Diabetes Care Volume 40, June 2017 751

Laura L. Ekblad,1 Juha O. Rinne,1,2 Insulin Resistance Predicts Pauli Puukka,3 Hanna Laine,4,5 Satu Ahtiluoto,6 Raimo Sulkava,6 Cognitive Decline: An 11-Year Matti Viitanen,4,7 and Antti Jula3 Follow-up of a Nationally Representative Adult Population Sample Diabetes Care 2017;40:751–758 | https://doi.org/10.2337/dc16-2001 PDMOOYHAT EVCSRESEARCH SERVICES EPIDEMIOLOGY/HEALTH

OBJECTIVE The aim of this study was to examine whether insulin resistance, assessed by HOMA of insulin resistance (HOMA-IR), is an independent predictor of cognitive decline.

RESEARCH DESIGN AND METHODS

The roles of HOMA-IR, fasting insulin and glucose, HbA1c, and hs-CRP as predictors of cognitive performance and its change were evaluated in the Finnish nationwide, population-based Health 2000 Health Examination Survey and its 11-year follow- up, the Health 2011 study (n = 3,695, mean age at baseline 49.3 years, 55.5% women). fl Categorical verbal uency, word-list learning, and word-list delayed recall were used 1Turku PET Centre, University of Turku, Turku, as measures of cognitive function. Multivariate linear regression analysis was per- formed and adjusted for previously reported risk factors for cognitive decline. 2Division of Clinical Neurosciences, Turku Uni- versity Hospital, Turku, Finland RESULTS 3National Institute for Health and Welfare, Turku, Finland Higher baseline HOMA-IR and fasting insulin levels were independent predictors of 4Turku City Hospital, University of Turku, Turku, poorer verbal fluency performance (P = 0.0002 for both) and of a greater decline in Finland verbal fluency during the follow-up time (P = 0.004 for both). Baseline HOMA-IR and 5Department of Medicine, Turku University Hos- pital, University of Turku, Turku, Finland insulin did not predict word-list learning or word-list delayed recall scores. There 6 « APOE34 University of Eastern Finland, Kuopio, Finland were no interactions between HOMA-IR and apolipoprotein E 4( ) genotype, 7Clinical Geriatrics, Karolinska Institutet, Karo- hs-CRP, or type 2 diabetes on the cognitive tests. Fasting glucose and hs-CRP levels at linska University Hospital, Huddinge, baseline were not associated with cognitive functioning. Corresponding author: Laura L. Ekblad, llekbl@ utu.fi. CONCLUSIONS Received 17 September 2016 and accepted 8 Our results show that higher serum fasting insulin and insulin resistance predict March 2017. poorer verbal fluency and a steeper decline in verbal fluency during 11 years in a This article contains Supplementary Data online representative sample of an adult population. Prevention and treatment of insulin at http://care.diabetesjournals.org/lookup/ resistance might help reduce cognitive decline later in life. suppl/doi:10.2337/dc16-2001/-/DC1. This article is featured in a podcast available at http://www.diabetesjournals.org/content/ Alzheimer disease, the most common type of dementia, has become a major public diabetes-core-update-podcasts. health concern in recent years. Targeting modifiable risk factors for cognitive decline in © 2017 by the American Diabetes Association. midlife could delay the onset of Alzheimer disease and thus help reduce the economic Readers may use this article as long as the work is properly cited, the use is educational and not burden associated especially with the late stages of the disease (1). Diabetes is an for profit, and the work is not altered. More infor- acknowledged risk factor for Alzheimer disease and cognitive impairment (2–4). A re- mation is available at http://www.diabetesjournals cent review concluded that individuals with diabetes typically perform 0.3–0.5 SD units .org/content/license. 752 Insulin Resistance and Cognitive Decline Diabetes Care Volume 40, June 2017

lower on cognitive tests across all age- associated with the metabolic syndrome hs-CRP values .10 mg/L in 2000 (n = groups compared with the general popula- (20), which is why we hypothesized that 105) were excluded to eliminate the con- tion. The mechanisms underlying these even the association of IR and cognition founding effects of an infectious disease. subtle cognitive decrements, however, are could be modified by inflammation grade. The mean age in 2000 was 49.3 years not yet well established. These decrements To test these hypotheses, we studied (range 30–86), and 55.5% were women. most likely do not indicate early stages of 3,695 individuals who participated in dementia, but it is possible that they lower the Finnish population-based Health Measurements the threshold for developing clinical symp- 2000 and Health 2011 studies. Blood pressure was measured at baseline toms of dementia later in life (5). in a sitting position from the right arm Insulin resistance (IR) is closely associ- RESEARCH DESIGN AND METHODS with a standard mercury manometer ated with obesity, chronic low-grade in- Study Population (Mercuro 300; Speidel & Keller, Jungin- flammation, and low levels of physical The data for this study were acquired from gen, ), and the average of two activity and it can be seen as the hallmark the Health 2000 Health Examination Sur- measurements was used for the analyses. of the metabolic syndrome and type 2 di- vey and its 11-year follow-up survey, Baseline fasting blood samples were abetes (DM2) (6). Epidemiological studies Health 2011. The surveys were conducted drawn, the duration of fasting time was have shown that the metabolic syndrome by the Finnish National Institute for recorded, and the samples were stored in midlife increases the risk for cognitive Health and Welfare in 2000–2001 and at 270°C until analyzed. decline (7,8). Recent evidence suggests 2011 (21,22). The Health 2000 survey Serum cholesterol, HDL cholesterol, tri- that brain IR could be an important trig- was a nationwide, comprehensive, popu- glycerides, glucose, hs-CRP, and insulin gering factor in the development of Alz- lation-based examination survey repre- values were determined from the frozen heimer disease neuropathology (9) and sentative of the Finnish adult population. samples. Cholesterol values were deter- possibly a key link between the metabolic A total of 8,028 individuals aged 30 years mined by a CHOD PAP test (Olympus syndrome and cognitive decline (10). or older were randomly selected from the System Reagent; OLYMPUS, Hamburg, Treatment with intranasally administered Finnish population register from 80 health Germany), HDL cholesterol values by a insulin has been shown to improve mem- service districts throughout Finland HDL-C Plus test (Roche Diagnostics, ory in patients with Alzheimer disease and using a two-stage stratified cluster sam- Mannheim, Germany), triglycerides by a mild cognitive impairment (11). Previous pling procedure. Of the study population, GPO PAP test (Olympus System Reagent), studies indicate that the response to in- 84% (n = 6,770) attended the health ex- and glucose values by a hexokinase test tranasal insulin varies according to the amination proper or the health examina- (Olympus System Reagent). Serum insulin patients’ sex and APOE«4 genotype (12). tion at home (21). was determined by a microparticle en- There are a number of cross-sectional In 2011 all the individuals alive who zyme immunoassay (Abbott Laboratories studies linking IR with poorer cognitive belonged to the Health 2000 study sam- Dainabot, Tokyo, Japan). HbA1c was de- performance (13–17). Only two previous ple, who still lived in Finland and who had termined with an immunoturbidimetric longitudinal studies, to our knowledge, not refused to participate in the upcom- method (Hemoglobin A1c assay; Abbott examined the effects of insulin or IR on ing follow-up studies, were invited to the Laboratories). Serum hs-CRP was ana- cognitive decline over time. A study on Health 2011 study (22). lyzed by an automated analyzer (Optima; 999 men showed that higher insulin levels Both studies were approved by the Thermo Electron Oy, Vantaa, Finland) and at age 50 years predicted lower cognitive Ethics Committee for Epidemiology and an ultrasensitive immunoturbidimetric test scores 20 years later (18). The Ath- Public Health in the hospital district of test (Ultrasensitive CRP; Orion Diagnos- erosclerosis Risk in Communities (ARIC) Helsinki and Uusimaa, Finland. All partic- tica, Espoo, Finland). The detection limit cohort study (19) examined 7,148 individ- ipants gave written informed consent for for quantitation of the CRP assay was uals at baseline for fasting insulin and participating in the studies. 0.20 mg/L (23). Those who fell below HOMA-IR. Cognitive performance was Altogether, 3,695 individuals who had this limit were given the value 0.2 mg/L evaluated 3 and 9 years after baseline. attended the health examination proper divided by 2 = 0.1 mg/L. Non-HDL choles- The study concluded that both fasting in- in 2000 (n = 6,354) and who attended the terol was counted as total cholesterol sulin and HOMA-IR were associated with health examination or the home health minus HDL cholesterol. HOMA-IR was poorer cognition 3 years after baseline examination in 2011 and thus had been used as a measure of IR and counted by and with a decline in cognitive test scores tested for cognition on both occasions the equation fasting insulin (mU/mL) during the 6-year follow-up for cognitive were included in this study. Participants times fasting glucose (mmol/L) divided functioning. who, at baseline, had fasted for ,4h(n = by 22.5 as previously described (24). The aim of this study was to assess 226); who had insulin treatment or un- whether IR is a risk factor for cognitive known diabetes medication (n = 59); and Cognitive Tests decline during 11 years. We hypothe- who had not completed the cognitive The participants were tested at baseline sizedthatIR,estimatedwithHOMA-IR, tests (n = 127) or had missing HOMA-IR and at follow-up for verbal fluency and is an independent risk factor for cogni- values (n =4)wereexcluded.Atotalof encoding and retaining verbal material tive decline and that, based on previous 789 individuals had died or were lost to according to the Finnish version of the studies, APOE«4 genotype (12,15) and sex follow-up. A flowchart of the study popu- Consortium to Establish a Registry for (12,13,15) might influence this risk. lation is provided in Supplementary Fig. 1. Alzheimer’s Disease (CERAD) test battery Also, low-grade inflammation possibly In the analysis of low-grade inflam- (22,23,25,26). In the verbal fluency test, modulates the risk of cognitive decline mation and cognition, individuals with the participants were asked to list as care.diabetesjournals.org Ekblad and Associates 753

many animals as possible during 1 min. week, 5 = four to six times a week, and 2 were performed additionally in the sub- This categorical fluency test is considered 6 = daily. Individuals using oral diabetes population who had fasted for 10 h or to represent both language skills and ex- medication or who had fasting glucose longer (35.8%, n = 1,321) prior to blood ecutive functions. In the word-list learning values .7.0 mmol/L were classified as sampling. test, 10 words were shown to the partic- having DM2. APOE genotype was assessed In additional analyses model 2 was fur- ipants. The participants were asked in 3,469 individuals (93.8% of the study ther adjusted for BDI score, physical activ- to read the words aloud, to memorize population) who gave their written con- ity, smoking, and alcohol consumption them, and to repeat the words they re- sent for DNA sampling. APOE« genotyping (model 3). member within 90 s. In 2000, if the partic- was performed with the MassARRAY sys- The analyses for change in cognition ipant remembered all 10 words after one tem (Sequenom, San Diego, California) over 11 years were adjusted for baseline round, the result was counted as a full with a modified protocol that has previ- cognitive test scores in each model. 30 points. In 2011, all three rounds were ously been described (28). APOE«4 geno- Statistical significance was set at P , repeated for all participants, regard- type was considered positive for subjects 0.05 for all other tests except for interac- less of whether they remembered all with one or two «4 alleles. tions, where P , 0.1 was considered sta- 10 words after the first round or not. tistically significant. The word-list learning test was used to Statistical Analysis There were no interactions for HOMA- assess verbal learning and memory. After The differences between individuals IR 3 APOE«4 genotype (all P values 5 min the participants were asked to recall who did not attend the health examina- .0.26), HOMA-IR 3 hs-CRP (all as many words as possible from the pre- tion proper or the home health exami- P values .0.16), or HOMA-IR 3 DM2 (all viously presented word list (word-list de- nation in 2011 (n = 1,451) and those P values .0.48) on any of the cognitive layed recall). The delayed recall test was included in this study (n =3,695)were tests. The interaction for HOMA-IR 3 sex used to evaluate verbal episodic memory. analyzed by Student t test for continu- was significant for word-list delayed recall The change in cognitive test scores during ous variables and x2 test for categorical at follow-up (P = 0.06). Thus, sex-stratified the follow-up was counted as cognitive variables. analyses were performed only for HOMA- test score in 2000 minus cognitive test For the description of the characteris- IR and this cognitive test. No other inter- score in 2011. tics of the study population, the study pop- actions were found for HOMA-IR 3 sex (all ulation was divided into three groups other P values .0.29). Covariates according to the baseline tertiles of The analyses were performed with JMP Previously reported risk factors for cogni- HOMA-IR. The cutoff for the second tertile Pro 11.0 (SAS Institute, Cary, NC). tive impairment (model 1: age, sex, years was 1.16 and for the third tertile 2.01. In all of education; model 2: also DM2, APOE«4 further analyses, HOMA-IR was analyzed RESULTS genotype, BMI, systolic blood pressure, as a continuous variable. Before the anal- The characteristics of the study popula- and levels of HDL and non-HDL choles- yses, the skewed distributions of HOMA- tion according to the tertiles of HOMA-IR terol and triglycerides; and model 3: all IR,glucose,andinsulin,HbA1c,hs-CRP,tri- are shown in Table 1. Individuals with previous covariates and Beck depres- glycerides, and BDI score were corrected higher levels of HOMA-IR were older sion inventory [BDI] score [27], level of by a logarithmic transformation (log ). The e (Ptrend , 0.0001), more often men than physical activity, smoking, and alcohol age- and sex-adjusted differences among women (Ptrend , 0.0001), had fewer years consumption), assessed at baseline, the tertiles of HOMA-IR for the character- of education (Ptrend , 0.0001), and were were used as covariates in the analyses. istics were examined by ANCOVA for con- less often smokers (Ptrend , 0.009) than Sex, DM2, APOE«4 genotype, physical ac- tinuous variables and by logistic regression those with lower levels of HOMA-IR (Ptrend tivity score, smoking, and alcohol con- analysis for categorical variables. indicates age- and sex-adjusted differences sumption were analyzed as categorical Linear regression analysis was used to among the tertiles of HOMA-IR). The mean variables, and all other covariates were examine the associations of continuous, cognitive test scores in 2000 and 2011, ac- added to the model as continuous log-linear baseline HOMA-IR, glucose, in- cording to the tertiles of HOMA-IR and ad- variables. Information on years of formal sulin, HbA1c, and hs-CRP and of the cogni- justed for age, sex, and years of education, education, current smoking, alcohol con- tive test scores in 2011 and the change in are shown in Fig. 1. sumption, and level of physical activity cognitive test scores from 2000 to 2011. The individuals who did not attend the was addressed by a questionnaire. Exces- First, the analyses were adjusted for age, health examination proper or the home sive alcohol consumption was classified sex, and education (model 1 [Table 2]). health examination in 2011 were older as .24/16 doses of alcohol (12 g alco- Model 1 was further adjusted for APOE«4 (P , 0.0001) and had fewer years of for- hol/dose) per week (for men/women). genotype, DM2, BMI, systolic blood pres- mal education (P , 0.0001) than those Self-reported physical activity was as- sure, HDL and non-HDL cholesterol, and included in this study. The proportion of sessed by asking the participants how of- triglycerides (model 2 [Table 3]). The in- women was similar in both groups (56.0% ten they exercised in their free time for at teractions of HOMA-IR 3 sex, HOMA-IR 3 vs. 55.5%, P = 0.75) (data not shown). least 30 min vigorously enough to cause hs-CRP, and HOMA-IR 3 APOE«4 geno- sweating and breathlessness to a mild type and HOMA-IR 3 DM2 on cognitive Multivariate Correlates of Cognitive extent. The results were classified as fol- test scores at follow-up and for the change Performance lows: 1 = a few times a year or more sel- in cognition from 2000 to 2011 were ana- In the linear regression analyses adjusted dom, 2 = two to three times a month, 3 = lyzed in model 2. Because of the variation for age, sex, and education (model 1), once a week, four = two to three times a in fasting times, the analyses for model higher levels of HOMA-IR, glucose, and 754 Insulin Resistance and Cognitive Decline Diabetes Care Volume 40, June 2017

Table 1—Characteristics of the study population at baseline and cognitive test In the linear regression model further scores at baseline and at follow-up according to the baseline tertiles of HOMA-IR adjusted for the previously recognized Tertile of HOMA-IR metabolic risk factors for cognitive decline and APOE«4 genotype (model 2), higher 1st 2nd 3rd P trend levels of HOMA-IR and insulin were inde- n (%) 1,246 (33.7) 1,221 (33.0) 1,228 (33.2) pendent predictors of poorer verbal flu- Female 784 (62.9) 677 (55.4) 589 (48.0) ,0.0001 ency at follow-up (P = 0.0002 for both). Age (years) 47.0 6 11.3 48.7 6 11.8 52.1 6 12.4 ,0.0001 Baseline HOMA-IR or insulin levels were Years of education 12.5 6 3.9 12.4 6 4.0 11.2 6 3.8 ,0.0001 not associated with word-list learning (P = Fasting time (h:min) 8:41 9:12 9:42 ,0.0001 0.60 and 0.69, respectively) or with HOMA-IR 0.83 6 0.23 1.55 6 0.24 3.56 6 2.21 ,0.0001 word-list delayed recall (P =0.38and DM2 1 (0.0) 10 (0.8) 77 (6.3) ,0.0001 0.37, respectively) at follow-up (Table 3). Fasting glucose In sex-stratified analyses, HOMA-IR did (mmol/L) 5.1 6 0.4 5.3 6 0.4 5.8 6 1.1 ,0.0001 not predict word-list learning perfor- Fasting insulin (mU/L) 3.7 6 1.0 6.6 6 1.1 13.6 6 6.2 ,0.0001 mance in men (P = 0.26) or in women

HbA1c % (mmol/mol) 5.1 6 0.3 5.2 6 0.3 5.5 6 0.6 ,0.0001 (P = 0.99) (data not shown). (32 6 3.3) (33 6 3.3) (37 6 6.6) Higher baseline HOMA-IR and insulin hs-CRP (mg/L) 0.91 6 1.43 1.18 6 1.62 1.83 6 2.07 ,0.0001 levels predicted a greater decline in ver- Systolic blood bal fluency (P = 0.004 for both) but not a pressure (mmHg) 125 6 18 131 6 19 139 6 20 ,0.0001 decline in word-list learning (P =0.55and BMI (kg/m2)24.16 3.1 26.3 6 3.6 29.7 6 4.6 ,0.0001 0.66, respectively) or word-list delayed HDL cholesterol recall (P = 0.96 and 0.88) (Table 3). (mmol/L) 1.51 6 0.37 1.36 6 0.35 1.19 6 0.31 ,0.0001 Baseline glucose (all P values .0.12), Non-HDL cholesterol HbA1c levels (all P values .0.29), or (mmol/L) 4.25 6 1.05 4.54 6 1.05 4.91 6 1.18 ,0.0001 hs-CRP levels (all P values .0.26) were Triglycerides not associated with any of the cognitive 6 6 6 , (mmol/L) 1.17 0.56 1.37 0.70 1.96 1.20 0.0001 tests in 2011 or with the change in cogni- « APOE 4* 391 (33.4) 368 (31.8) 361 (31.6) 0.73 tive performance from 2000 to 2011 (glu- BDI score 6.2 6 6.4 6.3 6 6.5 7.1 6 6.7 0.04 cose P . 0.22, HbA1c P . 0.44, and hs-CRP Alcohol consumption P . 0.15) in model 2 (Table 3). . 24/16 doses/week The results for the additional analyses (for men/women) 71 (5.7) 79 (6.5) 85 (6.9) 0.84 of model 3 and the values of the covariates Current smoking 277 (22.3) 226 (18.6) 196 (16.0) 0.009 as determinants of cognitive performance Physical activity score 3.7 6 1.4 3.7 6 1.4 3.6 6 1.4 0.0002 are shown in Supplementary Tables 1 and fl 6 6 6 , Verbal uency in 2000 25.9 7.0 25.2 7.0 23.9 6.9 0.0001 2. HOMA-IR (P = 0.01) and insulin (P =0.02) Word-list learning in were independent predictors of poorer 2000† 22.1 6 3.7 21.8 6 4.0 20.9 6 4.0 0.02 verbal fluency at the 11-year follow-up. Word-list delayed The results for a decline in verbal fluency recall in 2000‡ 7.6 6 1.8 7.5 6 1.8 7.2 6 1.8 0.52 from2000to2011werenolongerstatis- Verbal fluency in 2011 25.4 6 7.3 24.5 6 7.6 22.7 6 7.1 ,0.0001 tically significant for HOMA-IR (P = 0.051) Word-list learning in or insulin (P = 0.056, data not shown) in 2011† 21.8 6 4.3 21.5 6 4.4 20.3 6 4.6 0.02 this additional model of adjustment. Word-list delayed recall in 2011‡ 7.5 6 2.1 7.4 6 2.1 6.9 6 2.2 0.39 The proportion of individuals who were examined after an overnight fast is shown Data are mean 6 SD or n (%), unless otherwise indicated. Ptrend values for age- and sex-adjusted differences among the tertiles of HOMA-IR. The cutoff for the second tertile is 1.16 and for the in Supplementary Fig. 2. The results for third tertile 2.01. *APOE«4 genotype is considered positive for subjects with one or two e4 the analyses of the individuals who had alleles. †The maximum score for word-list learning is 30 points. ‡The maximum score for word- fasted for 10 h or longer (n = 1,321) are list delayed recall is 10 points. provided in Supplementary Table 3. In these participants, HOMA-IR (P = 0.046) and HbA1c (P = 0.01) were independent predictors of poorer verbal fluency in insulin predicted a lower verbal fluency P values .0.61) (Table 2). Higher base- 2011, and HOMA-IR (P = 0.02), insulin score (HOMA-IR P , 0.0001, glucose P = line HOMA-IR (P = 0.004) and insulin (P = 0.02), and HbA1c (P = 0.04) predicted 0.02, and insulin P , 0.0001). Levels of (P = 0.005) levels were associated a greater decline in verbal fluency from HbA1c (P = 0.07), or hs-CRP (P = 0.17) with a greater decline in verbal fluency 2000 to 2011. were not associated with verbal fluency from 2000 to 2011 but not with a de- scores 11 years later. None of these cline in word-list learning (P =0.30and CONCLUSIONS variables predicted performance on 0.44, respectively) or word-list delayed Here, we show that higher levels of insulin word-list learning (all P values .0.10) recall (P = 0.90 and 0.94, respectively) and IR are independent predictors of or word-list delayed recall tests (all (Table 2). poorer verbal fluency performance and care.diabetesjournals.org Ekblad and Associates 755

cognition (18,19). The follow-up study on 999 men showed that higher serum insu- lin concentrations at age 50 years were associated with lower cognitive z scores (based on Mini-Mental State Examination, Trail Making A, and Trail Making B tests) after 20 years, but this association was not statistically significant after adjustment for diastolic blood pressure. The longitu- dinal associations of IR and cognition were not reported (18). In the ARIC study IR was associated with a decline in delayed word recall and first-letter fluency (19). Com- pared with the ARIC study (19), the cur- rent study provides a longer follow-up period for cognitive decline, and unlike the ARIC study, adjustments were made for covariates closely associated with IR, such as BMI and level of physical inactivity, whichhavebeenreportedtoincreasethe risk for cognitive decline. In our study, IR did not predict word-list delayed recall after adjustment for age, sex, and educa- tion. Additional adjustments for metabolic covariates did not change our results. However, the delayed recall test that we used was slightly different from the test used in the ARIC study, which could be one possible explanation for the differ- ence in the results regarding delayed re- call. In both studies, the participants were askedtolearn10nounsandtorecallthem after 5 minutes. In our study the words were repeated three times as a list, and in the ARIC study the words were incor- porated in sentences. It is possible that the CERAD word-list delayed recall test that we used is not sensitive enough to assess a decline in delayed memory in middle-aged individuals. We (15) and others (13) have discov- ered sex differences in the cross-sectional association of IR and cognition, suggesting that IR would be a risk factor for poorer cognitive performance in women but not in men. One longitudinal study on the metabolic syndrome and cognition, with a follow-up of 16 years, found that the metabolic syndrome was associated with cognitive decline only in women Figure 1—Change in mean cognitive test scores from 2000 to 2011, adjusted for age, sex, and (29). These results could not be confirmed years of education, according to the tertiles of HOMA-IR. N = 3,695. Range for verbal fluency, in the present longitudinal study. Men and – – – 0 54; word-list learning, 0 30; word-list delayed recall, 0 10. women were both susceptible to the harmful effects of IR on cognition. Our of a greater decline in verbal fluency dur- did not modify these results. Baseline previous cross-sectional study showed ing 11 years. Insulin and IR were not as- fasting glucose or hs-CRP levels did not an interaction for APOE«4 andIRonverbal sociated with the word-list learning or predict cognitive test scores. fluency performance. In the stratified word-list delayed recall tests, which Our results are consistent with, and analyses the association of IR and poorer were used as assessments of memory. extend, the findings of the two previous verbal fluency was evident only in noncar- Sex, APOE«4 genotype, or hs-CRP levels longitudinal studies on insulin or IR and riers of APOE«4 (15). Studies on intranasal 756 Insulin Resistance and Cognitive Decline Diabetes Care Volume 40, June 2017

« Table 2—Age-, sex-, and education-adjusted associations of baseline IR, fasting APOE 4 differences have not been inves- tigated in most longitudinal studies on the glucose and insulin, HbA1c, and hs-CRP values with cognitive test scores at follow-up andwithchangeincognitivetestscoresfrom2000to2011 metabolic syndrome and cognition. It is Verbal Word-list Word-list delayed possible that the early effects of insulin fluency learning recall or IR on cognition are different from their b SE b SE b SE longitudinal effects. The possible modu- lating effect of sex and APOE«4 genotype Cognitive test score, 2011† HOMA-IR 20.81*** 0.17 20.14 0.09 0.01 0.05 on the association of insulin or IR and cog- Glucose 22.40* 1.02 20.91 0.56 20.06 0.27 nition should be further explored in future Insulin 20.86*** 0.19 20.14 0.10 0.01 0.05 studies, also by using sophisticated imag-

HbA1c 22.70 1.50 20.86 0.81 20.06 0.39 ing techniques. A genetic and/or a sex dif- hs-CRP 20.12 0.09 20.04 0.05 20.01 0.61 ference in the association between IR and Change in cognition, 2000–2011‡ cognition would necessitate personalized HOMA-IR 0.41** 0.14 0.09 0.08 0.005 0.04 preventive and therapeutic interventions Glucose 1.37 0.8 0.91 0.49 0.37 0.24 to reduce the risk for cognitive decline. Insulin 0.43** 0.15 0.07 0.09 20.003 0.04 In the current study a higher level of IR HbA1c 1.29 1.22 0.67 0.71 0.21 0.34 hs-CRP 0.07 0.07 0.03 0.04 20.007 0.02 was associated with poorer performance on the categorical verbal fluency test, N = 3,695, except for the analysis of hs-CRP and cognition, where n = 3,590. Estimates (b) and SEs are derived from linear regression analysis and adjusted for age, sex, and years of formal which represents the function of brain education. The analyses for change in cognition are adjusted even for baseline cognitive test frontal and temporal lobe regions (30) scores. Logarithmic transformation is used for HOMA-IR, fasting glucose and insulin, HbA1c,and and which can be considered as a mea- , , , † hs-CRP to achieve a normal distribution. *P 0.05, **P 0.01, ***P 0.001. Note that a surement of language skills and executive negative estimate for cognitive test score in 2011 indicates a lower cognitive test score for those with higher levels of IR, fasting glucose, etc. ‡A positive estimate for the change in cognition function (31). In line with our findings, from 2000 to 2011 indicates a greater decline in cognitive test score for those with higher levels cross-sectional studies (13–17) on IR of IR, fasting glucose, etc. and cognition have shown associations between higher levels of IR and poorer executive function. There are several pos- insulin and cognition suggest that possibly unclear. Here, the association of IR on cog- sible pathways to explain these findings, only noncarriers of APOE«4 would benefit nition 11 years later was similar in noncar- since the mechanisms between IR and from treatment with intranasal insulin riers and carriers of APOE«4. The previous cognitive decline are thought to be mul- (12). The explanation for these sex and longitudinal studies (18,19) did not assess tifaceted. Cross-sectional brain positron APOE«4 differences, however, is still sex or APOE«4 differences, and sex or emission tomography studies with 18F- fluorodeoxyglucose show that, consistent with the localization of verbal fluency in Table 3—Multivariate correlations of baseline IR, fasting glucose and insulin levels, the prefrontal and temporal cortices,

HbA1c, and hs-CRP values with cognitive test scores at follow-up and with change in higher levels of IR are associated with cognitive test scores from 2000 to 2011 lower regional glucose metabolism in fron- Verbal Word-list Word-list delayed tal, parietal, and temporal cortical areas fluency learning recall in cognitively normal adults (32) and in b SE b SE b SE adults at risk for Alzheimer disease (33). A similar pattern of 18F-fluorodeoxyglucose– Cognitive test score, 2011† HOMA-IR 20.86*** 0.23 20.07 0.13 0.05 0.06 positron emission tomography reduction Glucose 21.93 1.25 20.59 0.68 0.08 0.33 is seen in prodromal Alzheimer disease Insulin 20.91*** 0.25 20.05 0.13 0.06 0.37 (34). Brain MRI studies show that IR is HbA1c 22.01 1.89 0.18 1.02 0.34 0.50 associated with lower temporal lobe hs-CRP 20.11 0.10 20.01 0.05 0.01 0.03 gray matter volume (14,35) but also with Change in cognition, 2000–2011‡ lower volumes of wider areas, such as the HOMA-IR 0.55** 0.19 0.07 0.11 0.002 0.05 prefrontal cortices and precuneus (35). In Glucose 1.26 1.03 0.63 0.59 0.29 0.29 addition, brain white matter changes Insulin 0.58** 0.20 0.05 0.66 20.01 0.06 seen in MRI images are more common HbA1c 1.20 1.55 20.05 0.90 0.11 0.44 hs-CRP 0.12 0.08 0.02 0.05 20.001 0.02 in individuals with the metabolic syn- drome compared with those without N = 3,695, except for the analysis of hs-CRP and cognition, where n = 3,590. Estimates (b) and SEs are derived from linear regression analysis and adjusted for age, sex, years of education, APOE«4 (36), and these changes are associated status, DM2, BMI, systolic blood pressure, HDL and non-HDL cholesterol, and triglycerides. with poorer verbal fluency in patients The analyses for change in cognition are adjusted even for baseline cognitive test scores. with prodromal Alzheimer disease (37). Logarithmic transformation is used for HOMA-IR, fasting glucose and insulin, HbA1c, hs-CRP, and triglycerides to achieve a normal distribution. **P , 0.01, ***P , 0.001. †Note that a negative The symptoms of Alzheimer disease estimate for cognitive test score in 2011 indicates a lower cognitive test score for those with typically begin with a decline in episodic higher levels of IR, fasting glucose, etc. ‡A positive estimate for the change in cognition from memory, but this decline is only clinically 2000 to 2011 indicates a greater decline in cognitive test score for those with higher levels of IR, evident close to the onset of the disease fasting glucose, etc. (38). We did not find any association care.diabetesjournals.org Ekblad and Associates 757

between IR and the CERAD word-list de- subpopulation who had fasted for 10 h Alzheimer’s disease. Alzheimers Dement 2007; layed recall test, a commonly used test of or longer. The strength of our study is 3:186–191 episodic memory to screen for Alzheimer the large, population-based study cohort, 2.OttA,StolkRP,vanHarskampF,PolsHA, Hofman A, Breteler MM. Diabetes mellitus and disease. This could be due to our relative- based on a nationally representative sam- the risk of dementia: The Rotterdam Study. ly young study population. Late-onset Alz- ple of the Finnish adult population. The Neurology 1999;53:1937–1942 heimer disease begins after 65 years of long follow-up time allows us to examine 3. Biessels GJ, Staekenborg S, Brunner E, age, but neuropathological changes typi- the decline in cognition associated with IR. Brayne C, Scheltens P. Risk of dementia in di- cal to the disease can be detected years In conclusion, we show that IR is an in- abetes mellitus: a systematic review. Lancet Neurol 2006;5:64–74 or even decades before the onset of any dependent risk factor for cognitive decline 4. Cheng G, Huang C, Deng H, Wang H. Diabetes cognitive symptoms. Typically, the accu- and, more specifically, a decline in verbal as a risk factor for dementia and mild cognitive mulation of b-amyloid, the neuropatho- fluency. Although the differences between impairment: a meta-analysis of longitudinal logical hallmark of Alzheimer disease, can the tertiles of IR in verbal fluency perfor- studies. Intern Med J 2012;42:484–491 be detected first in the frontal and the mance are small, and not of clinical signif- 5. Biessels GJ, Strachan MW, Visseren FL, Kappelle LJ, Whitmer RA. Dementia and cogni- temporal cortices of the brain (39). This icance at the individual level, these subtle tive decline in type 2 diabetes and prediabetic is interesting, since categorical verbal flu- changes in cognition associated with IR stages: towards targeted interventions. Lancet ency is subserved by these brain regions. could lower the threshold for more severe Diabetes Endocrinol 2014;2:246–255 Our study has limitations. The golden changes in cognition over time. Longitudi- 6. Reaven GM. Banting lecture 1988. Role of standard for determining IR, the insulin nal studies involving more detailed insulin resistance in human disease. Diabetes 1988;37:1595–1607 clamp method, could not be used because neuropsychological assessment and brain 7. Yates KF, Sweat V, Yau PL, Turchiano MM, of the large, comprehensive health exam- anatomical and functional imaging are Convit A. Impact of metabolic syndrome on cog- ination nature of this epidemiological needed to explore the different cognitive nition and brain: a selected review of the liter- study. We could not control for all comor- domains and the neuroanatomical and ature. Arterioscler Thromb Vasc Biol 2012;32: – bid conditions associated with IR such as neuropathologic changes associated with 2060 2067 8. Siervo M, Harrison SL, Jagger C, Robinson L, sleep apnea and nonalcoholic fatty liver IR. Targeting therapeutic strategies, such Stephan BC. Metabolic syndrome and longitudi- disease, which have been associated as lifestyle interventions, at people with IR nal changes in cognitive function: a systematic with poorer cognitive performance. Also, in midlife could potentially reduce the in- review and meta-analysis. J Alzheimers Dis we defined diabetes according to fasting cidence of cognitive decline later in life. 2014;41:151–161 9. Correia SC, Santos RX, Perry G, Zhu X, glucose values and the use of oral antidia- Moreira PI, Smith MA. Insulin-resistant brain betes drugs and, thus, individuals with state: the culprit in sporadic Alzheimer’sdis- normal fasting glucose but elevated post- Acknowledgments. The authors sincerely ease? Ageing Res Rev 2011;10:264–273 prandial values might have been falsely thank all the volunteers who participated in the 10. Kim B, Feldman EL. Insulin resistance as a classified as not having diabetes. The Health 2000 and the Health 2011 surveys for their key link for the increased risk of cognitive im- generous contribution. The authors thank Kaisa pairment in the metabolic syndrome. Exp Mol word-list learning test was performed Silander, Mari Kaunisto, and the staff of the Med 2015;47:e149 slightly differently in 2000 and 2011, and Technology Centre, Institute for Molecular 11. Craft S, Baker LD, Montine TJ, et al. Intra- this might have affected the scores of Medicine Finland (FIMM), University of Helsinki, nasal insulin therapy for Alzheimer disease and both word-list learning and word-list de- for the APOE genotyping work. amnestic mild cognitive impairment: a pilot clin- layed recall. However, only 17 participants Funding. This study was supported by personal ical trial. Arch Neurol 2012;69:29–38 grants from the foundation of Yrjo¨ Jahnsson (L.L.E.), 12. Claxton A, Baker LD, Wilkinson CW, et al. Sex had received a full 30 points on the word- the Finnish Brain Foundation (Suomen Aivosa¨ati¨ o)¨ and ApoE genotype differences in treatment re- list learning test, and excluding these indi- (L.L.E.), the Paulo Foundation (L.L.E.), and Finnish sponse to two doses of intranasal insulin in adults viduals did not change our results. In the Governmental Research Funding for Turku Univer- with mild cognitive impairment or Alzheimer’s word-list delayed recall, the delayed recall sity Hospital and Turku City Hospital. disease. J Alzheimers Dis 2013;35:789–797 No potential conflicts of sectiononlylastedfor5min.Thistestis Duality of Interest. 13. Schuur M, Henneman P, van Swieten JC, interest relevant to this article were reported. et al. Insulin-resistance and metabolic syn- designed to screen for Alzheimer disease, Author Contributions. L.L.E. analyzed data drome are related to executive function in and it may be that in this middle-aged and wrote the manuscript. L.L.E., J.O.R., P.P., women in a large family-based study. Eur J Epi- population a test with a longer delay H.L., S.A., R.S., M.V., and A.J. planned this study. demiol 2010;25:561–568 would have been more sensitive to detect J.O.R., P.P., H.L., S.A., R.S., M.V., and A.J. critically 14. Benedict C, Brooks SJ, Kullberg J, et al. Im- revised and edited the manuscript for important delayed memory decline. The fasting paired insulin sensitivity as indexed by the intellectual content. P.P. helped with the statistical HOMA score is associated with deficits in verbal times of our study volunteers varied, analysis of data. S.A., R.S., and A.J. took part in de- fluency and temporal lobe gray matter volume which might have resulted in falsely signing the Health 2000 and Health 2011 surveys. All in the elderly. Diabetes Care 2012;35:488–494 fi higher HOMA-IR values for those with a authors approved the nal version of the manu- 15. Ekblad LL, Rinne JO, Puukka PJ, et al. Insulin shorter fasting time. However, the base- script to be published. L.L.E. is the guarantor of this resistance is associated with poorer verbal flu- work and, as such, had full access to all the data in line HOMA-IR values were not higher for ency performance in women. Diabetologia the study and takes responsibility for the integrity – – 2015;58:2545 2553 participants who had fasted for 4 10 h ofthedataandtheaccuracyofthedataanalysis. 16. Abbatecola AM, Paolisso G, Lamponi M, compared with those who had fasted Prior Presentation. Parts of this study were et al. Insulin resistance and executive dysfunc- ’ for .10 h. Thus, we are confident that presented in abstract form at the Alzheimer s tion in older persons. J Am Geriatr Soc 2004;52: including participants who had fasted Association International Conference, Toronto, 1713–1718 Ontario, , 24–28 July 2016. for 4–10 h did not give false positive 17. Geroldi C, Frisoni GB, Paolisso G, et al. In- sulin resistance in cognitive impairment: the associations between HOMA-IR and cog- References InCHIANTI study. Arch Neurol 2005;62:1067–1072 nitive performance. To confirm this, 1. Brookmeyer R, Johnson E, Ziegler-Graham K, 18. Kilander L, Nyman H, Boberg M, Hansson L, we provide additional analyses for the Arrighi HM. Forecasting the global burden of Lithell H. Hypertension is related to cognitive 758 Insulin Resistance and Cognitive Decline Diabetes Care Volume 40, June 2017

impairment: a 20-year follow-up of 999 men. neuropsychological assessment of Alzheimer’s 33. Willette AA, Bendlin BB, Starks EJ, et al. As- Hypertension 1998;31:780–786 disease. Neurology 1989;39:1159–1165 sociation of insulin resistance with cerebral glu- 19. Young SE, Mainous AG 3rd, Carnemolla M. 26. Hanninen¨ T, Pulliainen V, Salo J, et al. ja Suo- cose uptake in late middle-aged adults at risk for Hyperinsulinemia and cognitive decline in a men muistitutkimusyksikoiden¨ asiantuntijaryhma.¨ Alzheimer disease. JAMA Neurol 2015;72:1013– middle-aged cohort. Diabetes Care 2006;29: Kognitiiviset testit muistihairi¨ oiden¨ ja alkavan 1020 2688–2693 dementian varhaisdiagnostiikassa: CERAD- 34. Langbaum JB, Chen K, Lee W, et al.; Alz- 20. Yaffe K, Kanaya A, Lindquist K, et al. The tehtav¨ asarja.¨ Suom Laakaril 1999;54:1967– heimer’s Disease Neuroimaging Initiative. Cate- metabolic syndrome, inflammation, and risk of 1975 gorical and correlational analyses of baseline cognitive decline. JAMA 2004;292:2237–2242 27. Beck AT, Steer RA, Brown GK. Manual for fluorodeoxyglucose positron emission tomogra- 21. Aromaa A, Koskinen S. Health and func- the Beck Depression Inventory-II. San Antonio, phy images from the Alzheimer’s Disease Neu- tional capacity in Finland. Baseline results of TX, Psychological Corporation, 1996 roimaging Initiative (ADNI). Neuroimage 2009; the Health 2000 health examination survey [Inter- 28. Janis¨ MT, Siggins S, Tahvanainen E, et al. 45:1107–1116 net], 2004. Helsinki, Finland, National Public Health Active and low-active forms of serum phos- 35. Willette AA, Xu G, Johnson SC, et al. Insulin Institute. Available from http://urn.fi/URN:NBN:fi- pholipid transfer protein in a normal Finnish resistance, brain atrophy, and cognitive perfor- fe201204193452. Accessed 1 June 2016 population sample. J Lipid Res 2004;45:2303– mance in late middle-aged adults. Diabetes Care 22. Lundqvist A, Maki-Opas¨ T, Eds. Health 2011 2309 2013;36:443–449 survey – methods [Internet], 2016. Tampere, 29. McEvoy LK, Laughlin GA, Barrett-Connor E, 36. Segura B, Jurado MA, Freixenet N, Falcon´ C, Finland, Finnish National Institute for Health and et al. Metabolic syndrome and 16-year cognitive Junque´ C, Arboix A. Microstructural white mat- Welfare. Available from http://urn.fi/URN:ISBN: decline in community-dwelling older adults. ter changes in metabolic syndrome: a diffusion 978-952-302-669-8. Accessed 1 September 2016 Ann Epidemiol 2012;22:310–317 tensor imaging study. Neurology 2009;73:438– 23. Heistaro S, Ed. Methodology report. Health 30. Gourovitch ML, Kirkby BS, Goldberg TE, 444 2000 survey [Internet], 2008. Helsinki, Finland, et al. A comparison of rCBF patterns during let- 37. Makino T, Umegaki H, Suzuki Y, et al. Re- National Public Health Institute. Available from ter and semantic fluency. Neuropsychology lationship between small cerebral white matter http://urn.fi/URN:NBN:fi-fe201204193320. Ac- 2000;14:353–360 lesions and cognitive function in patients with cessed 1 June 2016 31. Shao Z, Janse E, Visser K, Meyer AS. What do Alzheimer’s disease and amnestic mild cognitive 24. Matthews DR, Hosker JP, Rudenski AS, verbal fluency tasks measure? Predictors of ver- impairment. Geriatr Gerontol Int 2014;14:819– Naylor BA, Treacher DF, Turner RC. Homeostasis bal fluency performance in older adults. Front 826 model assessment: insulin resistance and beta- Psychol 2014;5:772 38. Twamley EW, Ropacki SA, Bondi MW. Neu- cell function from fasting plasma glucose and 32. Baker LD, Cross DJ, Minoshima S, Belongia ropsychological and neuroimaging changes in insulin concentrations in man. Diabetologia D, Watson GS, Craft S. Insulin resistance and preclinical Alzheimer’s disease. J Int Neuropsy- 1985;28:412–419 Alzheimer-like reductions in regional cerebral chol Soc 2006;12:707–735 25. Morris JC, Heyman A, Mohs RC, et al. The glucose metabolism for cognitively normal 39. Braak H, Braak E. Frequency of stages of Consortium to Establish a Registry for Alz- adults with prediabetes or early type 2 diabetes. Alzheimer-related lesions in different age cate- heimer’s Disease (CERAD). Part I. Clinical and Arch Neurol 2011;68:51–57 gories. Neurobiol Aging 1997;18:351–357