Thyroid Dysfunction and the Risk of Clinically Relevant Depression: A Longitudinal Cohort Study

Abstract

Objective:

To examine the association between thyroid function, measured via Thyroid Stimulating Hormone (TSH), and depression, assessed by the Patient Health Questionnaire-9 (PHQ-9), in a longitudinal cohort.

Methods:

We conducted a longitudinal cohort study of euthymic, euthyroid adults at baseline (PHQ-9 < 5, TSH >0.3–4.2 mIU/L) from January 1, 2000, to August 30, 2021, with TSH and PHQ-9 assessed within six months and at least two paired measurements per participant. Data included demographics, medical comorbidities, thyroid function, psychotropic medications, thyroid disorders, hormone replacement, and mood disorder diagnoses (ICD-10-CM). Thyroid dysfunction was classified as “Low TSH ≤ 0.3 mIU/L” or “High TSH > 4.2 mIU/L” based on the first occurrence of abnormal values. The primary outcome was clinically relevant depression (CRD, PHQ-9 ≥ 10). Cox proportional hazards models assessed the association between time-varying thyroid dysfunction and time to CRD, including sex-based differences in the onset of thyroid dysfunction or CRD.

Results:

Among 6,191 adults (mean age 50.2, 67% female, 92.4% Caucasian), thyroid dysfunction was significantly associated with an increased CRD risk (HR = 1.33, p < 0.001). Low TSH (HR = 1.57, p < 0.001) and high TSH (HR = 1.25, p = 0.005) were both linked to elevated risk of CRD. Females had a higher incidence and earlier onset of thyroid dysfunction and CRD.

Conclusion:

Thyroid dysfunction is associated with an increased risk of CRD, with females showing a higher incidence and earlier onset of both conditions, underscoring the need for depression screening in affected individuals. Further research should validate these findings in diverse populations.

Background and Significance

Thyroid hormones play a vital role in regulating mood.¹ Research consistently indicates that individuals with thyroid dysfunction have an elevated risk of mood disturbances.^{2, 3} Thyroid dysfunction is linked to the development of mood disorders, and, while many mechanisms have been postulated, the exact mechanism remains uncertain.^{4, 5} Hypothyroidism, often indicated in most cases by elevated thyroid-stimulating hormone (TSH), is frequently associated with depressive symptoms⁶ and treatment resistance to antidepressants in patients with mood disorders.⁷ Hyperthyroidism, characterized by thyroid hormone excess, often results in anxiety, irritability, and mood swings.⁸ Increased awareness of mood changes among patients with thyroid dysfunction can lead to a more comprehensive healthcare delivery, ultimately improving clinical outcomes.^{9, 10}

In a retrospective cross-sectional study,² we recently reported an increased odds of Clinically Relevant Depression (CRD, defined as a PHQ-9 [Patient Health Questionnaire-9] score ≥ 10) among individuals with low TSH (≤0.3 mIU/L).¹¹ These findings were replicated in two large, independent cohorts at the University of Utah Health in Utah and the Marshfield Clinic Health System in Wisconsin.^{12, 13} However, the findings were limited by the cross-sectional nature of the cohorts, highlighting the need to investigate this relationship in a longitudinal study. A deeper understanding of this association could offer valuable insights into the temporal dynamics between thyroid dysfunction and the development of CRD.

To address this quandary, we conducted a longitudinal study to evaluate the relationship between thyroid function and depression in a cohort of euthymic and euthyroid individuals at baseline. We analyzed the association between thyroid dysfunction categorized as “Low TSH ≤ 0.3 mIU/L” or “High TSH > 4.2 mIU/L”, and the risk of CRD among individuals who were euthyroid (TSH > 0.3 to 4.2 mIU/L) and euthymic (PHQ-9 < 5) at baseline. We further conducted a sex-stratified analysis to examine the development of thyroid dysfunction and CRD.

Patients and Methods

Study design

We conducted a longitudinal cohort study that included adults (≥18 years) who received care at the Mayo Clinic in Rochester, Minnesota, between January 1, 2000, and August 30, 2021, and completed at least two distinct PHQ-9 assessments, each paired with a TSH measure within 6 months (Figure-1). When both measures were available, the earliest thyrotropin level date first paired with a PHQ-9 was September 7, 2007, and the earliest PHQ-9 date paired with a thyrotropin level was November 29, 2007. This allowed for potentially multiple TSH values to be paired with the same PHQ-9 score. While TSH is a test commonly ordered for symptomatic patients seeking medical care¹⁴, PHQ-9 is routinely administered during primary care visits and mental health evaluations in both outpatient and inpatient settings. We did not restrict our cohort based on the indication for either the TSH or PHQ-9 assessment. Baseline data was defined as the first pair of PHQ-9 and TSH values meeting these criteria in a patient’s electronic health record (EHR). The cohort was further restricted to individuals who were euthyroid (TSH levels > 0.3 to 4.2 mIU/L) and euthymic (PHQ-9 < 5) at baseline. Following cohort entry, PHQ-9 and TSH measurements were not required to be temporally paired, reducing bias related to differential testing frequency. Individuals who provided consent for the use of their EHR for research were included. The study was approved by the Mayo Clinic Institutional Review Board (IRB #21–010265). The authors have adhered to the Principles of Ethical Publishing.¹⁵

Figure 1.

The study flow diagram.

Data were electronically extracted from the EHR and included demographic information (age, sex, race, ethnicity, and marital status), body mass index (BMI), medical comorbidities (assessed using the Charlson Comorbidity Index [CCI])¹⁶, psychiatric conditions (e.g., anxiety disorders, psychotic disorders, and substance use disorders), psychotropic medications (such as Selective Serotonin Reuptake Inhibitors [SSRIs], Serotonin-Norepinephrine Reuptake Inhibitors [SNRIs], Tricyclic Antidepressants [TCAs], Monoamine Oxidase Inhibitors [MAOIs], Benzodiazepines [BZD], stimulants, and antipsychotics), thyroid hormone therapies (THT – liothyronine [T3] and levothyroxine [T4]), anti-thyroid medications [Methimazole and Propylthiouracil] and thyroid laboratory results (TSH, free T4 [FT4], total T4, free T3, total T3, thyroid peroxidase antibody (TPO Ab), thyroid receptor antibody, and thyroid-stimulating immunoglobulin).

Data Resources

We utilized various platforms, including the Unified Data Platform, Mayo Data Explorer, and Lab Information Management Systems, for data extraction, similar to our cross-sectional study.² To maintain data quality, we adhered to the standard operating procedures (SOPs) during the extraction process. Furthermore, previously validated automated digital algorithms were used to extract CCI scores.¹⁶

Ascertainment of the Mood Disorder and Thyroid Disorder Diagnoses

Diagnoses of mood and thyroid disorders were identified using the International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes, and corresponding ICD-9-CM codes.^{12, 13, 17} Mood disorders were identified using ICD-10-CM codes F30–F39, while thyroid disorders were identified using codes E00, E01, E02, E03, E04, E05, E06, E07, and E89.0.^{2, 17} To address any diagnostic overlap, we followed a SOP similar to that used in our cross-sectional study.²

Thyroid functioning

TSH categories were defined based on TSH levels: low TSH as ≤ 0.3 mIU/L, normal TSH as > 0.3 to 4.2 mIU/L, and high TSH as > 4.2 mIU/L. Thyroid dysfunction was classified as ‘low TSH (≤ 0.3 mIU/L)’ or ‘high TSH (> 4.2 mIU/L)’.

Statistical Analysis

Patient characteristics were summarized as mean ± standard deviation (SD) for continuous variables and frequency (percentage) for categorical variables. TSH was log-transformed using base 10 due to its right-skewed distribution.

We aimed to determine which event occurred first: CRD (PHQ-9 score ≥10)/an abnormal PHQ-9 score or thyroid dysfunction/an abnormal TSH. These two outcomes represent competing risks, as the occurrence of one precludes observing the other as the first event. Because event times were not precisely observed; rather, they were known only to fall between clinic visits, the data were interval-censored. During each clinic visit, typically only one type of event (either PHQ-9 or TSH) was assessed. When both events were recorded within overlapping intervals, we assumed that the event observed first occurred first. We analyzed these interval-censored competing risks using the ciregic function from the R package intccr.¹⁸ We also used Kaplan-Meier time-to-event analysis to estimate CRD onset and summarize time to abnormal TSH among participants who developed thyroid dysfunction prior to CRD, reporting the median and interquartile range (IQR), and proportions.

Cox proportional hazards models were used to assess the association between baseline TSH, age, and abnormal TSH levels with time to CRD. Three models were tested, each incorporating different sets of variables. Model 1 included baseline TSH (log10-transformed) and age, modeled using a penalized spline to allow for nonlinear associations. Model 2 expanded on this by incorporating a time-dependent indicator for the first occurrence of an abnormal TSH during the follow-up period prior to the occurrence of CRD. Once someone developed thyroid dysfunction, they were not reclassified as ‘euthyroid,’ even if follow-up TSH values returned to normal. Abnormal TSH status was evaluated in two ways: as a binary variable (normal vs abnormal) and as a three-level variable distinguishing low abnormal (≤ 0.3 mIU/L), high abnormal (>4.2 mIU/L), and normal. Model 3 further adjusted for baseline covariates, including BMI, CCI, thyroid diagnosis within 6 months of baseline TSH, and the most severe mood diagnosis within six months of baseline PHQ-9, categorized as bipolar disorder (BD), persistent mood disorder (PMD), major depressive disorder (MDD), other, or none. Additionally, Model 3 accounted for medication use within six months prior to the relevant measurement, with psychotropic medications assessed before the PHQ-9 date and THT medications before the TSH date. Psychotropic medications included mood stabilizing anticonvulsants, antidepressants, antipsychotics, and BZD.

To evaluate potential differences in associations by presence of any baseline mood disorder, thyroid disorders/THT, and by sex, we conducted stratified subgroup analyses. To strengthen causal interpretation, we also performed temporal sensitivity analyses by restricting the dataset to TSH-PHQ-9 pairs in which the TSH measurement preceded the PHQ-9 assessment within 6 months, 1 month (<=30 days), or 2 weeks. We performed analyses to evaluate whether the onset of CRD was associated with an increased subsequent risk of developing thyroid dysfunction, with CRD status included as a time-dependent covariate.

For all analyses, statistical significance was set at P<.05. Data management and analysis were performed using RStudio (RStudio Team 2021, Boston, Massachusetts).

Results

Figure 1 represents the study flow chart. The cohort comprised 6,191 euthyroid and euthymic individuals, each with at least 2 distinct PHQ-9 assessments, each paired with a TSH that occurred within 6 months. Across all PHQ-9 assessments during the study period (mean of 7.1 observations per person), the average time to the nearest TSH measurement was 28 days earlier. The mean age was 50.23±16.83 years, with 67.05% females, 92.4% Caucasian, and a mean BMI of 29.29 kg/m². The baseline mean TSH was 2.09±0.96 mIU/L (range 0.4 to 4.2 mIU/L), and the mean PHQ-9 score was 1.78±1.43 (range 0 to 4). The median duration of follow-up, during which PHQ-9 measurements were collected, was 3.4 years (range 0.4 days to 13.5 years). Psychotropic medication use at baseline within the previous 6 months ranged from 0.2% to 12.8%, depending on the medication type.

Table 1 presents the baseline characteristics and differences between females and males. Compared to males, females were younger (48.6 vs. 53.5 years, p< .001), less likely to be single (16.0% vs. 20.9%, p<0.001), and had a higher prevalence of thyroid disorders (13.8% vs. 4.3%, p< .001) and THT use (19.4 % vs. 7.2%, p< .001). Males had a higher BMI (29.9 kg/m² vs. 29.0 kg/m², p< .001), higher CCI Index score (3.8 vs. 2.6, p< .001), and more frequent use of any sedative/hypnotic/anxiolytic (17.3% vs. 10.6%, p< .001) and any antipsychotic (4.0% vs. 2.0%, p<.001).

Table 1.

Baseline patient characteristics (N=6191)

Characteristic	Female (N=4151)	Male (N=2040)	Total (N=6191)	p value
Overall	4151 (67.0%)	2040 (33.0%)	6191 (100.0%)	<.001 a
Age; Mean (SD)	48.6 (16.8)	53.5 (16.4)	50.2 (16.8)	<.001 b
Race				.04 c
American Indian	25 (0.6%)	8 (0.4%)	33 (0.5%)
Asian	83 (2.0%)	34 (1.7%)	117 (1.9%)
Black	84 (2.0%)	66 (3.2%)	150 (2.4%)
Native Hawaii/Pacific Islander	2 (0.0%)	1 (0.0%)	3 (0.0%)
White	3851 (92.8%)	1870 (91.7%)	5721 (92.4%)
Other/Unknown/Undisclosed/Unable	106 (2.6%)	61 (3.0%)	167 (2.7%)
Ethnicity				.09c
Hispanic or Latino	115 (2.8%)	69 (3.4%)	184 (3.0%)
Not Hispanic or Latino	3968 (95.6%)	1925 (94.4%)	5893 (95.2%)
Unknown/Undisclosed/Unable	68 (1.6%)	46 (2.3%)	114 (1.8%)
Marital status				<.001 c
Married/Life Partnership	2708 (65.2%)	1358 (66.6%)	4066 (65.7%)
Single	665 (16.0%)	426 (20.9%)	1091 (17.6%)
Widowed/Divorced/Separated	767 (18.5%)	252 (12.4%)	1019 (16.5%)
Unknown/Undisclosed	11 (0.3%)	4 (0.2%)	15 (0.2%)
BMI nearest, N=6149; Mean (SD)	29.0 (7.6)	29.9 (6.5)	29.3 (7.3)	<.001 b
Charlson Score, N=6181; Mean (SD)	2.6 (2.5)	3.8 (3.1)	3.0 (2.7)	<.001 b
Any antidepressant	208 (5.0%)	115 (5.6%)	323 (5.2%)	.30d
SSRI	129 (3.1%)	77 (3.8%)	206 (3.3%)	.18d
SNRI	48 (1.2%)	15 (0.7%)	63 (1.0%)	.14d
TCA	27 (0.7%)	12 (0.6%)	39 (0.6%)	.87d
Other	34 (0.8%)	18 (0.9%)	52 (0.8%)	.77d
Any Sedative/hypnotic/anxiolytic	439 (10.6%)	353 (17.3%)	792 (12.8%)	<.001 d
• Benzodiazepines	200 (4.8%)	151 (7.4%)	351 (5.7%)	<.001 d
• Beta-blockers	129 (3.1%)	162 (7.9%)	291 (4.7%)	<.001 d
• SARI/Trazodone	257 (6.2%)	188 (9.2%)	445 (7.2%)	<.001 d
• Z-Drug	35 (0.8%)	34 (1.7%)	69 (1.1%)	.006 d
Any Antipsychotic	84 (2.0%)	82 (4.0%)	166 (2.7%)	<.001 d
• Atypical antipsychotics	34 (0.8%)	37 (1.8%)	71 (1.1%)	<.001 d
• Typical antipsychotics	68 (1.6%)	67 (3.3%)	135 (2.2%)	<.001 d
Other medications
• MSAC	84 (2.0%)	53 (2.6%)	137 (2.2%)	.17d
• Dopaminergic Agonist	15 (0.4%)	7 (0.3%)	22 (0.4%)	>.99d
• Any Stimulants	6 (0.1%)	6 (0.3%)	12 (0.2%)	.23d
Thyroid disorder diagnosis	574 (13.8%)	87 (4.3%)	661 (10.7%)	<.001 d
Thyroid hormone therapies (T3, T4)	805 (19.4%)	146 (7.2%)	951 (15.4%)	<.001 d
Anti-thyroid medications	4 (0.1%)	2 (0.1%)	6 (0.1%)	>.99d
Mood disorder Diagnosis				.009 c
BD	97 (2.3%)	63 (3.1%)	160 (2.6%)
PMD	31 (0.7%)	17 (0.8%)	48 (0.8%)
MDD	502 (12.1%)	232 (11.4%)	734 (11.9%)
Other	28 (0.7%)	31 (1.5%)	59 (1.0%)
None	3493 (84.1%)	1697 (83.2%)	5190 (83.8%)

^aChi-squared test for equal probabilities

^bt-test

^cFisher's Exact Test for Count Data with simulated p-value

^dFisher's Exact Test for Count Data

Psychotropic medications are based on prescriptions six months prior to the PHQ-9 date. Thyroid hormone therapy is based on prescriptions six months prior to TSH lab date. Mood diagnoses are within 6 months of the PHQ-9 date.

Thyroid diagnosis is within 6 months of the TSH lab date.

BD, bipolar disorders; BMI, body mass index; MAOI, monoamine oxidase inhibitor; MDD, major depressive disorder; MSAC, mood stabilizer anticonvulsant agents; PMD, persistent mood disorder; SARI, Serotonin antagonist and reuptake inhibitors; SNRI, serotonin-norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor; TCA, tricyclic antidepressants

Incident CRD and abnormal TSH

Overall, 24.4% of individuals developed CRD, while 31.9% exhibited thyroid test abnormalities (abnormal TSH values) at some point during the follow-up, while the rest remained euthyroid or euthymic. The cumulative incidence plot (Figure 2A) illustrates the probability of individuals transitioning from a normal state (euthymic and euthyroid) to either CRD or abnormal TSH, with individuals allowed only one transition based on whichever event occurred first during the follow-up period. At baseline, all individuals were in the normal state (euthymic and euthyroid). The probability of remaining in the normal state steadily decreases, reaching approximately 0.20 by the end of the study period. This decline suggests that a significant proportion of individuals transition out of the normal state as time progresses.

Figure 2.

Time to first CRD and abnormal TSH: cumulative incidence estimates indicate that individuals are more likely to transition to abnormal TSH than to CRD, indicating that abnormal thyroid function is the more common first departure from a normal state (A). Sex-stratified analyses in females (B) and males (C). Blue solid lines represent normal states, yellow dotted lines indicate CRD, and dark grey dotted lines denote abnormal TSH.

Conversely, the cumulative incidence curves for abnormal TSH and CRD show early separation, with abnormal TSH increasing at a faster rate. This separation persists throughout the follow-up, suggesting that abnormal TSH is a relatively common outcome in individuals seeking medical care and that it precedes CRD over time. These results suggest that individuals initially in a normal (euthymic and euthyroid) state are more likely to progress to thyroid test abnormalities before CRD. In sex-stratified analyses, the curves were more distinct among females (Figure 2B), with women showing a higher cumulative incidence and earlier onset of both thyroid dysfunction and CRD compared to men.

Time to first CRD

The Kaplan-Meier curve (Figure 3) shows the probability of CRD over 12.5 years. The median time to CRD onset was 10.7 years (95% CI = 10.0–11.4 years). Before CRD or censoring, 1,386 participants (22.4%) had abnormal TSH levels, with 5.6% having low-TSH and 16.8% having high-TSH. Among these participants, the median time to the development of abnormal TSH levels was 1.9 years (IQR 0.99 to 3.5 years), with observed times ranging from 8 days to 11.6 years.

Figure 3.

Kaplan–Meier Curve Showing the Cumulative Probability of CRD Over Time.

In Model 1, baseline TSH was significantly associated with time to CRD (HR = 0.76; p = 0.01), suggesting a higher baseline value within the normal range is associated with a protective effect on time to CRD. Age was also significantly associated with time to CRD (p < .001). In Model 2, abnormal TSH was associated with an increased risk of CRD (HR = 1.33, 95% CI: 1.16–1.52, p < .001), indicating that those with abnormal TSH were at higher risk for CRD (Table 2, Figure 4A). When abnormal TSH was further categorized into “low-TSH” and “high-TSH,” both categories were significantly associated with an increased risk of CRD, with HRs of 1.57 (95% CI: 1.26–1.96, p < .001) for low-TSH and 1.25 (95% CI: 1.07–1.4, p = .005) for high-TSH (Figure 4B). The effect was stronger in the low-TSH compared to the high-TSH, but not statistically significant (p = .07). When adjusting for baseline characteristics in Model 3, the estimates for the effect of abnormal TSH on risk of CRD showed similar direction and strength (Table 2).

Table 2.

Hazard ratios for Associations of CRD with time-dependent covariate abnormal TSH.

	Model 2								Model 3
		Binary		3 levels					Binary		3 levels
		Abnormal TSH		Low TSH (≤ 0.3 mlU/L)		High TSH (>4.2 mlU/L)			Abnormal TSH		Low TSH (≤ 0.3 mlU/L)		High TSH (>4.2 mlU/L)
Criteria	No. With incident CRD/No. at Risk	HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P	No. With incident CRD/No. at Risk	HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P
Main analysis
Any TSH observation; First observed CRD	1510/6191	1.33 (1.16, 1.52)	<.001	1.57 (1.26, 1.96)	<.001	1.25 (1.07, 1.45)	.005	1504/6149	1.44 (1.25, 1.66)	<.001	1.71 (1.36, 2.14)	<.001	1.34 (1.14, 1.58)	<.001
Subgroups - Mood disorder/Thyroid disorder (at baseline)
Mood disorder	384/1001	1.52 (1.07, 2.14)	.02	1.64 (0.92, 2.92)	.09	1.47 (0.98, 2.19)	.06	381/988	1.55 (1.08, 2.22)	.02	1.70 (0.94, 3.05)	.08	1.49 (0.98, 2.26)	.06
No mood disorder	1126/5190	1.29 (1.11, 1.49)	<.001	1.54 (1.21, 1.95)	<.001	1.20 (1.02, 1.42)	.03	1123/5161	1.42 (1.22, 1.66)	<.001	1.72 (1.34, 2.19)	<.001	1.32 (1.11, 1.57)	.002
Thyroid dx or THT	295/1254	1.55 (1.18, 2.03)	.002	1.88 (1.31, 2.70)	<.001	1.40 (1.03, 1.90)	.03	294/1250	1.49 (1.13, 1.96)	.004	1.82 (1.26, 2.62)	.001	1.35 (1.00, 1.83)	.05
No thyroid dx or THT	1215/4937	1.41 (1.19, 1.67)	<.001	1.62 (1.20, 2.19)	.002	1.34 (1.10, 1.63)	.003	1210/4899	1.43 (1.20, 1.69)	<.001	1.59 (1.17, 2.15)	.003	1.37 (1.13, 1.67)	.001
Subgroups - Sex
Male	388/2040	1.53 (1.14, 2.06)	.005	1.33 (0.75, 2.34)	.33	1.60 (1.15, 2.23)	.005	384/2018	1.74 (1.28, 2.37)	<.001	1.54 (0.86, 2.75)	.14	1.81 (1.29, 2.54)	<.001
Female	1122/4151	1.28 (1.09, 1.49)	.002	1.60 (1.25, 2.03)	<.001	1.16 (0.98, 1.39)	.09	1120/4131	1.36 (1.16, 1.60)	<.001	1.73 (1.35, 2.21)	<.001	1.24 (1.03, 1.48)	.02
Temporal sensitivity
6 months prior	1077/5560	1.98 (1.65, 2.37)	<.001	2.28 (1.70, 3.06)	<.001	1.86 (1.51, 2.30)	<.001	1073/5523	1.99 (1.65, 2.39)	<.001	2.51 (1.86, 3.37)	<.001	1.81 (1.46, 2.24)	<.001
1 month prior	727/4281	2.28 (1.85, 2.80)	<.001	2.65 (1.87, 3.75)	<.001	2.15 (1.69, 2.72)	<.001	725/4251	2.42 (1.95, 2.99)	<.001	3.11 (2.18, 4.43)	<.001	2.21 (1.73, 2.81)	<.001
14 days prior	602/3849	2.11 (1.68, 2.64)	<.001	2.77 (1.94, 3.94)	<.001	1.86 (1.42, 2.43)	<.001	602/3828	2.30 (1.82, 2.91)	<.001	3.51 (2.45, 5.04)	<.001	1.93 (1.46, 2.54)	<.001

Model 2 adjusted for baseline log10 TSH and age. Model 3 further adjusted for baseline covariates, including BMI, CCI, thyroid diagnosis within 6 months of baseline TSH, and the most severe mood diagnosis within six months of baseline PHQ-9, categorized as BD, PMD, MDD, other, or none. Additionally, Model 3 accounted for medication use within six months prior to the relevant measurement, with psychotropic medications assessed before the PHQ-9 date and thyroid hormone therapy (THT) medications before the TSH date. Psychotropic medications included anticonvulsant mood stabilizers, antidepressants, antipsychotics, and benzodiazepines.

Temporal sensitivity explored subsets of the data only where TSH was observed prior to PHQ-9. CRD events were only determined based on this reduced data set.

Figure 4.

Model-based survival curves showing the estimated cumulative probability of remaining free of CRD over time, based on time-dependent TSH status for hypothetical patients with fixed covariate profiles. Each survival curve represents the estimated probability of CRD for an individual with 2 mIU/L TSH and age 50, stratified by abnormal TSH. Covariates were held constant from time 0 throughout the follow-up period for illustration purposes: (A) Abnormal vs normal TSH; (B) Low vs normal vs high TSH levels.

Subgroup Analyses

Among those with a mood disorders diagnosis within 6 months of their baseline PHQ-9 assessment (n=1,001) and those without (n=5,190), abnormal TSH was associated with an increased risk of CRD in both groups (Table 2). The effect was slightly stronger among those with a mood disorder (HR = 1.52 [95% CI: 1.07, 2.14]; p = .02) compared to those without a mood disorder (HR = 1.29 [95% CI: 1.11, 1.49]; p < .001). When considering TSH subtype, individuals without a mood disorder showed a statistically significant association for high TSH (HR = 1.20 [1.02, 1.42]; p = .03), whereas low TSH showed a slightly stronger association (HR = 1.54 [1.21, 1.95]; p < .001), suggesting a possible differential effect (p = 0.08). In the mood disorder subgroup, the HRs for low and high TSH were 1.64 and 1.47, respectively, but did not reach statistical significance.

Among individuals with a thyroid disorder diagnosed within ± six months of their baseline TSH measurement or THT use in the six months prior (n = 1254) and those without either (n = 4937), abnormal TSH was associated with increased CRD risk in those with (HR = 1.55; 95% CI: 1.18–2.03; p = .002) and without a thyroid disorder or THT use (HR = 1.41; 95% CI: 1.19–1.67; p < .001). Among those with and without a thyroid disorder or THT use, low and high TSH showed comparable associations.

Among both males and females, abnormal TSH was associated with increased risk of CRD, with similar effect sizes (HR = 1.53 [95% CI: 1.14, 2.06] for males; HR = 1.28 [95% CI: 1.09, 1.49] for females) (Table 2). However, the effect of TSH subtype varied by sex: in females, low-TSH was more strongly associated with CRD risk (HR= 1.60 [95% CI: 1.25, 2.03]) than high TSH (HR = 1.16 [95% CI: 0.98, 1.39]) (p= .02), whereas in males, high-TSH had a stronger association (HR = 1.60 [95% CI: 1.15, 2.23]) than low TSH (HR = 1.33 [95% CI: 0.75, 2.34]), though this difference was not statistically significant (p= .56).

Temporal sensitivity

Consistent associations were observed across all temporal windows (Table 2). When limited to pairs within 6 months, abnormal TSH was associated with a significantly increased risk of incident CRD (HR = 1.98, [1.65, 2.37]; p < .001). In the three-level model, both low TSH (HR = 2.28, [1.70, 3.06]; p < .001) and high TSH (HR = 1.86 [1.51, 2.30]; p < .001) showed elevated risks.

These associations remained robust when narrowing the time window. For TSH assessments within 1 month prior to PHQ-9, abnormal TSH was associated with higher CRD risk (HR = 2.28 [1.85, 2.80]; p < .001), with stronger effects for both low (HR = 2.65 [1.87, 3.75]; p < 0.001) and high TSH (HR = 2.15 [1.69, 2.72]; p < .001). Similarly for the 14-day window, abnormal TSH was associated with higher CRD risk (HR = 2.11 [1.68, 2.64]; p < .001), with low TSH showing the greatest risk increase (HR = 2.77 [1.94, 3.94]; p < .001), and high TSH maintaining a significant association (HR = 1.86 [1.42, 2.43]; p < .001).

Bidirectional Risk: From CRD to Abnormal TSH

Development of CRD was associated with a modest, non-significant increase in the risk of subsequent abnormal TSH (HR = 1.10 [0.97, 1.25]; p = .15) (Supplemental Table). Although the analysis included 997 individuals who developed CRD prior to abnormal TSH or censoring, the confidence interval includes the possibility of no effect, suggesting either a weak association or limited power to detect a small increase in risk.

Discussion

In this large longitudinal cohort study of 6,191 euthyroid and euthymic individuals at baseline, we found a significant association between abnormal TSH and an increased risk of CRD. Abnormal TSH levels were consistently associated with a significantly increased risk of CRD across all temporal windows and remained robust even when narrowing the time window to 1 month or 14 days prior to PHQ-9 assessments. Both low and high TSH levels were associated with a higher risk for CRD compared to normal TSH levels.

The finding of low TSH levels and a high risk of CRD is consistent with three recently published cross-sectional studies conducted in the United States: (i) our 2023 cohort study at the Mayo Clinic², (ii) a 2025 cohort study from the Marshfield clinic¹³, and (iii) a recent cohort study from the University of Utah Health¹², with white predominance >90% in all the cohorts. Our findings are also consistent with a 2024 study from the Netherlands,³ in their cross-sectional analysis of 9471 patients. While all 3 US studies used PHQ-9 to define CRD, this study used the Dutch version of the Centre for Epidemiologic Studies Depression (CES-D) scale score of 16 or more to define clinically relevant depressive symptoms. In the same cohort, the authors also reported a longitudinal analysis for 8366 patients, but did not find an association between TSH and incident depression, which is inconsistent with our longitudinal study results. Use of CES-D, lower number of females (54.7% vs 67%), higher mean age (64.7 vs 50.2 years) and a longer median follow up time (10.7 vs 3.6 years) were the notable differences in the longitudinal cohort compared to our cohort.

A 2014 Dutch study¹⁹ of 7893 Caucasians aged 55 years and older reported that low-normal TSH values (within the reference range) were associated with more depressive symptoms in both cross-sectional and longitudinal analyses. The authors used CES-D score of 16 or greater to define depression and the mean follow up time was 8 years. Another cohort study published in 2006 using the Third National Health and Nutrition Examination Survey database reported that acute depressive syndrome is associated with a lower TSH levels in men alone, who did not have a prior history of thyroid dysfunction.²⁰ Similarly, in our cohort, baseline low-normal TSH (within the normal reference range) was associated with an increased risk of CRD compared to high-normal TSH levels.

Several studies have identified high TSH as a risk factor for depression,^{6, 7, 13} a finding we replicated in our cohort. Our results suggest that the development of high TSH over time, indicating an underactive thyroid, is a risk factor for depression. These findings emphasize that euthyroid individuals who develop thyroid dysfunction are at an elevated risk of depression and should undergo clinical screening.

Our study highlights that individuals are more likely to develop thyroid test abnormalities prior to the onset of CRD, and this early thyroid dysfunction appears to increase the risk of subsequent CRD. Notably, the risk is higher in females, who not only show a greater likelihood but also an earlier onset of both thyroid dysfunction and CRD compared to males. These findings carry important clinical implications, particularly given that females are already known to be at higher risk for both depression and thyroid disorders, as consistently reported in prior studies.^{15, 21–25}

A 2016 cross-sectional study reported a U-shaped relationship between FT4 levels and depressive symptoms.²⁶ However, we could not analyze the association between FT4 and CRD due to the very small sample size of FT4 data in our cohort, with only 20.7% having at least one FT4 measurement, and just 23.2% of those including a baseline value. Studies have shown that elevated TPO Ab are associated with depression.²⁷ However, due to low sample size this could not be analyzed.

Recent meta-analyses have yielded discordant findings regarding the association between thyroid dysfunction and depression.^{28, 29} This could be due to differences in study designs—cross-sectional versus longitudinal—differences in diagnostic scales, variations in the measurement of thyroid lab assessment, and the short duration of the studies. Several studies suggest a significant correlation between TSH levels and the severity of depressive symptoms.^{9, 30, 31} This relationship is particularly pertinent in the context of treatment-resistant depression, indicating that perturbations in TSH levels may contribute to the persistence and severity of depressive episodes.^{7, 32}

Strengths and Limitations

Ours is one of the largest longitudinal studies conducted in the United States, employing robust methodology and rigorous statistical analysis to investigate the association between TSH abnormalities and depression. Despite this, it has several limitations that should be acknowledged. It includes only individuals evaluated at our institution for various concerns, having both TSH and PHQ-9 data in the Mayo Clinic EHR, which may introduce selection bias. Since TSH is commonly tested during depression evaluations, this could have increased the detection of abnormal values. At baseline, 10.7% of individuals had a thyroid disorder, which was adjusted for in the models, although residual confounding may remain. However, the use of ICD codes may have led to underestimation of true incidence because of possible coding inaccuracies, unrecognized subclinical disease, and exclusion of individuals who did not seek medical care, resulting in conservative incidence estimates. In subgroup analyses by thyroid disorder and THT use, low TSH was significantly associated with CRD risk in both groups, while high TSH was not—possibly due to smaller sample sizes.

The study period was limited by PHQ-9 data availability, reducing follow-up duration. As an observational study, it remains susceptible to confounding despite covariate adjustments. To mitigate this, we selected a baseline cohort of euthyroid and euthymic individuals. However, some may have had abnormal TSH values prior to the index date, leading to potential misclassification. Only 5% were on antidepressants at baseline, reducing the likelihood of active depressive symptoms influencing the results.

The cohort was predominantly white (92%), which may limit generalizability, though prior studies published from Olmsted County, Minnesota, support applicability to the Upper Midwest region.^{33, 34} CRD was defined using a PHQ-9 cutoff rather than clinical interviews, though a score ≥10 has demonstrated 88% sensitivity and specificity compared to interview-based diagnoses of clinical depression.³⁵ For our predicted cumulative incidence estimates, we assumed that the interval observation times were independent of the event occurrence. In practice, testing may be influenced by clinical suspicion or patient characteristics, which could bias cumulative incidence in either direction; for example, frequent testing of higher-risk patients could lead to overestimation, while delayed detection in some participants could lead to underestimation. Therefore, our estimates should be interpreted as approximations under the assumption of noninformative observation. Finally, we modeled time to CRD using the first abnormal TSH as a time-dependent covariate and did not reclassify participants if subsequent values normalized. This approach captures the potential impact of the initial abnormal TSH while avoiding noise from transient fluctuations, but it may overstate the effect on CRD, since any subsequent events are attributed to the first abnormal TSH even if thyroid function later returns to normal. The mood disorder subgroup included a smaller number of patients and events, which reduced precision and may have limited the ability to detect statistically significant associations.

Conclusions

This large longitudinal study emphasizes that euthyroid individuals who develop thyroid abnormalities are at an increased risk of CRD, highlighting the importance of screening for depression in adults with newly diagnosed thyroid dysfunction. Further research is needed to replicate these associations in other geographic areas and in non-white populations.

Abbreviations and Acronyms

BD

Bipolar disorder

BMI

Body mass index

BZDs

Benzodiazepines

CCI

Charlson comorbidity index

CRD

Clinically relevant depression

EHR

Electronic health record

FT4

Free T4

HR

Hazard Ratio

MAOIs

Monoamine Oxidase Inhibitors

MDD

Major depressive disorder

PHQ-9

Patient health questionnaire-9

PMD

Persistent Mood disorder

SNRIs

Serotonin-Norepinephrine Reuptake Inhibitors

SOP

Standard operating procedures

SSRIs

Selective Serotonin Reuptake Inhibitors

SD

Standard deviation

T3

Liothyronine

T4

Levothyroxine

THT

Thyroid hormone therapies

TSH

Thyroid stimulating hormone