Association of Combined Healthy Lifestyle Factors With Incident Dementia in Patients With Type 2 Diabetes

Standard Protocol Approvals, Registrations, and Patient Consents

The UK Biobank study was approved by the North West Multi-Centre Research Ethics Committee. All participants provided written informed consent for the study.

Study Design and Population

The UK Biobank is a prospective, population-based cohort of more than 500,000 participants aged 40–70 years recruited between 2006 and 2010.¹⁸ Participants attended 1 of 22 assessment centers across the United Kingdom, where they completed touchscreen and nurse-led questionnaires, had physical measurements taken, and provided biological samples. Analyses were restricted to individuals aged at least 60 years at baseline, given that dementia generally occurs at older ages. Participants with type 1 diabetes identified based on self-reported medical history and hospital inpatient records were excluded to capture primarily people with type 2 diabetes. We also excluded participants with missing data on glycated hemoglobin (HbA_1c), lifestyle variables, or covariates, as well as those with self-reported dementia at baseline or prevalent dementia diagnosis identified via hospital records (eFigure 1 in the supplement, links.lww.com/WNL/C329).

Healthy Lifestyle Score

We constructed a healthy lifestyle score based on 7 modifiable behavioral factors, including 4 traditional factors (smoking, physical activity, alcohol consumption, and diet) and 3 emerging factors (sleep duration, sedentary behavior, and social contact) assessed at baseline using a touchscreen questionnaire. Body mass index (BMI) was not included in the lifestyle score because weight change is a typical phenotype during diabetes progression and medical treatment could also result in weight change,¹⁹ and thereby, BMI may not serve as a valid measure for weight management. Details of the assessments of individual lifestyle factors can be found in eMethods in the supplement (links.lww.com/WNL/C329).

Participants scored 1 point for healthy category of each lifestyle factor on the basis of national recommendations if available. For smoking status, no current smoking was classified as healthy category. Regular physical activity was defined as at least 150 min/wk of moderate activity or 75 min/wk of vigorous activity or an equivalent combination.²⁰ In accordance with previous UK Biobank studies,^7,9 a healthy diet was defined as an adequate consumption of at least 4 of 7 food groups (an increased amount of fruits, vegetables, whole grains, and fish and a reduced amount of refined grains and processed and unprocessed meats) following recommendations on dietary priorities for cardiometabolic health (eTable 1 in the supplement, links.lww.com/WNL/C329).²¹ Given the reported U-shaped relationship between alcohol consumption and dementia risk,^22,23 healthy drinking was defined as moderate alcohol consumption with up to 1 drink (14 g)/day for women and up to 2 drinks (28 g)/day for men.²⁴ Likewise, because short and long sleep have been reported with a higher risk of dementia,^11,25 adequate sleep duration (7–9 h/d) was classified as healthy category. We used television watching as the proxy for leisure sedentary behavior and defined a healthy level as <4 h/d²⁶ The social contact index was assessed using information on the number in the household, frequency of friend/family visits, and participation in leisure/social activity.²⁷ Participants who were active and moderately active were considered as having frequent social contact. The lifestyle score ranged from 0 to 7, with a higher score indicating higher adherence to an overall healthy lifestyle. This score has been used in a recently published study among individuals with type 2 diabetes from the UK Biobank.¹⁵ Because of small numbers of participants in extreme groups with low scores (0, 1, and 2), these groups were combined into 1 category. In sensitive analysis, we further constructed a weighted lifestyle score considering effect size (β coefficient) for individual lifestyle factors derived from the adjusted model: weighted lifestyle score = (β1× factor1 + β2 × factor2 +…+ β7 × factor7) × (7/sum of the β coefficients).

Ascertainment of Outcome

Incident dementia was ascertained using linkage with hospital inpatient records and death registry data. Date and cause of hospital admissions were obtained from record linkage to Health Episode Statistics (England and Wales) and the Scottish Morbidity Records (Scotland). The information on death was provided by the National Health Service Digital (England and Wales) and the National Health Service Central Register (Scotland). Detailed information regarding the linkage procedure is available online.²⁸ Additional cases were retrieved from primary care records. Dementia was defined according to the ICD-9 and ICD-10 for Alzheimer disease, vascular dementia, and other dementia classifications (eTable 2, links.lww.com/WNL/C329).

Assessment of Prediabetes and Diabetes

The UK Biobank algorithms for identification of prevalent type 2 diabetes based on self-reported medical history and medication have been described.²⁹ Admissions and diagnoses data of the hospital inpatient records were used to ascertain type 2 diabetes with the ICD-9 (250.00, 250.10, 250.20, or 250.90) and ICD-10 codes (E11). Participants who had self-reported or hospital diagnoses of type 2 diabetes were identified as diagnosed patients. Undiagnosed patients were determined according to the American Diabetes Association (ADA) criteria as HbA_1c level ≥48.0 mmol/mol (6.5%).³⁰ Prediabetes was defined as the HbA_1c level of 39.0–47.0 mmol/mol (5.7%–6.4%) in participants without diabetes.

Covariates

Area-based socioeconomic status was derived from the postcode of residence using the Townsend deprivation index, a composite measure of data on car ownership, household overcrowding, owner occupation, and unemployment.³¹ Ethnicity and educational attainment were self-reported and coded as dichotomized variables. Prevalent noncommunicable diseases (hypertension, heart diseases, stroke, head injury, hyperlipidemia, depression, schizophrenia, Parkinson disease, chronic obstructive pulmonary disease, chronic asthma, chronic liver diseases, and cancer) were ascertained during a nurse-led interview at baseline. We calculated morbidity count as 0, 1, 2, or ≥3. Medication use for diabetes, cholesterol, and hypertension was collected by self-reported touchscreen questionnaire or verbal interviews. BMI was calculated as weight in kilograms divided by the square of height in meters. The HbA_1c level was measured by high-performance liquid chromatography using the VARIANT II Turbo analyzer (Bio-Rad Laboratories). APOE haplotypes were extracted from the genetic data and categorized into APOE ε4 carriers (any ε4 allele) and noncarriers. Diabetes duration was calculated from age at diagnosis of diabetes and age at baseline assessment for diagnosed patients and assumed 0 years for undiagnosed patients. Tests for cognitive function (reaction time test of symbol matching and visual pairs memory test) were completed through a touchscreen tool.

Statistical Analyses

Baseline characteristics of the analytic sample were summarized as percentage for categorical variables and mean (SD) or median (interquartile range) for continuous variables. Cox proportional hazards models were used to estimate the hazard ratios (HRs) and 95% CIs for incident dementia associated with diabetes status and healthy lifestyle score. Follow-up time was calculated from the baseline date to diagnosis of outcome, death, or the censoring date (July 30, 2021), whichever came first. We then examined the association of the combination of diabetes and healthy lifestyle score (12 categories with no diabetes and a lifestyle score of 7 as reference) with dementia. Statistical interaction between diabetes and healthy lifestyle score was tested by fitting an interaction term in the models. We also estimated dementia risk according to diabetes and the number of traditional and emerging lifestyle factors. Models were adjusted for age, sex, education, ethnicity, Townsend index (in quintiles), morbidity count, BMI, medication for hypertension and cholesterol, APOE ε4 carrier status, and cognitive performance at baseline. For analyses on diabetes status, models were additionally adjusted for lifestyle score and vice versa. The proportional hazards assumption was assessed using the Schoenfeld residuals method and satisfied.

The association between healthy lifestyle score and incident dementia was further investigated among participants with and without diabetes, respectively. Absolute risk over a 10-year period was calculated as the cumulative incidence of dementia occurring in a given group using competing risk regression. Among participants with diabetes, models were additionally adjusted for HbA_1c level (<53 or ≥53 mmol/mol [7.0%]), diabetes duration (<5, 5–9, or ≥10 years), and diabetes medication use (oral antidiabetic drugs only, insulin use, or neither). The dose-response relationship between healthy lifestyle score and dementia risk was tested using penalized splines. Stratified analyses among participants with diabetes were performed by age at diabetes diagnosis (<55 or ≥55 years), sex, education, BMI category (<30 or ≥30 kg/m²), APOE ε4 carrier status, morbidity count (≤1 or >1), HbA_1c level, diabetes duration, and diabetes medication. Furthermore, the population attributable fraction was calculated to estimate the proportion of individuals with incident dementia that would be attributable to lifestyle factors, assuming a causal relationship.

Several sensitivity analyses were conducted to test the robustness of the results. Residual confounding by socioeconomic status was evaluated by further adjusting for employment and household income. The models were repeated after excluding participants who developed dementia within the first 3 years of follow-up to reduce potential reverse causality or excluding participants with prevalent diseases. The analyses were also run after excluding participants with undiagnosed diabetes in case that substantial change in lifestyle might occur after the awareness of diabetes. Proportional subdistribution hazards regression models were used to account for competing risk of death. To address bias from complete case analysis, missing values for exposure and covariates were imputed by multiple imputation with chained equations. All statistical analyses were performed with SAS, version 9.4 (SAS Institute Inc), and R, version 4.1.1 (R Foundation). A 2-sided p value < 0.05 was considered statistically significant.

Data Availability

The data used in this current study are available from the UK Biobank data resources, which can be found online.³² Permissions are required to gain access to the data resources, subject to successful registration and application process.