Statistical considerations in the choice of endpoint for drug use disorder trials

Abstract

Background To date, the US Food and Drug Administration (FDA) requires drug use disorder trials developing new medications to use abstinence, a clinically meaningful endpoint, as the primary outcome. Although abstinence is the gold standard, only a relatively small percentage of participants in drug use disorder trials ever achieve this endpoint. This has prompted clinical trialists to consider quantitative measures of frequency of use, recognizing that some reductions in drug use that fall short of complete abstinence may potentially represent clinically important improvements. While much of the discussion concerning alternative outcomes to abstinence has focused on their clinical relevance, there are important statistical considerations that should also be taken into account. Purpose In this paper, we demonstrate and highlight the degree to which use of a quantitative measure of frequency of use, relative to a binary measure of abstinence, yields a very discernible increase in statistical power for assessing efficacy or effectiveness of treatments for drug use disorders. Conclusion While it is well established that dichotomizing a quantitative measure invariably results in loss of statistical power, what is less well recognized is the degree of loss in the context of drug use disorder trials. In some cases, the required sample size must be almost doubled to achieve the same level of power.

Distress tolerance trajectories following substance use treatment.

“…outcome research (Fitzmaurice, Lipsitz, & Weiss, 2017). Specifically, we regressed the observed DT variables at each posttreatment wave (i.e., T 2-5 ) on a wave-specific covariate that represented frequency of use within the interval since the last assessment (i.e., the percentage of days on which substance use occurred during the intervals between T 1-2 , T 2-3 , T 3-4 , and T 4 -5 ).…”

“…A second model was used to investigate the impact of a time-varying covariate, frequency of use between each assessment time point, on time-specific fluctuations in DT. Utilizing this variable not only allowed us to quantify the amount of substance use but also optimized statistical power when compared to binary measurement approaches of use historically used in treatment outcome research (Fitzmaurice, Lipsitz, & Weiss, 2017). Specifically, we regressed the observed DT variables at each posttreatment wave (i.e., T 2–5 ) on a wave-specific covariate that represented frequency of use within the interval since the last assessment (i.e., the percentage of days on which substance use occurred during the intervals between T 1–2 , T 2–3 , T 3–4 , and T 4–5 ).…”

What defines a clinically meaningful outcome in the treatment of substance use disorders: reductions in direct consequences of drug use or improvement in overall functioning?

“…This may be due in part to the relatively short duration of most clinical trials, in comparison to the 12-month timeframe required for meeting the criteria for sustained full remission in the former DSM-IV (41), which would have indicated clear evidence of improvement. It also may be reflective of researchers’ preference for continuous outcome measures, rather than dichotomous or categorical outcomes, as they are generally known to have greater statistical power to detect a treatment effect (42). Nevertheless, the lack of inclusion of post-treatment/follow-up assessment of DSM diagnostic criteria in clinical trials is notable, despite its direct measurement of an individual’s functioning with respect to drug-related problems or consequences.…”

Consumption outcomes in clinical trials of alcohol use disorder treatment: Consideration of standard drink misestimation

“…Dichotomizing continuous variables also has numerous statistical consequences (3–5), including the obscuring of individual differences, loss of reliability, reduced effect sizes, and loss of power. Researchers have cautioned against collapsing continuous drinking data (e.g., percentage of heavy drinking days; PHDD) into more coarse categories (6,7), noting potentially reduced effect sizes, which may be particularly detrimental for AUD clinical trials that often yield relatively small effect sizes (8,9). There is ample statistical evidence to conclude that dichotomizing continuous outcomes has a detrimental impact on effect size estimation; however, we do not know how much of a detriment this creates specifically when the continuous PHDD variable is dichotomized into NHDD.…”