Longitudinal missing data strategies for substance use clinical trials using generalized estimating equations: an example with a buprenorphine trial

Abstract

Objective: A review of substance use clinical trials indicates that sub-optimal methods are the most commonly used procedures to deal with longitudinal missing information.

Methods: Listwise deletion (i.e., using complete cases only), positive urine analysis (UA) imputation, and multiple imputation (MI) were used to evaluate the effect of baseline substance use and buprenorphine/naloxone tapering schedule (7 or 28 days) on the probability of a positive UA (UA+) across the 4-week treatment period.

Results: The listwise deletion generalized estimating equations (GEE) model demonstrated that those in the 28-day taper group were less likely to submit a UA+ for opioids during the treatment period (odds ratios (OR) = 0.57, 95% confidence interval (CI): 0.39-0.83), as did the positive UA imputation model (OR = 0.43, CI: 0.34-0.55). The MI model also demonstrated a similar effect of taper group (OR = 0.57, CI: 0.42-0.77), but the effect size was more similar to that of the listwise deletion model.

Conclusions: Future researchers may find utilization of the MI procedure in conjunction with the common method of GEE analysis as a helpful analytic approach when the missing at random assumption is justifiable.

Keywords: generalized estimating equations; longitudinal missing data; multiple imputation; positive urine analysis imputation; psychopharmacology clinical trials; substance use disorder treatment.

Figure 1

Percent of participants with a positive opioid UA using all available data across the 4-week trajectory while receiving one of two tapering schedules

Figure 2

Percent of participants with a positive opioid UA assuming all missing UAs were positive across the 4-week trajectory while receiving one of two tapering schedules