We analyzed changes in medication costs during an 18 month clinical trial. The design, methods, and 6 month outcomes of the study have been described previously [11]. Participants (n = 79) were recruited from two primary care internal medicine practices at the University of Colorado. Participants responded after receiving a recruitment letter. They were initially screened by telephone for eligibility, and then were evaluated in-person to ensure appropriateness for participation in an extended behavioral weight loss intervention. Participants had an average age of 56.3 years, an average BMI of 39.5 kg/m2, and were 75.7 % female. All participants had at least one of the following co-morbidities: diabetes/pre-diabetes (34 %), hypertension (58 %), abnormal cholesterol (65 %), or sleep apnea (30 %). Participants did not have to be taking medication for the weight-related condition to be eligible. Among all participants, the average number of weight-related co-morbidities was 1.9.
All participants received intensive behavioral treatment and subsidized access to portion-controlled foods for weight loss during the first 6 months. Thus, months 0 to 6 formed the active weight loss phase of the trial. At month 6, participants were randomized to one of two treatment conditions to help them maintain weight loss: “Standard Maintenance” or “Intensified Maintenance.” All participants continued to have access to subsidized portion-controlled foods, but only those in the Intensified Maintenance group continued in-person visits during months 7–18. A total of 84 participants were randomized to a treatment condition at month 6, and of these, 79 completed 18 months of treatment; thus, the attrition rate was 6.0 %. All participants lost a clinically significant amount of weight during the first 6 months,9 and those assigned to Intensified Maintenance kept off significantly more weight at month 18 (data submitted for publication). The trial was approved by the Colorado Multiple Institutional Review Board (protocol #10-0719) and was registered at www.clinicaltrials.gov (NCT01220089). All participants gave informed consent prior to enrolling in the trial.
Medication names and doses were reviewed with study participants at each of the major assessment points (month 0, month 6, and month 18). Costs were assigned to each medication at the specific dose, using the ‘Big 4’ Federal Supply Schedule, the pharmaceutical price list used by the Coast Guard, the Department of Defense, the Public Health Service, and the Veterans Administration [12]. We opted not to use the average wholesale price from the “Red Book”, as U.S. government reports have suggested that these prices have been inflated by the pharmaceutical industry [13, 14].
In addition to testing changes in total medication costs, we classified medications as being used for diabetes, hypertension, or hyperlipidemia (i.e., a medication to treat a weight-related condition), or as being used for another condition. We hypothesized that if medication reductions were observed, they would most likely be seen in medications used to treat weight-related conditions. We considered classification of any medication used to treat a weight-related condition, but this was not a clear separation (e.g., non-steroidal anti-inflammatory medications and anti-depressants could potentially be classified as being medications for weight-related conditions, or could be classified as treating non weight-related conditions). Finally, we conducted post-hoc analyses to examine whether certain subsets of participants had greater reductions in medication costs. Specifically, we examined individuals with at least a 10 % weight loss, as well as those with diabetes or hypertension, who were likely to be taking the largest number of medications at baseline.
The primary outcome was cost and the main explanatory variable was treatment group. We log-transformed cost and used linear mixed models with random intercept to account for the repeated outcome at three time points (month 0, month 6, and month 18). Time was used as an explanatory variable as well as an interaction between time and treatment group to allow for differential effect of the treatment group at each time point. The smearing approach for back transforming the mean in the log scale was used to obtain an estimate of the mean cost in the original scale [15–17]. To obtain standard errors of the mean in the original scale, the bootstrap method with 2000 samples (with replacement) was used. As the distribution of the differences of cost in the original scale seemed fairly symmetric, p values were obtained assuming a normal distribution for the empirical distribution of the mean estimate. Results based on the linear mixed model were also obtained for cost of medications non-weight-related as well as weight-related. All analyses were conducted in the statistical program R [18]. We present results below for re-transformed data, in order to compare them to other published studies on this topic. Analyses using median data showed similar results. Costs are presented in 2013 U.S. dollar amounts.