This observational, matched cohort study used data spanning the time frame from May 1, 2000 through March 30, 2016 from the Optum Clinformatics™ database. The database contains information on over 150 million individuals and laboratory results for over 30 million people. Before release, the data were verified, adjudicated, and adjusted. Data elements include patient demographics, inpatient and outpatient services, and outpatient prescription drugs. The information is fully de-identified and HIPAA compliant. Given the retrospective nature of the study design and de-identified data, the study was exempt from internal review board evaluation.
For inclusion in the study, patients were required to initially have compliant use of Synthroid® for a 6-month lead-in period, where compliant use was defined as receipt of at least 146 days’ supply of Synthroid® (i.e., greater than 80% medication possession ratio) for the 6-month lead-in period with no filling of prescriptions for any other LT4 formulation. In the subsequent 6-month identification period, patients were defined as continuous users of Synthroid® or switchers from Synthroid® to an alternative LT4 formulation. Specifically, continuous users filled at least one prescription for Synthroid® and did not fill any prescriptions of non-Synthroid® formulations of LT4, while switchers filled at least one prescription for a non-Synthroid® formulation during the identification period. The non-Synthroid formulations considered in the study included Levothroid, Levoxyl, Tirosint, Unithroid, and generic levothyroxine. For continuous users, the index date was defined as the date of first fill of a Synthroid® prescription during the identification period, while for switchers the index date was defined as the date of first fill for non-Synthroid® LT4 prescription during the identification period. The post-period was defined as the time from the index date through 2 years following the index date. Figure 1 presents the study design.
In addition to being identified as a continuous user or switcher, patients were required to have received at least one diagnosis of hypothyroidism (ICD-9-CM codes 243, 244.0, 244.1, 244.8, 244.9 or ICD-10-CM of E030, E031, E038, E039, E890) at some point from the start of the lead-in period through the end of the post-period (i.e., the study period), to have at least one TSH laboratory test result recorded during the last 6 months of the post-period, and to have had continuous insurance coverage over the study period. Patients were excluded if they were identified as pregnant (ICD-9-CM of 630.xx-679.xx, V22.xx, V23.xx or ICD-10-CM of Oxxxx, Z33xx, Z34xx) or if a diagnosis of thyroid cancer (ICD-9-CM of 193 or ICD-10-CM of C73), iodine hypothyroidism, or other iatrogenic hypothyroidism (ICD-9-CM of 244.2, 244.3 or ICD-10-CM of E00xx, E01xx, E02xx, E032) was included in the patient’s record at any time over the study period. In addition, patients were excluded if they filled a prescription for liothyronine or a liothyronine–levothyroxine combination therapy during the study period or if they were younger than 18 years old at the index date. Figure 2 shows how these inclusion/exclusion criteria affected sample size.
Given the above cohort, a 1:1 nearest neighbor, greedy matching algorithm without replacement was applied to match continuous users and switchers on the basis of propensity scores. The propensity score model estimated the probability that a patient would switch therapy while controlling for patient-level characteristics. Differences in post-period outcomes between the propensity score matched groups were then examined using multivariable logistic regressions.
The key outcomes of interest were TSH laboratory values that were not within the range recommended by expert bodies and, also, negative clinical outcomes. The primary recommended TSH laboratory value range (0.45–4.12 mIU/L) was based upon AACE/ATA guidelines (2012) while an alternative TSH range, (0.4–4.0 mIU/L) based upon ATA task force guidelines (2014), was used in a sensitivity analysis. A negative clinical outcome was defined as a composite endpoint, operationalized as receipt of a diagnosis of at least one of the following: chronic kidney disease (CKD), depression, fatigue, heart failure, hyperlipidemia, hypertension, or obesity, during the post-period. In addition, the likelihoods of an individual negative clinical outcome in the post-period for any of the above conditions were also examined.
Both the propensity score model and the logistic regressions controlled for patient characteristics and general health status in the 6 months prior to the index date (i.e., the pre-period). Specifically, the analyses controlled for patient age, sex, region of residence, type of insurance coverage, pre-period Charlson Comorbidity Index (CCI) score, and visits to an endocrinologist. In addition, the propensity score model also controlled for the dose of Synthroid® and the average copayment for Synthroid® in the lead-in period, while the models that examined outcomes controlled for index LT4 prescription dose and copayment. Multivariable logistic models that examined patient outcomes also included a lagged dependent variable to control for prior presence of the condition of interest. As exploratory analyses, the associations between the key outcomes and the number of switches among LT4 medications in the post-period were examined. In these analyses, a subsequent switch could have been to any of the other LT4 formulations, including a switch back to Synthroid®.
All analyses were conducted using SAS, Version 9.4 (Cary, NC), and a P value less than 0.05 was considered, a priori, to be statistically significant.