Thyroid Stimulating Hormone Stability in Patients Prescribed Synthetic or Desiccated Thyroid Products: A Retrospective Study

Study Design and Setting

This retrospective matched cohort study was conducted at Kaiser Permanente Colorado (KPCO) and approved by the KPCO institutional review board. Electronic medical, pharmacy, and laboratory records were used to identify patients and treatments, and outcomes were obtained from the KPCO Virtual Data Warehouse.

Study Patients

Patients aged ≥18 years receiving levothyroxine or desiccated thyroid for the treatment of hypothyroidism between January 1, 2005 and December 31, 2015 were eligible for inclusion (Table 1). To ensure chronic medication use, we first identified the earliest thyroid product dispensed during the evaluation period. We then identified second dispensing of the same product occurring at least 1 year later, defined as the index date. Data was collected for 3 years after the index date. Patients were required to maintain KPCO membership for the duration of the study. Patients were excluded if they had the following comorbid conditions: panhypopituitarism, post-radioiodine therapy, history of thyroid cancer, pregnancy, Graves’ disease, or Hashimoto’s thyroiditis. Patients were also excluded if they were prescribed more than 1 agent for treatment of hypothyroidism. Patients prescribed levothyroxine who met inclusion criteria were matched 1:1 on age, sex, and race/ethnicity to patients prescribed desiccated thyroid.

Outcome Variables

The primary outcome was the percent of euthyroid (0.320-5.500 uIU/mL) TSH values. Secondary outcomes included visit-to-visit TSH variability, percent of patients with TSH values in-range throughout the entire follow-up period, and number of TSH values.

Statistical Analysis

Mean (standard deviation) or count (percent) was used to describe the baseline characteristics of the final study population at index date. Differences in baseline characteristics between the 2 cohorts were compared using χ² tests for categorical end points and 1-way ANOVA for continuous end points. Multivariable-adjusted regression models were calculated to determine the association of desiccated thyroid product type and each outcome. All models were adjusted for the following covariates: hypertension, previous myocardial infarction, diabetes, heart failure, number of TSH laboratory values during follow up, and use of a medication known to cause a significant increase or decrease in the serum concentration of thyroid products.

Logistic regression was used to determine the association with in-range TSH values; linear regression was used to determine the association with visit-to-visit TSH variability; Poisson regression was used to determine the association with count of TSH laboratory values obtained during the 3-year follow-up period. Because each patient could contribute more than 1 TSH value in the follow-up period to the analysis, standard errors were calculated allowing for intragroup correlation when applicable. To improve the interpretability of regression coefficients, after fitting each regression model we obtained predicted values of the outcomes for synthetic levothyroxine and desiccated thyroid users.