Endogenous and Exogenous Thyrotoxicosis and Risk of Incident Cognitive Disorders in Older Adults

Key Points

Question  Is excess thyroid hormone associated with a higher risk of cognitive disorders in older adults?

Findings  In this cohort study among 65 931 patients 65 years and older receiving primary care within a single health care system, a low thyrotropin level from either endogenous or exogenous thyrotoxicosis was associated with increased risk of incident cognitive disorder, with an adjusted hazard ratio of 1.39.

Meaning  Practice patterns favoring aggressive case detection and treatment with thyroid hormone in older adults should be reconsidered in light of the frequency of overtreatment and the potential for harm associated with excess thyroid hormone.

Importance  Thyroid hormone is among the most common prescriptions in the US and up to 20% may be overtreated. Endogenous hyperthyroidism may be a risk factor for dementia, but data are limited for iatrogenic thyrotoxicosis.

Objective  To determine whether thyrotoxicosis, both endogenous and exogenous, is associated with increased risk of cognitive disorders.

Design, Setting, and Participants  This cohort study performed a longitudinal time-varying analysis of electronic health records for patients receiving primary care in the Johns Hopkins Community Physicians Network between January 1, 2014, and May 6, 2023. Patients 65 years and older with at least 2 visits 30 days apart to their primary care physicians were eligible. None of the 65 931 included patients had a history of low thyrotropin (TSH) level or cognitive disorder diagnoses within 6 months of their first visit. Data analysis was performed from January 1 through August 5, 2023.

Exposure  The exposure variable was a low TSH level, characterized based on the clinical context as due to endogenous thyrotoxicosis, exogenous thyrotoxicosis, or unknown cause, excluding those attributable to acute illness or other medical factors such as medications.

Main Outcomes and Measures  The outcome measure was cognitive disorders, including mild cognitive impairment and all-cause dementia, to improve sensitivity and account for the underdiagnosis of dementia in primary care.

Results  A total of 65 931 patients were included in the analysis (median [IQR] age at first visit, 68.0 [65.0-74.0] years; 37 208 [56%] were female; 46 106 [69.9%] were White). Patients exposed to thyrotoxicosis had cognitive disorder incidence of 11.0% (95% CI, 8.4%-14.2%) by age 75 years vs 6.4% (95% CI, 6.0%-6.8%) for those not exposed. After adjustment, all-cause thyrotoxicosis was significantly associated with risk of cognitive disorder diagnosis (adjusted hazard ratio, 1.39; 95% CI, 1.18-1.64; P < .001) across age groups. When stratified by cause and severity, exogenous thyrotoxicosis remained a significant risk factor (adjusted hazard ratio, 1.34; 95% CI, 1.10-1.63; P = .003) with point estimates suggestive of a dose response.

Conclusions and Relevance  In this cohort study among patients 65 years and older, a low TSH level from either endogenous or exogenous thyrotoxicosis was associated with higher risk of incident cognitive disorder. Iatrogenic thyrotoxicosis is a common result of thyroid hormone therapy. With thyroid hormone among the most common prescriptions in the US, understanding the negative effects of overtreatment is critical to help guide prescribing practice.

Several large cohort studies have suggested that excess thyroid hormone is a risk factor for dementia.^1-8 In meta-analyses, low serum thyrotropin (TSH) level had an adjusted risk ratio of 1.67 (95% CI, 1.04-2.69) for incident dementia in 1 study,⁹ while a similar point estimate was not significant in a second.¹⁰ Design limitations in these studies include a lack of repeated measures (where exposure misclassification can lead to negative results), limited follow-up periods (some only 1 year), and the almost universal exclusion of iatrogenic thyrotoxicosis, the most common cause of excess thyroid hormone. Two cross-sectional studies of cognitive function in patients with thyroid cancer with suppressed TSH level had limited power and reported mixed results.^11,12 These data contribute to uncertainty about the clinical significance of thyrotoxicosis for cognitive function and implications for thyroid hormone use.

We have therefore used a large electronic health record (EHR) database to examine the time-varying association between endogenous and exogenous sources of thyrotoxicosis and mild cognitive impairment (MCI) or dementia, collectively referred to as cognitive disorder, an outcome that accounts for the underdiagnosis of dementia in primary care.¹³ Since iatrogenic exposure results from treatment for hypothyroidism, there are alternative causal hypotheses that our approach is equipped to analyze: (1) that hypothyroidism is the risk factor for dementia, with confounding by indication, or (2) reverse causation, with the preclinical phase of dementia resulting in an elevated TSH level through central stressors,¹⁴ leading to higher rates of treatment and consequent overtreatment.

Understanding whether excess thyroid hormone, particularly exogenous thyroid hormone, is associated with dementia risk has profound public health implications. Thyroid hormone is among the most common prescriptions in the US,¹⁵ with the trend toward more aggressive prescribing in both the US¹⁶ and Europe.¹⁷ However, surveys found that only 60% of those receiving therapy are euthyroid.^18,19 Furthermore, for older adults (age ≥65 years), there is growing controversy about the appropriate interpretation of mild TSH elevations above population reference ranges, traditionally labeled subclinical hypothyroidism (4.5-10 mIU/L in most assays), as such increases are often transient²⁰ and may be accompanied by rising free thyroxine (FT₄) levels.²¹ The goal of this research is to investigate whether aggressive treatment practices, which can cause iatrogenic thyrotoxicosis, risks cognitive harm.

This study was conducted using EHRs for patients of Johns Hopkins Community Physicians, a network of primary care clinics throughout Maryland and Washington, DC. The current Epic Systems EHR was adopted in mid-2013; thus, we included data from January 1, 2014, to May 6, 2023. Patients were included beginning with their first completed internal or family medicine visit at which they were aged 65 years or older and were excluded if there were fewer than 30 days between the first and last such visits. Patients were excluded for a baseline diagnosis of MCI or dementia, or a low TSH measurement, documented before or within 6 months after the first eligible visit. This 6-month window accounts for diagnoses added to a patient’s medical record following an initial visit due to the process of care and avoids overcounting these as incident events. Included patients were not required to have a TSH measurement at or before their first included visit. This study was approved by the Johns Hopkins University Institutional Review Board (Nos. IRB00228485, IRB00258816, and IRB00269466), and a waiver of consent was granted on the grounds that all data were collected as part of routine care, no changes to care were made, and the risk of a breach of privacy was appropriately minimized.

All International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes and specific medications are presented in eTable 1 in Supplement 1. Patient characteristics including age, sex, race, and Hispanic ethnicity were determined from clinical intake forms. Race was categorized as Asian, Black, White, other (American Indian, Pacific Islander, Native Hawaiian, multiracial, and explicitly listed as other), and unknown, due to small numbers among each of the other categories. Additional information on patients listing their race explicitly as other was not available as part of the data for this study.

Low TSH measurements were identified based on the assay-specific reference range. Laboratory values missing a reference range were considered low if below 0.45 mIU/L by convention. Where the sample collection date was unavailable (1.6% of all TSH measurements), we used the result date (entered into the EHR). We excluded TSH measurements ordered during a hospital inpatient stay or emergency department visits, as acute medical stress can cause a low TSH. Additionally, we excluded all TSH measurements in patients with a diagnosis code for pituitary disease or adrenal insufficiency in the year prior or 30 days after the TSH measurement, as this context is unreliable for the diagnosis of thyrotoxicosis. Since FT₄ measurement was not available frequently enough to determine subclinical and overt thyrotoxicosis, we categorized severity based on TSH alone as either moderate (≥0.1 mIU/L to less than the lower limit) or severe (≤0.1 mIU/L).

A patient was considered to have hypothyroidism if they had an encounter diagnosis, hospital billing, or hospital admission code for hypothyroidism or if they were prescribed a thyroid hormone replacement. We included all medications ordered or recorded by a Johns Hopkins Health System clinician but do not capture outside clinicians or pharmacy data. We used the earlier of prescription or diagnosis date as the date of disease. We used an analogous definition for endogenous hyperthyroidism based on diagnosis, immunotherapy exposures, and/or antithyroid prescriptions.

Where a patient had both hyperthyroidism and hypothyroidism diagnosis codes, these could represent either (1) temporary adverse effects of treatment, (2) permanent adverse effects of treatment, or (3) progression of thyroiditis across phases. In such cases, the first diagnosis continued to be active until the last instance of that diagnosis, when a second diagnosis could dominate. However, if the patient had a prescription for thyroid hormone or an antithyroid medication, they were considered to have hypothyroidism or hyperthyroidism, respectively, regardless of the proximate diagnosis codes, since treatment excesses are sometimes coded using disease signifiers.

Using the patient’s medical record as above, we identified low TSH values caused either by endogenous hyperthyroidism or overtreatment in those with hypothyroidism. Where the attribution was clear, with no other proximate explanation, we categorized the measurement as confident endogenous or confident exogenous. In cases where the patient had no relevant prescriptions or multiple diagnoses, we categorized the TSH measurement as likely exogenous or likely endogenous according to the active diagnosis. Low TSH values with no proximate thyroid-related diagnosis codes or prescriptions were categorized as unknown cause. See the eMethods in Supplement 1 for complete details.

A patient was considered to have a cognitive disorder if an MCI or any-cause dementia ICD-10 code appeared as a visit diagnosis code, hospital billing code, or hospital admission code. This relatively simple phenotype has been shown to outperform more complex phenotypes that require multiple codes or include medication information.²²

To estimate the associations between thyrotoxicosis and cognitive disorder, we used a time-varying Cox model. The time of event was the age at which the patient first met the criteria for cognitive disorder. We considered patients left-truncated until the age of their first visit and right-censored after their most recent visit. While death is a competing risk for cognitive disorder, we did not have access to complete death records; thus, patients who died were treated as right-censored. We fit 3 models targeting different levels of exposure granularity: all-cause thyrotoxicosis (model 1); endogenous, exogenous, and unknown-cause thyrotoxicosis (model 2); and moderate and severe thyrotoxicosis by exposure type (model 3). In all models, thyrotoxicosis was encoded as a binary time-varying exposure indicating past exposure. In models 2 and 3, endogenous, exogenous, and unknown-cause exposures were not mutually exclusive (eg, patients may have had both endogenous and exogenous exposures). In contrast, indicators for moderate thyrotoxicosis were set to zero once the patient experienced severe thyrotoxicosis of the same type. Thus, hazard ratios (HRs) for moderate thyrotoxicosis reflect the risk associated with only having moderate thyrotoxicosis. To account for potential confounding, we estimated adjusted versions of each model including sex, race, Hispanic ethnicity, and age at first visit as time-fixed covariates and diagnosed hypothyroidism, hyperthyroidism, and number of primary care visits in the prior year as time-varying covariates. Standard errors for all parameters were estimated using the observed Fisher information matrix, and P values were calculated based on the Wald statistic and 2-tailed with significance at P < .05. We tested the proportional hazards assumption by fitting linear regression models to the Schoenfeld residuals of each feature. Evidence supported nonproportionality for the utilization feature; however, the degree of deviation from proportionality was deemed acceptable based on visual examination (eTable 3 in Supplement 1). All analyses were conducted in Python (Python Software Foundation) using the lifelines package, version 0.27.0.

We repeated our primary adjusted analysis including interaction terms between age, sex, and race and all exposure variables in models 1 and 2. Due to sample size, we grouped patients older than 80 years when testing for interactions with age, and patients with other or unknown race when testing for interactions with race. Hypothesis tests for these interaction terms were adjusted using Bonferroni correction. To examine whether the association between hypothyroidism and cognitive disorder was mediated by overtreatment, we estimated adjusted HRs (aHRs) for hypothyroidism with and without thyrotoxicosis exposure. To account for variability in the way MCI codes are used in primary care, we repeated our primary analysis using only dementia as our outcome. To test the sensitivity of our results to possible thyrotoxicosis misclassification, we repeated our primary analysis limited to confident attributions.

Because thyroid testing is common when a clinician suspects dementia, a greater number of low TSH measurements may be observed because of increased screening just before dementia diagnoses due to undocumented clinical suspicion. To test for this, we repeated our primary analysis with offsets added to the time of all TSH laboratory test results, thereby shifting laboratory tests that were taken just before a cognitive disorder diagnosis to after the diagnosis. We varied these offsets from 30 to 180 days to judge the degree to which testing around the time of diagnosis affected our primary analysis.

Of the 81 682 patients 65 years and older with a primary care visit during the study period, 9107 were excluded because they did not have a second visit after at least 30 days, and 6644 were excluded because they had either a cognitive disorder diagnosis or thyrotoxicosis documented within 6 months of their first visit, leaving 65 931 included patients (Table 1). Of these, 37 208 (56.4%) were female, and 28 723 (43.6%) were male. A total of 258 (0.4%) were American Indian; 3018 (4.6%) were Asian; 12 692 (19.3%) were Black; 284 (0.4%) were Pacific Islander; 46 106 (69.9%) were White; 2233 (3.4%) listed their race explicitly as other; and 1336 (2.0%) were of unknown race. The median (IQR) age at the first recorded visit was 68.0 (65.0-74.0) years, with 76% in the 65- to 75-years age range and only 5% older than 85 years. The median (IQR) time from first to last visit was 3.9 (1.6-7.1) years, and patients had a median (IQR) of 12.0 (6.0-23.0) primary care visits during the study period.

A total of 24 867 low TSH measurements among 2710 (4.1%) patients were recorded during the study period. Figure 1 shows a waterfall diagram illustrating the attribution of these TSH measurements to exogenous, endogenous, or unknown cause. The majority of low TSH measurements were exogenous (14 875 of 24 867 [60%]), followed by unknown cause (5833 of 24 867 [24%]) and endogenous (4159 of 24 867 [17%]). Endogenous hyperthyroidism was more likely to be severe, with 42% (1753 of 4159) of these values below 0.1 mIU/L, compared with 31% (4643 of 14 875) of exogenous thyrotoxicosis and 28% (1638 of 5833) of those with unknown cause. An initial crude association analysis conducted without ordering events or accounting for censoring showed that having a low TSH level was significantly associated with sex, age, Black race, years of follow-up, total number of visits, and cognitive disorder diagnosis. The cumulative incidence curve for thyrotoxicosis by age is in Figure 2A. Stratification by sex revealed a strong female predominance for iatrogenic overtreatment.

During follow-up, 4779 patients (7.2%) received a new cognitive disorder diagnosis, with the majority (3702 patients) eventually receiving a dementia diagnosis. A cognitive disorder diagnosis was significantly associated with sex and age, Black or unknown race, years of follow-up, total number of visits, and low TSH measurements. The cumulative incidence curve for cognitive disorder and dementia by patient age is in Figure 2B.

Unadjusted cognitive disorder incidence was higher among patients with prior thyrotoxicosis (HR, 1.48; 95% CI, 1.27-1.73; P < .001) as shown in Table 2 and Figure 2C, with cumulative incidences in exposure vs comparison groups of 11.0% (95% CI, 8.4%-14.2%) vs 6.4% (95% CI, 6.0%-6.8%) at age 75 years and 34.2% (95% CI, 29.2%-39.9%) vs 25.9% (95% CI, 25.0%-26.8%) at age 85 years. Among the covariates, Black race (aHR, 1.28; 95% CI, 1.18-1.38; P < .001), Hispanic ethnicity (aHR, 1.38; 95% CI, 1.11-1.72; P = .004), primary care visits in the past year (aHR, 1.08; 95% CI, 1.07-1.09; P < .001) were significantly associated with cognitive disorder by age. All covariate aHRs are in eTable 2 in Supplement 1. After adjustment, all-cause thyrotoxicosis (aHR, 1.39; 95% CI, 1.18-1.64; P < .001) was significantly associated with incident cognitive disorder (Table 2 and Figure 3). Stratifying by cause, exogenous thyrotoxicosis remained significantly associated with incident cognitive disorder (aHR, 1.34; 95% CI, 1.10-1.63; P = .003), while endogenous thyrotoxicosis was not significant with similar point estimates. When further stratified by severity, there was a suggestion of a dose response for exogenous thyrotoxicosis with an increase from an aHR of 1.23 (95% CI, 0.97-1.55), P = .08, for moderate disease to an aHR of 1.65 (95% CI, 1.20-2.28), P = .002, for severe disease (Figure 3).

We did not find any significant interactions (eTable 4 in Supplement 1). Hypothyroidism as a diagnosis code was not significantly associated with cognitive disorder after adjusting for utilization and thyrotoxicosis from overtreatment (aHR, 1.03; 95% CI, 0.95-1.12; P = .50). Limiting the analysis to confident cause attributions for TSH measurements or dementia diagnoses did not change the direction or significance of any associations (eTable 5 in Supplement 1). Finally, even after including offsets for TSH measurements of up to 180 days, the associations between all-cause or exogenous thyrotoxicosis and cognitive disorders remained significant, while the association with endogenous thyrotoxicosis remained positive though nonsignificant (eFigure in Supplement 1).

Using a large clinical database and rigorous phenotyping, we found that low TSH levels from either endogenous hyperthyroidism or overtreatment with thyroid hormone is associated with increased risk of incident cognitive disorder. Cumulative incidence of cognitive disorder in exposure and comparison groups was 11% vs 6% at age 75 years and 34% vs 26% at age 85 years. After adjustment for potential confounders, thyrotoxicosis was associated with a relative increase in risk of 39% across age groups. However, it is important to note that our population was care-seeking, and therefore, rates of disease and associations with exposures may be inflated relative to a true population-based sample.

While this is not a randomized clinical trial, there are several reasons to believe the observed associations may be causal. First, we observed similar increases in risk from 2 different causes of thyrotoxicosis (ie, consistency across exposure mechanisms). Second, our results are consistent with the existing literature on thyroid function and cognitive disease, where lower TSH and higher FT₄ levels have been associated with increased dementia risk in multiple cohort studies^1-4,6,7,9,12 and mendelian randomization,^23,24 with exceptions that may be due to statistical issues.^10,11 This diversity of data sources and study designs producing similar results decreases the likelihood that these results may be due to bias in any individual study (ie, consistency across settings and designs). Finally, for exogenous thyrotoxicosis, our observations were suggestive of a dose response, with severe exposure (aHR, 1.65) associated with greater risk of cognitive disorder than moderate exposure (aHR, 1.23) (ie, biological gradient).

A strength of our study was the size and complexity of the data, which allowed us to examine a large number of potential confounders and sources of bias through sensitivity analyses. We found that rates of both low TSH levels and dementia varied by sex and race as expected based on prior literature. Women were more likely to have low TSH levels than men and were especially more likely to be overtreated. Older women were at increased risk of dementia. Additionally, Black patients had a left-shifted distribution of endogenous TSH, which is consistent with prior work in National Health and Nutrition Examination Survey (NHANES) III,²⁵ and were more likely to be diagnosed with dementia. However, these covariates did not explain the association between low TSH level and dementia, which remained significant in fully adjusted models. Further, we accounted for potential confounding by indication as a result of an association with hypothyroidism, and we accounted for potential ascertainment bias in multiple ways. First, patients were excluded if their first instance of thyrotoxicosis or diagnosis of cognitive disorder occurred within 6 months of their first recorded visit. Second, we adjusted for frequency of visits prior to diagnosis, reducing the risk of ascertainment bias associated with disparities in frequency of health system utilization. Finally, we found that adding temporal offsets to dates of TSH laboratory tests, effectively ignoring tests within up to 6 months before the cognitive diagnosis, did not change our conclusions.

Limitations in EHR data mean that we were not able to analyze in this study whether the indication for thyroid hormone treatment affected the risk associated with TSH suppression, and whether lesser degrees of TSH suppression might be associated with increased risk if the individual being treated does not have underlying hypothyroidism. A limitation of all studies relying primarily on EHRs is the possibility for measurement error and missing data biases. For example, dementia diagnosis codes have been shown to have 70% sensitivity and 77% positive predictive value.²² Additionally, misused or missing diagnosis codes or alternate causes of abnormal TSH level—such as low T3 syndrome—that occur concurrently with the explanations considered here may have resulted in exposure misclassification. We observed only 469 confident endogenous thyrotoxicosis cases out of over 6000 patients with hyperthyroidism diagnoses. This likely contributes to the lack of statistical significance for endogenous thyrotoxicosis. However, limiting our exposure and outcomes variables to the most definitive criteria did not substantially alter the point estimates. Together with our evaluation of ascertainment bias, this supports the validity of these findings. Future studies should be expanded to assess cumulative exposure measures, which will increase the power to detect the type of dose-response association suggested by this study and support a causal association.

In this cohort study, we report that an increased risk of cognitive disorders is among the potential negative consequences of thyroid hormone excess, a common consequence of thyroid hormone treatment. Clinicians considering thyroid hormone therapy in older adults should first confirm that treatment is indicated and carefully avoid overtreatment by using age-appropriate treatment strategies.²⁶

Accepted for Publication: August 15, 2023.

Published Online: October 23, 2023. doi:10.1001/jamainternmed.2023.5619

Corresponding Author: Jennifer S. Mammen, MD, PhD, Asthma and Allergy Center, Johns Hopkins University School of Medicine, 5501 Hopkins Bayview Cir, 2A62, Baltimore, MD 21224 (jmammen1@jhmi.edu).

Author Contributions: Dr Adams had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Adams, Oh, Lyketsos, Mammen.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Adams, Oh, Mammen.

Critical review of the manuscript for important intellectual content: Adams, Oh, Yasar, Lyketsos.

Statistical analysis: Adams, Mammen.

Obtained funding: Lyketsos, Mammen.

Administrative, technical, or material support: Adams, Oh.

Conflict of Interest Disclosures: Dr Lyketsos reported personal fees from Karuna, MapLight Therapeutics, Axsome Therapeutics, GIA, GW Research Limited, Merck, EXCIVA GmbH, Otsuka, IntraCellular Therapies, and Medesis Pharma for consulting for treatment development in Alzheimer disease outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by the Richman Family Precision Medicine Center of Excellence in Alzheimer’s Disease, including significant contributions from the Richman Family Foundation, the Rick Sharp Alzheimer’s Foundation, the Sharp Family Foundation, and others. The work was also supported by National Institutes of Health grants R01AG064256 (Dr Mammen), P30 AG066507 (Drs Adams and Lyketsos), and R01AG076525 and R01AG057725 (Dr Oh).

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data Sharing Statement: See Supplement 2.