Using health check data to investigate cognitive function in Aboriginal and Torres Strait Islanders living with diabetes in the Torres Strait, Australia

Abstract Background Type 2 Diabetes (T2DM) has a subtle deleterious effect on cognition and imposes a higher lifetime risk of cognitive impairment and dementia. In populations where both T2DM and dementia are highly prevalent, understanding more about the early effects of T2DM on cognition may provide insights into the lifetime risks of this disease. Methods In 2016, 186 Australian Aboriginal and/or Torres Strait Islander residents of the Torres Strait (54% female, mean age =38.9 years, SD =15.9, range =15–74) participated in a community health check. The effect of diabetes (Type 1 or Type 2) on speed of thinking and working memory was assessed with the Cogstate Brief Battery (CBB) during the health check. Results One third of participants had diabetes (n = 56, 30.1%). After adjusting for age, education and previous iPad/Tablet experience, participants with diabetes had a small, yet significant reduction in accuracy on the One Back working memory task (β = −.076, p = .010, r 2 = .042). The effect was most pronounced among participants with diabetes aged 20–49 years (n = 20), who also had evidence of poorer diabetes control (eg HbA1c% ≥6.5, 76.6%), relative to participants with diabetes aged 50 years and over (n = 31) (HbA1c% ≥6.5, 32.0%, p = .005). Conclusions Early and subtle decrements in working memory may be a potential complication of diabetes among Aboriginal and Torres Strait Islander residents of the Torres Strait. Several potentially influential variables were not captured in this study (eg medication and diabetes duration). Greater preventative health resources are required for this population, particularly given the emerging elevated dementia rates linked to chronic disease.


| INTRODUC TI ON
Type 2 Diabetes (T2DM) is a major global public health concern.
In 2017, T2DM was the ninth leading cause of mortality and affected 6.38% of the world's population. This represents a prevalence milestone in a rising trend that shows no signs of abating. 1 In Australia, T2DM was the 12th largest contributor to the country's 2015 burden of disease and affected 5.3% of adults in 2017-2018. 2 Among Aboriginal and Torres Strait Islander inhabitants of Australia, referred to here respectfully as Indigenous Australians, the rate of T2DM is 4.3 times higher than in the non-Indigenous population. 3 For individuals, T2DM is associated with irreparable damage to multiple body organs and systems, including the brain. For older adults, it is well established that a history of T2DM is associated with a greater likelihood of cognitive impairment and an increased risk for dementia from both cerebrovascular disease and Alzheimer's Disease. 4 In adults without dementia, cognitive impairment associated with T2DM is subtle and expressed most reliably as decreases in processing speed, 5 although this may extend to executive function 6 and working memory. 7 While studies examining T2DM have been conducted mainly in older adults, qualitatively similar but quantitatively less severe cognitive impairment has been observed in adults with T2DM aged less than 60 years. 8,9 For Indigenous Australians, the prevalence of dementia is 3-5 times higher than non-Indigenous Australians. [10][11][12][13] Elevated levels of childhood trauma, stroke, head injury, and lower levels of skilled employment have been identified as predictors of dementia among mainland Aboriginal populations, while more education has been identified as protective. Recent evidence suggests that vascular risk factors, including diabetes, may be associated with higher rates of dementia among Indigenous Australians living in the Torres Strait. 13,14 Understanding more about the cognitive implications of diabetes in these vulnerable populations could therefore improve understanding of the development of diabetesrelated brain disease and assist with targeting preventative health activities.
The Zenadth Kes Health Partnership (ZKHP) was a communitybased health-screening program of residents aged 15-78 years, undertaken on two islands in the Torres Strait between October and December 2016. 15 In this population, where diabetes incidence is almost four times the general Australian population, 16 the purpose of the ZKHP was to provide a health service for the community and simultaneously explore the association between the metabolic syndrome and other chronic health conditions. The aim of this current study was to understand whether diabetes, defined as either Type 1 or Type 2, was having a subclinical effect on cognition in this population. It was hypothesized that speed and accuracy on cognitive tasks would reduce as a function of age and increase with years of education and iPad/Tablet experience. It was also hypothesized that participants with diabetes would have slower reaction times and lower accuracy on cognitive tasks, after accounting for age, education and iPad/Tablet experience.  (4) having a physical or sensory disability (eg vision) preventing valid assessment. A total of 200 participants underwent cognitive assessment and the sample for the current study was limited to the 186 participants who had a complete and valid result on at least one of the cognitive tests. 17 The 14 participants excluded at this stage were older than those who were retained (ie mean age of 62.8 and 38.9 years, respectively, p < .001) and more likely to have diabetes (p = .002).

| Data collection
Demographic information, diet, health behaviours, depressive symptoms and blood samples for several routine and research blood tests were collected. Demographic data were age (years), gender (male, female), total years of education, employment status when of working age (15-64 years) defined as a person having a paid job (yes, no) and island of residence. Information on history of use of electronic devices (iPad or computer tablet (yes, no)) and hand dominance was also collected. Self-reported use of alcohol and tobacco was collected, and participants also completed a food questionnaire, which included consumption of take-away food and sugary drinks in the week preceding their health check.
Measures of cardiovascular health included heart rate, systolic and diastolic blood pressure, hypertension (ie systolic ≥140 mmHg or diastolic ≥90 mmHg) and urinary albumin creatinine ratio (urinary ACR). Metabolic markers were triglycerides, high-density lipoprotein (HDL), low-density lipoprotein (LDL), total cholesterol, HDL/total cholesterol ratio, glucose, glycosylated haemoglobin (HbA1c%), body mass index (BMI), waist circumference (centimetres), waist/height ratio. Glucose, triglycerides, lipoproteins and cholesterol were analysed at a commercial pathology service (Sullivan Nicolaides, Cairns, Australia). A measure of omega-3 (n − 3) long-chain polyunsaturated fatty acid (LCPUFAs) was obtained by a whole blood collection on a validated dried blood spot system. Fatty acid composition was analysed by capillary gas chromatography. 18 An adapted Patient Health Questionnaire 9 (aPHQ-9) depression screening tool was used to measure depressive symptoms. This instrument has been specifically designed for use with Indigenous Australians in primary healthcare settings. 19

| Diabetes
Participants were defined as having diabetes if they self-reported being treated for diabetes or had a HbA1c greater than or equal to 6.5% based on blood pathology analyses. The definition of diabetes used in this study encompassed both Type 1 and Type 2. The presence of gestational diabetes (GDM) was not captured. However, given that T2DM is highly prevalent in this population, 16 it likely accounted for almost all of the diabetes cases. Participants who did not report being treated for diabetes and had a HbA1c ≥ 6.5% were considered as 'newly diagnosed diabetes'. Information on which medication participants with diabetes had been prescribed was obtained by reviewing participant medical records. Collection of this data was limited to participants from only one of the island sites. As a result, medication information was available for approximately half of the participants with diabetes and was not included in the analyses.

| Cognitive assessment
Cognition was assessed in the study using the Cogstate Brief Battery (CBB), using an iPad platform, predominantly by a member of the team who was a provisional psychologist (FT). The CBB was selected as the cognitive screening tool as it has high sensitivity to cognitive dysfunction associated with dementia and MCI. 20 The tool also has some usability, acceptability and validity among Indigenous Australians, [21][22][23] including Indigenous residents of the Torres Strait. 17 Further details on the appropriateness of the CBB and its administration during the current study are described elsewhere. 17 A description of the four CBB tasks and their 'built-in' Completion and Integrity Criteria is provided in Table S1. In brief, the four tasks

| Outcome measures
The primary outcome measures in this study were reaction time and accuracy on the CBB tasks. The Detection, Identification and One Back tasks provided three measures of reaction time in milliseconds and log 10 transformed milliseconds. The One Card Learning and One Back tasks provided two measures of accuracy as a percentage and an Arcsine transformed percentage. In addition, each participant had a z-score derived for each of these transformed measures, using normative data in provided by Cogstate PTY LTD. The normative data comprised means and standard deviations for the four CBB tasks for males and females combined by six age groups (18-34, 35-49, 50-59, 60-69, 70-79 and 80-89). Z-scores that were one standard deviation (SD) below the respective Cogstate mean were flagged using binary yes/no variable (ie z-score < 1SD), as were z-score that were two SDs below the mean (ie z-score < 2SD). In addition, scores that were 0.5 SD below the mean were also flagged, as this cut-off has been used elsewhere as a more sensitive measure of cognitive impairment. 7 For each participant, a 'cognitive speed' domain was derived (ie an average log-transformed reaction time measure from three tasks

| Statistical analyses
All analyses were undertaken using STATA 15 Statistical software (StataCorp. 2017, College Station, TX). Prior to statistical analysis, the distribution of the data was assessed for normality assumptions and outliers. The distribution of study variables (eg demographic and cardio-metabolic) by diabetes status was tested with independent-sample t tests and rank-sum tests for parametric and non-parametric continuous data, respectively, and with Chi2 tests for categorical data (Table S2). For the CBB outcome variables, the distribution of parametric measures (ie log-transformed reaction times and Arcsine transformed accuracy) was examined by diabetes status using means and linear regressions (Table S3). Nonparametric CBB outcomes (ie untransformed reaction time and accuracy measures, number of errors and continuous z-scores) were examined by diabetes status using medians and Wilcoxon rank-sum tests. Binary variables that identified whether a Z-score was one or two standard deviations below a Cogstate mean were examined by diabetes status using proportions and Pearson chi-squared tests.
Univariate regression analyses were used to examine the distribution of CBB outcome variables by study variables (eg demographic, cardio-metabolic and immune markers; Table 1). As age, education and iPad/Tablet experience were all associated with CBB outcomes and diabetes status at a univariate level, these factors were adjusted for in three multivariate regression models examining the relationship between diabetes and each CBB outcome (  diabetes status, is provided in Figure 1 and Figure S1. Interaction terms were used to analyse the effect of diabetes on One Back out-

| Univariate analyses
When performance on CBB tasks was examined in univariate analyses, participants with diabetes were significantly slower on all reaction time measures and had significantly lower accuracy on the One Back task (Table 1). Reaction times on the three measures increased as a function of age and decreased with years of education. Previous experience with an iPad/Tablet was associated with faster response times. Accuracy on One Card Learning task was not related to any study variables and these results are not tabled.

| Multivariate analyses
After adjusting for age, education and previous iPad/Tablet experience (Model 4), none of the three reaction time measures were significantly associated with diabetes status (  Figure 1A) and three age groups  the One Back task was higher in the diabetes group compared to the no diabetes group (3 and 2, respectively, p = .008) and the median z-score for accuracy on the One Back task was lower among participants with diabetes (−0.39 and 0.09, respectively, p = .002). The proportion of participants with a z-score ≤ 0.05 standard deviations below the age normative mean was also higher among people with diabetes (47.1% and 30.2%, respectively, p = .039; Table S3). There were no significant differences in z-score measures on the other tasks. Other cardiovascular risk factors, HbA1c, health behaviours, depression and gender were not related to Cogstate outcomes during univariate and/or multivariate modelling (Table 2).  There were no differences in speed measures between the participants with and without diabetes after adjusting for these variables.
The One Back task used in the current study is a measure of working memory, which is a cognitive ability underpinned by executive functioning. 24 Our results therefore correspond with research showing lower working memory 7 and executive function 6,25 among people with T2DM. As the reduction we observed was primarily among participants in their young to mid-adult life (ie 20-49 years), our results are most consistent with findings from a recent metaanalysis of 12 studies that showed working memory and executive functioning decrements are observable in midlife among people with T2DM. 9 Our study supports a growing consensus that cognition may be affected in early to mid-adulthood during T2DM and adds to existing knowledge by showing this may represent a potential complication of diabetes for Indigenous residents of the Torres Strait.
The magnitude of reduction observed in One Back accuracy was much lower than reported elsewhere. For example, while Pelimanni and Jehkonen 9 reported medium effect sizes, in our study, the effect was small. Diabetes status accounted for approximately 4% of the variance in One Back performance and participants with diabetes made, on average, one additional error on this task compared to participants without diabetes. This very modest reduction in performance would be considered, at most, a "diabetes associated cognitive decrement". 4 These are subtle decrements that may give rise to selfreported complaints, but are unlikely to affect social or occupational functioning or diabetes self-management. 26 These decrements have been noted to progress slowly over many years, 27,28 particularly in the context of chronically elevated blood glucose. 6,29 While this may eventually lead to deficits that are clinically or practically important for everyday functioning, 30 it remains unclear whether dementia represents an 'end point' on the same continuum. 4 Among older Indigenous residents of the Torres Strait, vascular risk factors such as diabetes have been identified as potential drivers of the elevated dementia rates. 14 The results of the current study may provide some evidence of the detrimental effect of a vascular risk early in this process. However, this proposition remains far from conclusive given the small effect size in our study and the broader uncertainty about the continuum between diabetes-associated cognitive decrements and later dementia.
The difference in One Back accuracy between participants with and without diabetes was most evident in younger age groups and became less pronounced with older age. There may be several study design reasons for this finding. First, Indigenous Australians have an earlier age of mortality, which has resulted in a healthy survivor effect in other research. 31,32 In our study, older participants with diabetes were relatively healthy compared to their younger diabetes counterparts in terms of cardio-metabolic indicators. Our results suggest a healthy survivor effect could extend to cognition, at least in terms of working memory. Our results may have also been influenced by selection bias, where older residents with diabetes and cognitive difficulties may have been less likely to self-select into a community health screen initially. We also excluded data from 14 participants who were unable to complete any of the CBB tasks, most of whom were older and had diabetes. Had these participants been retained and appropriately supported to undertake the CBB tasks, as noted elsewhere, 17 then poorer performance on the One Back task among people with diabetes may have also been observed in the older age groups.
In addition to study design factors, the lower performance of younger participants with diabetes on the One Back task may reflect their increased morbidity relative to their older counterparts.
Poorer glycaemic control in the younger group, as evidenced by higher HbA1c, was particularly evident. A systematic review of 86 articles indicated that high HbA1c had a weak negative association with cognitive function in older people (ie 51-85 years of age) with T2DM without dementia. 33 Our study adds to the body of knowledge by suggesting this effect may also be present among younger people who have both diabetes and notable indicators of morbidity.
Our limited sample size of 51 people with diabetes, unfortunately, prevented us from exploring this with further modelling.
In contrast to many studies, 9  The small sample size reduced our confidence in regression modelling with interaction terms for multiple age groups. While aggregating participants into three age groups improved this modelling, it prevented examining differences between the younger age groups. This study also did not differentiate by diabetes into Type 1 and 2 Diabetes and GDM. However, given that T2DM is particularly prevalent in this population, 16 it is likely this accounted for most of the diabetes cases.

| CON CLUS IONS
In a global context where diabetes remains an important risk factor for cognitive decline, our findings suggest that early and subtle decrements in working memory may be a potential complication of diabetes among Indigenous Australians living in the Torres Strait. In this population, which has elevated dementia rates linked to chronic disease, our results highlight the need for more preventative health resourcing. Our results suggest that early identification of younger people with diabetes, targeted education and supported glycaemic control could be important for protecting cognitive health.

ACK N OWLED G EM ENTS
The authors wish to thank the residents of the Torres Strait who participated in this study. This study was made possible by the con-

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are not publicly available due to privacy or ethical restrictions. The data are available on request from the corresponding author. Additional institutional approvals, such as ethics approval, would be required to enable sharing of these data.