Latinos: HCHS/SOL Results Neil Schneiderman10 Diabetes Care 2018;41:1501–1509 |

Diabetes Care Volume 41, July 2018 1501

Hector M. Gonzalez,´ 1 Wassim Tarraf,2 Metabolic Syndrome and Priscilla Vasquez,´ 1 Ashley H. Sanderlin,3 Natalya I. Rosenberg,4 Sonia Davis,5 Neurocognition Among Diverse Carlos J. Rodr´ıguez,6 Linda C. Gallo,7 Bharat Thyagarajan,8 Martha Daviglus,4 Middle-Aged and Older Hispanics/ Tasneem Khambaty,9 Jianwen Cai,5 and Latinos: HCHS/SOL Results Neil Schneiderman10 Diabetes Care 2018;41:1501–1509 | https://doi.org/10.2337/dc17-1896

OBJECTIVE Hispanics/Latinos have the highest risks for metabolic syndrome (MetS) in the U.S. and are also at increased risk for Alzheimer disease. In this study, we examined associations among neurocognitive function, MetS, and inflammation 1Department of Neurosciences and Shiley- ’ among diverse middle-aged and older Hispanics/Latinos. Marcos Alzheimer s Disease Research Center, University of California, San Diego, La Jolla, CA 2Institute of Gerontology and Department of RESEARCH DESIGN AND METHODS Healthcare Sciences, Wayne State University, Cross-sectional data (2008–2011) from theHispanic Community Health Study/Study Detroit, MI of Latinos (HCHS/SOL) were analyzed to examine associations between neuro- 3Section of Gerontology and Geriatric Medicine, cognition and MetS among diverse Hispanics/Latinos (N = 9,136; aged 45–74 years). Department of Internal Medicine, Wake Forest School of Medicine, Winston-Salem, NC 4Institute for Minority Health Research, College RESULTS of Medicine, University of Illinois at Chicago, MetS status was associated with lower global neurocognition, mental status, verbal Chicago, IL learning and memory, verbal fluency, and executive function. Age significantly 5Collaborative Studies Coordinating Center, De- modified the associations between MetS and learning and memory measures. partment of Biostatistics, The University of North Carolina at Chapel Hill, Chapel Hill, NC Significant associations between MetS and neurocognition were observed among 6Departments of Medicine and Epidemiology, RISK METABOLIC AND CARDIOVASCULAR middle-aged Hispanics/Latinos, and all associations remained robust to additional Wake Forest School of Medicine, Winston-Salem, covariates adjustment. NC 7Institute for Behavioral and Community Health CONCLUSIONS and Graduate School of Public Health, San Diego State University, San Diego, CA WefoundthatMetSwasassociatedwithlowerneurocognitivefunction,particularly 8Department of Laboratory Medicine and Pathol- inmidlife.Ourfindingssupportandextendcurrenthypothesesthatmidlifemaybea ogy, University of Minnesota Medical Center particularly vulnerable developmental period for unhealthy neurocognitive aging. Fairview, Minneapolis, MN 9Department of Psychology, University of Mary- land, Baltimore County, Baltimore, MD 10Department of Psychology, University of Miami, The metabolic syndrome (MetS) is a cluster of five overlapping risk factors for diabetes, Coral Gables, FL heartdisease,andneurocognitiveimpairmentanddisorders(1,2).MetSemergesinmid- Corresponding author: Hector M. Gonzalez,´ life and is related to the development of diabetes, heart disease, and stroke (3). MetS [email protected]. affects about one-third of U.S. adults and occurs earlier and more commonly among Received 11 September 2017 and accepted 7 Hispanics/Latinos compared with other racial/ethnic groups (4). Like MetS, neuro- April 2018. cognitive decline, mild cognitive impairment, and Alzheimer disease and related de- This article contains Supplementary Data online mentias (ADRD) are currently thought to emerge in midlife and progress with age (5–7). at http://care.diabetesjournals.org/lookup/ Although age and apoE4 genotypes are leading risks of ADRD, they are not modifiable. suppl/doi:10.2337/dc17-1896/-/DC1. Thus, there is heightened interest in preventing neurocognitive decline and ADRD by © 2018 by the American Diabetes Association. identifying and modifying cardiovascular disease (CVD) risks earlier in the life course. As Readers may use this article as long as the work is properly cited, the use is educational and not such, MetS prevention and control may also serve as effective means for reducing not only for profit, and the work is not altered. More infor- CVD, but also potentially neurocognitive decline and dementias (8), particularly among mation is available at http://www.diabetesjournals Hispanics/Latinos who are at high risk for MetS, ADRD, and low health care access (9). .org/content/license. 1502 Metabolic Syndrome and Neurocognitive Function Diabetes Care Volume 41, July 2018

Previous MetS studies reported asso- to this age-group given the age distribu- assistants. Four neurocognitive tests ciations with neurocognitive impairment tions of Hispanics/Latinos in the four com- were used in this study: 1)Six-Item and ADRD in older adults (10,11), but munitiesofinterest.Adetaileddiscussion Screener (SIS) (18), 2) Brief Spanish En- principally among those with high levels of the study objectives, design, and im- glish Verbal Learning Test (B-SEVLT) ofinflammatorybiomarkers(12–14).Less plementation has been published else- (19,20), 3) Word Fluency (WF) test of is known about middle age, when MetS where (16,17). the Multilingual Aphasia Examination emerges and neurocognitive function is (21,22), and 4) Digit Symbol Subtest thought to become vulnerable (8,15). In Neurocognitive Tests (DSS) (23). Three neurocognitive tests terms of primary prevention, middle age Neurocognitive tests were administered were only available in English (SIS, WF, is likely when effective behavioral and in quiet testing rooms in each of the andDSS).Therefore,theyweretranslated pharmacological interventions are indi- four HCHS/SOL clinics. Participants were by a certified translator and reviewed by cated to arrest MetS from further de- tested in their preferred language (En- the HCHS/SOL Translation Committee veloping into diabetes, CVD events, and glish or Spanish) during face-to-face in- to ensure the Spanish translations were brain pathologies; that is, provided ad- terviews by trained, bilingual research appropriate for diverse, Pan-American equate health care is available, especially to the most vulnerable populations. — In this study, we examined associa- Table 1 Characteristics of the HCHS/SOL target population by MetS status tions among neurocognitive function, No MetS MetS Total (n = 4,270) (n = 4,866) (n = 9,136) P value MetS, and inflammation among diverse middle-agedandolderHispanics/Latinos. Sex (%) Given that current opinions suggest that Female 52.9 56.3 54.7 0.0221 neurocognitive decline and ADRD pathol- Education (%) 0.000 Less than HS 35.0 43.4 39.5 ogy begins in midlife, we sought to com- HS or equivalent 21.7 20.7 21.2 pare midlife and later life associations More than HS 43.3 35.9 39.3 among neurocognitive function, MetS, Background (%) 0.0039 and inflammation in this important and Dominican 9.8 8.4 9.0 vulnerable population of Hispanics/Latinos. Central American 6.4 6.7 6.6 Cuban 25.2 29.9 27.7 RESEARCH DESIGN AND METHODS Mexican 32.8 29.4 30.9 Puerto Rican 17.5 18.7 18.1 Study Sample South American 6.2 4.7 5.4 We used data from the Hispanic Com- Other 2.1 2.3 2.2 munity Health Study/Study of Latinos Language preference (HCHS/SOL). The HCHS/SOL is a large, Spanish 83.3 88.3 86.0 0.000 epidemiological study of Hispanic/Latino Antihypertensives (%) men and women from diverse back- Yes 12.5 39.9 27.3 0.000 grounds.Recruitmentanddatacollection Antidiabetics (%) occurred between 2008 and 2011 and Yes 4.9 26.7 16.7 0.000 included 16,415 participants from four Age, years (mean) 54.8 (9.4) 57.8 (9.9) 56.5 (9.9) 0.000 major U.S. communities (Bronx, NY; Chi- CES-D-10 (mean) 7.2 (7.7) 7.8 (7.9) 7.5 (7.8) 0.001 cago, IL; Miami-Dade, FL; and San Diego, MetS components (%) 0.000 CA) with known large Hispanic/Latino Abdominal obesity 67.5 100.0 85.0 concentrations. Study participants were Impaired TG 13.9 59.2 38.3 sampled to ensure representation of Impaired HDL 15.1 60.0 39.2 Hispanic/Latino adults, aged 18–74 years Impaired BP 35.7 76.3 57.6 at recruitment into the study, in the tar- Impaired fasting glucose 23.9 73.3 50.5 get populations. Sample size was set to MetS components count (%) 0.000 None 8.6 0.0 4.0 allow for appropriate inferences to six 1 33.4 0.0 15.4 major Hispanic/Latino groups, including 2 52.9 0.0 24.4 Dominicans, Central Americans, Cubans, 3 3.7 48.6 27.8 Mexicans, Puerto Ricans, South Ameri- 4 1.4 34.4 19.2 cans, and other. Briefly, the HCHS/SOL 5 0.0 17.1 9.2 was designed to collect detailed demo- MetS components sum (mean) 1.6 (0.9) 3.7 (0.9) 2.7 (1.6) 0.000 graphic, sociocultural, and health data Cognitive outcomes through survey questionnaires and rich B-SEVLT-sum (mean) 22.9 (6.9) 22 (6.8) 22.4 (6.9) 0.000 biological specimens (e.g., blood and B-SEVLT-recall (mean) 8.2 (3.6) 7.9 (3.5) 8.1 (3.5) 0.000 urine) from consenting participants. WF (mean) 19.2 (9) 17.6 (8.6) 18.3 (8.8) 0.000 DSS (mean) 36.1 (16.7) 32 (15.8) 33.9 (16.4) 0.000 The design of the study included a two- Global cognition (mean) 51.1 (11.3) 49.3 (11.2) 50.2 (11.3) 0.000 stage probability sampling approach SIS #4 (%) 14.1 17.5 16.0 0.003 that oversampled adults 45–74 years MetS and its components are defined using the IDF specifications. HS, high school. of age to allow appropriate inferences care.diabetesjournals.org Gonzalez´ and Associates 1503

Latinos. In addition, the Neurocognitive for HDL abnormalities; 3) elevated blood To examine differential effects due to Reading Center principal investigator pressure (BP) (systolic BP $130 mmHg or inflammation, we used a joint modeling (H.M.G.) reviewed the original and trans- diastolic BP $85 mmHg), based on the approach (equivalent to split modeling lated tests to ensure translation accuracy average of three 1-min separated BP for stratified analyses) interacting inflam- and test fidelity. The neurocognitive test- readings,oruseofantihypertensivemed- mation groups (high vs. normal) hsCRP ing andscoring procedures usedin HCHS/ ications; and 4) impaired fasting glu- levels (hsCRP ,3 vs. $3 mg/L) with all SOL have been previously described (24). cose ($100 mg/dL), previous diabetes model covariates. This hsCRP threshold Briefly, the SIS (range 0–6) is a mental diagnosis, or use of antidiabetic medi- is associated with high CVD risk (26). In status test that was scored dichoto- cations. additional sensitivity analysis, we also con- mously with a reported value of #4 trolled for nonsteroidal anti-inflammatory representing cognitive impairment. Covariates drug use because they are associated with The cut point reflects previous validation We accounted for four sociodemographic inflammation reduction. work in non-Latino white patients with variables:age,sex,levelofeducation(less dementia (18). However, because the than high school, high school or equiv- Analytic Procedures SIS was not validated for dementia in alent,andmorethanhighschool),andthe All analyses were done using procedures Hispanics/Latinos, we will refer to a low six Hispanic/Latino backgrounds (i.e., Do- for complex surveys in the Stata software SIS score plainly as low mental status. The minican, Central American, Cuban, Mex- package (14.1; StataCorp, College Station, B-SEVLT is an episodic learning and mem- ican, Puerto Rican, and South American, TX). HCHS/SOL sampling weights were ory test with two scores: 1) the summed and other). We also controlled for de- adjusted for nonresponse and calibrated total correctly learned items across three pressive symptoms using the 10-item to the 2010 U.S. Census. All estimates and learning trials (B-SEVLT-sum; range 0–45) Center for Epidemiological Studies De- inferences were generatedusing methods fi and 2) total correctly recalled items (B- pression Scale (CES-D-10) and for eld appropriate for subpopulation analyses fi SEVLT-recall; range 0–15) following an center site to account for potential var- and accounted for the strati cation, clus- interference trial. WF is a language and iations due to locales or study protocol tering, and probability weighting in HCHS/ verbal fluency test (continuous) of implementation. SOL to allow correct generalizations to the summed correctly generated words (beginning with letters F and A) within 1 min — (i.e., for each letter). DSS is a mental Table 2 Association of MetS with neurocognitive function among Hispanics/Latinos of diverse backgrounds processing speed and executive function b (SE) test that was a continuous totalof correct responses (range 0–90). Model 1 Model 2 Model 3 We used generalized structural equa- B-SEVLT-sum tion modeling for survey data to fita MetS 20.07 (0.025)** 0.42 (0.162)* 0.53 (0.153)*** confirmatory factor using the four con- Age 20.03 (0.002)*** 20.02 (0.002)*** 20.02 (0.002)*** 2 2 tinuous neurocognitive measures and MetS*age 0.01 (0.003)** 0.01 (0.003)*** the dichotomous mental status measure B-SEVLT-recall MetS 20.02 (0.022) 0.47 (0.158)** 0.58 (0.149)*** and subsequently derive a factor score Age 20.03 (0.001)*** 20.02 (0.002)*** 20.02 (0.002)*** (mean 50; SD 10) for a latent global MetS*age 20.01 (0.003)** 20.01 (0.003)*** neurocognition measure. In addition to WF modeling global neurocognition as a pri- MetS 20.15 (0.028)*** 20.15 (0.185) 20.09 (0.029)** mary outcome, we also independently Age 20.01 (0.002)*** 20.01 (0.003)*** 20.004 (0.002)* examinedeachofthefourneurocognitive MetS*age 0.00 (0.003) NA↟ testsaswellaslowmentalstatus(SIS#4). DSS All continuous neurocognitive outcomes MetS 20.15 (0.022)*** 20.23 (0.185) 20.09 (0.021)*** 2 2 2 were standardized (Z scored) to facilitate Age 0.04 (0.002)*** 0.04 (0.003)*** 0.03 (0.001)*** MetS*age 0.0015 (0.003) NA↟ cross-test result comparisons and interpretation of effect sizes in SD terms. Global cognition MetS 20.07 (0.023)** 0.37 (0.158)* 0.50 (0.144)*** Age 20.03 (0.002)*** 20.03 (0.002)*** 20.02 (0.002)*** MetS MetS*age 20.01 (0.003)** 20.01 (0.003)*** The MetS was measured following In- ternational Diabetes Federation (IDF) OR (95% CI) fi speci cations (25). A participant was # fi Low mental status (SIS 4) classi ed as meeting MetS if they satis- MetS 1.13 (0.95; 1.34) 1.71 (0.48; 6.16) 1.07 (0.90; 1.29) fied criteria for abdominal obesity ($94 cm Age 1.05 (1.04; 1.06)*** 1.05 (1.03; 1.07)*** 1.05 (1.04; 1.06)*** for men and $80 cm for women) plus any MetS*age 0.99 (0.97; 1.02) NA↟ two of four factors including 1)elevated Estimates based on survey-generalized linear regression models assuming a Gaussian distribution triglyceride (TG) level ($150 mg/dL) or for continuous outcomes (Z scores) and logistic distribution for low mental status. MetS was treatment for TG abnormalities; 2)low defined using the IDF specifications. Model 1: age and sex adjusted; model 2: model 1 plus age*MetS HDL cholesterol (,40 mg/dL for men interaction; model 3: model 2 plus education, Hispanic/Latino background, and CES-D-10 scale. *P , 0.05; **P , 0.01; ***P , 0.001; ↟Interactionexcludedbecausenotsignificant in model 2. and ,50 mg/dL for women) or treatment 1504 Metabolic Syndrome and Neurocognitive Function Diabetes Care Volume 41, July 2018

target population. Our analytic sample test for differential nonadditive effects of middle-aged (45–64 years; n = 7,899) included 9,136 Hispanic/Latino partici- of continuous age on the associations and older individuals (65–74 years; n = pants; Supplementary Fig. 1 details the between MetS status and neurocognition. 1,237). Using 65 years as a threshold study’s inclusion criteria. The study pro- To facilitate the interpretation of results, corresponds with and enables descriptive tocol was reviewed and approved by the we calculated the marginal means of the comparisons with published results on institutionalreviewboardsatUniversityof continuous neurocognitive outcomes and the associations between MetS and neu- California, San Diego, and all other par- marginal probabilities for the dichotomous rocognition from other cohort studies ticipating sites. SIS by MetS status and plotted them across (13,14). The splines tests and P values are Our data analyses were conducted in the age continuum (Fig. 1). In sensitivity presented in Table 3, the estimated slopes three steps. First, we generated descrip- analyses, we independently adjusted underlying these tests are presented in tive statistics to characterize the target for 1) antidiabetic and antihypertensive Supplementary Table 1, and the predicted subpopulation overall and by MetS status medications and 2) nonsteroidal anti- marginal means and probabilities by MetS (Table 1). We tested for significant differ- inflammatory drug use. Adjustment for status over age are presented in Fig. 2. ences by MetS status using survey- these medications did not have any mea- In order to provide additional infor- adjusted x2 tests for the categorical surableeffectonthereportedresults mation that could inform clinical and covariates (e.g.,education) andtwo-tailed (H.M.G., W.T., unpublished observations). prevention practice, we examined indi- t tests for the continuous measures (e.g., Additionally, we explicitly modeled vidual MetS components in relation to CES-D-10). Second, we used survey gener- and tested differential age slopes by neurocognitivefunction.Additionally,we alizedlinearmodels(assumingaGaussian MetS status for middle-aged and older examined neurocognitive performance distribution for continuous outcomes and adults using linear age splines with knots as a function of a higher count of com- logisticforthedichotomouslowmental setat65years.Linearsplinesarepiecewise ponent indicators. To do so, we used status indicator) to examine the associ- regressionmodelstoexamineprespecified similar survey-generalized linear models ations between the MetS and neuro- hypotheses on differential nonlinear asso- to separately examine the covariate- cognition (Table 2). For each outcome, ciations between outcomes and predic- adjusted associations among the individ- we fit two incremental models to 1) test tors.Theoreticaldiscussions(27–29)and ual components of MetS, as well as a sum age- and sex-adjusted associations and 2) an applied perspective (30) that guided ofthe five individual components andour examine attenuations due to control for work in this manuscript have been pub- neurocognitive outcomes (Table 4). As education, Hispanic/Latino background, lished elsewhere. Briefly, the linear splines with the overall syndrome models, we depressive symptoms, and study site. allow us to separately model, and sub- used interactions to test for differential Given the importance of age to both sequently test, the intercepts and slopes nonadditive effects of continuous age on neurocognition and metabolic health, for age and age by MetS status inter- the associations between the individual we used interactions to account and actions in the two distinct age groupings predictors and neurocognition. Finally, to test for possible differentia- tion by inflammation severity, we refitthe models as described above and tested for differential effects in the association between MetS and neurocognition by normal (,3)andhigh($3) CRP groups. To do so, we used a joint modeling approach interacting the primary predictor and all covariateswiththebinaryCRPindicator.The results for these analyses are presented in Supplementary Table 2, and the predicted marginal means and probabilities for each CRP group by MetS status over age are presented in Supplementary Fig. 2.

RESULTS Descriptive Statistics More than half of Hispanics/Latinos in the target subpopulation satisfied criteria for MetS (54%), and the prevalence varied by age (50.4 vs. 67.2% among 45–64 and 65–74-year-olds,respectively).Descriptive characteristics for the target subpopula- Figure 1—A–F: Estimated marginal means (probabilities) of neurocognitive function over age by tion and by MetS status are provided in MetS status. Estimates based on survey-generalized linear regression models assuming a Gaussian Table 1. Slightly more than half of the tar- distribution for continuous outcomes (Z scores) and logistic distribution for low mental status. MetS defined using the IDF specifications. Models include sex, age*MetS interaction, education, get subpopulation (54.6%) was female, Hispanic/Latino background, and CES-D-10. Pr, predicted probability. and the average age was 56.4 years. Indi- viduals meeting criteria for MetS were care.diabetesjournals.org Gonzalez´ and Associates 1505

Table 3—Tests of differences in slopes of neurocognitive function by age splines associations between MetS and global b and MetS status neurocognition ( Main = 0.50 [SE 0.144]; b 2 D SE P value Interaction= 0.01[SE0.003]),B-SEVLT-sum (bMain = 0.53 [SE 0.153]; bInteraction = 20.01 B-SEVLT-sum b Slope differences age ,65 20.012 0.004 0.0066 [SE0.003]),andB-SEVLT-recall( Main =0.58 b 2 Slope differences age $65 0.038 0.020 0.0561 [SE 0.149]; Interaction = 0.01 [SE 0.003]). No MetS: slope differences age ,65 vs. age $65 0.055 0.016 0.001 The associations between MetS and WF MetS: slope differences age ,65 vs. age $65 0.005 0.013 0.711 (b = 20.09 [SE 0.029]) and DSS (b = 20.09 B-SEVLT-recall [SE0.021]) were attenuatedby 36and39%, Slope differences age ,65 20.013 0.004 0.0027 respectively. The estimated coefficients, SEs Slope differences age $65 0.024 0.020 0.2291 (for the continuous outcomes), and CIs (for , $ No MetS: slope differences age 65 vs. age 65 0.031 0.016 0.053 thedichotomousSIS)arepresentedinTable2. MetS: slope differences age ,65 vs. age $65 20.007 0.014 0.62 Figure 1F presents the fully adjusted mar- WF ginal means for the neurocognitive scores Slope differences age ,65 20.001 0.005 0.7519 Slope differences age $65 20.017 0.025 0.5041 and probability of low mental status over No MetS: slope differences age ,65 vs. age $65 0.001 0.021 0.97 age by MetS status and their 95% CIs. MetS: slope differences age ,65 vs. age $65 0.016 0.016 0.307 DSS Age (Linear) Splines Models Slope differences age ,65 20.004 0.004 0.2787 We found that the effect of MetS on Slope differences age $65 0.014 0.017 0.3929 neurocognition was modified by age- No MetS: slope differences age ,65 vs. age $65 0.001 0.014 0.968 groups (45–64 vs. $65 years) for global , $ 2 MetS: slope differences age 65 vs. age 65 0.018 0.010 0.069 neurocognition and in particular for mem- Global cognition ory outcomes. We found that MetS status Slope differences age ,65 20.013 0.004 0.0036 Slope differences age $65 0.035 0.020 0.0722 was associated with more pronounced age No MetS: slope differences age ,65 vs. age $65 0.047 0.016 0.003 slopes among younger individuals (45–65 MetS: slope differences age ,65 vs. age $65 20.001 0.014 0.926 years), but not so among older individuals Low mental status ($65 years). Specifically, the age slope Slope differences age ,65 20.002 0.018 0.9017 differences among younger individuals Slope differences age $65 20.161 0.078 0.0383 with MetS were steeper relative to those No MetS: slope differences age ,65 vs. age $65 20.152 0.068 0.025 not meeting MetS criteria with respect to , $ MetS: slope differences age 65 vs. age 65 0.007 0.040 0.854 global cognition (D = 20.013 [SE 0.004]), , $ Age 65 vs. age 65 (age65*MetS interaction) B-SEVLT-sum(D=20.012[SE0.004]),and B-SEVLT-sum 0.0204 B-SEVLT-recall (D = 20.013 [SE 0.004]). B-SEVLT-recall 0.0162 The differences in slopes between indi- WF 0.7379 viduals with and without MetS were not DSS 0.1223 statisticallydistinguishableamongthosein Global cognition 0.0409 the$65yearsgroup.Furthermore,among Low mental status 0.8723 those not meeting criteria for MetS, the Estimates based on survey-generalized linear regression models assuming a Gaussian distribution age slopes were more pronounced among for continuous outcomes (Z scores) and logistic distribution for low mental status. MetS older adults ($65 years) relative to youn- was defined using the IDF specifications. Models include sex, age splines (at age 65)*MetS ger individuals with respect to global interaction, education, Hispanic/Latino background, and CES-D-10 scale. cognition (D = 0.047 [SE 0.016]), B- SEVLT-sum (D = 0.055 [SE 0.016]), and, older andmorelikelytobefemale,haveless B-SEVLT-sum (b = 20.07 [SE 0.025]), WF to a slightly lesser extent, B-SEVLT-recall than a high school education, and use anti- (b = 20.15 [SE 0.028]), and DSS (b = 20.15 (D = 0.031 [SE 0.016]). The interaction tests hypertensive and antidiabetic medications. [SE 0.022]) but not with B-SEVLT-recall or and their P values for the age-groups and mental status (Table 2). Given our stated the slope contrasts for the linear age splines Crude Differences in Neurocognitive interest in midlife associations, we ac- by MetS status are included in Table 3, and Function counted for interactions between age the corresponding slopes are plotted in Fig. Individuals with MetS had lower average and MetS, which provided evidence for 2. The estimated slopes for the linear age global neurocognitive performance and significant modification effects with global splines, their SEs, and P values are included were more likely to satisfy criteria for low neurocognition (b = 0.37 [SE 0.158]; in Supplementary Table 1 to facilitate a mental status. Additionally, MetS was Main b = 20.01 [SE 0.003]), particu- clearer reading of the plots included in Fig. 2. uniformlyassociatedwithlowerperform- Interaction larly the memory measures, namely ances on B-SEVLT-sum, B-SEVLT-recall, B-SEVLT-sum (b = 0.42 [SE 0.162]; WF, and DSS (Table 1). Main Associations with MetS Components bInteraction = 20.01 [SE 0.003]) and There were notable variations in the Adjusted Associations B-SEVLT-recall (bMain = 0.47 [SE 0.158]; associations betweentheindividual com- In age- and sex-adjusted models, MetS bInteraction = 20.01 [SE 0.003]) (Table 2). ponents and neurocognitive outcomes status was inversely associated with global Adjusting for the additional covaria- (Table 4). We found significant age- neurocognition (b = 20.07 [SE 0.023]), tes did not considerably change the modified effects of fasting glucose in 1506 Metabolic Syndrome and Neurocognitive Function Diabetes Care Volume 41, July 2018

(NHANES) (4). It is noteworthy that nationally, the rates of MetS among Hispanics/Latinos have steadily increased over the past decade (4). At the population level, our findings suggest that these increasing rates of MetS could have negative implications for neurocognitive aging among Hispanics/Latinos. Previous studies have report mixed results on lower neurocognitive function among older adults meeting criteria for MetS (13,14,31–33). Authors of a recent systematic review concluded that the inconsistent findings could be explained by the age-modifying effects of MetS on lower neurocognitive function. That is, the MetS effects on neurocognition were pronounced among the young-old (i.e., ,70 years) compared with older Figure 2—Estimated marginal means (probabilities) of neurocognitive function by linear age (i.e.,$70years)adults(34).Furthermore, splines over age by MetS status. Estimates based on survey-generalized linear regression models among adults .70 years, MetS was not assuming a Gaussian distribution for continuous outcomes (Z scores) and logistic distribution for low mental status. MetS defined using the IDF specifications. Models include sex, age splines (at found to be associated with lower neu- age 65)*MetS interaction, education, Hispanic/Latino background, and CES-D-10. Pr, predicted rocognitive function and was reportedly probability. protective against neurocognitive decline (35,36). We found that those meeting criteria for MetS age-associated terms of performance on global neuro- (Supplementary Table 2). The adjusted neurocognitive decrements were consis- cognition (b = 0.47 [SE 0.156]; marginal means for the neurocognitive Main tent in both middle age and older adult- b = 20.01 [SE 0.003]), learning scores and probability of low mental Interaction hood. In contrast, among participants memory (b = 0.53 [SE 0.164]; status over age by MetS status for each Main not meeting criteria for MetS, there b =20.01[SE0.003]),andrecall CRP group and their 95% CIs are pre- Interaction were more marked neurocognitive dec- (b = 0.42 [SE 0.159]; b = 20.01 sented in Supplementary Fig. 2. Main Interaction rements among older adults relative to [SE 0.003]). We also found age-modified those in middle age. This suggests that effectsofelevatedBPonlearningand CONCLUSIONS MetSisatriggerforneurocognitivechange memory scores (b = 0.34 [SE 0.161]; Main Based on this large sample of diverse and decline starting in middle age. An b =20.01[SE0.003]).ElevatedTGs Interaction middle-agedandolderHispanics/Latinos, alternative explanation for the less pro- (b = 20.05 [SE 0.021]), HDL (b = 20.06 [SE we found that MetS was related to lower nounced neurocognitive MetS effects 0.020]), BP (b = 20.06 [SE 0.021]), and neurocognitive function in midlife and, in very old adults comes from a recent fastingglucoselevels(b=20.06[SE0.020]) in a less pronounced way, among older- meta-analysis of aggregated (1.3 million were linked to lower, non–age-modified aged adults. Our finding supports and participants), longitudinal cohort data of performance on DSS. HDL (b = 20.07 [SE extends current hypotheses that midlife BMI (a MetS component). Firstly, higher 0.027]), BP (b = 20.09 [SE 0.026]), and may be a particularly vulnerable period midlife BMI was associated with in- fasting glucose levels (b = 20.07 [SE for the development of neurocognitive creased all-cause mortality, whereas 0.029]) were also linked to lower perfor- decline. Secondly, we found that midlife higher BMI was related to lower all-cause mance on the WF test. Finally, a higher episodic learning and memory were par- mortality among very old adults ($85 count of MetS components was consis- ticularly vulnerable to MetS effects, years) (37). Secondly, weight loss is com- tentlyassociated with a more pronounced which has direct implications for the monly observed clinically in very old decrease over age in global cognition and development of memory impairment, patients with concomitant neurocogni- the two memory measures (Table 3) and which is the hallmark of AD. Thirdly, tive decline, particularly among those in with lower, non–age-modified scores on we found no evidence that inflammation terminaldecline.Bothexplanationscould the DSS (b = 20.03 [SE 0.008]) and WF (hsCRP) moderated associations be- provide insights into the lack of a robust (b = 20.04 [SE 0.010]). As with the overall tween MetS and neurocognitive perfor- negative association between neurocog- syndrome, there were no statistically mance in middle age or in later life, which nition and MetS we observed among significant associations between the in- differs from previous studies of older older adults in this study. The age-related dividual components or their count and Hispanics/Latinos (14). In our large and MetS associations with neurocognition lower mental status scores. representative sample of middle-aged findings in our Hispanic/Latino cohort CRP-Differentiated Models and older Hispanics/Latinos, over half studywereconsistentwiththeconclusion We found no effect modifications by met MetS criteria, which is consistent of this systematic review and other inflammation severity in the relation- with national estimates from National previous studies. Middle age has been ships between MetS and neurocognition Health and Nutrition Examination Survey often overlooked in the existing literature care.diabetesjournals.org Gonzalez´ and Associates 1507

Table 4—Association of individual MetS components and components count with neurocognitive function among Hispanics/Latinos of diverse backgrounds b (SE) OR (95% CI) B-SEVLT-sum B-SEVLT-recall WF DSS Global cognition Low mental status Component 1 Abdominal obesity 0.45 (0.232) 0.33 (0.216) 20.01 (0.033) 0.05 (0.028) 0.4 (0.212) 0.9 (0.72; 1.12) Age 20.02 (0.003)*** 20.02 (0.003)*** 20.00 (0.002)** 20.03 (0.001)*** 20.02 (0.003)*** 1.05 (1.04; 1.06)*** Obesity*age 20.01 (0.004) 20.01 (0.004) NA NA 20.01 (0.004) NA Component 2 TGs 0.06 (0.165) 0.37 (0.160)* 20.04 (0.024) 20.05 (0.021)* 0.14 (0.154) 0.93 (0.77; 1.11) Age 20.02 (0.002)*** 20.02 (0.002)*** 20.00 (0.002)** 20.03 (0.001)*** 20.02 (0.002)*** 1.05 (1.04; 1.06)*** TGs*age 20.002 (0.003) 20.01 (0.003)* NA NA 20.003 (0.003) NA Component 3 HDL cholesterol 20.08 (0.152) 0.14 (0.156) 20.07 (0.027)** 20.06 (0.020)** 20.04 (0.145) 1.03 (0.87; 1.23) Age 20.02 (0.002)*** 20.02 (0.002)*** 20.00 (0.002)** 20.03 (0.001)*** 20.03 (0.002)*** 1.05 (1.04; 1.06)*** HDL*age 0.001 (0.003) 20.002 (0.003) NA NA 0.00 (0.003) NA Component 4 Elevated BP 0.34 (0.161)* 0.06 (0.152) 20.09 (0.026)*** 20.06 (0.021)** 0.24 (0.150) 1.08 (0.89; 1.31) Age 20.02 (0.003)*** 20.02 (0.002)*** 0 (0.002) 20.03 (0.001)*** 20.02 (0.002)*** 1.05 (1.04; 1.06)*** Elevated BP*age 20.01 (0.003)* 20.002 (0.003) NA NA 20.01 (0.003) NA Component 5 IFG 0.53 (0.164)** 0.42 (0.159)** 20.07 (0.029)* 20.06 (0.020)** 0.47 (0.156)** 1.12 (0.92; 1.36) Age 20.02 (0.002)*** 20.02 (0.002)*** 20.00 (0.002)* 20.03 (0.001)*** 20.02 (0.002)*** 1.05 (1.04; 1.06)*** IFG*age 20.01 (0.003)*** 20.01 (0.003)** NA NA 20.01 (0.003)** NA Sum components Sum 0.16 (0.058)** 0.17 (0.058)** 20.04 (0.010)*** 20.03 (0.008)*** 0.15 (0.055)** 1.01 (0.95; 1.08) Age 20.01 (0.003)*** 20.01 (0.003)*** 20.00 (0.002)* 20.03 (0.001)*** 20.02 (0.003)*** 1.05 (1.04; 1.06)*** Sum*age 20.003 (0.001)** 20.003 (0.001)** NA NA 20.003 (0.001)** NA MetS MetS 0.53 (0.153)*** 0.58 (0.149)*** 20.09 (0.029)** 20.09 (0.021)*** 0.50 (0.144)*** 1.07 (0.90; 1.29) Age 20.02 (0.002)*** 20.02 (0.002)*** 20.00 (0.002)* 20.03 (0.001)*** 20.02 (0.002)*** 1.05 (1.04; 1.06)*** MetS*age 20.01 (0.003)*** 20.01 (0.003)*** NA NA 20.01 (0.003)*** NA Estimates based on survey-generalized linear regression models assuming a Gaussian distribution for continuous outcomes (Z scores) and logistic distribution for low mental status. MetS and its components are defined using IDF specifications. Models adjust for age, sex, education, Hispanic/Latino background, and CES-D-10 scale. IFG, impaired fasting glucose; NA, interaction not included in the model. *P , 0.05; **P , 0.01; ***P , 0.001. due to common misconceptions that neu- learning and memory, the associations this cluster ofsymptoms in relation to brain rocognitiveproblemsarerestrictedtoolder between executive function and MetS function and health. As a whole, MetS ef- age. Our findings provide additional ev- were not significantly modified by age. fect sizes in relation to neurocognition idence and support for recognizing that That is, the associations between MetS were small. However, MetS was asso- midlife may be a particularly vulnerable with executive function were similar be- ciatedwiththeequivalentof$2yearsof period of neurocognitive aging. As such, tween middle-aged and older Hispanics/ global neurocognitive aging when com- our study results indicate that there may Latinos in HCHS/SOL. This finding also pared with MetS-free participants in this be a window of opportunity for MetS suggests executive function may be study. At the population level, such sav- intervention in midlife. Although prelim- more continuously vulnerable to MetS ings in neurocognitive aging in this highly inary, additional biomarker (e.g., neuro- throughout middle age and older adult- vulnerable population could potentially imaging)andclinicalinformationwillhelp hood. translate into more years free of neuro- elucidate the nature of our age-related A closer examination of the relation- cognitive impairment and its associ- observations among Hispanics/Latinos. ships between the individual neurocog- ated social, physical, and economic In this study, episodic learning and nitive tests and MetS components revealed burdens. memory were most notably related to a pattern in which the associations between In contrast to previous studies (13,14), MetS, which is important because mem- ageandlowerepisodiclearningandmemory inflammation was not related to neuro- ory impairment is a hallmark symptom of scores were most strongly modified by cognition and MetS in HCHS/SOL. It is AD,themostcommonformofdementing impaired fasting glucose. Additionally in unclear why inflammation had no effects disorders. As noted above, the relation- ourstudy,verbalfluency,processingspeed, on the relationships between MetS and ship between neurocognitive function and executive function were associated neurocognitivefunctioninthisstudy.One and MetS was modified by age. Secondly, withmultipleMetScomponents.Neverthe- likely explanation derives from the rel- executive function was lower among less, our findingthathavingahighercount ative youth of the HCHS/SOL sample individuals with MetS compared with of impaired MetS components was asso- compared with previous studies, which the MetS-free group, which is consistent ciated with lower neurocognitive perform- primarily consisted of older adults. It may with previous work (34). Unlike episodic ances emphasizes the value of examining be that MetS is an early and primary 1508 Metabolic Syndrome and Neurocognitive Function Diabetes Care Volume 41, July 2018

catalystforlowerneurocognitivefunction, population representing nearly one-fifth the study and takes responsibility for the whereas inflammation is a secondary re- of the U.S. population that has a very high integrity of the data and the accuracy of the sponse that may be more apparent later prevalence of MetS. Future studies are data analysis. in life. Clearly, our single cross-sectional planned to examine Hispanic/Latino di- References study must be replicated in other pop- versity in relation neurocognition and 1. Grundy SM, Cleeman JI, Daniels SR, et al.; ulations. However, the biological brain MetS, including sex differences by His- American Heart Association; National Heart mechanism in middle age merits further panic/Latino background. Diverse His- Lung, and Blood Institute. Diagnosis and man- investigation in humans and nonhuman panics/Latinos have been overlooked agement of the metabolic syndrome: an Amer- models, which may provide additional and long neglected in any health science, ican Heart Association/National Heart, Lung, and fi insights for earlier preventive and ther- but particularly neurocognitive aging and Blood Institute Scienti c Statement: Executive Summary. Circulation 2005;112:e285–e290 apeutic opportunities. Nevertheless, our ADRD research. 2. Grundy SM, Brewer HB Jr,CleemanJI,SmithSC results point to MetS as a potential Jr, Lenfant C; National Heart, Lung, and Blood target for interventions aimed at main- Conclusion Institute; American Heart Association. Definition Based on this representative sample of taining neurocognitive health in diverse of metabolic syndrome: report of the National middle-aged and older Hispanics/Latinos, Hispanics/Latinos beginning in middle Heart, Lung, and Blood Institute/American Heart .54% met MetS criteria (24). We found Association conference on scientific issues re- age. that MetS was associated with lower neu- lated to definition. Arterioscler Thromb Vasc Biol There are several limitations that read- 2004;24:e13–e18 rocognitive function, particularly among ers should note when interpreting these 3. Arenillas JF, Moro MA, Davalos´ A. The met- middle-agedadults(4,9).Assuch,prevent- study results. Firstly and as mentioned abolic syndrome and stroke: potential treatment ing and managing MetS in this vulnerable approaches. Stroke 2007;38:2196–2203 above, this was a cross-sectional study, population has important public health 4. Aguilar M, Bhuket T, Torres S, Liu B, Wong RJ. which precludes causal inferences. As implications. Prevalence of the metabolic syndrome in the such, we were unable to observe neuro- United States, 2003-2012. JAMA 2015;313: cognitive decline. Additionally, we did 1973–1974 not ascertain mild cognitive impairment 5. Petersen RC. Mild cognitive impairment as a Acknowledgments. The authors thank the staff diagnostic entity. J Intern Med 2004;256:183–194 or ADRD. Secondly, although large, our of and participants in HCHS/SOL for important 6. Mueller SG, Weiner MW, Thal LJ, et al. Ways sample was representative of Hispanics/ contributions (see http://www.cscc.unc.edu/ toward an early diagnosis in Alzheimer’s disease: Latinos in our targeted communities. Our hchs/ for a list of investigators). the Alzheimer’s Disease Neuroimaging Initiative sample is not nationally representative. Funding. H.M.G. and W.T. received support for (ADNI). Alzheimers Dement 2005;1:55–66 HCHS/SOL is a large prospective cohort this work from National Institute on Aging (NIA) 7. Vemuri P, Knopman DS, Lesnick TG, et al. grants R01-AG048642, RF1-AG054548, and R21- Evaluation of amyloid protective factors and study that is focused primarily on heart AG053760. H.M.G. also received support from Alzheimerdiseaseneurodegenerationprotective disease, and the neurocognitive battery NIA grant P30-AG005131 and W.T. from NIA factors in elderly individuals. JAMA Neurol 2017; was limited. This limitation contributed to grant P30-AG053760. They previously received 74:718–726 our inability to identify participants meeting support from National Heart, Lung, and Blood 8. Misiak B, Leszek J, Kiejna A. Metabolic syn- ’ criteria for mild cognitive impairment or Institute (NHLBI) grant HC65233. The HCHS/SOL drome,mildcognitiveimpairmentandAlzheimer s was carried out as a collaborative study sup- disease–the emerging role of systemic low-grade ADRD. Thirdly, the HCHS/SOL cohort is ported by contracts from the NHLBI to the inflammationandadiposity.BrainResBull2012;89: relatively young (18–74 years), and the University of North Carolina (N01-HC65233), 144–149 number of participants .65 years of age University of Miami (N01-HC65234), Albert 9. Henry J. Kaiser Family Foundation. Uninsured is relatively small compared with those in EinsteinCollegeofMedicine(N01-HC65235),North- rates for the nonelderly by race/ethnicity, U.S. western University (N01-HC65236), and San Diego [Internet]. Available from https://www.kff.org/ middle age, which could have limited our State University (N01-HC65237). The following uninsured/state-indicator/rate-by-raceethnicity. ability to detect significant associations institutes/centers/offices contributed to the Accessed 12 March 2018 between neurocognition and MetS HCHS/SOL through a transfer of funds to the 10. Bokura H, Nagai A, Oguro H, Kobayashi S, among older participants. Fourthly, we NHLBI: National Institute on Minority Health and Yamaguchi S. The association of metabolic syn- did not objectively measure English/ HealthDisparities, NationalInstituteon Deafness drome with executive dysfunction independent of fi and Other Communication Disorders, National subclinical ischemic brain lesions in Japanese adults. Spanish language pro ciency and relied Institute of Dental and Craniofacial Research, Dement Geriatr Cogn Disord 2010;30:479–485 exclusively on study participants’ pre- National Institute of Diabetes and Digestive and 11. PanzaF,FrisardiV,CapursoC,etal.Metabolic ferred language for test administration, Kidney Diseases, National Institute of Neurolog- syndrome and cognitive impairment: current which could introduce some imprecision ical Disorders and Stroke, and the National epidemiology and possible underlying mecha- fi – to our results. It is noteworthy, however, InstitutesofHealth(NIH)Of ceofDietarySupple- nisms. J Alzheimers Dis 2010;21:691 724 . ments. This work was supported by the NIH. 12. Taylor VH, MacQueen GM. Cognitive dys- that 85% of participants preferred test- The contents of this work are solely the re- function associated with metabolic syndrome. ing in Spanish. Fifthly, we only examined sponsibility of the authors and do not necessarily Obes Rev 2007;8:409–418 one inflammatory marker (hsCRP) to rep- represent the official views of the NIH. 13. Yaffe K, Haan M, Blackwell T, Cherkasova E, licate previous findings, and other mark- Duality of Interest. No potential conflicts of Whitmer RA, West N. Metabolic syndrome and fi ers could yield divergent results. Sixthly, interest relevant to this article were reported. cognitive decline in elderly Latinos: ndings from Author Contributions. H.M.G. conceived and the Sacramento Area Latino Study of Aging study. HCHS/SOL did not include older adults researched the field. W.T. cowrote the manu- J Am Geriatr Soc 2007;55:758–762 .74 years of age. Consequently, we may scriptandconductedtheanalyses.P.V.andA.H.S. 14. Yaffe K, Kanaya A, Lindquist K, et al. The have been unable to find any MetS- assistedindraftingandeditingthemanuscript. metabolic syndrome, inflammation, and risk of protective effects reported in other stud- N.I.R., S.D., C.J.R., L.C.G., B.T., M.D., T.K., and cognitive decline. JAMA 2004;292:2237–2242 N.S. reviewed and edited the manuscript. J.C. 15. Feng L, Chong MS, Lim WS, et al. Metabolic iesofveryoldadults.Furthermore,HCHS/ edited the manuscript and provided statistical syndrome and amnestic mild cognitive impair- SOLisauniquepopulationofmiddle-aged expertise. H.M.G. is the guarantor of this work ment: Singapore Longitudinal Ageing Study-2 and older Hispanics/Latinos, a diverse and, as such, had full access to all of the data in findings. J Alzheimers Dis 2013;34:649–657 care.diabetesjournals.org Gonzalez´ and Associates 1509

16. Lavange LM, Kalsbeek WD, Sorlie PD, et al. Community Health Study/Study of Latinos. Arch 31. Liu CL, Lin MH, Peng LN, et al. Late-life Sample design and cohort selection in the His- Clin Neuropsychol 2015;30:68–77 metabolic syndrome prevents cognitive decline panic Community Health Study/Study of Latinos. 25. Alberti KG, Eckel RH, Grundy SM, et al.; among older men aged 75 years and over: one- Ann Epidemiol 2010;20:642–649 International Diabetes Federation Task Force year prospective cohort study. J Nutr Health 17. Sorlie PD, Aviles-Santa´ LM, Wassertheil- on Epidemiology and Prevention; National Heart, Aging 2013;17:523–526 Smoller S, et al. Design and implementation Lung, and Blood Institute; American Heart As- 32. LiuM,HeY,JiangB,etal.Associationbetween of the Hispanic Community Health Study/ sociation; World Heart Federation; International metabolic syndrome and mild cognitive impair- Study of Latinos. Ann Epidemiol 2010;20:629–641 Atherosclerosis Society; International Associa- ment and its age difference in a Chinese com- 18. Callahan CM, Unverzagt FW, Hui SL, Perkins tion for the Study of Obesity. Harmonizing munity elderly population. Clin Endocrinol (Oxf) AJ, Hendrie HC. Six-item screener to identify the metabolic syndrome: a joint interim state- 2015;82:844–853 cognitive impairment among potential subjects ment of the International Diabetes Federation 33. Watts AS, Loskutova N, Burns JM, Johnson for clinical research. Med Care 2002;40:771–781 Task Force on Epidemiology and Prevention; DK. Metabolic syndrome and cognitive decline in 19. GonzalezHM,MungasD,ReedBR,MarshallS,´ National Heart, Lung, and Blood Institute; Amer- early Alzheimer’s disease and healthy older Haan MN. A new verbal learning and memory ican Heart Association; World Heart Federation; adults. J Alzheimers Dis 2013;35:253–265 test for English- and Spanish-speaking older International Atherosclerosis Society; and Inter- 34. Siervo M, Harrison SL, Jagger C, Robinson L, people. J Int Neuropsychol Soc 2001;7:544–555 national Association for the Study of Obesity. Stephan BC. Metabolic syndrome and longitu- 20. Gonzalez´ HM, Mungas D, Haan MN. A verbal Circulation 2009;120:1640–1645 dinal changes in cognitive function: a systematic learningandmemorytestforEnglish-andSpanish- 26. Ridker PM. Clinical application of C-reactive review and meta-analysis. J Alzheimers Dis 2014; speaking older Mexican-American adults. Clin protein for cardiovascular disease detection and 41:151–161 Neuropsychol 2002;16:439–451 prevention. Circulation 2003;107:363–369 35. Harrison SL, Stephan BC, Siervo M, et al. Is 21. Lezak M, Howieson DB, Loring DW. Neuro- 27. Suits DB, Mason A, Chan L. Spline functions there an association between metabolic syn- psychological Assessment. New York, Oxford ﬁtted by standard regression methods. Rev Econ drome and cognitive function in very old adults? University Press, 2004 Stat 1978;60:132–139 The Newcastle 85+ Study. J Am Geriatr Soc 2015; 22. Benton AL, Hamsher K. Multilingual Aphasia 28. Smith PL. Splines as a useful and convenient 63:667–675 Examination. 2nd ed. Iowa City, IA, AJA Asso- statistical tool. Am Stat 1979;33:57–62 36. van den Berg E, Biessels GJ, de Craen AJM, ciates, 1989 29. Marsh LC, Cormier DR. Spline regression Gussekloo J, Westendorp RGJ. The metabolic syn- 23. Wechsler D. WAIS-R Manual: WechslerAdult models [Internet], 2002. Available from http:// drome is associated with decelerated cognitive de- Intelligence Scale - Revised. San Antonio, TX, methods.sagepub.com/book/spline-regression- cline in the oldest old. Neurology 2007;69:979–985 Psychological Corporation, 1981 models. Accessed 10 March 2018 37. Kivimaki¨ M, Luukkonen R, Batty GD, et al. 24. Gonzalez´ HM, Tarraf W, Gouskova N, et al. 30. Mitchell MN. Interpreting and Visualizing Body mass index and risk of dementia: analysis of Neurocognitivefunctionamongmiddle-agedand Regression Models Using Stata. Philadelphia, individual-level data from 1.3 million individuals. older Hispanic/Latinos: results from the Hispanic Taylor & Francis, 2012 Alzheimers Dement 2018;14:601–609