From a total of 32,656 treatment episodes captured in the study, 25,545 were noncensored, from which 14,602 were identified as having initiated a taper at some point during the episode. Among these episodes, 4,917 were decreasing or had decreased below 5mg/day in the final four weeks of the episode, thus meeting study inclusion criteria, and ended prior to July 1, 2005, thus allowing 18 months follow-up to verify our definition of sustained success. These episodes were counted amongst 4,183 unique individuals.
From the study sample of 4,917 treatment episodes, 1,305 (27%) episodes were completed with a daily dose ≤ 5mg/day, however among episodes reaching this criteria, 659 were followed by at least one of the following events: treatment re-entry within 18 months (458 episodes, 35.1% of all completed tapers), opioid-related hospitalization (319 (24.4%)), mortality (29 (2.2%)). As such, 1305 − 659 = 646 episodes (13% of the study sample) were defined as sustained successful tapers.
Summary statistics stratified by treatment outcome were presented in . Individuals with treatment episodes resulting in a sustained successful taper were younger, more likely to be male, had lower CCI scores, better treatment adherence, lower maximum mean weekly doses, and longer taper durations.
Results of multivariate analyses on the odds of sustained success in the three models tested were presented in . Patient demographics and treatment adherence variables were included in each model, and their effects were largely consistent across model formulations. Individuals of age 25–34 and 35–49 had 27–40% lower odds of sustained success in tapering compared to individuals younger than 25. Females also had lower odds of sustained success, with the most conservative estimated odds ratio suggesting 19% lower odds of sustained success compared to males (model 3:Odds Ratio:0.81; 95% Confidence Interval:(0.67–0.99)). Individuals with higher levels of medical comorbidity also had lower odds of sustained success in tapering, however this effect was not statistically significant in model 2, when tapering dynamics were represented with the taper duration (0.83(0.68–1.00). Later subsequent treatment episodes had lower odds of sustained success, while episodes initiated in more recent calendar years, controlling for all other measured covariates, had higher odds of sustained success in tapering.
Results of multivariate regression analysis of sustained success in dose tapering
Treatment adherence was included in each of the model formulations, and was positively associated with higher odds of sustained success. Our results suggest that a 1% increase in adherence resulted in a minimum 1.9% increase in the odds of success (model 2: 1.02(1.00–1.03)).
Results of model 1 suggest that individuals reaching a maximum dose of between 60–100mg and more than 100mg had 44% and 60% lower odds of sustained success in tapering compared to those maintained on lower doses. Univariate results were similar to those obtained in the multivariate model. An early taper start week (before week 12) had higher unadjusted odds of sustained success; however the adjusted effect was not statistically significant. Finally, more aggressive tapers had lower odds of sustained success, as episodes with median percentage decreases greater than 4% were nearly 27% less likely to result in sustained success (0.73(0.61–0.88)) in comparison to less aggressive tapers.
Model 2 indicated that taper duration was strongly associated with sustained success. Episodes in which the taper lasted 12–52 weeks were 3.58 times more likely to result in sustained success than those lasting <12 weeks (3.58(2.76–4.65)), while tapers lasting longer than 52 weeks were 6.7 times more likely to result in sustained success (6.68(5.13–8.70)).
Model 3 provides a refinement of model 1 results, indicating that tapers with 25–50% of weeks where the dose was decreasing had the highest odds of success (1.61(1.22–2.14)). In weeks during which the mean dose was decreasing, a median percentage change < 5% had lower odds of success (0.53(0.41–0.71) than episodes in which the median percentage change was 5%–15%, while higher percentage changes in mean dose when the dose was decreasing did not have statistically significantly different odds of sustained success.
Of the three models tested, Model 2, with taper dynamics represented by the taper duration covariate, provided the highest AIC and BIC values, indicating greatest model fit and thus highest explanatory power. Each model formulation was tested using a generalized linear mixed regression model; however given the low number of repeated taper episodes, results using this type of model were identical to the pooled analysis presented.