|Home | About | Journals | Submit | Contact Us | Français|
It is accepted that markedly elevated thyroid-stimulating hormone (TSH) levels are associated with impaired cognitive function. However, the findings regarding the association between mildly elevated TSH levels and cognition are equivocal. The objective of this study was to assess the relation between TSH levels in the normal to mildly elevated range (0.3–10.0mIU/L) and several domains of cognitive function.
A healthy, community-based sample of 489 men and women (40–88 years old, mean=60.5 years) enrolled in the B-Vitamin Atherosclerosis Intervention Trial were studied. A neuropsychological test battery was used to assess a broad array of cognitive functions. Four uncorrelated neuropsychological factors were extracted by principal component analysis. Using multivariable linear regression, performance on each factor was examined in relation to TSH levels, controlling for age, gender, race-ethnicity, education, homocysteine levels, low-density lipoprotein cholesterol levels, and smoking status.
TSH levels were not associated with any of the four factor scores in the total sample or in younger (age<60) or older (age≥60) subjects, although there was a trend for older subjects with higher levels of TSH to do more poorly on paragraph recall (p=0.06). Gender-stratified analyses showed that TSH was positively associated with scores on word list learning for females only (p=0.003).
In this community-based sample of middle-aged to older individuals, increasing TSH levels were not associated with significantly reduced cognitive performance in any domain. Further exploration of the effects of gender on the association between TSH and cognition is warranted.
Thyroid hormones (TH) are essential for brain development and function, and TH levels have been associated with cognitive performance (1). Elevated thyroid-stimulating hormone (TSH) levels in the range found in patients with overt hypothyroidism are associated with impaired function in many cognitive domains. The association between subclinical hypothyroidism (mild elevation of TSH levels) and cognition is less clear (2).
Given the equivocal association between mildly elevated TSH levels and some areas of cognition, we sought to assess the relation between TSH levels in the normal to mildly elevated range (0.3−10.0mIU/L) and cognition in otherwise healthy, community-based, middle-aged and older men and women who had volunteered for a clinical trial on B-vitamin supplementation.
The B-Vitamin Atherosclerosis Intervention Trial is a randomized, double-blind, placebo-controlled trial designed to test whether B-vitamin supplementation will reduce the progression of early carotid artery atherosclerosis in subjects with elevated fasting plasma homocysteine levels, but without clinically evident cardiovascular disease. A secondary aim of this trial was to evaluate the impact of the B-vitamin intervention on cognitive performance. Briefly, potential subjects were prescreened by telephone and met initial screening eligibility if they were at least 40 years of age, postmenopausal (for women), and reported no history of diabetes, heart disease, stroke, or cancer.
A total of 2395 subjects met telephone prescreening criteria and attended at least one clinic screening visit. Of these, 545 met additional screening criteria (fasting plasma homocysteine>8.5μmol/L; no history of cardiovascular disease, no life threatening disease with prognosis<5 years, alcohol intake<5 drinks per day, or no substance abuse), and received neuropsychological testing before randomization to the study intervention. All subjects provided written informed consent. The study was approved by the Institutional Review Board of the University of Southern California.
Before randomization, a trained psychometrist (J.A.S.) (trained and overseen by a neuropsychologist [C.A.M.]) administered a battery of neuropsychological tests, designed to assess an array of cognitive functions and abilities. These tests were administered in a fixed order, and most subjects were able to complete the battery in a little over an hour (median=70 minutes, range 43−131 minutes). Tests and cognitive skills assessed were the following:
Participants fasted for 8 hours before blood draws. The blood draw for TSH was done in the morning, on the same day, and before neuropsychological testing. TSH was measured using the Roche Elecsys 2010 TSH method, third generation, with a functional sensitivity (20% interassay coefficient of variation over 6–8 weeks, determined in our lab [C.A.S.]) confirmed at 0.01mIU/L. Thyroid peroxidase antibodies (TPOAb) were measured by a highly sensitive direct radioassay system (Kronus, Boise, ID) (13). Normal levels for TPOAb were<1.0IU/mL; subjects with levels>1.0IU/mL were considered to be positive, and subjects with levels<1.0IU/mL were considered to be TPOAb negative. The presence of TPOAb is suggestive of thyroid dysfunction.
Demographic information (age, gender, race-ethnicity, education, and income) and medication use (including estrogen, thyroid supplements, and thyroid suppressants) were obtained by structured questionnaires. Fasting total plasma cholesterol and triglyceride levels were measured using an enzymatic method of the Standardization Program of the National Centers for Disease Control and Prevention. High-density lipoprotein cholesterol levels were measured after lipoproteins containing apolipoprotein B were precipitated in whole plasma using heparin manganese chloride. Fasting low-density lipoprotein (LDL) cholesterol was estimated using the Friedewald equation (14). Fasting total plasma homocysteine was determined according to the method of Araki and Sako (15). Mood was assessed using the Center for Epidemiological Studies Depression (CES-D) Scale (16). Estimated premorbid mental ability was assessed with the American National Adult Reading Test (AMNART) (17).
Data reduction of neuropsychological test results was undertaken to reduce the likelihood of spurious associations due to multiple hypothesis testing. A factor analysis with orthogonal rotation was performed among the entire cohort (n=545) on the 14 cognitive tests using the raw test scores. A four-factor solution was selected that accounted for 66% of the variance. The solution yielded four uncorrelated factors that can be broadly described based on their factor loadings. The first factor (1) (eigenvalue=5.40, percentage of variance=38%) generally represented General Cognition (namely, cognitive skills apart from episodic memory), with high factor loadings on Shipley Abstraction, Block Design, Judgment of Line Orientation, Trails-B, Boston Naming Test, Symbol Digit Modalities Test, Letter–Number Sequencing, and Animal Naming. The remaining three factors generally represented aspects of episodic memory: (2) Factor two—Word List Learning (including verbal episodic memory and conceptual ability; eigenvalue=1.55, percentage of variance=11%), with high factor loadings on CVLT immediate and delayed recall; (3) Factor three—Logical Memory (eigenvalue=1.38, percentage of variance=9%), with high factor loadings on immediate and delayed recall of the East Boston Memory Test story; and (4) Factor four—Visual Memory (eigenvalue=1.02, percentage of variance=7%) with high factor loadings on Faces I and II. Factor scores for each of the four factors were calculated for each subject. For subjects with missing data on cognitive tests (64 out of 8720 scores or<1.0% of all test scores), values were imputed using mean scores derived from the appropriate age (40–49, 50–59, 60–69, 70–79, and 80+years), education (high school or less, some college, bachelors degree, and graduate/professional degree), and gender group. Reductions (0.2–1.4%) in test variances from these imputations were small. As a sensitivity analysis, models were analyzed excluding subjects with missing data; the results were not altered.
Of the 545 subjects with neuropsychological test data, one subject did not have TSH data. Because the focus of this analysis was on cognitive function among individuals with normal to mildly elevated TSH, individuals with low (<0.3mIU/L) (n=3) and markedly high (>10.0mIU/L) (n=7) TSH levels [based on current laboratory practice guidelines (18)] were excluded from the analysis. Also excluded were subjects on thyroid supplements (n=34), and individuals with limited English language skills, such that the validity of the test was questioned by the psychometrist (n=11). TSH was transformed with a natural logarithmic function before statistical comparisons, because the distribution of TSH was skewed. Pearson correlations between continuous demographic variables and cognitive factor scores were computed to identify possible confounding variables. Because CES-D did not meet the criteria for confounding (i.e., did not correlate with both TSH levels and the factor scores), CES-D scores were not included as an adjusting covariate. Independent t-Tests were used to assess differences on factor scores by TPOAb status (positive or negative); these tests were conducted separately in elevated (3.1–10.0mIU/L) and normal (0.3–3.0mIU/L) TSH groups. As no differences on average factor scores were found by TPOAb status, all subjects were used, regardless of TPOAb status. This left a total of 489 subjects in the analyses.
The association between TSH and cognitive function was investigated using multivariable linear regression methods, modeling each of the four cognitive factors separately. The dependent variable was the factor score. The primary independent variable was a continuous log TSH variable. In building the regression model, the AMNART and education were highly and statistically significantly correlated. Further, education was a stronger correlate of cognitive function than the AMNART. To avoid co-linearity in regression models, education was chosen to control for expected intelligence and educational experience. Additional independent variables in the regression model included age, gender, race-ethnicity, plasma LDL-cholesterol, total homocysteine, and current smoking status (yes/no). These variables have been shown to be associated with cognition and TSH in other studies (19). To explore whether the association between TSH and cognitive performance differed by age or gender, we tested an interaction term for age×TSH, using age as a continuous variable, and gender×TSH. We did not explore for three-way interactions or for other two-way interactions. Data were analyzed using SAS statistical software (8.02 release; SAS Institute, Cary, NC).
Table 1 presents the demographic characteristics of the 489 subjects included in this analysis. More than half (55%) of the subjects were older than 60 years, male (64%), and non-Hispanic white (64%). TPOAb prevalence and TSH levels were higher in women than in men. The majority were highly educated, having a bachelor's, graduate, or professional degree (63%). Table 2 presents the neuropsychological raw test scores of the subjects. Men had slightly higher mean scores on all neuropsychological tests except for Trails B, episodic visual memory (Faces I and II), and one of two tests measuring episodic verbal memory (CVLT immediate and delayed).
In the total sample (n=489) log TSH was not associated with cognitive performance on any of the four cognitive factors (Table 3). A significant age by TSH interaction was found for factor 3 (reflecting Logical Memory; p=0.05). There was also a significant gender by TSH interaction for factor 2 (reflecting Word List Learning; p=0.003). No other significant age or gender interactions were apparent for any of the other factors.
Despite the significant interaction between age and factor 3 performance, age-stratified analyses did not show any significant associations between log TSH and factor 3 for either younger (age<60) or older (age≥60) subjects, although there was a trend for older individuals with higher levels of TSH to do more poorly on this factor (β=−0.21, p=0.06) (immediate recall β=−0.27, p=0.04; delayed recall β=−0.07, p=0.56, data not shown). Gender-stratified analyses showed that in women, log TSH was positively associated with scores on factor 2 (p=0.003).
Because of the possibility that subtle changes in cognition might be missed by reducing the data to four factors, an exploratory analysis was performed examining the effect of log TSH on the individual cognitive tests. The results were essentially the same as the regression on the factor scores. Similar to the main analyses, there was a significant gender–TSH interaction for the immediate (p=0.003) and delayed (p=0.05) recall of the word list learning test (data not shown). Gender-stratified analyses showed that among women, higher levels of TSH were associated with significantly better performance on the immediate recall of the word list learning test (β=2.17, p=0.009), and marginally significantly better performance on the delayed recall of the word list learning test (β=0.75, p=0.08, data not shown).
In this community-based sample of 489 volunteers 40–88 years old, there was no association between TSH levels (0.3–10.0mIU/L) and any cognitive factor when evaluated in the total sample, or when evaluated in younger (age<60) or older (age≥60) groups. Gender-stratified analyses showed that females with higher levels of TSH recalled significantly more words on a verbal learning test than females with lower levels of TSH.
Because of equivocal findings among studies, possibly due to differences in sample populations, cognitive domains studied, and tests utilized, it has been suggested that tests specifically assessing processing speed, sustained attention, and working memory, using populations of older men and women, might better capture deficits that may occur with subclinical hypothyroidism (2). Although we did use tests measuring these subdomains (e.g., Symbol Digit and Letter–Number Sequencing), in a middle-aged and older population of men and women, we did not detect any association between mildly elevated TSH levels and these cognitive domains. Our null finding for the association between TSH and processing speed and sustained attention (Symbol Digit Modalities Test [SDMT]) is in agreement with others who have examined these domains using the SDMT (or Digit Symbol Modalities Test) in mildly elevated TSH populations (20–23). Our null finding for the association between TSH and working memory, using Letter–Number Sequencing, is in agreement with some studies (20–22), but not others (23). It is not in agreement with two recent studies that both used an N-back task as a measure for working memory. Zhu et al. (24) found that newly identified, untreated, subclinical hypothyroid clinic patients (n=11, 10 female, age 17–47 years old [mean=29.67±7.67], TSH=14.67+7.13mIU/L) scored significantly lower on a two-back task compared to euthyroid subjects (p<0.012). This finding was supported by functional MRI results showing reduced activity in the frontal cortex of these patients. Samuels et al., using a randomized, double-blind, cross-over trial of usual dose versus lower dose of L-T4 on hypothyroid patients (n=19 females, 20–75 years old), found that patients had significantly lower scores on a three-back task on the lower dose (TSH=17.37+3.04), compared to their scores on the usual dose (TSH=2.19±0.35) (p=0.02) (25). The major difference between our study and these two studies is population: both used younger, predominantly female, subclinical, or hypothyroid patients, with markedly higher TSH levels. In our older population of subjects with mildly elevated TSH, we used Letter–Number Sequencing to assess working memory because it is viewed as sensitive to working memory deficits, better standardized, and more widely used than N-back tasks.
The significant positive association we found for women between increasing levels of TSH and immediate recall of word list learning is novel and in the opposite direction commonly hypothesized, but has been previously reported (26). The significant negative association between TSH and performance on the immediate recall of a paragraph among older individuals is in agreement with two other studies conducted among young to middle-aged women, with higher levels of TSH [5.6–28.6mIU/L (22);>5.0mIU/L, mean=8.8±1.5mIU/L (23)]. However, it was not reported if this deficit was in both the immediate and delayed recall of the paragraph. Our significant findings could be due to a chance association given the number of analyses that were performed. Even if replicated, the clinical significance of these two findings may be questioned, because there was no significant association between levels of TSH and either the delayed recall of the paragraph or the delayed recall of the word list.
Our generally null findings are in overall agreement with the recent large community- or population-based studies (n=281–5865) of older individuals (mean age>60, approximately equal numbers of males and females) examining the effects of mildly elevated TSH and/or subclinical hypothyroidism on cognition (20,27–29). Two of these studies used a test battery similar to our own, measuring several cognitive domains [attention, psychomotor speed, cognitive flexibility, fluency, and immediate and delayed memory (20,29)]. In contrast, a small study of community-dwelling elderly (mean age=74; n=97) found that individuals with mildly elevated TSH levels (4.0–8.78mIU/L) performed significantly worse on the immediate and delayed recall of a word list and the Mini-Mental State Exam, but not differently on tests measuring working memory or processing speed, compared to individuals with normal TSH levels (0.4–4.0mIU/L), after controlling for depression (21).
We found no association between log TSH levels and depression, as assessed by the CES-D (β [SE]=0.68 [0.54], p=0.21). Other studies examining the relationship between subclinical hypothyroidism and depression or depressive symptoms have been equivocal. However, our null finding between TSH levels and depressive symptoms is in agreement with two recent randomized placebo-controlled, double-blind studies examining the effect of thyroxine replacement in subclinical hypothyroidism, using a similar TSH range (≤10.0mIU/L). Both studies found no differences between thyroxine-treated and placebo groups on depressive symptoms (20,30). Two recent studies that have reported levels of well-being to be significantly lower in hypothyroid patients versus the general population, or for depression to occur significantly more often in subclinical hypothyroid patients than in individuals with overt hypothyroidism, did not find significant associations between levels of well-being or depression and TSH levels (31,32).
Our overall null findings are at variance with research on mechanisms by which TH affects cognition and mood (33). However, such studies have included young to middle-aged animal or human subjects with much higher TSH levels (>10.0mIU/L) indicative of marked hypothyroidism. Extrapolating these results to populations with lower TSH levels, other age groups, and equally to men and women may not be appropriate, as suggested by our results and others noted above.
Our study has several limitations. Participants were chosen based on their fasting homocysteine level (≥8.5μmol/L). Although they were in otherwise good health and although we adjusted for homocysteine levels in our analysis, homocysteine is also associated with cognition (34), and our results may not generalize to individuals with lower homocysteine levels. Additionally, hyperhomocysteinemia may be associated with hypothyroidism (35), although in our study homocysteine and TSH levels were not correlated (r=0.039, p=0.40).
Secondly, although we did measure TPOAb status, the prevalence of which is comparable to (although slightly lower than) a large population-based study (36), we did not measure thyroxine and triiodothyronine and thus were not able to accurately determine whether subjects with mildly elevated TSH had subclinical hypothyroidism and/or mild thyroid failure. Additionally, TSH values were drawn only once; laboratory protocol specifies multiple testing to accurately measure thyroid status. Thus, the poor association we found between TSH levels and the cognitive data could be due to reduced TSH reliability. However, our results are in agreement with other studies that were able to classify thyroid status, suggesting that our population was no different. Further, although we were interested in mildly elevated TSH levels, restricting the range of TSH values may have lead to a loss of power in identifying TSH-cognitive associations. Finally, because this was a cross-sectional study, we were not able to ascertain the associations of cognition with chronically elevated TSH levels. Our ability to detect associations between cognition and mild TSH elevation is reduced if cognitive outcomes depend on the chronicity of TSH elevations.
The strengths of our study include the relatively large sample size of both middle-aged and older adults and a diverse racial-ethnic population. This large sample size enabled us to detect correlations as small as r=0.13, with 80% power for n=489 (r=0.17 for n=267; r=0.19 for n=222). Additionally, we used a comprehensive neuropsychological battery that assessed multiple cognitive domains. Finally, we were able to measure and control for several important confounders.
In sum, we found that elevated TSH levels within the range of 0.3–10.0mIU/L were not associated with notably reduced cognitive performance in middle-aged and older individuals. These results challenge the notion that the elderly may be more cognitively vulnerable to higher TSH levels. Our exploratory analyses raise the possibility of a gender effect that may be of interest for future studies.
The authors wish to thank the B-Vitamin Atherosclerosis Intervention Trial participants and their families for their time. This study was supported by NIH Grants RO1AG-017160 and 5-T32 AG00037.
No competing financial interests exist.