Executive Functioning in Extreme Obesity: Contributions from Metabolic Status, Medical Comorbidities, and Psychiatric Factors.

Participants were recruited from the University of Michigan bariatric surgery clinic (prior to surgical intervention) between March 2015 and June 2018. Inclusion criteria included (a) age 18 years or older, (b) body mass index (BMI) ≥ 40 or BMI ≥ 35 with at least one comorbid medical condition, and (c) able and willing to provide written informed consent. Lean control participants with no metabolic syndrome components were recruited through a University website. Lean control participants were excluded if they were taking medications for blood pressure, cholesterol, diabetes, or triglycerides. In total, 138 patients pursuing bariatric surgery and 46 lean control participants consented to the study and completed all baseline visits. Cognitive assessments were performed on 123/138 individuals with obesity and all 46 lean control participants. Some participants did not complete all NIH Toolbox ( n =4) and Wide Ranging Achievement Test–Fourth Edition (WRAT-4; n =2) assessments and were excluded from the study. In sum, 117 participants with obesity and 46 lean control participants were retained for analysis. This study was approved by the University of Michigan Institutional Review Board and all participants signed informed consent documents. Patient height, weight, and waist circumference were measured at study entry. Waist circumference was measured as the National Cholesterol Education Program (NCEP) defined waist, at the top of the iliac crest. Participants underwent blood pressure evaluations and a fasting lipid panel. Additionally, all participants without a prior diagnosis of diabetes underwent glucose tolerance testing. HbA 1C was obtained only for participants with obesity. Diabetes and pre-diabetes were defined according to the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus [ 22 ]. Additional medical comorbidities were cataloged using a modified Charlson Comorbidity Index [ 23 ] based on participant medical history, with the removal of diabetes in order to avoid double counting this condition. Participants were asked to report their lifetime tobacco smoking history on a study enrollment questionnaire. Problematic use of alcohol was evaluated using raw scores from the Alcohol Use Disorder Identification Test (AUDIT) [ 24 ]; the AUDIT has demonstrated good test-retest reliability ( r =.84) and excellent construct and predictive validity (PPV>.90 in most studies) [ 25 ]. Diagnosis of obstructive sleep apnea (OSA), including severity and method of treatment, was evaluated using sleep study reports from polysomnograms and home OSA tests completed no more than 2 years before participant surgery date. The number of medications (both CNS and non-CNS) participants were taking at study entry was also calculated. CNS medications included selective serotonin reuptake inhibitors (SSRIs), serotonin and norepinephrine reuptake inhibitors (SNRIs), tricyclic antidepressants (TCAs), antipsychotics, anxiolytics, mood stabilizers, antiepileptic medications, opioids, and other CNS-related medications (ondansetron, metoclopramide, diphenhydramine, naltrexone, prednisone, zolpidem, and cyclobenzaprine). Psychiatric symptoms were measured using norm-referenced subscale T-scores from the Minnesota Multiphasic Personality Inventory–2–Restructured Form (MMPI-2-RF) [ 26 ] measuring depression (RC2), anxiety (RC7), and somatic complaints (RC1). The MMPI-2-RF has demonstrated equivalent or improved reliability, convergent validity, and divergent validity when compared to prior versions of the Minnesota Multiphasic Personality Inventory; comprehensive psychometric characteristics are available elsewhere [ 27 ]. Potential for binge eating disorder was assessed using raw scores from the Questionnaire of Eating and Weight Patterns–Revised (QEWP-R) [ 28 ], the most frequently used questionnaire in bariatric surgery candidates; the QEWP-R has demonstrated fair construct and criterion validity (kappa=.26–.37) but limited research has evaluated reliability [ 29 ]. The number of chronic overlapping pain conditions (COPC) [ 30 ] was tabulated, included chronic lower back pain, endometriosis, irritable bowel syndrome, temporomandibular disorder, urologic chronic pelvic pain syndrome, migraine/chronic tension-type headache, myalgic encephalomyelitis/chronic fatigue syndrome, fibromyalgia, and vulvodynia. Variables of interest were classified into risk factor categories to reflect common classifications of variables that may impact cognition, as outlined above. These include metabolic risk factors, other medical comorbidities, OSA, medications, and psychiatric symptoms. Continuous variables were used wherever possible (e.g., waist circumference, MMPI-2-RF T-scores), and categorical variables were used where appropriate (e.g., smoking status, diabetic status). Specific variables contributing to each risk factor category are also reflected in Tables 3 and 4 . The primary cognitive outcomes were domain performances on the NIH Toolbox cognitive assessment and Rey AVLT. The NIH Toolbox is a brief, computerized assessment of global cognitive functioning comprised of subtests measuring selective attention/inhibition (Visual Flanker), set-shifting (Dimensional Change Card Sort), working memory (List Sorting), processing speed (Pattern Processing Speed), and episodic visual memory (Picture Sequence Memory). These five subtests are combined to yield a global composite score of global cognitive ability. The NIH Toolbox has documented strong test-retest reliability, with most intraclass correlation coefficients above 0.90. In adults, convergent validity statistics with other commonly used cognitive measures range from 0.46 to 0.65, whereas discriminant validity statistics range from 0.02 to 0.15, suggesting that the NIH Toolbox tasks appropriately measure the intended cognitive constructs [ 31 ]. The Rey AVLT is a word-list memory test with well-established normative databases that examine an individual’s acquisition, retention, and retrieval of verbal information [ 32 ]. The Rey AVLT has shown statistically significant but low correlations with tests assessing verbal intellect ( r =.12–.22), non-verbal intellect ( r =.07–.28), and full scale IQ ( r =.12–.29) [ 33 ]. Each NIH Toolbox domain score and Rey AVLT outcome was standardized based on age according to test normative data and further standardized to lean control participants based on educational equivalency using the WRAT-4 [ 34 ], a test of single word reading ability, for all regression analyses. Educational equivalency was preferred over years of formal education to better account for the quality of education on neuropsychological test performance [ 35 ]. Descriptive statistics were used to characterize study participants in terms of demographic information, medical and psychiatric comorbidities, and cognitive outcomes. Pearson’s chi-square or Fisher’s exact tests were used to compare participants with extreme obesity and lean control participants in terms of categorical risk factors. Two sample t -tests were used to assess group differences in continuous risk factors and cognitive outcomes. Cohen’s d was used to assess the standard effect size of group differences for continuous risk factors and Cohen’s ω was used to assess the standardized effect size of group differences for categorical risk factors. A p -value of .05 was used for significance testing in all analyses. While a Bonferroni-style adjusted p -value was considered to control for type I error due multiple comparisons, the sample sizes and small effects in the present analysis would have significantly increased the risk of type II error. In an effort to balance these competing considerations [ 36 , 37 ], we retained a traditional p -value of .05 and evaluated significant findings in the context of prior research and patterns of effects related to the variables of interest. As such, our exploratory findings should be interpreted with caution until confirmatory studies can be conducted using larger samples. A two-stage multivariable linear regression approach was used to assess associations between individual risk factors and cognition, after accounting for demographic factors. The first-stage regression model fit the age and WRAT-4 standardized cognitive outcomes as a function of race and sex. Then, studentized residuals (approximately normally distributed with mean 0 and variance 1) were calculated from the first-stage model. The second-stage model fit the studentized residuals as a function of each individual risk factor. To assess the relative adjusted importance that each risk factor category had on cognition, R 2 was calculated for the full model and for the full model holding each risk factor category out, one at a time. ANOVA F -tests were used to determine statistical significance of the R 2 increases corresponding to each risk factor category. Regression assumptions were confirmed via statistical tests and diagnostic plots. Each model was free of collinearity by evaluating variance inflation factors. All analyses were completed using R version 3.5.2.

Demographic factors, including medical and psychiatric variables, are presented in Table 1 . Groups did not differ on age, sex, or ethnicity, but the group of participants with obesity included more African American participants (15.4% compared to 2.2%, p =0.02, Cohen’s ω =0.23) and the group of lean control participants was more highly educated ( p <0.01, Cohen’s ω =0.32). Race and educational equivalency were used as covariates in all regression analyses to control for group differences. Compared to lean control participants, those with obesity had a worse metabolic profile: higher weight, waist circumference, SBP, DBP, triglyceride levels, fasting and 2-h glucose, and lower HDL (all p s<0.01). Participants with obesity also had a higher Charlson Comorbidity Index, took more CNS and non-CNS medications, more frequently had OSA, had a higher number of COPC, and reported increased somatic symptoms (all p s≤0.01). Participants with obesity had lower NIH Toolbox composite scores compared to lean control participants after accounting for the effects of age using test-specific normative data ( p <0.01). Specifically, cognitive deficits in the group of participants with obesity were most significant on measures of Visual Flanker ( p <0.01, Cohen’s d =0.67), Dimensional Change Card Sort ( p <0.01, Cohen’s d =0.63), Pattern Processing Speed ( p <0.01, Cohen’s d =0.72), and Rey AVLT total learning ( p =0.01, Cohen’s d =0.61) ( Table 2 ). Results from the multivariable linear regression models for the NIH Toolbox outcomes are displayed in Table 3 . Waist circumference was the only individual risk factor significantly associated with NIH Toolbox composite score (Point Estimate (PE): −0.07, 95% confidence interval (CI): −0.13, −0.02). When looking at specific subtests of the NIH Toolbox, larger waist circumference was also associated with worse performance on Visual Flanker (PE: −0.09, 95% CI: −0.14, −0.04) and Dimensional Change Card Sort (PE: −0.08, 95% CI: −0.13, −0.03). Additionally, Dimensional Change Card Sort performance significantly worsened as the number of CNS medications increased (PE: −0.24, 95% CI: −0.43, −0.06). Somatic symptom complaints (PE: −0.02, 95% CI: −0.04, −0.0001) and systolic blood pressure (PE: 0.14, 95% CI: 0.01, 0.27) were associated with performance on the List Sorting test and a history of smoking was associated with worse Picture Sequence Memory (PE: −0.46, 95% CI: −0.90, −0.03) performance. No risk factors were significantly associated with performance on the Pattern Processing Speed task. Regression results for the Rey AVLT outcomes are displayed in Table 4 . Higher levels of depression were associated with lower recognition performance (PE: −0.03, 95% CI: −0.06, 0.01). In contrast, higher anxiety scores were associated with improved performance on each Rey AVLT outcome: learning (PE: 0.03, 95% CI: 0.01, 0.06), recall (PE: 0.04, 95% CI: 0.01, 0.06), and recognition (PE: 0.04, 95% CI: 0.02, 0.07). The only non-psychiatric risk factor associated with any Rey AVLT outcomes was triglyceride levels, which were significantly associated with lower verbal learning performance (PE: −0.12, 95% CI: −0.24, −0.003). The combination of metabolic, psychiatric, medication use, OSA, and other medical comorbidity variables accounted for between 13.8 and 23.0% of the variance in the cognitive outcomes. The differences in model R 2 between the full models and when leaving each risk factor out one at a time are displayed in Fig. 1 . Metabolic factors had a significant effect on the Visual Flanker ( R 2 difference: 0.13, p =0.02) and Dimensional Change Card Sort (R 2 difference: 0.11, p =0.047) tasks. Medication use had a significant effect on the Dimensional Change Card Sort ( R 2 difference: 0.07, p =0.02) task. Psychiatric factors emerged as the most important risk factors for the Rey AVLT recognition ability, as evidenced by an R 2 increase of 0.13 ( p =0.01).

The present study builds on previously reported findings [ 21 ] by evaluating domain-specific cognitive functions and relevant medical and psychiatric factors that contribute to the relationship between waist circumference and reduced cognition in individuals with obesity. Consistent with the stated hypotheses, tasks measuring processing speed and executive functioning were significantly different between groups, while cognitive functions with less frontal-subcortical involvement, such as memory recall and recognition, were not significantly different between groups. Contrary to hypotheses, a measure of working memory did not reveal significant group differences, whereas a verbal list learning measure was significantly different between groups. Waist circumference among participants with extreme obesity was significantly related to the magnitude of executive deficits in selective attention/inhibition and set-shifting. Additionally, use of CNS medication contributed significantly to set-shifting deficits. Surprisingly, processing speed performances were not significantly associated with any variable of interest, despite group differences on this measure. In addition, results showed significant changes in some cognitive domains related to vascular factors including systolic blood pressure, smoking history, and triglycerides as well as psychiatric factors including somatic concerns, anxiety, and depression. Study findings generally support trends in the literature indicating that executive functions (here measured by selective attention/inhibition and set-shifting) are the cognitive domains most affected by obesity and likely account for most of the global declines in cognition observed in this population. Results demonstrated that these findings are driven primarily by waist circumference (and CNS medication use in the case of set-shifting) and not by a host of other common obesity-related medical and psychiatric conditions with known cognitive effects. Group differences in verbal learning, but not consolidation or recall of learned information, similarly reflect attentional and executive deficits impacting cognition. Together, these findings suggest that abdominal obesity itself is a primary contributor to cognitive deficits in this sample and that these deficits primarily consist of frontal/executive processes. Furthermore, it was only the contributions of metabolic and medication factors that were significantly associated with performance on these executive tasks. While significant subcortical processing speed differences were noted for the group of participants with obesity, these results were not significantly associated with any variable of interest, including waist circumference. This finding may reflect statistical limitations of the study (e.g., restricted range due to high rates of medical and psychiatric conditions in participants with obesity) or may indicate the presence of an unmeasured variable accounting for these differences. Given that responses on timed tasks from the NIH Toolbox were recorded by manual selection made on a tablet, it is possible that speed differences were the result of peripheral motor functioning and not cognitive processing [ 38 ]. Vascular factors of systolic blood pressure, smoking history, and triglycerides showed sporadic and somewhat unexpected effects on aspects of cognitive functioning. Surprisingly, systolic blood pressure was positively associated with performance on a working memory task despite that participants with obesity were, on average, mildly hypertensive. Smoking history contributed to declines in episodic visual memory but not verbal memory, and higher triglycerides were associated with reductions in verbal learning but not recall and recognition. Given the lack of consistent associations with multiple cognitive measures and the large number of comparisons, these results should be interpreted with caution and should be considered hypothesis generating. Psychiatric factors, including somatic concerns, depression, and anxiety, were also associated with cognitive outcomes. Somatic concerns were associated with reduced working memory performance in this sample. Somatic concerns are frequently elevated in individuals with obesity and are often attributed to chronic pain, gastrointestinal distress, cognitive concerns, and generalized malaise [ 17 , 18 , 20 , 39 ]. Higher levels of chronic somatic symptoms may divert attention and executive capacities making them less available for external demands (i.e., cognitive testing), again suggesting a frontal-subcortical contribution to this finding. Consistent with well-established memory deficits associated with depression, participants in this sample with higher levels of depression showed lower rates of recognition discriminability, a function of attentional and executive deficits affecting memory [ 40 ]. Greater levels of anxiety, on the other hand, were consistently related to better learning, recall, and recognition memory. Given that participants showed average rates of anxiety, the association between increased anxiety and improved memory functioning is interpreted to reflect the Yerkes–Dodson principle that mild levels of anxious arousal improve cognitive performance [ 41 ]. Overall, the strength of associations in these results is relatively small, with each model R 2 ranging from 0.138 to 0.230, even while accounting for many prominent obesity-related medical and psychiatric comorbidities with known effects on cognition. These findings suggest that additional unmeasured factors are likely making significant contributions to deficits in cognitive functioning. Furthermore, individual differences and interaction effects across variables of interest may result in multicausal and cumulative pathways to cognitive differences that are obscured when variables of interest are evaluated individually. Findings from the present study offer several important clinical considerations, including pre- and post-operative intervention opportunities that may improve clinical outcomes associated with surgical weight loss interventions. First among these is the clinical value of individualized cognitive assessment, especially of executive functions, for individuals with obesity seeking bariatric surgery. Identification of cognitive deficits, if present, allows for individually tailored interventions that may improve implementation of behavioral changes necessary to achieve healthy eating behaviors and sustain treatment adherence following bariatric surgery [ 7 , 42 , 43 ]. Such interventions may include compensatory accommodations aimed at reducing demand on executive functions or may focus on enhancing executive abilities directly [ 7 ]. Cognitive rehabilitation techniques geared toward remediating executive deficits have demonstrated mixed but promising results [ 7 , 44 , 45 ]. Indeed, bariatric surgery itself has been shown to be an effective intervention for improving executive functioning in individuals with obesity, and these cognitive changes following surgery have been associated with higher rates of sustained weight loss at 2- and 3-year post-operative follow-ups [ 12 , 43 , 46 – 49 ]. Furthermore, sustained weight loss in mid-life may help ameliorate pathological cognitive decline associated with aging, and adherence to diet and exercise interventions among elderly individuals with obesity has been shown to significantly improve cognition over a 1-year period [ 50 , 51 ]. Importantly, the clinical implications of the present findings suggest that individualized cognitive assessment, and executive functioning in particular, is essential for improving treatment adherence and post-operative outcomes for individuals presenting for bariatric surgery. While identifying an optimal cognitive assessment battery for this clinical application is beyond the scope of the present study, results indicate that even a brief, computerized assessment, such as the NIH Toolbox, serves as an important and clinically useful screener of executive functioning. Study findings should be interpreted in light of some limitations. First, the groups in this study differed on demographic variables of race and education such that, compared to controls, the group of participants with obesity included a higher proportion of racial minorities and attained lower levels of education overall, despite the fact that both groups were highly educated when compared to national trends. To address this limitation, differences in race and education were statistically adjusted in all regression analyses. Second, the cognitive battery used in this study does not represent a comprehensive neuropsychological evaluation that would account for additional measures of attention, visuospatial processing, language, and manual motor speed which may have further clarified the nature of cognitive differences observed (e.g., processing speed, visual memory functions). Third, as a cross-sectional study, longitudinal analysis of the relationship between obesity status and cognition over time or the role of weight loss on changes in cognition was not possible. However, a report of 2-year follow-up data on these participants after bariatric surgery is planned. Finally, this study included small sample sizes, modest effect sizes, and comparisons across multiple medical and psychiatric variables, thus complicating interpretation of results based on null hypothesis significance testing. Nevertheless, the pattern of results related to abdominal obesity and executive functioning is largely consistent with the literature on obesity and cognition, while accounting for the most common medical and psychiatric comorbidities associated with obesity. Larger replication studies will help to confirm and clarify these results.

While many studies have found cognitive differences associated with obesity, they have often failed to adequately account for medical and psychiatric confounds [ 3 ]. The present study replicated the most consistently documented cognitive differences associated with obesity (i.e., reduced executive functioning) while including a precise and extensive assessment of obesity-related metabolic, psychiatric, and comorbid medical conditions with known cognitive effects. Specifically, the present study found that waist circumference and use of CNS medications were associated with significantly lower executive functions of set-shifting and selective attention/inhibition. While processing speed was significantly slower in participants with obesity, this finding was unrelated to any variable of interest in this study, indicating a need for future research. Taken together, these findings support a relationship between abdominal obesity and mild cognitive deficits primarily driven by executive functions and independent of multiple medical and psychiatric conditions commonly co-occurring with obesity. Study findings underscore the clinical importance of assessing for executive deficits in individuals with obesity and tailoring individualized interventions that promote long-term post-operative clinical outcomes.

Obesity across the lifespan has been associated with deficits in cognitive functioning and higher incidence of dementia in later life [ 1 , 2 ]. These cognitive differences have been reported to affect diverse cognitive domains, including general intellectual ability, attention, processing speed, visuospatial skills, language, learning, and memory, with impairments in executive functions being most consistently documented [ 2 – 7 ]. While some data indicate a relationship between obese adiposity and cognitive deficits independent of commonly co-occurring medical conditions [ 2 , 8 ], additional obesity-related factors have also been shown to make independent contributions on cognition in this population [ 3 , 5 , 6 , 9 ]. Metabolic dysregulations such as disrupted insulin signaling in diabetes mellitus, microvascular changes due to hypertension, and systemic inflammatory processes associated with the accumulation of adipose tissue reflect causal mechanisms that may contribute to cognitive deficits in individuals with obesity [ 3 – 5 , 9 – 12 ]. In addition to these processes, multiple risk factors commonly associated with obesity, such as chronic pain, poor sleep, psychiatric symptoms, and medications with central nervous system (CNS) effects, have also been shown to contribute to cognitive inefficiencies, weight gain, or both [ 13 – 20 ]. A recent review by Gunstad et al. [ 7 ] examined the moderating contributions of cognitive deficits, especially executive dysfunction, in behavioral factors contributing to overeating, excess weight gain, and implementation of weight-loss intentions that present important considerations when seeking to optimize long-term clinical outcomes following bariatric surgery. Nevertheless, very few studies have comprehensively evaluated the relationship between extreme obesity and cognitive functioning while controlling for the most common medical and psychiatric factors that may be contributing to this relationship. The study authors previously evaluated global cognitive status in patients seeking bariatric surgery compared to healthy controls [ 21 ]. Adjusting for age and education, abdominal obesity was significantly associated with a lower global composite score on a computerized measure of broad cognitive functioning (NIH Toolbox Composite) but not with scores on a measure of verbal memory (Rey Auditory Verbal Learning Test (AVLT) Delayed Recall). Specifically, 21.4% of participants with obesity demonstrated generalized cognitive deficits, with waist circumference being the primary predictor of poorer performance. The present study seeks to evaluate specific cognitive domains that may account for the previously identified relationship between global cognition and obesity, while simultaneously accounting for the contributions of metabolic factors, medical comorbidities, and psychiatric symptoms on cognitive outcomes. It was hypothesized that frontal-subcortical networks would be most affected by obesity and related medical and psychiatric conditions, resulting in lower performances on standardized neuropsychological tests of executive functioning, attention, and processing speed in individuals with extreme obesity. Specifically, it was hypothesized that NIH Toolbox scores for Visual Flanker (selective attention and response inhibition), Dimensional Change Card Sort (set-shifting), List Sorting (working memory), and Pattern Processing Speed (processing speed) would be negatively associated with metabolic, medical, and psychiatric variables of interest. Furthermore, it was hypothesized that Rey AVLT scores for total learning (verbal encoding), delayed recall (memory retrieval), and delayed recognition (memory consolidation) would not reveal group differences as these diffusely organized memory networks would be less reliant on specific frontal-subcortical processes.