The Amsterdam Aging Cohort is an ongoing longitudinal cohort study which includes patients from the outpatient geriatric clinic at the Amsterdam University Medical Center, location VUmc (Rhodius-Meester et al., 2021). We included 893 patients with brain imaging who attended the memory clinic seeking medical care between February 2016 and June 2021. During this period, almost a patients visiting the memory clinic (89%) were willing to participate in the study. All patients were given a complete standardized comprehensive geriatric assessment (CGA) by trained nurses and doctors. This included an assessment of multiple geriatric domains, including cognition, physical function, nutrition, revision of medication in use and detailed medical history. Cognitive diagnosis – such as dementia (McKhann et al., 1984; Román et al., 1993; Neary et al., 1998; McKeith et al., 2005; Dubois et al., 2007; Rascovsky et al., 2011), mild cognitive impairment (MCI) (Albert et al., 2013), or subjective cognitive decline (SCD) (Studart and Nitrini, 2016) – were evaluated in a multidisciplinary consensus meeting. Our analysis included only patients who underwent brain Magnetic Resonance Imaging (MRI) or Computed Tomography (CT) as part of the diagnostic work-up. All patients gave written informed consent for their data to be used and the study was approved by the local Medical Ethics Committee.

Diabetes mellitus (DM) was defined either as having a history of diabetes or using antidiabetic medication. Other cardiovascular diseases – including coronary disease, heart failure, atrial fibrillation, and peripheral artery disease – were assessed on the basis of medical history and double checked with the patient and their family or carer. We dichotomized smoking status (never smoked v. ever smoked). Blood pressure and gait speed, as well as patient height and weight, were measured during the visit (Odden et al., 2012). Venous blood was drawn from all patients to measure cholesterol levels and non-fasting glucose. Medication as provided by the patient's pharmacy was reviewed with the patient, and with a partner, family member or carer if necessary.

Brain imaging was performed during a patient's first visit using CT (n = 238), 1.5T MRI (n = 162) and 3T MRI (n = 478) devices. The scans were reviewed and scored visually by two trained experts supervised by a clinical radiologist. Atrophy was scored on T1 sequence using visual rating scales ranging from 0 to 4 for medial temporal lobe atrophy (MTA), and from 0 to 3 for global cortical atrophy (GCA) (Harper et al., 2015). The average of the left and right side was used for MTA. White matter hyperintensities were scored on FLAIR/T2 sequence using the Fazekas scale (0–3) (Fazekas et al., 1987), and the number of microbleeds (on susceptibility-weighted imaging) and lacunes were counted. In this manuscript, we refer to either brain atrophy, white matter hyperintensities, lacunes or microbleeds collectively as “brain structure abnormalities.”

To assess whether the observed differences in brain structure also had a functional impact, cognition was included in the analysis. Cognitive performance was assessed in a standardized manner by trained neuropsychologists and divided into four domains: memory, language, executive function, and processing speed. All patients were assessed using the Mini Mental State Examination (MMSE) and the Geriatric Depression Scale (GDS). Memory was tested with the auditory verbal learning test (Van Der Elst et al., 2005) and Visual Association Test (VAT) (Lindeboom et al., 2002). Language was tested using the Category Fluency Animals Test (Van Der Elst et al., 2006) and the VAT naming test, a component of the VAT. Processing speed was examined with the Stroop Color-Word test (SCWT) (Van der Elst et al., 2006) and the Trail Making Test-A (TMT-A) (Reitan, 1955). Finally, executive function was assessed with the Behavioral Assessment of the Dysexecutive Rule-changing test (BADS) (Burrell and Piguet, 2015) while correcting for speed using the SCWT and the TMT. For the purposes of the analysis, all test results were converted to Z-scores or inverse Z-scores. A higher Z-score indicates poorer performance.

Baseline characteristics for men, women, and the total population are reported as mean (SD), or median (interquartile range) for categorical variables. Differences between groups were analyzed using Student's T-test, the Mann-Whitney U-test, the Kruskal Wallis test, ANOVA and chi-square testing where appropriate. First, logistic regression analyses were performed to assess the association of diabetes with brain structures separately for men and women. For the logistic regression analysis, we dichotomized the scores of the visual rating scale and the values for microbleeds and lacunes. A cut-off value of two or more was used for the imaging scores of atrophy (MTA and/or CGA) and WMH (Rhodius-Meester et al., 2017). Microbleeds were dichotomized as present or not present, and a value of one or more was adopted a cut-off for lacunes (Henneman et al., 2009; Jokinen et al., 2011). Second, linear regression analyses were performed to assess the association of diabetes with cognitive functioning separately for men and women. All analyses were adjusted for age (model 1), and additionally for smoking and alcohol consumption (model 2), and hypertension and cardiovascular disease (coronary disease, heart failure, atrial fibrillation, stroke or TIA and peripheral arterial disease) (model 3). In addition, we adjusted for presence of subjective complaints, mild cognitive impairment, or dementia (data not shown). For the analyses of functional cognitive measures, furthermore, we corrected all models for level of education. A p-value < 0.05 was considered statistically significant. Data were analyzed with SPSS software, version 26 (IBM Corp, Armonk, NY, USA).

To determine whether male or female sex and CVD was associated with a higher risk of cSVD to a greater degree than these factors individually, we added an interaction term to the regression analysis, testing multiplicative interaction. To assess additive interaction, we calculated RERI (Relative Risk due to Interaction) (Knol et al., 2007; Knol and VanderWeele, 2012). For this analysis, when the combined risk of sex and CVD was higher than the sum of the risks associated with the individual factors, the interaction between sex and CVD was considered to constitute an additional risk factor. A RERI above zero indicated that interaction between female sex and cardiac disease had an additional effect on the outcome; a RERI below zero indicated that this was the case for the interaction between male sex and cardiac disease. In the RERI analysis, we also corrected for age, smoking, and alcohol consumption, analogous to the logistic regression analyses. These analyses were performed in R (R Core Team, 2020).

The sex-specific logistic regression analyses of the relation between diabetes and brain structures showed that in women, the presence of diabetes was significantly associated with an increased risk of having brain atrophy and lacunes (Table 2). Age-adjusted odds ratios were 2.16 (95% CI 1.00–4.67) and 2.60 (95% CI 1.24–2.46). However, in men, diabetes was not associated with an increased risk of having brain structure abnormalities. Additional adjustments for cardiovascular risk factors and disease (model 2 and 3) did not change these effect estimates (Table 2). Adjusting for cognitive diagnosis (subjective complaints, mild cognitive impairment, or dementia) did not change the effect estimates (data not shown). Diabetes was not associated with an increased risk of WMH and microbleeds. When adding an interaction term to the regression, we did not find a significant interaction of sex with diabetes. When assessing additive interaction using RERI analysis, a trend was seen toward an increased risk for women with diabetes of atrophy and lacunes (RERI 0.45 for the presence of atrophy and female sex, and 0.48 for the presence of lacunes and female sex) (Supplementary Table 2).

The sex-specific linear regression analyses of the relation between diabetes and cognitive performance showed that diabetes in women was associated with a significantly lower score of executive function (beta z-score 0.07; 95% CI 0.00–0.14), processing speed (beta 0.06; 95% CI 0.90–0.95), and language (beta 0.07; 95% CI 0.01–0.12) (Table 3). In men, diabetes was not associated with cognitive performance. Additional adjustments for cardiovascular risk and disease, and cognitive diagnosis did not change the effect estimates (data for cognitive diagnosis not shown). We observed an interaction of sex and diabetes: women with diabetes were at increased risk for impaired processing speed (B 0.17 (0.03–0.33), p = 0.04). We observed no interaction between diabetes and sex in the other cognitive domains.

Mechanisms which may affect susceptibility to the vascular complications of diabetes include altered coagulation, oxidative stress, endothelial dysfunction and impaired vasodilation (Kautzky-Willer et al., 2016; de Ritter et al., 2020). Women with diabetes might be in a more pro-thrombotic state than men, which may lead to lacunes and atrophy, and a more general decline in brain health, even when the prevalence of diabetes is similar in both sexes (Smith et al., 2012; Neergaard-Petersen et al., 2014). They generally also have greater levels of systemic inflammation and more oxidative stress than men with diabetes, leading to impaired vascular reactivity, which is specifically associated with the occurrence of lacunes (Mrgan et al., 2018). Sex-dependent differences in vascular physiology may therefore render women more susceptible to the cerebrovascular complications of diabetes, and also lead to functional decline.

Levels of central adiposity in men and women with diabetes may also be of importance. There is evidence to suggest that women have a poorer cardiovascular risk profile than men when they are diagnosed with diabetes, especially when central adiposity is measured (Paul et al., 2012; Peters et al., 2016a). This may be the result of a longer period of development of diabetes in women: women are more insulin-sensitive in middle age, and their insulin sensitivity deteriorates more than in men before they reach the diagnosis of diabetes. A longer period of time before a formal diagnosis of diabetes can be made may also lead to an increased occurrence of other risk factors such as abdominal adiposity, and to higher levels of subclinical damage mediated by hyperglycemia (Woodward et al., 2015; Peters et al., 2016b). Abdominal adiposity is independently related to brain structure abnormalities, including silent lacunary infarcts, and therefore may mediate the increased prevalence in women with diabetes by comparison with men (Yamashiro et al., 2014).

Furthermore, there is little awareness in the field of geriatrics about the long-term cardiovascular effects of pregnancy-related complications, as well as other women-specific factors such as timing of menopause and gestational diabetes (Keskin et al., 2015; Kuh et al., 2018). These important cardiovascular parameters are therefore often not registered in medical files, as is the case in our dataset (Wilkins-Haug et al., 2015). It has long been known that menopausal status and the timing of menopause influence cardiovascular risk, abdominal obesity, occurrence of DM, and clinical course of dementia (Archer, 2009; Gong et al., 2021; Hickey and Mishra, 2021). They may therefore also have affected the clinical outcome in our study. Further studies investigating the long-term effects of DM should therefore include sex-specific cardiovascular risk factors to assess their impact on brain structure and cognitive function.

As a consequence of ongoing lower inclusion rates of women in studies investigating the long-term effects of diabetes, at least in part, it is unclear which mechanisms lead to the poorer clinical outcome of women with diabetes (Norhammar and Schenck-Gustafsson, 2013). However, there are several scientific findings which may play a role. Social gender norms have a profound influence on patients' disease perception, moment of referral, interpretation of symptoms and the likelihood of receiving guideline-recommended treatment. In the case of diabetes, evidence shows that women diagnosed with diabetes attain glycemic targets less often, and are screened less for the complications of diabetes (Choe et al., 2018). Risk factor targets for co-morbid cardiovascular disease are also less often achieved in women (Ferrara et al., 2008; Wannamethee et al., 2012; Rossi et al., 2013; de Jong et al., 2020). Since early intervention in diabetes improves long-term outcome, this may also have implications for the incidence of (cerebral) complications (Group UPDS, 1998). The same truth holds for cognitive impairment: women with dementia are generally referred later than men (Howard et al., 1998; Sourial et al., 2020). Women might therefore experience delay in receiving adequate supporting care when experiencing cognitive impairment. Hence, inequities in the recognition and treatment of (cardiovascular) risk factors for dementia, as well as the recognition and treatment of dementia itself, may play a role in the occurrence of sex-related differences in the cerebral complications of diabetes.

A major strength of our study is the standardized work-up of a large group of “real-life” patients. We included a multi-domain assessment which was part of medical routine care. Because of this integration in regular care practice, routinely used measurements and tools were used, facilitating the translation of our research to clinical practice. In addition, we combined these clinically used parameters with imaging markers as well as extensive neuropsychological testing, bridging the gap between etiological research and clinical practice. Our study has also several limitations. Firstly, because of our cross-sectional design, we cannot draw conclusions about the causality of our findings. We balanced this with logistic regression models in which we corrected for confounding factors. Related to this, we have no data showing possible differences between brain structure at the time of diagnosis of diabetes, and later life. It is possible that women with diabetes already have worse brain health at this time, and that it is not a direct consequence of diabetes, but merely coexists due to other pathological processes. Furthermore, HbA1c and time since diagnosis of DM were not included in our dataset. We were therefore unable to assess the influence of glycemic regulation. We did include non-fasted glucose, which was similar for men and women, in our baseline characteristics. Also, it is remarkable that the mean age in our sample is similar for men and women. Since the life expectancy in women exceeds the life expectancy in men, we would expect a higher mean age for women. The similar age for men and women may reflect underlying gender bias in referral to our clinic. However, it may also be explained by other forms of sampling bias not related sex or gender. Furthermore, as mentioned before, women-specific pathology was hardly registered at all in our patient files. A history of gestational diabetes, polycystic ovary syndrome and premature menopause contribute to the excess risk of diabetic complications and diabetes (Soedamah-Muthu et al., 2004; Huxley et al., 2006; Peters et al., 2014). Furthermore, preeclampsia is related to structural brain damage later in life (Siepmann et al., 2017). Lastly, premature menopause is additionally associated with poorer cognitive performance and a higher risk of dementia later in life (Ryan et al., 2014). However, the precise relationship between cerebrovascular disease and pregnancy-related cardiovascular disease remains unclear, since a recent review concluded that gestational hypertension is not related to cerebral stroke (Lo et al., 2020). Finally, other factors that affect the social position of patients and therefore their quality of care – such as class, cultural background, and variables associated with poorer referral and poorer health care provision in general – were not available in our study (Vaccarino et al., 2002; LaVeist et al., 2003).

In conclusion, this sex-specific analysis of the association between diabetes and its cerebrovascular complications shows that diabetes is significantly associated with brain structure abnormalities and function in women but not in men. We can only speculate on the nature of these differences, and whether they are dependent on gender or sex. Although we did not find a statistically significant interaction between female sex and diabetes, the differences in associations for men and women are striking and underline the importance of the sex-specific analysis of clinical data. Further research should at least include data on abdominal obesity and female-specific risk factors such as pregnancy-related complications and menopause, and more studies are needed to elucidate the mechanisms which contribute to the association between diabetes, brain structure and cognition in women. Elucidating the sex-specific relationships between diabetes and cSVD may help to understand the gap in burden of dementia and help to achieve more equity in the care for this group of patients.