Thyroid hormone changes are central to leading paradigms of depression; however, the ability of thyroid hormones to predict clinical outcomes of depression is still unclear. A study by Qiao and colleagues examined the predictive power of pretreatment free triiodothyronine (FT3), free thyroxine (FT4), and thyroid stimulating hormone (TSH) to explain posttreatment response variability in patients with depression using a data-driven precision medicine approach. Thyroid hormone level data from inpatients diagnosed with depression and no other disorder (N=2,086; 69% female) were used for this aim. Peripheral venous blood samples were drawn in a fasting state in the mornings following admissions between 2014 and 2020. Clinical outcomes were evaluated by changes in symptom severity, as measured by the Hamilton Depression Scale (HAMD) and Hamilton Anxiety Scale (HAMA), including response (HAMD reduction of =50% from admission) and remission (HAMD scores <8). Prediction models were trained to detect clinical outcomes (remitted or not) using support vector machine learning. Treatments during hospitalization included antidepressants (98%), antipsychotics (74%), mood stabilizers (53%), benzodiazepines (92%), repetitive transcranial magnetic stimulation (55%), and psychotherapy (97%). The average hospital admission was 14.83 ± 4.396 days. Lower values of FT3 and FT4, but not TSH, within the normal range predicted poorer response to antidepressant treatment. The model with the best accuracy (0.86; sensitivity=0.91; specificity=0.79) combined values of FT3 and FT4 with HAMD and HAMA scores on admission to predict clinical improvement. Together these findings indicate a crucial role for thyroid function in depression and support the use of thyroid measurements, particularly FT3 and FT4, as predictors for clinical outcomes. External validation of this prediction model is needed and investigation of multidimensional markers as features to construct predictive models is recommended.
Qiao D et al. J Affect Disord 2021;299:159-65. Abstract.