We carried out a population based, retrospective cohort study with a nested case-control design using population health databases for Ontario, Canada, linked anonymously by encrypted health card numbers. The Ontario drug benefit database records all publicly funded drugs dispensed to Ontarians aged 65 or older.37
The national ambulatory care reporting system database details visits to emergency departments, and the Canadian Institute for Health Information discharge abstract database provides information on hospital admissions. Independent comparisons have validated the discharge abstract database against hospital medical records, determining accuracy of the database for procedures, diagnoses for primary admission, and major complications,38 39 40 41
and have been used extensively for data on fractures.42 43 44
The Ontario health insurance plan database identifies claims for physician services, and the registered persons database provides information on demographics and death. Diabetes status was obtained from the Ontario diabetes database45
and history of thyroid cancer from the Ontario cancer registry.46
The cohort consisted of Ontario residents aged between 70 and 105 years with at least one prescription for levothyroxine between 1 April 2002 and 31 March 2007. People entered the cohort on the date of their first prescription. Although the Ontario drug benefit database records prescription data for patients starting at age 65 years, we excluded those under 70 to provide at least five years of drug history for all participants.
We excluded people if they had received dialysis or palliative care in the six months before entry to the cohort because of differing goals of care or bone physiology. People were also excluded if they had a history of thyroid cancer or previous hyperthyroidism before entry to the cohort, which might prompt higher dose targets for levothyroxine beyond mere replacement.
The primary outcome was any fracture, defined as the first visit to an emergency department or admission to hospital for any fracture of the wrist or forearm, shoulder or upper arm, thoracic spine, lumbar spine and pelvis, hip or femur, or lower leg or ankle. We excluded fractures that occurred in the context of seizure, trauma, bone malignancy (primary or secondary), multiple myeloma, or disease. The secondary outcome was a hip or femur fracture alone, with the same contextual exclusions. (See web extra for codes according to international classification of diseases, ninth and 10th revisions.)
Participants were followed until the first occurrence of an outcome of interest, death, initiation of dialysis or palliative care, diagnosis of thyroid cancer, or the end of the study period (31 March 2008).
Cases, controls, and matching
We defined a case as any cohort member who had at least one relevant fracture during follow-up, with the date of admission to hospital for the first fracture serving as the index date. For each case we selected up to five potential controls from the cohort who were still at risk for an event on the index date. Controls were assigned the same index date as their respective case. Participants could serve as a control more than once and were later eligible to become a case.
We matched controls to cases on age (within one year), sex, and duration in cohort (follow-up 30 days either way).
Cases and controls were defined, according to use of levothyroxine, as current users, recent past users, or remote users, based on the timing of their most recent prescription for levothyroxine before the index date. The duration of a prescription was calculated as the number of days supplied for the most recent prescription in the Ontario drug benefit database plus an additional 14 days to allow for non-adherence. We classified participants as current users if the duration of their prescription encompassed the index date, recent past users if the prescription ended 15-180 days before the index date, or remote users if the prescription ended more than 180 days before the index date.
The current user subgroup was further categorised based on the cumulative dose of levothyroxine in the year before the index date. We determined this dose by calculating the total number of milligrams dispensed from all prescriptions filled in the previous 15 months intended for drug use in the year before the index date. If participants had been supplied two different concurrent prescriptions for levothyroxine, we used the more recent dose for the overlapping days. We determined the distribution of the cumulative daily dose to be a minimum of 0.018 mg to a maximum of 158.5 mg (median 24.94 mg). Low dose users were those with a cumulative daily dose in the lowest 25th centile of this distribution, medium dose users between the 25th and 75th centiles, and high dose users above the 75th centile.
For current users, we examined all the doses of levothyroxine prescribed in the year before the index date. If the dose changed, we specified the direction of the most recent dose change before the index date. An increase in dose was considered to have occurred if the most recent dose before the index date was higher than the second most recent dose before the index date. A dose reduction was noted if the most recent dose before the index date was lower than the second most recent dose before the index date.
We used conditional logistic regression to estimate associations between levothyroxine use and fracture outcomes in the matched cases and controls, controlling for potential confounders such as fifth of neighbourhood income (from 1 to 5 for the lowest to highest income fifth), Charlson comorbidity index, and history of health service use at the index date. This logistic regression also controlled for other confounders, including previous fractures, risk factors for fracture, and comorbid conditions in the five years before the index date, and the use of other prescription drugs in the year before the index date. The analysis was done using SAS version 9.2.
For our primary analysis we compared the association between any fracture and current use of levothyroxine compared with remote use. We also examined the association between recent past use compared with remote use and fractures to determine whether a residual effect remained after recent discontinuation of levothyroxine. We then examined the association between levothyroxine dose and any fracture in the subgroup of current users. We compared the risk of fracture in the high and medium dose groups with that in the low dose group.
For our first secondary analysis, we examined the outcome of hip or femur fracture specifically as most of these fractures are non-traumatic and carry the greatest morbidity and mortality for older people. We created a new set of cases and controls, whereby cases were all cohort members who had experienced a hip or femur fracture during the observation period and controls were those who did not experience a hip or femur fracture. We repeated the primary analyses for this outcome. Secondly, we compared the risk of fracture between people who had changed dose in the past year with those who had not, considering any dose change as well as a dose increase or decrease. Thirdly, we stratified the cases and controls by sex and repeated all the primary and secondary analyses for men and women separately. We also tested for an interaction between sex and levothyroxine use for each analysis, to assess whether sex modified the effect of levothyroxine on the risk of fractures.47