooth curve tting, we performed an interaction test between HbA1c and sex in different HbA1c groups. All analyses were performed with the statistical software packages R (http://www.Rproject.org, The R Foundation) and Empower Stats (http://www.empowerstats.com, X&Y Solutions, Inc, Boston, MA). P values less than 0.05 (two-sided) were considered statistically signi cant. Abbreviations LDL: low-density lipoprotein; HDL: high-density lipoprotein; LH: quotient of LDL and HDL; TG: triglyceride; BS: blood glucose level; SBP: systolic blood pressure; DBP: diastolic blood pressure; BMI: body mass index; PS: carotid plaque score; Met-syn: metabolic syndrome; Med-bp: medication to reduce blood pressure; Med-sugar: medication to reduce blood sugar or insulin injection; Med-cho: medication to reduce the level of cholesterol; Drink-V: amount of drinking per day; WMLs: white matter lesions Declarations HL and JY contributed to the drafting of the manuscript, and analysis and interpretation of the data. SG contributed to the conception and critical revision of the manuscript, analysis and interpretation of the data and approved the nal version of the submitted manuscript. Both authors read and approved the nal manuscript.

Additional information
Competing interests: The authors declare that they have no competing interests.