Trajectories of Cognition and Daily Functioning Before and After Incident Diabetes

Diabetes constitutes a serious health burden affecting ∼537 million adults worldwide, projected to rise to 783 million by 2045 (1). Hyperglycemia predisposes individuals to cerebral small-vessel disease, which contributes to cognitive decline by causing neuronal ischemia, disturbances in the cerebral white matter, or aiding the leak of neurotoxic mediators (2). Since the increment of blood glucose starts years before the onset of diabetes (3), patients with diabetes might have accelerated cognitive decline before the diabetes diagnosis (4,5). However, previous research on the relationship between diabetes and cognitive dysfunction is mainly based on cohorts with diabetes and, therefore, can only use cognitive measurements after diabetes. Given that the accelerated cognitive decline is a prodromal feature of dementia that provides a time window for dementia prevention (6), understanding the trajectory of cognition before and after diabetes may help in the early identification of adults at risk for future dementia onset.

Hyperglycemia also causes systemic chronic inflammation and loss of skeletal muscle strength and quality, related to physical dysfunction (7). Patients with diabetes tend to experience a higher risk of physical disability (8). Besides, former studies reported that people at the prediabetes stage already have an increased risk of disability, even before progression to diabetes (9,10). To date, however, no studies have measured the changes in daily functioning occurring in the years before and after diabetes onset; thus, the pattern of its trajectory remains largely unknown. Given the substantial disability-free life-year loss for patients with diabetes (11), this evidence can help tailor the timing and design of future intervention strategies.

Therefore, using data from a nationally representative cohort, the English Longitudinal Study of Ageing (ELSA), we explored the magnitude of pre- and postdiabetes diagnosis change in cognition and daily functioning as well as acute change during diabetes diagnosis among older adults (≥50 years old).

We used data from the ELSA waves 2 to 9 (2004–2018). The detailed study design of this cohort has been described elsewhere (12). Briefly, the ELSA randomly sampled community-dwelling older adults (>50 years) living in England and performed the follow-up resurvey biennially. The London Multicenter Research Ethics Committee approved the ELSA, and all participants provided written informed consent.

The flowchart for participant selection of the current study is shown in Supplementary Fig. 1. The baseline survey (wave 2) recruited 9,432 participants, of which 3,090 were excluded for the following reasons: <50 years (n = 261), prevalent diabetes or unavailable glycemic status (n = 863), prevalent stroke (n = 361), prevalent dementia, Alzheimer or Parkinson disease (n = 86), unavailable cognition or disability assessment (n = 112) or covariates (n = 499) at baseline, or loss to follow-up (n = 908). Hence, 6,342 participants were finally included with complete baseline data and at least one reassessment of cognition and daily functioning during follow-up.

Structured questionnaires were administered by trained field workers. A computer-assisted personal interview system was used to collect demographic, lifestyles, and medical history. Education level was classified with the ELSA simplification of the 1997 International Standard Classification of Education as less than lower secondary education, upper secondary and vocational training, and tertiary education (13). BMI was calculated as body weight divided by the square of height (kg/m²). Blood pressure level was measured three times by nurses, with the means used, and hypertension was defined as a systolic blood pressure of ≥140 mmHg and/or a diastolic blood pressure of ≥90 mmHg, self-reported doctor-diagnosed hypertension, or use of blood pressure-lowering medication. Smoking status was recorded and classified into never, former, and current. Alcohol intake was calculated from participant-reported drinking frequency during the past 12 months (weekly drinking vs. occasional or never) (14). Physical activity was dichotomized into inactive (no moderate or vigorous activity on a weekly basis) and moderate-vigorous activity (more than once per week) (15). Depressive symptoms were measured using the 8-item version of the Center for Epidemiologic Studies Depression (CESD) Scale, and participants scoring >4 were defined as having depressive symptoms (16). Information about self-reported doctor-diagnosed heart disease, chronic lung disease, and cancer was also collected.

Cognition and daily functioning were assessed in each wave. Participants underwent a battery of three cognitive tests for the cognitive assessment, including orientation, memory, and executive function, with higher scores indicating better cognitive function. Orientation was assessed by asking questions regarding the date (year, month, day of month, and day of week) and season, and then allocated 1 point to each correct answer, with the sum score ranging from 0 to 5. Memory was determined by testing immediate and delayed recall for 10 unrelated words. The sum of words that were successfully recalled in these two was used as the composite memory score, ranging from 0 to 20. Executive function was assessed by asking participants to name as many animals as they could in 1 min, and the number of animal names was counted as the executive score. Both the reliability and validity of these tests have been well documented in the elderly population (14,17,18).

The z scores were calculated to allow direct comparisons across different cognitive tests. Specifically, we standardized the follow-up score by subtracting the mean of the baseline score and then dividing it by the baseline SD. The global cognitive z score was estimated by averaging the z scores from the three tests and then standardizing it to baseline using the mean and SD of the global cognitive z score. Therefore, a unit of z score would mean the 1 SD above the mean baseline score.

Daily functioning was assessed by the activities of daily living (ADL: walking across a room, dressing, bathing and showering, eating, getting in or out of bed, using the toilet) and instrumental ADL (IADL: using the phone, taking medications, managing money, shopping for groceries, preparing meals) (19). Each item was scored as 0 (did not report any problems with the activity) or 1 (some difficulty with the activity). The ADL and IADL scores were the sum of their components, and the level of disability was assessed as the sum of these two, ranging from 0 to 11, with higher scores indicating worse ability.

Diabetes was defined as self-reported doctor-diagnosed diabetes, current use of glucose-lowering medications, or glycated hemoglobin A_1c (HbA_1c) ≥6.5% (47.5 mmol/mol) (20). Data about self-reported doctor-diagnosed incident diabetes and glucose-lowering medication were collected in each wave. At the same time, the HbA_1c measurements were available in certain waves (waves 2, 4, 6, 8, and 9). The date of diabetes diagnosis was recorded as the middle date between the date of the last interview and that of the interview reporting an incident event.

Baseline characteristics are presented as mean (SD) or median (interquartile range) for continuous variables and frequency (percentage) for categorical variables. We used linear mixed models to analyze repeated measurements since this model can use all available follow-up data and accounts for the within-participant correlation. Linearity was visually explored with cubic splines for each exposure, with no evidence of deviation from linearity.

To investigate the before-diabetes mean difference in annual change of cognition and daily functioning between participants who did and those who did not have incident diabetes, we fitted the first model including incident diabetes (yes or no), time (years since baseline to incident diabetes or the end of follow-up), and time × diabetes interaction as fixed effects. The “time × diabetes” interaction term indicated a differential changing rate from baseline to incident diabetes or the end of follow-up. The second model was performed by additionally including the follow-up data after diabetes onset. The calendar time (years from baseline to the end of follow-up) was used to replace the time variable in the first model. We added a time-varying variable to evaluate the effect of incident diabetes on acute change in cognition and daily functioning (the value changed from 0 to 1 on the date of the incident diabetes). Then, we added a time after diabetes to evaluate the effect of incident diabetes on the postdiabetes decline of cognition and daily functioning during follow-up. This variable indicates the rate of change (slope) in cognition and daily functioning after incident diabetes compared with the before-diabetes stage. To evaluate the modifying effects of age and sex, the z test was used to compare the difference between subgroups (21). All models, unless specified, were adjusted for baseline covariates, including age, sex, BMI, education, marital status, depression symptoms, smoking status, alcohol consumption, physical activity, hypertension, chronic lung disease, heart disease, and cancer. More detailed descriptions of the statistical models are provided in the Supplementary Materials.

For demonstration purposes, we calculated participant-specific (conditional) values for cognition and daily functioning over time for a 65-year-old woman with the average values of all covariates at baseline (married, upper secondary education, BMI 27.5 kg/m², former smoker, ≤1 day/week alcoholic drink, moderate-vigorous activity level, prevalent hypertension, and without depressive symptoms, chronic lung disease, heart disease, or cancer) conditional on her experiencing or not experiencing an incident diabetes midway through the follow-up period (at year 7) (18). For this exemplar individual, we selected covariate values representative of the included population.

To test the robustness of the main findings, we performed the following sensitivity analyses: 1) restricting analyses of acute change after versus before diabetes and postdiabetes diagnosis change to 576 participants with incident diabetes; 2) restricting analyses to those with no incident stroke during follow-up; 3) restricting analyses to those with no incident Parkinson disease during follow-up; 4) repeating analyses by additionally adjusting for time-varying CESD scores; 5) repeating analyses of postdiabetes diagnosis change by additionally adjusting for the use of glucose-lowering medication; and 6) imputing missing data using multiple imputations.

Data were handled and analyzed with SPSS Statistics 25.0.0.1 (IBM Corp., Armonk, NY) and R, CRAN 4.1.2 software. All analyses were performed at the significance level of 0.05 (two-tailed), unless specified.

The original ELSA data sets are available at https://www.elsa-project.ac.uk/. The full data set used in this analysis is available from the corresponding author upon reasonable request.

Among the 6,342 participants (mean age, 65.0 [SD 9.4] years; 57.8% women), 576 (9.1%) incident diabetes events were identified during follow-up (median follow-up duration, 13.3 [interquartile range 6.4–14.0] years). The distributions of baseline characteristics for participants with and without incident diabetes are shown in Table 1. The available measurements for cognition and daily functioning in each wave are shown in Supplementary Tables 1 and 2. Compared with their diabetes-free counterparts, participants who experienced incident diabetes during follow-up tended to be men, had higher BMI, less alcohol consumption and physical activity, and were more likely to have depressive symptoms and hypertension at baseline.

As shown in Supplementary Table 3, the annual rate of cognitive decline before diabetes diagnosis in participants with future incident diabetes was similar to rate in participants who remained diabetes-free throughout follow-up, except for the memory function, which somehow showed a slower decline rate (β = 0.013 SD/year; 95% CI 0.002–0.024). We did not observe any significant acute cognitive change during diabetes onset (Table 2). While comparing the rates of cognition change between before and after the diabetes diagnosis, global cognition declined significantly faster in the years after diabetes onset than before the event (β = −0.035 SD/year; 95% CI −0.054 to −0.015). Similarly, the rates of change in orientation (β = −0.031 SD/year; 95% CI −0.060 to −0.002), memory (β = −0.016 SD/year; 95% CI −0.029 to −0.003), and executive function (β = −0.027 SD/year; 95% CI −0.042 to −0.013) were also accelerated after diabetes diagnosis (Table 2 and Fig. 1A). Subgroup analyses showed that after diabetes onset, older individuals had faster decline rates of global cognition and orientation, while no sex-specific difference was observed (Supplementary Tables 4 and 5).

Similar to the cognitive trajectory, there was no significant difference in the before-diabetes annual decline of daily functioning between participants with and without future diabetes (Supplementary Table 3). Those with incident diabetes also did not experience acute functioning decline during the diagnosis (Table 2). In the years after diabetes diagnosis, daily functioning declined significantly faster than before diabetes (β = 0.093 points/year; 95% CI 0.056–0.131). Likewise, the postdiabetes annual changes in ADL (β = 0.037 point/year; 95% CI 0.016–0.057) and IADL (β = 0.046 point/year; 95% CI 0.027–0.066) were also accelerated compared with the before-diabetes stage (Table 2 and Fig. 1B). Meanwhile, subgroup analyses showed that after its onset, older individuals had faster decline rates of daily functioning, while no sex-specific difference was observed (Supplementary Tables 4 and 5).

When we restricted the analyses to the 576 participants with incident diabetes, the postdiabetes diagnosis annual changing rates in cognition and daily functioning were still significant and with relatively larger effect sizes (Table 3). Similar results were observed when we restricted the analyses to those without incident stroke (Supplementary Table 6) and those without incident Parkinson disease during follow-up (Supplementary Table 7). In further analyses additionally adjusting for time-varying CESD scores (Supplementary Table 8), use of glucose-lowering medication after diabetes diagnosis (Supplementary Table 9), and analyses based on imputed data (Supplementary Table 10), results remained consistent with our main findings.

Among the 7,749 participants who had complete data at baseline, 1,407 (18.2%) were excluded because of missing covariates or loss of follow-up. Compared with the included participants, those excluded were older, less educated, had less alcohol consumption and physical activity, and had a higher prevalence of depressive symptoms, hypertension, heart problem, lung disease, and cancer at baseline (Supplementary Table 11).

Using data from a well-designed cohort of individuals >50, we found that incident diabetes was significantly associated with faster postdiabetes cognition and daily functioning decline but not with acute changes during the event. Compared with those who did not, those who experienced incident diabetes during follow-up had a similar changing rate of cognition and daily functioning before its onset.

The influence of incident diabetes on cognitive trajectory before the event has scarcely been investigated. A prior study based on the Chicago Health and Aging Project reported no significant difference in the cognitive slopes in the years before and after incident diabetes (22), which is consistent with our findings. In addition, we observed a slower rate of memory decline over the years before diabetes. Similar cognitive trends have also been reported by the Reasons for Geographic and Racial Differences in Stroke (REGARDS) study, which investigated the trajectory of cognitive decline before and after an incident stroke and reported significant increases in global cognition, new learning, and verbal memory over time before stroke (18). These aberrant associations may be attributable to practice effects due to repeated evaluation with the same test materials. Another possible explanation is that the natural change of cognition declined linearly between ages 45 and 65 years, followed by steeper declines after ages 65–70 years (23), while the REGARDS study and our study both used the linear mixed model, assuming the cognition declines linearly across different diagnostic phases. Therefore, for the event-free participants, the estimated cognitive slope among the entire follow-up duration might be overestimated at the younger stage.

Our study adds to previous research by describing the acute effect of incident diabetes on cognitive performance after the diagnosis. We did not find any acute cognitive decline during the onset of diabetes. This is reasonable; unlike any sudden cerebrovascular disease, such as stroke, the increase in blood glucose already occurs years before the diagnostic threshold of diabetes is reached (3,24). Therefore, the cerebral small-vessel damage induced by hyperglycemia might be subtle and progressive. Among participants with incident diabetes, cognitive decline after the diabetes diagnosis was significantly accelerated compared with its rate before the diagnosis. Although the 95% CIs of global cognition decline would meet the clinical threshold of cognition change (a decline of ≥0.5 SD) in ∼10 years since baseline, even minuscule cognitive function decline can result in a substantially increased risk of dementia over a long-term (25). Our findings support careful monitoring for cognitive impairment among patients with diabetes, especially after the diagnosis.

To the best of our knowledge, this study is the first prospective study to evaluate the trajectory of daily functioning before and after incident diabetes among the general population. Embedded in a nationally representative cohort with repeated functional assessments over a long follow-up, our study revealed a reliable and robust trajectory of functional change before and after diabetes onset. Although comparing with previous studies is difficult given the discrepancies in the study design, our finding is, generally, consistent with former studies reporting a significant trend for accelerated functional decline after the diabetes diagnosis (10,26). Besides, our study adds to previous research by investigating the change in daily functioning before the diagnosis and acute change after versus before diabetes onset. The rate of functional decline before the diabetes diagnosis in participants with future incident diabetes was similar to those in participants without diabetes, and no significant acute change was observed during its onset. This is reasonable given that the increment of blood glucose before the onset of diabetes tends to be gradual and might be insufficient to cause a significant functional decline in the short-term. However, it is also possible that the measurements of daily functioning are not sensitive enough to detect the minor deficits present during its onset.

Elevated depressive symptoms are often seen in adults with diabetes. The combined association of depression and diabetes with the risk for all-cause dementia is even stronger than the additive association (27). Besides, older adults with newly diagnosed diabetes and elevated depressive symptoms have a clinically meaningful and faster disablement trajectory than those without elevated depressive symptoms (28). Here we observed smaller changing rates for the postdiabetes functional decline when adjusting for time-varying CESD scores. Depression is also related to poorer glucose control and worse medication adherence (29), which could, in turn, affect cognitive and functional health. Although our results remain robust after further adjusting for glucose-lowering medication during follow-up, former studies reported that the association between treatment for diabetes and dementia is differential according to drug class, potentially mediated by hypoglycemic risk (30). Future studies with detailed data on glucose-lowering medication use and hypoglycemic events are needed to untangle their associations further.

The underlying mechanisms of linking diabetes to cognitive and physical dysfunction remain unclear. Subtle diabetes-related brain changes, such as disturbed white matter integrity and vascular lesions, could accumulate onwards, thus reducing the reserve capacity of the brain (5). Diabetes is also associated with certain atherosclerotic diseases, such as stroke, as important contributors to cognitive decline and physical disability (7,31). Additionally, insulin resistance and chronic inflammation underpin diabetes-related dysfunction (7,32,33), another important mechanism involving the shared risk factors. Many prevalent risk factors, such as obesity, hypertension, and depression, are not only associated with diabetes incidence but also contribute to cognitive and functional decline (34–38).

A major strength of our study is the large, well-designed population-based cohort with repeated outcome measurements that enabled us to generate a reliable and accurate trajectory of cognition and daily functioning. The long follow-up duration and relatively short follow-up interval allowed us to explore participants’ cognitive and functional decline before diabetes diagnosis and to calculate the acute change during diagnosis. Taken together, our study filled in a specific knowledge gap about the trajectory of cognition and daily functioning among people with diabetes.

Our study, however, has some limitations that should be acknowledged. First, the ELSA mainly included White participants, limiting our generalizability of our findings to other races/ethnicities.

Second, although we excluded participants with prevalent dementia and Alzheimer disease at baseline, data about cognitive impairment were not available. This might overestimate our findings, given that individuals with impaired cognitive status experience a higher risk of incident diabetes and have accelerated cognitive and functioning decline (39,40).

Third, although the moderately nonlinear changes of cognition and motor function were suggested among the general population (23), here we found no evidence of deviation from linearity, which might be due to the different measurements we used.

Fourth, only questionnaire-based outcome measurements were used here, which may not be sensitive enough to detect the minor deficits before diabetes onset. Prediabetes is also related to cognitive and functional decline (4,10). However, limited by the data availability, we cannot address these questions. Future studies using more comprehensive neuropsychological and functional assessments and detailed prediabetes information are needed.

Finally, only those with complete baseline information and at least one repeated measurement were eligible for the current study, leading possibly to selection bias. Results from the nonresponse analysis show that the included participants were relatively healthier than those excluded. However, as our results using multiple imputations remain robust, this might not significantly bias our findings.

In summary, our study suggests that incident diabetes was significantly associated with faster postdiabetes-diagnosis decline of cognition and daily functioning, but not prediabetes diagnosis or acute change during the event. Further studies are needed to determine the mechanisms linking incident diabetes to cognitive decline and physical disability.

Acknowledgments. The authors thank the staff and the participants of the ELSA study.

Funding. This work was supported by the Research Project of Changning District Health Committee of Shanghai Municipality, China (20214Y032) to H.G. and the Domestic Cooperation Project of Science and Technology Commission of Shanghai Municipality, China (20015800300) to D.S. K.W. was supported by a scholarship from the China Scholarship Council.

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. X.J. and H.G. wrote the manuscript. X.J., H.G., K.W., and F.A. were responsible for the study concept and design. K.W. composed the statistical data set and performed the statistical analyses. D.S., J.Z., and Y.F. revised the manuscript critically for intellectual content. K.W. is the guarantor of this work and has full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.