Association of Thyroid Dysfunction With Cognitive Function

This individual participant data analysis assesses the cross-sectional and longitudinal associations of baseline thyroid dysfunction with cognitive function and dementia.

T hyroid dysfunction is considered a potentially reversible cause of cognitive decline; hence, thyroid function screening tests are described in guidelines as an essential component of the workup for the diagnosis of dementia. [1][2][3] Thyroid dysfunction is frequently observed in individuals with suspected dementia. 4 However, the outcomes of treatment of overt hypothyroidism and hyperthyroidism and subclinical hyperthyroidism on cognitive function are not fully clarified. 5-7 For subclinical hypothyroidism, 4 of 5 recent randomized clinical trials and a meta-analysis on levothyroxine treatment did not find evidence for an improvement in cognitive function. [8][9][10][11][12][13] Moreover, meta-analyses of observational studies have yielded inconsistent results on associations of subclinical and overt thyroid dysfunction with cognitive impairment and risk of dementia. 14-17 An individual participant data analysis of cohort studies might help clarify the conflicting results of previous studies, as it allows for uniform definitions of thyroid dysfunction and it can assess the differential associations by age groups, sex, and thyroid medication in subgroup analyses. 18 In the present study, we investigated cross-sectional and longitudinal associations of thyroid dysfunction with cognitive function and dementia in an individual participant data analysis of multiple cohorts.

Study Population
We first approached the coordinating center of the Thyroid Studies Collaboration, a collaborative project of 25 existing longitudinal studies with information on thyroid status. 18 The Medical Ethics Committee of the Leiden University Medical Center waived the need for review owing to the retrospective nature of the study using only previously collected data; no individuals underwent interventions for the present study. Each participant gave informed consent to the original study they participated in, which was oral or written depending on the original study design and legislation at the time of data collection. All 15 Thyroid Studies Collaboration cohorts that had collected data on cognitive function or dementia joined the project. The study designs for all cohorts participating in the current study have been described previously in more detail. 19- 33 We approached 14 additional cohorts that were extracted from 4 recent meta-analyses on subclinical thyroid dysfunction and cognitive function or dementia. 14-17 Six of these cohorts consented to collaborating and sharing data. [34][35][36][37][38][39] Lastly, we included publicly available data of the National Health and Nutrition Examination Survey waves of 1999 to 2002 and 2011 to 2012, which simultaneously collected thyroid and cognitive function among many other parameters. 40

Thyroid Function
Thyroid dysfunction was determined biochemically by measurements of thyrotropin and free thyroxine (FT 4 ) concentrations in all cohorts. Cohort-specific cutoff values were used for FT 4 levels (eTable 1 in the Supplement). In accordance with previous projects in the Thyroid Studies Collaboration, participants were classified as euthyroid if thyrotropin level was 0.45 to 4.49 mIU/L. 18 Overt hyperthyroidism was defined as a thyrotropin level less than 0.45 mIU/L and FT 4 level above the reference range. Subclinical hyperthyroidism was defined as a thyrotropin level less than 0.45 mIU/L and FT 4 levels within the reference range, or only as thyrotropin level less than 0.45 mIU/L in absence of an FT 4 measurement (n = 896 among 10 cohorts) because overt hyperthyroidism is rare. 41 A combination of thyrotropin level of 4.50 to 20 mIU/L and FT 4 levels within the reference range was defined as subclinical hypothyroidism. Individuals who had missing FT 4 measurements with mildly elevated thyrotropin levels (4.50-20 mIU/L) were considered subclinically hypothyroid (n = 523 among 8 cohorts) because chances of overt hypothyroidism in this patient category are low. 41 A thyrotropin level of 20 mIU/L or greater or thyrotropin level of 4.50 mIU/L or greater combined with FT 4 levels below the reference range was defined as overt hypothyroidism.  [46][47][48][49][50] The minimal clinically important difference for executive function was defined as a difference of 4 points in LDST. 51 Memory was measured using either Rey's Auditory Verbal Learning Test (also referred to as Word Learning Test or Verbal Learning Test), Digit Span Test, or Visual Association Test. [52][53][54][55] No minimal clinically important difference for memory tests was found in the literature. MMSE were excluded from this analysis because cognitive function tests are insufficient to diagnose dementia. 56 Prevalence of dementia at baseline was available for 11 cohorts; 431 participants had a diagnosis of dementia at baseline, but only 78 of them were classified as noneuthyroid. Owing to the small number of participants with thyroid dysfunction at baseline, no cross-sectional analyses for dementia were performed.

Statistical Analyses
We used a 2-stage individual participant data analysis approach, which accommodates uniform definitions and analyses for each cohort while keeping complexity to a minimum. 18,57 The first stage consisted of study-level analysis of thyroid dysfunction and cognitive function or dementia conducted on the original data sets with participant-level data. In the second stage, the effect estimates from the first stage were pooled using a random-effects model based on restricted maximum likelihood. Heterogeneity across studies was quantified using the I 2 statistic: less than 40% was considered low heterogeneity; 40% to 75%, moderate heterogeneity; and greater than 75%, high heterogeneity.
For both the cross-sectional and longitudinal analyses between thyroid dysfunction and cognitive function, we used multivariable linear regression models. To facilitate combination of different scales, the results were transformed to standardized mean differences. In the prospective analysis of cognitive decline, we calculated the difference between the last available measurement of cognitive function and baseline cognitive function. The difference was divided by the follow-up time in years to obtain an annual decline, irrespective of duration of follow-up. The annual decline was subsequently standardized, resulting in a standardized mean difference in annual change in cognitive function allowing comparison of changes over time.
The risk of developing dementia during follow-up was assessed using Cox regression models. In these analyses, participants with dementia at baseline were excluded. For studies without precise registration of the date of dementia diagnosis, it was assumed that dementia developed halfway between the registration date and the last date that absence of dementia was ascertained.
Thyroid dysfunction (overt hyperthyroidism, subclinical hyperthyroidism, subclinical hypothyroidism, and overt hypothyroidism) was included as a categorical variable with the euthyroid group serving as reference. All analyses were adjusted for age and sex. The longitudinal analyses of cognitive decline were adjusted for baseline cognitive function. Prespecified subgroup analyses were performed by stratification and interaction analysis for sex and for age younger or older than 75 years. Additional analyses were performed with adjustment for educational attainment, though this variable was not available in all cohorts. In sensitivity analyses, participants with missing FT 4 measurements in the subclinical hyperthyroid and subclinical hypothyroid groups were excluded, as were those who used antithyroid medication or thyroid hormone replacement therapy at baseline. Furthermore, we assessed robustness of the associations by pooling the estimates using fixed-effect models and by excluding stud-ies with strata of fewer than 10 participants. To assess whether effects were dependent on degree of disruption of thyrotropin, analyses were repeated with thyrotropin categories of less than 0.10 mIU/L, 0.10 to 0.44 mIU/L, 4.5 to 6.9 mIU/L, 7.0 to 10 mIU/L, and greater than 10 mIU/L, in which participants with thyrotropin between 0.45 and 4.49 mIU/L served as reference. Lastly, instead of using biochemical cutoff points, thyrotropin and FT 4 were analyzed continuously across the full range with cognitive function. Thyrotropin was transformed using the natural logarithm; for both natural log-transformed thyrotropin and FT 4 , models were constructed per standard deviation. Continuous models were performed minimally adjusted by age and sex and with additional adjustment for educational attainment. For sensitivity purposes, the analyses were also conducted excluding the participants who used antithyroid medication or thyroid hormone replacement therapy at baseline. Cohorts with greater than 10% missing measurements for FT 4 were excluded for the continuous analyses on FT 4 . All P values were 2-tailed; statistical significance was set at P < .05.
Study-level analyses were performed using SPSS Statistics, version 25 (IBM). Effect estimates were pooled and summarized in forest plots using R, version 3.6.1 and metafor package (R Foundation for Statistical Computing). 58 This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
A total of 38 144 participants from 21 cohorts provided data on a measure of cognitive function (Table). The median age varied from 58 to 93 years, and 18 089 (47.4%) participants were women. Follow-up for cognitive decline was available for 14 cohorts, with a median follow-up duration varying from 1.7 to 11.3 years, accumulating 114 267 person-years.
Eight cohorts provided follow-up for dementia incidence on 46 606 participants (eTable 2 in the Supplement). Among these participants, 28 820 (61.8%) were women, and the median age at baseline was between 57 and 85 years. During follow-up, 2033 (4.4%) cases of incident dementia were identified. Median follow-up duration ranged from 3.8 to 15.3 years, accumulating 525 222 person-years.

Thyroid Dysfunction and Global Cognitive Function
Cross-sectionally, thyroid dysfunction was not associated with global cognitive function among 18 cohorts ( Figure 1; eFigure 1 in the Supplement). The largest observed difference was   decline per year based on the SD in the largest cohort for this analysis. Additional adjustment for educational attainment did not materially change the results (eFigure 2 in the Supplement). Stratification by age and sex did not show any differential effects for global cognitive function (eTable 3 in the Supplement). No statistically significant associations were found when individuals were categorized by severity of thyrotropin abnormality (eFigure 3 in the Supplement). Reanalyzing the data with a fixed-effects model or without strata with fewer than 10 participants did not yield different results (eTable 4 in the Supplement). Leaving out participants with missing FT 4 measurements or those using antithyroid medi-  cation or thyroid hormone replacement therapy at baseline also did not change the results. A positive association was found between continuous thyrotropin and global cognition only when thyroid supplementation and antithyroid medication users were excluded (0.028 higher standardized mean difference per SD; 95% CI, 0.003 to 0.053; P = .03; eTable 5 in the Supplement). No association between continuous FT 4 levels and global cognitive function was found. Heterogeneity across studies was low for the cross-sectional main analyses (I 2 = 0%-40%), while heterogeneity was low to moderate for the longitudinal and sensitivity analyses (I 2 = 0%-70%).

Thyroid Dysfunction and Executive Function and Memory
No negative association was observed cross-sectionally between thyroid dysfunction and executive function or memory among 11 and 8 cohorts, respectively (Figure 1; eFigures 4 and 5 in the Supplement). Participants with overt hyperthyroidism had 0.20 standardized mean difference (95% CI, 0.07 to 0.33; P = .002) higher executive function score compared with participants who were euthyroid; transformed, this would account for 1.6 more correct substitutions within 60 seconds for the LDST based on the largest cohort in this analysis. In both executive function and memory, participants with subclinical hypothyroidism performed better than participants who were euthyroid (executive function: 0.07 standardized mean difference; 95% CI, 0.01 to 0.13; P = .03; memory: 0.08 standardized mean difference; 95% CI, 0.01 to 0.15; P = .03). Longitudinally, no association was found between thyroid dysfunction at baseline and decline in executive function among 7 cohorts or memory among 6 cohorts; all differences were smaller than 0.1 standardized mean difference (Figure 2). Additional adjustment for educational attainment did not materially change the results (eFigure 2 in the Supplement). No statistically significant interaction with sex or age was present (P > .05 for all; supporting data in eTable 3 in the Supplement). Using a fixed-effects model or excluding strata with fewer than 10 participants did not change the results for executive function or memory (eTable 4 in the Supplement). The association of subclinical hypothyroidism and better executive function was attenuated when participants with missing FT 4 measurements were left out, while the association with memory was unchanged. The positive association between overt hyperthyroidism and executive function disappeared when participants using thyroid medication were removed. No association was found when individuals were categorized by severity of thyrotropin abnormality or when thyrotropin level was analyzed continuously (eFigure 3 and eTable 5 in the Supplement). Continuous analysis of FT 4 levels showed a positive association with executive function (0.019 higher standardized mean difference per SD; 95% CI, 0.002 to 0.036; P = .03), which was attenuated when participants using thyroid medication were left out. Heterogeneity across studies was low for the cross-sectional main analyses (I 2 = 0%-40%), while heterogeneity was low to moderate for the longitudinal analyses (I 2 = 0%-70%) and up to high heterogeneity in the sensitivity analyses (I 2 ≤ 73%).

Thyroid Dysfunction and Dementia
Cross-sectional analysis of thyroid dysfunction and dementia were unfeasible owing to few participants who were not euthyroid with dementia at baseline (78 participants among 11 cohorts). In longitudinal analyses among 12 cohorts, no association was found between thyroid dysfunction and incident dementia (Figure 3; eFigure 6 in the Supplement). The hazard ratio of dementia ranged from 1.54 (95% CI, 0.76 to 3.10) for overt hyperthyroidism to 0.79 (95% CI, 0.48 to 1.28) for overt hypothyroidism. Continuous analysis of thyrotropin and FT 4 levels also did not provide evidence for an association; hazard ratio, 0.96 per SD increase of natural log-transformed thyrotropin (95% CI, 0.91 to 1.02; P = .16); hazard ratio, 1.05 per SD increase of FT 4 (95% CI, 0.98 to 1.13; P = .16). Heterogeneity between studies was low (I 2 = 0%-40%).

Discussion
In this individual participant data analysis of 74 565 participants from 23 cohorts, there was no association between subclinical thyroid dysfunction and cognitive function, cognitive decline, or the onset of dementia. Owing to uncertainty of the results for overt hypothyroidism and hyperthyroidism, no rigorous conclusions can be drawn regarding the association between overt thyroid dysfunction and cognitive decline and dementia. While prior study-level meta-analyses also reported no association between subclinical hypothyroidism and cognitive function, cognitive decline, or dementia, they were limited by heterogeneity in definitions of thyroid dysfunction and choices of covariates in the statistical models. 14-17 Because we performed an individual participant data analysis, we could standardize definitions of thyroid function categories and of cog- nitive function and cognitive decline and standardize the statistical models. By addressing these limitations and reaching the same results, the present study provides the strongest observational evidence to date suggesting that subclinical hypothyroidism is not associated with cognitive function or cognitive decline. Various studies and 2 meta-analyses did show an association between subclinical or overt hyperthyroidism or low thyrotropin level within the reference range and a higher risk of dementia. 14,17,20,26,59-61 Although our findings for subclinical and overt hyperthyroidism and dementia did not reach statistical significance, they are directionally consistent with the literature. Despite combining 8 cohorts comprising more than 45 000 participants, the number of individuals with subclinical and overt hyperthyroidism and the number of individuals who developed dementia during follow-up are limited. Therefore, we cannot exclude a higher risk of dementia in individuals with hyperthyroidism. In addition, individuals with overt hyperthyroidism had a slightly higher rate of cognitive decline, though not statistically significant. Considering the existing literature and the other results in the present study, the observed cross-sectional association between overt hyperthyroidism and better executive function was most likely a chance finding. Moreover, the observed difference in executive function was less than half the minimal clinically important difference, making it a clinically insignificant finding regardless of the P value.
Higher vulnerability among subgroups has been proposed; younger adults and women might be more susceptible to cognitive dysregulation associated with thyroid dysfunction. 16,62 Moreover, cognitive decline might only be present in individuals with more extreme values of thyrotropin, 21,63 or variation in FT 4 instead of thyrotropin levels could be associated with dementia risk. 22 In the present multicohort study, we did not observe differential associations for participants younger and older than 75 years or for men and women, nor any association with variation in FT 4 level or more extreme values of thyrotropin. Therefore, subgroup associations reported in prior studies might not be generalizable outside the original cohorts.
As mentioned before, all but 1 randomized clinical trial on levothyroxine treatment for subclinical hypothyroidism also did not provide evidence for improvement of cognitive function. 8-12 Moreover, both undertreatment and overtreatment with levothyroxine are common, estimated at 27% and 14%, respectively. 64 Overtreatment is associated with increased risk of atrial fibrillation and atherosclerosis 65,66 and, via cerebrovascular damage, might be associated with increased risk of cognitive decline. Therefore, screening for subclinical thyroid dysfunction in older adults to prevent cognitive impairment and dementia does not appear to be effective.
The current individual participant data analysis has several strengths. The use of individual participant data from cohorts from all over the globe enhances generalization while allowing standardized definitions and relevant subgroup analyses. All but 5 of the included studies had a median age of 70 years or older, which is essential but often not the case in research concerning outcomes that are most relevant for older adults. 67 The present study approached cognition comprehensively; we assessed multiple domains of cognitive function, cross-sectionally and longitudinally, and incidence of dementia.

Limitations
Some limitations need to be acknowledged. Thyroid function categorization was based on biochemical characteristics. For 20% to 30% of the participants who were categorized as subclinical hypothyroid or hyperthyroid, we could not confirm subclinical thyroid dysfunction owing to the absence of FT 4 measurement. This may have led to some misclassification, yet sensitivity analyses excluding those participants with missing FT 4 data yielded similar results. We could not include educational attainment in our main analysis because 5 out of 18 cohorts did not collect these data. Even though the sensitivity analyses with adjustment for educational attainment yielded similar results as the main analysis, education is a possible confounder that could not be accounted for. For most cohorts, only 1 measurement of thyroid function was available, which is why only baseline thyroid function was used in the present individual participant data analysis. This study could therefore not capture any changes in cognitive function that might occur at the transition of one thyroid status to another. Moreover, for the vast majority of study participants, a maximum of 2 measurements of cognitive function was available, which precluded advanced modeling of change over time including nonlinear trajectories. In addition, the interpretation of longitudinal studies of cognitive function can be complicated by practice effects. 68 Standardization of change over time might not fully alleviate this; hence, residual practice effects may still be present. Furthermore, because dementia is clinically difficult to diagnose, some misclassification could have occurred, which may have led to an underestimation of the association. In addition, the number of incident dementia cases in the included cohort studies was low; we therefore cannot rule out a clinically relevant association between thyroid dysfunction and risk of dementia. The heterogeneity between studies may have been increased by the use of different cognitive function tests, different durations of followup, differences in age and sex distribution, different lifestyles across continents, and different inclusion criteria. As heterogeneity was expected a priori, we performed all metaanalyses with random effects. Nonetheless, results for fixedeffects meta-analyses were not materially different. The observed heterogeneity was larger in the longitudinal analyses heterogeneity (I 2 = 0%-70%) than in the cross-sectional analyses (I 2 = 0%-40%), likely owing to the additional variation of follow-up duration. We hypothesize that the minor differences in I 2 estimates between different cross-sectional analyses are attributable to differences in sample size per exposure. Because individuals with thyroid disease generally receive medical treatment, we cannot address the question of whether long-term untreated hyperthyroidism or hypothyroidism is associated with cognitive function and dementia risk. Moreover, these results only apply to objectifiable cognitive decline, which is not synonymous with the more subjective cognitive complaints.

Conclusions
In this individual participant data analysis combining the individual participant data of 74 565 participants from 23 cohorts, subclinical thyroid dysfunction was not associated with cognitive function, cognitive decline, or risk of dementia. Hence, it is unlikely that treatment for otherwise undetected subclinical thyroid dysfunction would improve cognitive func-tion. Moreover, the chance of overtreatment is considerable, which increases the risk of atrial fibrillation, atherosclerosis, and cerebral infarction and thereby might increase the risk of cognitive decline. Whether treatment of overt hypothyroidism or hyperthyroidism is associated with cognitive decline and risk of dementia remains uncertain. Existing clinical guidelines that prescribe screening of subclinical thyroid dysfunction for prevention of cognitive decline or dementia should therefore be revisited.