Aims/hypothesis The aim of this study was to examine whether cognitive function in early and later life, and decline in cognitive function from age 70 to 79 years, are associated with high blood glucose, as measured by HbA , atbaseline(age70),and 1c changes in blood glucose from age 70 to 79. Methods Participants (n = 1091) in the Lothian Birth Cohort of 1936 were examined. Fourteen tests were used to assess cognitive functions, grouped into four domains: visuospatial ability, processing speed, memory and crystallised ability. Test results, and measurements of HbA and other health variables, were collected at each of four waves of assessment: at the mean age of 70, 73, 1c 76 and 79 years. Data on cognitive function at age 11 was also available for this cohort. Latent growth curve modelling was performed and statistical controls for known risk factors were introduced. Results Higher age 11 cognitive function predicted lower HbA level at age 70 (p < 0.001). Higher cognitive function at age 70 1c was related to a comparatively smaller increase in HbA levels from age 70 to 79 (p < 0.001). HbA from age 70 to 79 did not 1c 1c have any consistent association with change in cognitive function from age 70 to 79. These associations survived adjustments for age, sex, education, APOE*ε4, smoking history, cardiovascular disease history, hypertension history, BMI and corrections for multiple testing. Conclusions/interpretation Our results show that, among older individuals, high blood glucose is consistently predicted by lower cognitive function. Clinical care that examines and tracks cognitive function, while also taking the positive effects of maintaining cognitive function and emulating healthy behaviours associated with higher cognitive function into account, may be one approach for protecting at-risk individuals from elevated blood glucose and subsequent type 2 diabetes mellitus. . . . . . Keywords Blood glucose Cognitive decline Cognitive function HbA Older age Type 2 diabetes 1c Abbreviations LGCM Latent growth curve model CVD Cardiovascular disease WAIS Wechsler Adult Intelligence Scale LBC1936 Lothian Birth Cohort of 1936 WMS Wechsler Memory Scale Electronic supplementary material The online version of this article (https://doi.org/10.1007/s00125-018-4645-8) contains peer-reviewed but unedited supplementary material, which is available to authorised users. Introduction * Drew M. Altschul firstname.lastname@example.org As the prevalence of type 2 diabetes mellitus has grown world- wide, so has the need to understand related health implications Department of Psychology, The University of Edinburgh, 7 George such as the link between diabetes and cognitive function. Type Square, Edinburgh EH8 9JZ, UK 2 diabetes is associated with an increased risk of cognitive Centre for Cognitive Ageing and Cognitive Epidemiology, The decline and dysfunction, including dementia . Diabetes- University of Edinburgh, Edinburgh, UK associated cognitive impairment is expected to become more Geriatric Medicine Unit, Western General Hospital, Edinburgh, UK prevalent as life expectancy for individuals with diabetes Diabetologia increases worldwide, highlighting a need to understand the re- variables, cognitive and health-related, in the same individuals ciprocal dynamic progression of type 2 diabetes and cognitive across discrete waves of assessment. HbA is a useful marker 1c decline. The present study investigates the relationships be- for studying the development of diabetes in this way, as it has tween cognitive change and HbA measurements, one of the good specificity and sensitivity to detect type 2 diabetes and 1c most important risk metrics in type 2 diabetes. also measures the severity of the disease [6, 14–16]. Impaired glucose control, confirmed by HbA measure- This study’s aim was to characterise, across the eighth de- 1c ment, is present in and predicts the development of type 2 cade of life, parallel changes in cognitive functions and high diabetes, even though the syndrome may be asymptomatic blood glucose (measured by HbA ), and to investigate asso- 1c and go undetected for years . High HbA is also associated ciations between baseline levels and change in these measures 1c with cognitive decline in older populations, in both healthy over time. Our study used four waves of data from the Lothian individuals  and those diagnosed with type 2 diabetes [4, 5]. Birth Cohort of 1936 (LBC1936), a narrow age cohort of over Additionally, HbA levels are associated with micro- and 1000 community-dwelling people tested in four waves from 1c macrovascular complications [2, 6], reduction in brain volume age 70 to 79 years. LBC1936 also provides cognitive function [7, 8], and dementia . The HbA –cognitive function rela- data from age 11, as well as other control variables. This 1c tionship could be causative: the toxic generation of free radi- sample allowed us to assess two non-mutually exclusive hy- cals which accompanies increased HbA may cross over into potheses: higher HbA at age 70 is associated with relatively 1c 1c the brain, affecting cognitive functions . greater cognitive decline from age 70 to 79 [17–19]; and the Previous work on cognitive functioning and diabetes has reverse causative hypothesis, i.e. that lower initial cognitive focused on controlling for known risk factors such as vascular function and relatively greater cognitive decline are associated disease and the APOE*ε4 allele, as well as identifying indepen- with subsequently higher HbA levels [13, 20]. 1c dent risk factors such as glucose peaks  and insulin resis- tance  that could be linked to cognitive decline. However, whereas type 2 diabetes and high blood glucose are associated Methods with poorer cognitive function later in life, evidence for reverse causation also exists, suggesting that lower prior cognitive Study population LBC1936 is a community-dwelling sample of function might contribute to type 2 diabetes aetiology . 1091 initially healthy individuals. All were born in 1936 and Many samples are unable to address these types of questions were followed up in four waves of one-to-one cognitive and because they lack a longitudinal design that tracks relevant health testing between 2004 and 2017, at mean ages 70, 73, 76 Diabetologia and 79 years. Details on the background, recruitment and data and genetic testing: high BP history (no/yes), cardiovascular collection procedures are available [21, 22]. Participants provid- disease (CVD) history (no/yes), smoking history (current/for- ed written informed consent. Ethics permissions were obtained mer/never), and the presence of an APOE*ε4 allele. The third from the Multicentre Research Ethics Committee for Scotland set contained only BMI (z scored), recorded at every testing (wave 1, MREC/01/0/56), the Lothian Research Ethics wave and introduced to our analyses separately from the other Committee (wave 1, LREC/2003/2/29) and the Scotland A cardiovascular variables because BMI is strongly associated Research Ethics Committee (waves 2–4, 07/MRE00/58). with HbA and can have a confounding effect. 1c Cognitive function When they were approximately 11 years Statistical analyses All four LBC1936 waves were analysed in old, 1028 members of the LBC1936 sat Moray House Test latent growth curve models (LGCMs), a structural equation No. 12, a broad cognitive ability test that included word clas- modelling technique that allows the user simultaneously to sification, proverbs, spatial items and arithmetic. The test cor- define and analyse multiple latent and measured variables related about 0.8 with the Terman–Merrill revision of the . For each variable modelled longitudinally, LGCMs Stanford–Binet test, providing concurrent validity . model level (i.e. an intercept) and slope (i.e. a trajectory of In older age, cognitive function is known to decline across change) variables. We modelled cognitive function using a multiple domains . Age effects also act on overall cognitive hierarchical ‘factor of curves’ model, previously established function, a superordinate factor of the lower function domains with these data . For each cognitive test we modelled a . We included tests to measure three important cognitive level (essentially the age 70 baseline) and a linear slope (the function domains that decline with age: visuospatial ability, change between age 70 and 79, taking all four measurement processing speed and memory. We also tested crystallised abil- occasions into account); for each cognitive domain (see ity, which remains relatively stable in later life . All four Cognitive function, above) a latent level and linear slope var- domains also contribute to overall cognitive ability . These iable were correspondingly composed of the individual tests’ relationships among cognitive tests and domains are described latent level and latent slope variables. At the top of this cog- in the Statistical Analyses section below. nitive hierarchy, latent levels and slopes for overall cognitive Cognitive functions in waves 1–4were assessed using14 function were formed from the level and slope variables of individually administered cognitive tests at the same clinical each cognitive domain. For HbA variables, the same ap- 1c research facility and using the same equipment and procedure proach was taken, but only a pair of level and slope variables for all four waves. The tests are fully described and referenced were defined for this outcome variable, as no hierarchical in an open-access protocol article . The visuospatial do- structure exists. A structural diagram to illustrate these mea- main consisted of matrix reasoning and block design from the sured and latent variables is presented in Fig. 1.We also in- Wechsler Adult Intelligence Scale (WAIS) , and spatial vestigated quadratic slope variables for both cognitive func- span forward and backward from the Wechsler Memory tion and HbA , but models including quadratic components 1c Scale (WMS) . Processing speed was measured through either would not converge successfully, or they fit very poorly symbol search and digit symbol substitution from the WAIS, and could not be trusted to produce reliable estimates. plus four-choice reaction time  and inspection time [30, Put simply, HbA level is analogous to blood glucose at 1c 31]. Memory was assessed using verbal paired associates and the study’s beginning (age 70 years), and HbA slope is 1c logical memory from the WMS , and the letter–number analogous to the magnitude at which blood glucose increases sequencing and digit span backward subtests of the WAIS over the following decade. Cognitive function level is analo- . Crystallised ability was measured through the National gous to cognitive function at age 70 and is known to be highly Adult Reading Test , Wechsler Test of Adult Reading  correlated with cognitive function at age 11 , demonstrat- and a phonemic verbal fluency test . ing the stability of cognitive function across a lifetime. The slope of cognitive function represents the rate at which cogni- Clinical diabetes measures For each individual during each tive functions change (mostly decrease, in fact) over subse- wave, HbA concentrations were measured using a quent waves, which is otherwise known as cognitive decline. 1c Menarini HA-8160 analyser (Wokingham, UK). Diabetes di- Control covariates that applied to all waves of data (sex, age agnosis status was independently recorded. Twelve individ- 11 cognitive function, years of education, APOE*ε4 status and uals with type 1 diabetes were excluded from further analyses. smoking history) were regressed on to the intercept and slope variables for latent levels and slopes, including those for the Covariates Three variable sets were evaluated as potential domains of cognitive function, but not on to the levels or slopes confounders. The first set defined our baseline control covar- of individual cognitive tests. Control covariates that were iates: sex, age (z scored), age 11 cognitive function (z scored), related to physical condition and could differ between waves and years of education (z scored). The second set included the (age, history of high BP, history of CVD, and BMI) were available cardiovascular risk factors, obtained from interview regressed on to the individual HbA measurements. 1c Diabetologia HbA HbA HbA HbA 1c 1c 1c 1c Time− Time− invariant Age 70 Age 73 Age 76 Age 79 varying covariate covariate 9.79 1 6.72 1 2.96 1 0 HbA HbA 1c 1c level slope Overall Overall level slope Domain 1 Domain 1 Domain 2 Domain 2 level slope level slope Test A Test A Test B Test B Test C Test C Test D Test D level slope level slope level slope level slope 1 0 1 0 1 0 1 0 1 1 1 1 2.96 2.96 2.96 2.96 1 6.72 1 6.72 1 6.72 1 6.72 1 9.79 1 9.79 1 9.79 1 9.79 Test A Test A Test A Test A Test B Test B Test B Test B Test C Test C Test C Test C Test D Test D Test D Test D Age 70 Age 73 Age 76 Age 79 Age 70 Age 73 Age 76 Age 79 Age 70 Age 73 Age 76 Age 79 Age 70 Age 73 Age 76 Age 79 Fig. 1 Simplified path diagram of the hierarchical factor of curves growth indicate the relationships between covariates that were fixed across waves model of cognitive function and HbA . Circles represent latent variables and the overall and domain latent variables. Dotted lines indicate the 1c and squares represent measured variables. Growth curves, including a relationships between covariates that differed between waves and indi- latent level and slope factor, were estimated for each cognitive test, and vidual measurements at different waves. To preserve interpretability, not the intercepts and slopes were analysed in a hierarchical model which all covariate relationships are shown. Solid double-arrowed lines indicate contained domain and overall factors of both level and slope. Basis coef- correlations between latent variables. Solid single arrowed lines indicate ficients (loadings on the individual test slopes) were fixed at 0, 2.96, 6.72 the variables that load on latent variables. Although only two domains and 9.79 to precisely represent the amount of time passing between as- and four tests are shown here, the full model used all 14 tests in the four sessments. All loadings on latent level were fixed at 1. Dashed lines domains, as described in Methods Model coefficients were estimated using full information Table 1. Electronic supplementary material (ESM) Table 1 maximum likelihood, i.e. all data were used for all partici- presents wave 1–3 descriptions of study completers only, i.e. pants, even individuals who did not complete all waves. those individuals who remained present in wave 4. Standard errors were calculated using the robust Huber– Individuals with type 2 diabetes were more likely than White method, and p values were computed using the Yuan– healthy individuals to have a history of CVD or high BP Bentler scaled test statistic . The false discovery rate cor- (Table 1). The type 2 diabetes group included more men, as rection for multiple testing was applied to the variables in each well as more current and former smokers. Individuals with model that were not control covariates, which included the type 2 diabetes had higher BMI and HbA levels; in all waves 1c associations between the latent variables of cognitive function their mean HbA was above the diagnostic threshold of 1c and its subdomains and HbA . All analyses were conducted 48 mmol/mol (6.5%). Individuals with type 2 diabetes had 1c in the R programming language, version 3.3.2 (R Foundation lower age 11 cognitive function and scored more poorly than for Statistical Computing, Vienna, Austria), using the latent healthy individuals on all cognitive tests across all four waves. variable analysis (lavaan) package for modelling . To illustrate cognitive decline across domains, we plotted the discrete overall cognitive function and domain scores across all four waves, with individuals categorised as either healthy or Results showing a clinical or biochemical (HbA above diagnostic 1c threshold) type 2 diabetes diagnosis at wave 1 (Fig. 2). Figure Cohort characteristics Demographic and clinical characteris- 2 represents study completers only, because including data from tics of the study population at all four waves are presented in all study participants would bias the means of later waves in the Diabetologia Table 1 Descriptive statistics for cognitive, demographic and clinical variables Variable Wave 1 (n = 1091) Wave 2 (n = 866) Wave 3 (n =697) Wave 4 (n =548) T2D No T2D T2D No T2D T2D No T2D T2D No T2D Participants, n (%) 76 (6.97) 1015 (93.0) 84 (9.70) 782 (90.3) 75 (10.8) 622 (89.2) 66 (12.0) 482 (88.0) Female, n (%) 27 (2.47) 516 (47.3) 28 (3.23) 390 (45.0) 24 (3.44) 313 (44.9) 26 (4.74) 247 (45.1) Ever a smoker, n (%) 51 (4.67) 539 (49.4) 50 (5.77) 401 (46.3) 46 (6.60) 301 (43.2) 39 (7.12) 212 (38.7) Ever had CVD, n (%) 32 (2.93) 236 (21.6) 35 (4.04) 215 (24.8) 34 (4.88) 202 (29.0) 25 (4.56) 179 (32.7) Ever had high BP, n (%) 56 (5.13) 377 (34.6) 58 (6.70) 367 (42.4) 57 (8.18) 321 (46.1) 51 (9.31) 264 (48.2) 30.5 (4.52) 27.6 (4.28) 30.1 (5.04) 27.7 (4.32) 29.6 (4.76) 27.5 (4.39) 29.7 (5.52) 27.1 (4.34) BMI, kg/m HbA , mmol/mol 59.0 (13.7) 40.0 (6.08) 53.1 (9.71) 37.9 (5.01) 53.2 (9.30) 39.4 (4.95) 55.1 (11.6) 38.4 (4.30) 1c HbA , % 7.5 (3.4) 5.8 (2.7) 7.0 (3.0) 5.6 (2.6) 7.0 (3.0) 5.8 (2.6) 7.2 (3.2) 5.7 (2.5) 1c Years of education 10.5 (1.09) 10.7 (1.13) 10.4 (1.02) 10.8 (1.15) 10.5 (1.06) 10.8 (1.15) 10.5 (1.09) 10.9 (1.18) Age 11 cognitive function 94.4 (16.8) 100.0 (14.8) 96.5 (17.1) 101.0 (15) 96.7 (16.9) 102.1 (15.0) 96.5 (17.2) 103.0 (14.9) Visuospatial ability domain Matrix reasoning 11.9 (5.28) 13.6 (5.10) 12.1 (4.51) 13.3 (5.00) 12.0 (5.06) 13.2 (4.88) 12.0 (4.78) 13.0 (5.06) Block design 32.3 (11.0) 33.9 (10.3) 30.3 (9.18) 34.0 (10.1) 30.6 (8.95) 32.4 (10.1) 29.9 (8.68) 31.4 (9.75) Spatial span 7.07 (1.65) 7.55 (1.44) 6.65 (1.70) 7.11 (1.59) 6.80 (1.74) 7.08 (1.57) 6.71 (1.48) 6.74 (1.62) Crystallised ability domain NART 31.1 (9.63) 34.7 (7.98) 31.9 (8.70) 34.6 (8.08) 32.5 (8.76) 35.3 (7.89) 34.2 (9.38) 35.8 (8.01) WTAR 37.3 (8.96) 41.3 (6.95) 38.8 (7.49) 41.2 (6.87) 39.0 (7.67) 41.3 (6.91) 40.2 (8.06) 41.8 (6.87) Verbal fluency 40.1 (13.9) 42.6 (12.4) 39.4 (12.6) 43.6 (12.9) 40.1 (13.2) 43.2 (12.7) 41.7 (14.2) 43.9 (13.2) Memory domain Logical memory 68.2 (19.0) 71.7 (17.9) 70.2 (19.6) 74.7 (17.6) 71.9 (20.6) 74.9 (19.0) 71.4 (22.6) 72.9 (20.1) VPA 25.4 (9.08) 26.5 (9.14) 25.8 (9.52) 27.3 (9.45) 24.3 (8.87) 26.7 (9.61) 26.7 (9.96) 27.2 (9.51) Digit span 7.30 (2.47) 7.77 (2.24) 7.27 (2.10) 7.87 (2.30) 7.19 (2.39) 7.84 (2.36) 7.08 (1.79) 7.62 (2.22) LNS 10.2 (3.58) 11.0 (3.12) 10.3 (3.07) 11.0 (3.07) 9.7 (2.99) 10.6 (2.98) 9.4 (2.82) 10.2 (2.90) Processing speed domain Symbol search 22.3 (7.45) 24.9 (6.27) 23.2 (6.49) 24.8 (6.13) 23.7 (5.89) 24.7 (6.52) 21.6 (6.37) 22.9 (6.66) Digit–symbol coding 50.8 (12.0) 57 (12.9) 51.3 (13.2) 56.9 (12.1) 49.5 (13.6) 54.3 (12.8) 48.2 (13.5) 51.6 (12.9) Inspection time 110 (10.4) 112 (11.0) 108 (11.5) 112 (11.8) 108 (13.0) 110 (12.5) 104 (14.3) 107 (13.5) Reaction time 0.676 (0.107) 0.640 (0.084) 0.687 (0.122) 0.645 (0.085) 0.700 (0.116) 0.676 (0.101) 0.706 (0.112) 0.706 (0.114) MMSE 28.1 (2.00) 28.8 (1.37) 28.4 (1.93) 28.8 (1.35) 28.1 (1.74) 28.7 (1.68) 28.2 (1.86) 28.6 (2.08) Data are presented as n (%) or mean (SD) LNS, letter–number sequencing; MMSE, Mini Mental State Examination; NART, National Adult Reading Test; T2D, type 2 diabetes, defined as self- reported physician diagnosis of diabetes; VPA, visual paired associates; WTAR, Wechsler Test of Adult Reading positive direction, making plots unrepresentativeofthe sample. slope estimates are also presented in Fig. 3, whereas disease- As expected, with the exception of crystallised ability, the relevant effects of control variables, measured at each wave, curves show clear declines across all other cognitive domains are presented separately in ESM Table 2. Model fit statistics of in individuals with and without type 2 diabetes; however, cog- all LGCMs indicated a good fit (root mean square error of nitive function was worse in individuals with type 2 diabetes. approximation [RMSEA] <0.33; standardised root mean re- Figure 2 is for illustration purposes only. sidual [SRMR] <0.62; comparative fit index [CFI] >0.939; Tucker–Lewis index [TLI] >0.936; see ESM Table 3 for com- Age 70–79 cognitive performance and HbA Figure 3 shows plete fit statistics). 1c the relationships between the level and slope estimates of Age 11 cognitive function made significant contributions HbA and overall cognitive function. Three models are pre- to both cognitive function level (β > 0.4, coefficient of partial 1c 2 2 sented, with control variables added cumulatively, as de- determination ρ >0.35) and HbA level (β > −0.066, ρ > 1c scribed in Methods. The diagrams present the results for over- 0.007) at age 70; that is, higher childhood cognitive function all cognitive function; domain-specific results can be found in was associated with higher cognitive function and lower ESM Table 2. The effects of control variables on the level and HbA concentration at age 70. More education was also 1c Diabetologia Fig. 2 Trajectories of overall and domain-specific cognitive 0.4 abilities in individuals with and without type 2 diabetes in the Lothian Birth Cohort 1936. Diabetes was defined by reported physician diagnosis or HbA 1c level >48 mmol/mol (6.5%) at wave 1. (a) Overall cognitive ability, composed of the four -0.4 specific cognitive domains: visuospatial ability (b), processing speed (c), memory (d) and crystallised ability (e). Red line and 95% confidence region, -0.8 individuals without diabetes; blue 70 73 76 79 line and 95% confidence region, Mean age (years) individuals with diabetes. These plots only include data from bc individuals with data available from all four waves of the Lothian 1 1 Birth Cohort 1936 study, since including data from all 0 0 individuals in these plots would bias the means towards a positive direction at older ages, making -1 -1 them unrepresentative of the whole sample -2 -2 70 73 76 79 70 73 76 79 Mean age (years) Mean age (years) de -1 -1 -2 -2 70 73 76 79 70 73 76 79 Mean age (years) Mean age (years) associated with better cognitive function in older age, though remain in the presence of all control covariates. The associa- to a lesser degree (β >0.12, ρ > 0.026). These findings are tion between cognitive function level and HbA slope was as 1c consistent with those previously reported [13, 20, 38]. large as r = −0.166 (ρ = 0.03), which was found after control- The relationship between cognitive function level at age 70 ling for the influence of age 11 cognitive function and educa- and cognitive function slope was robust and did not change tion. The relationship between cognitive function slope and 2 −6 when covariates were added (r =0.002, ρ =4 × 10 ), but the HbA slope, while consistently positive and significant, was 1c 2 −5 association was small enough that we can only infer that cog- small: at most it was r = 0.008 (ρ = 6.4 × 10 ) in the first nitive function level and slope have no strong relationship. model (Fig. 3a), which includes the fewest controls. This is likely due to low covariance between cognitive func- Associations between HbA level and cognitive function 1c tion level and slope; or, in other words, regardless of an indi- slope were significant but were not consistent across models vidual’s starting cognitive function level, cognitive function and were relatively small. Moreover, the association between will consistently decline over the course of the eighth decade. cognitive function level and HbA level was not always sig- 1c Significant associations of the same sign were found be- nificant, nor was it in the same direction across models. We tween HbA slope and both level and slope of cognitive found no significant associations between HbA level and 1c 1c function in models. These results indicate that individuals slope in any of these models. with higher cognitive function at age 70 maintain lower blood Associations between HbA growth curve estimates and 1c glucose over the following decade, and these associations cognitive function at the domain level generally followed the Memory score Visuospatial ability score Overall cognitive function score Crystallised ability score Processing speed score Diabetologia -0.080 (0.024) CF level 0.002 (0.001) -0.010 (0.002) CF slope Cognitive Cognitive Sex -0.004 (0.059) HbA level function function 1c -0.009 (0.006) HbA slope level slope 1c -0.166 (0.005) 0.554 (0.020) CF level Age 11 † † † -0.005 (0.001) CF slope 0.080 0.008 cognitive † -0.111 (0.034) HbA level (0.012) (0.001) 1c function -0.001 (0.003) HbA slope 1c 0.018 (0.002) 0.176 (0.016) CF level Years HbA HbA 1c 1c -0.021 (0.003) CF slope of level slope -0.018 (0.030) HbA level 1c education 0.002 (0.008) 0.000 (0.003) HbA slope 1c -0.176 (0.024) CF level CF level -0.080 (0.028) 0.002 (0.0005) 0.010 (0.002) CF slope CF slope -0.001 (0.002) Cognitive Cognitive Sex APOE 4 -0.005 (0.059) HbA level HbA level -0.071 (0.061) function function 1c 1c -0.008 (0.006) HbA slope HbA slope 0.014 (0.005) 1c level slope 1c -0.013 (0.001) 0.424 (0.019) CF level Age 11 † 0.012 (0.002) CF slope 0.004 0.0005 cognitive † -0.095 (0.033) HbA level (0.012) (0.0001) 1c function -0.003 (0.005) HbA slope 1c -0.009 (0.001) 0.128 (0.016) CF level CF level -0.043 (0.018) Years HbA HbA 0.016 (0.002) CF slope 1c 1c CF slope -0.012 (0.002) Smoking of level slope -0.010 (0.030) HbA level HbA level 0.106 (0.043) history 1c 1c education 0.001 (0.003) HbA slope -0.002 (0.008) HbA slope -0.014 (0.005) 1c 1c -0.090 (0.024) CF level CF level -0.087 (0.028) 0.002 (0.0004) -0.003 (0.002) CF slope Cognitive Cognitive CF slope -0.002 (0.002) Sex APOE ε4 0.014 (0.057) HbA level HbA level -0.073 (0.060) function function 1c 1c -0.007 (0.006) HbA slope HbA slope 0.002 (0.006) level slope 1c 1c -0.004 (0.001) 0.400 (0.018) CF level Age 11 -0.001 (0.001) CF slope † † -0.033 0.002 cognitive -0.066 (0.033) HbA level 1c (0.012) (0.0002) function -0.003 (0.003) HbA slope 1c -0.008 (0.001) 0.121 (0.016) CF level CF level -0.051 (0.018) Years HbA HbA 1c 1c -0.005 (0.001) CF slope CF slope 0.004 (0.001) Smoking of level slope 0.012 (0.029) HbA level HbA level 0.118 (0.042) history 1c 1c education -0.001 (0.003) HbA slope -0.001 (0.008) HbA slope -0.012 (0.005) 1c 1c Fig. 3 Path diagram of relationships between level and slope estimates of covariate in ESM Table 2). Numbers presented are standardised estimates overall cognitive function (CF) and HbA in the Lothian Birth Cohort with SEM in parentheses. Double-headed arrows between latent variables 1c 1936 from age 70 to 79 years. In (a), the model includes the covariates (circles) indicate correlations. Single-headed arrows from covariates shown, as well as age. In (b), the model includes the covariates shown, as (rectangles) indicate regression effects on latent variables. Dashed lines well as age and history of CVD and high BP, history of smoking and the indicate non-significant correlations, whilst solid lines and indicate sig- presence of an APOE*ε4 allele. In (c), the model includes all the covar- nificance; The false discovery rate correction for multiple testing was iates shown, age, and history of CVD, history of high BP, smoking and applied to all tested relationships. Additional estimates for domains and APOE*ε4, as well as BMI (the latter is demonstrated as a time-varying controls are presented in ESM Table 2 associations found with overall cognitive function level and We conducted sensitivity analyses on our three structural slope (Fig. 3;ESMTable 2), with the exception of crystallised equation models; specifically, we fit the same models, exclud- ability, which, even in older individuals, tends not to decline ing individuals from a wave if they had been given any type of over time. In the first model, with the fewest controls, all diabetes diagnosis. The resultant models were very similar to cognitive domain variables were significantly associated with the models presented in Fig. 3 (ESM Table 4); effect sizes HbA level and slope, and this held true across subsequent were generally reduced, and no previously small and/or un- 1c models, except for the slope variables of memory and stable associations became robust in these models. We also crystallised ability. Decline in memory and crystallised ability tested whether change in cognitive function that occurred be- was not associated with either HbA level or slope after con- tween age 11 and age 70 was associated with HbA levels, 1c 1c trols were included. but while age 11 cognitive function significantly predicted Diabetologia HbA , change in cognitive function did not (ESM Results, Other associations between slope and level estimates of 1c ESM Table 5). cognitive functions and HbA were generally small and were 1c inconsistent when different covariates were added. We found Post hoc analyses of diagnosis status interacting with cogni- no consistent relationships between HbA level and cognitive 1c tive function and HbA Individuals might change their behav- function slope, which represents cognitive decline. If HbA 1c 1c iour once they are aware that they have type 2 diabetes. To test had a causative, negative effect on cognitive function, we whether or not higher functioning individuals may be better at would expect to find this, but the largest association was caring for themselves after diagnosis, we ran a repeated- 0.018 and in the opposite direction from what we would ex- measures ANCOVA on the first two study waves, treating type pect. The associations discussed thus far are visible in Fig. 2, 2 diabetes diagnosis, wave and overall cognitive function as where one can see that the initial level of cognitive function is independent variables and HbA as the dependent variable. lower in individuals with type 2 diabetes, though the rate of 1c Type 2 diabetes diagnosis (ω =0.338, p < 0.0001), cognitive decline is not obviously different. 2 2 function (ω = 0.004, p < 0.0001) and wave (ω = 0.023, p < We found a significant association between age 11 cogni- 0.0001) all had individual main effects, as did the type 2 diag- tive function and HbA level starting at age 70 and cognitive 1c nosis and wave interaction (ω =0.006, p < 0.0001). However, function level at age 70, but we did not find any relationships type 2 diagnosis and cognitive function did not significantly between HbA level and age 70 cognitive function level or 1c interact (ω <0.001, p = 0.742), suggesting that knowledge of change in cognitive function from ages 11 and 70. This sug- one’s diagnosis is not related to higher functioning individuals gests that early life cognition, which is stable over the lifespan, taking better care of themselves. No additional interactions in drives the cross-sectional association between cognitive func- the model were significant either. These results were robust to tion and HbA at age 70. In 1947, there was no HbA mea- 1c 1c the inclusion of control covariates (ESM Table 6). surement at age 11, so we cannot decide here whether early In much the same way that cognitive function might cognitive function is causal to worsening blood glucose across behaviourally impact post-diagnosis HbA , individuals with the life course to age 70, or whether cognitive functioning and 1c type 2 diabetes who maintain low HbA measurements may blood glucose track each other through the life course. 1c show cognitive differences from individuals with type 2 dia- It is notable that age 11 cognitive function was not associ- betes who do not maintain healthier HbA .Wealsotestedthis ated with HbA slope, but cognitive function level at age 70 1c 1c is an ANCOVA framework, making cognitive function the was. This was surprising, given that cognitive function level at outcome variable, and including HbA , wave(1or2)and age 70 was not reliably related to HbA level starting at the 1c 1c type 2 diabetes diagnosis as independent variables. We found same age. However, these findings are consistent with a lead- main effects of type 2 diabetes diagnosis (ω = 0.014, p < lag effect: better cognitive functioning earlier in life predicts 0.0001) and HbA (ω = 0.006, p = 0.0001), but no effects lower blood glucose, but only later in life. 1c of wave or any interactions. These effects were robust to the Early cognitive function is a major life course variable that inclusion of covariates (ESM Table 7). Together, these find- influences blood glucose and type 2 diabetes progression. In ings suggest that type 2 diabetes diagnosis and HbA are both this sample, sex had few effects on level and slope outcomes. 1c associated with cognitive function, but neither changes in di- The strongest was with cognitive function level—men tended agnosis nor whether an individual with type 2 diabetes has to have higher cognitive function level in this sample. Years of relatively low or high HbA appear to be related to cognitive education display similar relationships: very small effects on 1c function. all level and slope estimates other than cognitive function level, with which education showed positive relationships (β >0.12). Discussion When introduced in model B, disease variables including APOE*ε4 and hypertension and CVD history greatly reduced The results show that in an older narrow age cohort, tested on the effect of cognitive function on HbA slope. Smoking 1c four occasions between 70 and 79 years, lower cognitive func- history, for example, had a notable effect on HbA level, 1c tion at 70 is associated with increases in HbA over the fol- indicating that smokers were more likely to have higher 1c lowing decade. Cognitive function level is negatively corre- HbA at age 70. The impact of these variables suggests that 1c lated with the slope of HbA , and this effect is robust to the the association between cognitive function and later measure- 1c inclusion of all covariates, including age 11 cognitive func- ments of HbA is related to health behaviours. One interpre- 1c tion. Age 11 cognitive function itself is consistently and neg- tation of the results is that individuals with higher cognitive atively related to HbA level, which is consistent with previ- function take better care of themselves: they smoke less, are 1c ous reports of this sample [20, 39]. Together, these results more active and have a healthier diet . These associations suggest that cognitive function consistently predicts blood between cognitive function and health later in life have been glucose later in life. extensively investigated [41–43], including through diabetes Diabetologia epidemiology [13, 20]. However, our post hoc analyses dem- the disease could lead to later impairment in cognitive func- onstrated that the relationship between cognitive function and tion. Our results also suggest that good cognitive function HbA could not be explained by participants with higher continues to protect an individual from developing high blood 1c cognitive function altering their behaviour more than those glucose, emphasising its possible importance in ameliorating with lower cognitive function when these participants become type 2 diabetes progression throughout the lifespan. Further aware of their type 2 diabetes diagnosis. research into the mechanisms whereby cognitive function im- The extensive clinical and cognitive data that were available pacts and is impacted by elevated blood glucose is warranted across all four waves, over a decade, are clear strengths of this in the pursuit of underused strategies for identifying at-risk study, as was the availability of a reliable cognitive function older individuals and protecting them from cognitive decline measure at an early age. Moreover, the broad range of validated and type 2 diabetes. cognitive tests allowed us to investigate both overall and do- Acknowledgements The LBC1936 study is funded by Age UK (The main level associations between cognitive function and blood Disconnected Mind project). We thank the Scottish Council for glucose. As the LBC1936 is a narrow age cohort, neither cohort Research in Education for allowing access to SMS1947. We thank the nor age effects could significantly bias our findings. LBC1936 study participants and research team members. There were also limitations to this study that ought to be considered. First, our sample featured relatively few individ- Data availability The datasets analysed during the current study are not publicly available due to factors of individual privacy and health security, uals diagnosed with type 2 diabetes, which limits the power of but are available from the study director of the Lothian Birth Cohorts of the statistical analyses. This issue was somewhat ameliorated 1921 and 1936 on reasonable request. by our use of HbA as an outcome, which captures fine- 1c grained information about diabetic and prediabetic status; nev- Funding The LBC1936 data were collected using a Research into Ageing programme grant; this research continues as part of the Age ertheless, comparatively few individuals had clinically elevat- UK-funded Disconnected Mind project. DMA, JMS and IJD are mem- ed blood glucose. Second, and similarly, dropout (most likely bers of The University of Edinburgh Centre for Cognitive Ageing and due to mortality and frailty) was significant; almost half of the Cognitive Epidemiology, which is funded by the Biotechnology and original sample did not return by the fourth wave. Whereas Biological Sciences Research Council and Medical Research Council (MR/K026992/1). our models take existing and missing data into account, ulti- mately these missing data limit our statistical power and Duality of interest The authors declare that there is no duality of interest would particularly impact our analyses of slopes. For instance, associated with this manuscript. we might not have possessed the statistical power to detect Contribution statement DMA performed the analyses and drafted the small changes in cognitive function that could have been driv- manuscript. JMS assisted in the study design, and critically reviewed en by high HbA . Third, we did not have any HbA mea- 1c 1c the analyses and manuscript. IJD conceptualised the study, assisted in surements from earlier in life, so we could not control for the the study design, data analysis, drafting and critical review of the manu- influence of type 2 diabetes precursors that might have existed script. All authors approved the final manuscript. DMA is responsible for the integrity of the work as a whole. earlier in life. However, the prevalence of type 2 diabetes in children and young adults is low  and would be lower still in a cohort born in 1936 . Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http:// HbA not only predicts type 2 diabetes but is also associ- 1c creativecommons.org/licenses/by/4.0/), which permits unrestricted use, ated with vascular complications [2, 6] and declines in both distribution, and reproduction in any medium, provided you give appro- cognitive function [3, 4, 9] and brain volume [7, 8]. However, priate credit to the original author(s) and the source, provide a link to the some prior findings are counterintuitive. For example, total Creative Commons license, and indicate if changes were made. brain volume was increased in a group receiving targeted glycaemic control therapy, but in spite of this no differences were found in cognitive outcomes . In general, it is difficult to determine the direction of causality in the association be- References tween type 2 diabetes and cognitive decline, but the current literature suggests that having type 2 diabetes causes greater 1. Biessels GJ, Strachan MW, Visseren FL, Kappelle LJ, Whitmer RA (2014) Dementia and cognitive decline in type 2 diabetes and pre- cognitive decline [11, 12, 19, 45, 46]. diabetic stages: towards targeted interventions. Lancet Diabetes Our results support an alternative hypothesis, i.e. that cog- Endocrinol 2:246–255 nitive function predicts high blood glucose; however, we 2. Bennett C, Guo M, Dharmage S (2007) HbA as a screening tool 1c stress that the two hypotheses are not mutually exclusive. for detection of type 2 diabetes: a systematic review. Diabet Med The best explanation for our results and the wider body of 24:333–343 3. Ravona-Springer R, Moshier E, Schmeidler J et al (2012) Changes research is that early life cognitive function contributes to in glycemic control are associated with changes in cognition in non- impairment of glucose tolerance and onset of type 2 diabetes. diabetic elderly. J Alzheimers Dis 30:299–309 Once an individual has developed diabetes, complication of Diabetologia 4. Yaffe K, Falvey C, Hamilton N et al (2012) Diabetes, glucose 25. Horn JL, Cattell RB (1967) Age differences in fluid and crystallized intelligence. Acta Psychol 26:107–129 control, and 9-year cognitive decline among older adults without dementia. Arch Neurol 69:1170–1175 26. Ritchie SJ, Tucker-Drob EM, Cox SR et al (2016) Predictors of 5. Awad N, Gagnon M, Messier C (2004) The relationship between ageing-related decline across multiple cognitive functions. impaired glucose tolerance, type 2 diabetes, and cognitive function. Intelligence 59:115–126 J Clin Exp Neuropsychol 26:1044–1080 27. Wechsler D (1997) Adult Intelligence Scale. Psychological 6. Saleem T, Mohammad KH, Abdel-Fattah MM, Abbasi AH (2008) Corporation, San Antonio Association of glycosylated haemoglobin level and diabetes 28. Wechsler D (1997) Wechsler Memory Scale. Psychological mellitus duration with the severity of coronary artery disease. Corporation, San Antonio Diab Vasc Dis Res 5:184–189 29. Deary IJ, Der G, Ford G (2001) Reaction times and intelligence 7. Launer LJ, Miller ME, Williamson JD et al (2011) Effects of inten- differences: a population-based cohort study. Intelligence 29:389– sive glucose lowering on brain structure and function in people with type 2 diabetes (ACCORD MIND): a randomised open-label 30. Deary IJ, Johnson W, Starr JM (2010) Are processing speed tasks substudy. Lancet Neurol 10:969–977 biomarkers of cognitive aging? Psychol Aging 25:219–228 8. Kerti L, Witte AV, Winkler A, Grittner U, Rujescu D, Flöel A 31. Deary IJ, Simonotto E, Meyer M et al (2004) The functional anat- (2013) Higher glucose levels associated with lower memory and omy of inspection time: an event-related fMRI study. NeuroImage reduced hippocampal microstructure. Neurology 81:1746–1752 22:1466–1479 9. Ramirez A, Wolfsgruber S, Lange C et al (2015) Elevated HbA is 1c 32. Nelson HE, Wilson J (1991) National Adult Reading Test (NART). associated with increased risk of incident dementia in primary care Nfer-Nelson, Windsor patients. J Alzheimers Dis 44:1203–1212 33. Wechsler D (2001) Wechsler Test of Adult Reading: WTAR. 10. McCrimmon RJ, Ryan CM, Frier BM (2012) Diabetes and cogni- Psychological Corporation, San Antonio tive dysfunction. Lancet 379:2291–2299 34. Lezak MD, Howieson DB, Loring DW, Fischer JS (2004) 11. Rawlings AM, Sharrett AR, Mosley TH, Ballew SH, Deal JA, Neuropsychological assessment. Oxford University Press, New Selvin E (2017) Glucose peaks and the risk of dementia and 20- York year cognitive decline. Diabetes Care 40:879–886 12. Ekblad LL, Rinne JO, Puukka P et al (2017) Insulin resistance 35. Preacher KJ (2008) Latent growth curve modeling. Sage predicts cognitive decline: an 11-year follow-up of a nationally Publications, Los Angeles representative adult population sample. Diabetes Care 40:751–758 36. Yuan K-H, Bentler PM (1997) Mean and covariance structure 13. Mõttus R, Luciano M, Sarr JM, McCarthy MI, Deary IJ (2015) analysis: theoretical and practical improvements. J Am Stat Assoc Childhood cognitive ability moderates later-life manifestation of 92:767–774 type 2 diabetes genetic risk. Health Psychol 34:915–919 37. Rosseel Y (2012) Lavaan: an R package for structural equation 14. Khan H, Sobki S, Khan S (2007) Association between glycaemic modeling. J Stat Softw 48:1–36 control and serum lipids profile in type 2 diabetic patients: HbA 1c 38. Čukić I, Mõttus R, Luciano M, Starr JM, Weiss A, Deary IJ (2015) predicts dyslipidaemia. Clin Exp Med 7:24–29 Do personality traits moderate the manifestation of type 2 diabetes 15. Tesfaye S, Stevens L, Stephenson J et al (1996) Prevalence of dia- genetic risk? J Psychosom Res 79:303–308 betic peripheral neuropathy and its relation to glycaemic control and 39. Spauwen P, Martens R, Stehouwer C et al (2016) Lower verbal potential risk factors: the EURODIAB IDDM Complications Study. intelligence is associated with diabetic complications and slower Diabetologia 39:1377–1384 walking speed in people with type 2 diabetes: the Maastricht 16. Shpirer I, Rapoport MJ, Stav D, Elizur A (2012) Normal and ele- Study. Diabet Med 33:1632–1639 vated HbA levels correlate with severity of hypoxemia in patients 1c 40. Deary IJ, Weiss A, Batty GD (2010) Intelligence and personality as with obstructive sleep apnea and decrease following CPAP treat- predictors of illness and death: how researchers in differential psy- ment. Sleep Breath 16:461–466 chology and chronic disease epidemiology are collaborating to un- 17. Strachan MW, Reynolds RM, Marioni RE, Price JF (2011) derstand and address health inequalities. Psychol Sci Public Interest Cognitive function, dementia and type 2 diabetes mellitus in the 11:53–79 elderly. Nat Rev Endocrinol 7:108–114 41. Calvin CM, Batty GD, Deary IJ (2011) Cognitive epidemiology: 18. Hassing LB, Hofer SM, Nilsson SE et al (2004) Comorbid type 2 concepts, evidence, and future directions. In: Chamorro-Premuzic diabetes mellitus and hypertension exacerbates cognitive decline: T, von Stumm S, Furnham A (eds) The Wiley-Blackwell handbook evidence from a longitudinal study. Age Ageing 33:355–361 of individual differences. Blackwell Publishing, Chichester, pp 19. Cukierman T, Gerstein H, Williamson J (2005) Cognitive decline 427–460 and dementia in diabetes—systematic overview of prospective ob- 42. Batty GD, Deary IJ (2004) Early life intelligence and adult health: servational studies. Diabetologia 48:2460–2469 associations, plausible mechanisms, and public health importance 20. Mõttus R, Luciano M, Starr JM, Deary IJ (2013) Diabetes and life- are emerging. BMJ 329:585–586 long cognitive ability. J Psychosom Res 75:275–278 43. Deary IJ, Batty GD (2007) Cognitive epidemiology. J Epidemiol 21. Deary IJ, Gow AJ, Taylor MD et al (2007) The Lothian Birth Community Health 61:378–384 Cohort 1936: a study to examine influences on cognitive ageing 44. American Diabetes Association (2000) Type 2 diabetes in children from age 11 to age 70 and beyond. BMC Geriatr 7:28 and adolescents. Pediatrics 105:671–680 22. Deary IJ, Gow AJ, Pattie A, Starr JM (2011) Cohort profile: the Lothian Birth Cohorts of 1921 and 1936. Int J Epidemiol 41:1576– 45. Geijselaers SL, Sep SJ, Claessens D et al (2017) The role of hyper- 1584 glycemia, insulin resistance, and blood pressure in diabetes- 23. Deary IJ, Whalley LJ, Starr JM (2009) A lifetime of intelligence: associated differences in cognitive performance—the Maastricht follow-up studies of the Scottish mental surveys of 1932 and 1947. Study. Diabetes Care 40:1537–1547 American Psychological Association, Washington, DC 46. Feinkohl I, Keller M, Robertson CM et al (2015) Cardiovascular 24. Verhaeghen P, Salthouse TA (1997) Meta-analyses of age–cogni- risk factors and cognitive decline in older people with type 2 dia- tion relations in adulthood: estimates of linear and nonlinear age betes. Diabetologia 58:1637–1645 effects and structural models. Psychol Bull 122:231–249
Diabetologia – Springer Journals
Published: Jun 2, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera