Effect of Levothyroxine Therapy on the Development of Depressive Symptoms in Older Adults With Subclinical Hypothyroidism

Key Points Question Does levothyroxine treatment have an effect on the development of depressive symptoms in older adults with subclinical hypothyroidism? Findings In this ancillary study of a randomized, placebo-controlled clinical trial of 427 participants with subclinical hypothyroidism, there was no statistically significant difference in the adjusted between-group difference in the mean Geriatric Depression Score at 12 months between the levothyroxine and placebo groups. Meaning These results do not provide evidence in favor of levothyroxine therapy in older persons with subclinical hypothyroidism to reduce the risk of developing depressive symptoms.


Introduction
Subclinical hypothyroidism is defined as elevated thyroid-stimulating hormone (TSH) levels in combination with free thyroxine (T4) levels within the reference range. 1 Subclinical hypothyroidism is a frequent condition in the general population, in particular among women and older adults, 2 with a prevalence up to 10% to 15%. 3,4 Increasing evidence suggests that patients with subclinical hypothyroidism should not routinely be treated with levothyroxine. [5][6][7] Current guidelines recommend levothyroxine therapy for adults with TSH levels greater than 10 mIU/L and for people with lower TSH values who are young, symptomatic, or have specific indications for prescribing. 8,9 Levothyroxine has become the most prescribed drug in the US in 2014 and the second most prescribed drug in the UK in 2019. 10,11 More frequent testing or lower TSH thresholds could be potential reasons for the increase in levothyroxine prescription. 1 Among other indicators, depressive symptoms are a common reason for starting levothyroxine therapy in patients with subclinical hypothyroidism. 12 A recent meta-analysis of 4 trials on the effect of levothyroxine therapy in adult patients with subclinical hypothyroidism found no benefit for depressive symptoms, 7 but data were limited by small sample sizes (N = 57-94) and potential biases, such as risk of bias assessment and publication bias. 7 The Thyroid Hormone Replacement for Untreated Older Adults with Subclinical Hypothyroidism (TRUST) trial was a randomized, multicenter, double-blind placebo-controlled trial on the benefits of levothyroxine for patients 65 years or older with subclinical hypothyroidism. 6,13 Our study aimed, first, to assess the possible effect of levothyroxine therapy on depressive symptom scores in older adults with subclinical hypothyroidism from the TRUST trial, and second, to update the previous meta-analysis 7 on the effect of levothyroxine therapy on depressive symptoms scores including the results from the TRUST trial. 6

Methods
This ancillary study is nested in a large, international study on levothyroxine therapy in older adults with subclinical hypothyroidism (the TRUST trial), which was conducted from April 2013 to October 31, 2016. 6,13 This ancillary study on depressive symptoms was predefined and registered in May 2013 separately from the main TRUST trial. 14 In participants from 2 countries of the TRUST trial, Switzerland and the Netherlands, depressive symptoms were measured at baseline and at 12-month follow-up using the 15-item Geriatric Depression Scale (GDS-15). In a secondary analysis on incidence of mild depression, participants from the TRUST site in Ireland were included. In Ireland, depressive symptoms were measured by using the Center for Epidemiologic Studies Depression 20-item scale ancillary study was accepted by the TRUST publication committee. This study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline for clinical trials. 15

The TRUST Trial
The main goal of the TRUST trial was to determine whether levothyroxine provides clinical benefits in older persons with subclinical hypothyroidism. 6 The TRUST trial protocol is available in Supplement 1. In the TRUST trial, patients underwent randomization in a 1:1 ratio, with stratification according to country, sex, and starting dose, with the use of randomly permuted blocks. 6 The study medication consisted of levothyroxine sodium tablets and matching placebo tablets taken orally once daily. The active intervention started with levothyroxine at a dose of 50 μg daily (or 25 μg in patients with a body weight of less than 50 kg or with known coronary heart disease, ie, previous myocardial infarction or symptoms of angina pectoris), and the placebo group started with a matching placebo for 6 to 8 weeks. The dose of levothyroxine was adjusted in 25-μg increments based on TSH levels measured 6 to 8 weeks after starting the intervention, 6 to 8 weeks after each dose adjustment, and at a 12-month follow-up in order to reach a TSH level within the reference range (0.4-4.6 mIU/L) in the levothyroxine group. An identical schedule for adjusting the dosage of placebo with mock titration was used to achieve an approximately equal frequency of dosage adjustments between the groups to maintain blinding. The participants, investigators, and treating physicians were unaware of the results of TSH measurements throughout the course of the trial and remained blinded for treatment allocation. 6 Participants Potential participants aged 65 or older were identified from clinical and general practitioner laboratory databases and recruited from the community in Switzerland, the Netherlands and Ireland.
Subclinical hypothyroidism was defined as the presence of persistently elevated TSH levels (4.6-19.9 mIU/L) at a minimum of 2 occasions at least 3 months apart with free T4 within the reference range. Participants were followed up for a minimum of 12 months. Exclusion criteria included the use of levothyroxine, antithyroid medication, amiodarone, or lithium; recent hospitalization for major illness; recent acute coronary syndrome, acute myocarditis, or pancreatitis; and terminal illness.

Depressive Symptoms
Depressive symptoms were measured using the 15-item Geriatric Depression Scale (GDS-15) at baseline and 12-month follow-up. The GDS-15 is a well-validated test for depression screening in older age, with validity to measure longitudinal changes. [16][17][18] The GDS-15 score ranges from 0 to 15, with higher scores indicating higher likelihood of depression. Score values from 0 to 2 indicate no depressive symptoms; 3-5, mild depressive symptoms; and 6 or greater, severe depressive symptoms. 19 The minimal clinically important between-group difference on the GDS-15 is 2 points. 18 In this study, we used the cutoff of 3, suggesting mild depression. 18 One of the advantages of the GDS-15 is that it relies less on the somatic symptoms of depression and has been proven valid for use in patients with chronic physical diseases, as it is often the case in older adults. 16  We defined the primary outcome as GDS-15 score at 12 months, adjusted for scores at baseline, age, sex, country, and starting dose of levothyroxine. As a secondary analysis, we dichotomized the depression scale using established cutoff values for depressive symptoms; for the GDS-15 score, a cutoff score greater than 3 suggests mild depression. 16 For this analysis, we additionally included data from the TRUST site in Ireland, where depressive symptoms were measured using the CESD-20.
This is a self-rating scale that measures depressive symptoms during the past week. The scale ranges from 0 to 60, with higher scores indicating more depressive symptoms. CESD-20 values from 16 to 20 indicate mild depressive symptoms, values from 20 to 25 indicate moderate symptoms, and values of 25 and greater indicate severe depressive symptoms. 20 In this study, we used the cutoff of 20 to define the presence of depression. 21 The CESD-20 has been shown to be a valid and reliable instrument to measure depressive symptoms and has been widely used in older populations. 22

Statistical Analysis
Analyses were conducted from December 1, 2019, to September 1, 2020. For the primary outcome analysis, we performed a modified intention-to-treat analysis using those participants having depressive symptoms outcomes (Figure 1). Using the available data set, we performed a power calculation with 427 participants with a standard deviation of 2 and a mean GDS-15 score in control of 1, which resulted in 100% power for detecting a mean difference of 2.0 points at a 1-sided α-level of .05. We used a multivariable linear regression analysis to assess the mean difference and relative 95% CI in GDS-15 scores at the 12-month follow-up between the levothyroxine and placebo groups.
We adjusted the results for GDS-15 scores at baseline, age, sex, country, and starting dose of levothyroxine.
In a secondary analysis, we first excluded participants with mild depression at baseline (indicated by GDS-15 scores >3 or CESD-20 scores >20) 16,23 and analyzed those who developed mild depression during follow-up in both the levothyroxine and placebo groups. Second, we included only participants with mild depression at baseline and analyzed those who recovered from mild depression. We conducted both secondary analyses (1) only in participants from Switzerland and the Netherlands and (2) including participants from Ireland (n = 87). We used a multivariable logistic regression model to assess odds ratios (ORs) and 95% CIs adjusted for depressive symptoms scores at baseline, age, sex, country, and starting dose of levothyroxine.
In subgroup analyses, we assessed the difference in GDS-15 scores at 12 months separately for men and women, for participants above and below 75 years of age, for different TSH levels (4.5 to <7.0, 7.0 to <10.0, and 10.0-20 mIU/L), and for participants with GDS-15 scores above and below 2, as  the minimal clinically relevant difference on the GDS-15 scale is 2 (first clinically relevant difference from 0). 18 We excluded participants without depressive symptom measurements ( Figure 1). We used ipsative imputation for participants with partially incomplete GDS-15 questionnaires, 24  with more than 20% missing answers was included).
We conducted several sensitivity analyses to test the robustness of the results. In the main analysis, we used a parametric model, because it has several advantages compared to nonparametric models and because parametric models have been shown to perform statistically well if the study size is big enough (n > 25). 25 To verify, we also applied a nonparametric model (Kernel regression) in a sensitivity analysis. 26 In another sensitivity analysis, we excluded participants with antidepressant treatment (Anatomical Therapeutic Chemical Classification System: N06A, n = 26). In a sensitivity analysis, we imputed missing GDS-15 score at baseline or follow-up (n = 45). We imputed data under the assumption of missing at random using multivariate regression imputation based on an individual's complete responses on the variables of sex, age, country, and starting dose of levothyroxine. We additionally simulated a scenario assuming all missing GDS-15 scores at 12 months to be 75% higher (imputed values multiplied by 1.75) and a worst-case scenario assuming all participants with missing GDS-15 score in the placebo group to have a maximal score of 15 points at 12 months, whereas we assumed that participants with missing scores in the treatment group had a score of 0 at 12 months.

Update of Meta-analysis
As this study is, to our knowledge, the biggest trial to investigate the effect of levothyroxine on depressive symptoms in patients with subclinical hypothyroidism, we also updated a recent metaanalysis on 278 participants with the results from this ancillary study on the TRUST trial on 427 participants. 7 Because studies used different scores to assess depressive symptoms, we pooled standardized mean differences (SMDs) using a random effects model. A negative SMD indicates benefit of levothyroxine therapy, with −0.2, −0.5, and −0.8 corresponding to small, moderate, and large effects, respectively. 27 Heterogeneity was assessed with I 2 statistics (might not be important: 0%-40%; moderate: 30%-60%; substantial: 50%-90%; considerable: >75%). 27 Two reviewers (L.W. and M.F.) independently evaluated the quality of evidence using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) tool. 28 The certainty of evidence reflects the extent to which the confidence in the estimate of the effect is adequate to support a recommendation regarding levothyroxine therapy in participants with subclinical hypothyroidism and depressive symptoms.
In a sensitivity analysis, we excluded studies that were only conducted in elderly participants from the meta-analysis.
All analyses were conducted with STATA version 16 (StataCorp LLC) for Windows.

Results
A total of 427 participants were included in the main analysis (mean age  (Table 2).
Of the 472 participants from the Netherlands and from Switzerland, 236 were allocated to levothyroxine and 236 to the placebo group ( Figure 1). Forty-five participants were excluded from the analysis (5 participants died before the 12-month visit, 6 had a missing GDS-15 score at baseline or There was no clinically relevant difference in depressive symptoms at 12 months in subgroup analyses by age, sex, and TSH levels (Figure 2), nor in participants with GDS-15 scores of 2 or higher at baseline (adjusted between-group difference, 0.61; 95% CI, −0.32 to 1.53; P = .20).
In a secondary analysis, we calculated the incidence of mild depression. At baseline, 188 patients in the levothyroxine group and 206 in the placebo group had no mild depression . During   Results stayed robust in a sensitivity analysis excluding participants with antidepressant medication at baseline (adjusted between-group difference, 0.14; 95% CI, −0.17 to 0.46) or using the Kernel-regression model (adjusted between-group difference, 0.12; 95% CI, −0.38 to 0.41). The results stayed robust when we imputed missing GDS-15 scores at baseline and follow-up and when we simulated missing values assuming the missing values to be 75% higher than nonmissing values.

JAMA Network Open | Diabetes and Endocrinology
When we simulated a worst-case scenario for missing values at 12 months, the difference between the groups was below minimal clinically important difference (−1.17; 95% CI, −1.74 to −0.61).
To pool the results from this ancillary study on depressive symptoms within the TRUST trial to the previous meta-analysis, 7 we transformed the adjusted mean difference in depressive symptoms between the 2 groups of 0.15 (95% CI, −0.15 to 0.46) to an adjusted standardized mean difference (SMD) of 0.08 (95% CI, −0.08 to 0.24). When we pooled the adjusted SMD with the previous meta-analysis, 7 we found an overall SMD of 0.09 (95% CI, −0.05 to 0.22) (Figure 3). There was no statistical heterogeneity (I 2 = 0.0%). In a sensitivity analysis including only 3 studies that were not only conducted in elderly participants, we found an overall SMD of 0.09 (95% CI, −0.20 to 0.38). [29][30][31] This ancillary study on depression within a randomized controlled trial had a weight of 67.6% to the overall effect. The quality of evidence assessed with the GRADE tool was high, indicating that further research is very unlikely to change our confidence in the estimate of effect (eTable 4 in Supplement 2).

Discussion
In this ancillary study within a randomized controlled trial among adults aged 65 years or older with subclinical hypothyroidism, levothyroxine therapy for 12 months had no clinical effect on depressive  This study has some limitations. First, GDS-15 scores were low in our study population at baseline as reflected in the general population. Therefore, we cannot exclude that patients with subclinical hypothyroidism and moderate or severe depressive symptoms might benefit from levothyroxine therapy. Second, participants lost to follow-up had a higher mean GDS-15 score at baseline compared to those with a completed follow-up, which suggests that participants with higher scores were more likely lost to follow-up. In a sensitivity analysis in which we imputed missing scores, assuming participants with missing scores to have 75% higher scores, results stayed robust. Third, as we only included participants aged 65 years or older, our results are not generalizable to younger patients with subclinical hypothyroidism. Fourth, this study on depressive symptoms was an ancillary study of the TRUST trial; therefore, we were not able to include all participants from the original study, as depressive symptoms were not measured at all sites of the TRUST trial. Fifth, depressive symptoms were self-reported.