Utilization of Propensity Score Weighting and Targeted Maximum Likelihood Estimation in the Post Hoc Analysis of a Randomized Controlled Trial for the Treatment of Cocaine Dependence (NIDA-MDS-Modafinil-0001)

Weiwei Shan, The University of Texas School of Public Health


Cocaine is one of the world’s most addictive and abused substances. According to the National Institute on Drug Abuse (NIDA) and National Survey on Drug Use and Health (NSDUH), in 2016, an estimated 2.1 million people aged 12 and above were past-month cocaine users (0.7% of the US population), and 43.1 million people 12 years and older admitted using cocaine in their lifetime (14.4% of the population). Moreover, NSDUH reported that in 2016, 913,000 Americans were cocaine addicts by the standards of the Diagnostic and Statistical Manual of Mental Disorders. Cocaine use and related complications cause more than 15,000 deaths in the US each year. A stunning 580 million dollars is attributed to direct costs of cocaine associated healthcare interventions each year in the US, imposing a tremendous burden on US taxpayers. Cocaine dependence is characterized by feeding compulsive need to the brain, creating a rewarding loop that overrides the attempts to abstinence. Despite the constant trial to develop treatments for cocaine dependence, currently there has been no Food and Drug Administration (FDA) approved effective medical interventions. Randomized control trials (RCTs) are the gold standard for modern pharmacology and clinical science to appropriately assess treatment effects. RCTs offer a means to attribute differences seen in outcomes solely to the corresponding treatments by providing completely balanced baseline covariate profiles between treatments. This, in turn, is achieved by properly randomized and rigorously controlled design of the RCTs. Previously, modafinil had been shown to attenuate attention deficit disorder via acting on the hypocretin/orexin and glutamate/GABA systems. Additionally, modafinil had exhibited stimulatory neurological effects with appreciable safety profile. In 2009, Anderson et al. conducted a double-blind placebo-controlled RCT for the effect of modafinil in reducing cocaine dependence (NIDA-MDS-Modafinil-0001). In the phase II trial, placebo, 200 mg or 400 mg of modafinil were self-administered by cocaine-dependent subjects for 12 weeks. Cocaine dependence, as evaluated by self-report, laboratory measurements and psychological assessments, was compared between treatment groups. Primary findings from this study suggest that no statistically significant difference was achieved in reducing cocaine use between the control and either of the modafinil treatment arms. Motivation In the study conducted by Anderson et al, ‘adaptive urn randomization’ was used to balance the treatment groups for covariates ‘site of study’, ‘gender’ and ‘cocaine use in past 30 days’. The authors openly acknowledged that covariates ‘race’ and ‘lifetime cocaine use’ exhibited significant difference between at least two of the three treatment groups. In addition, it remains unclear how many other baseline covariates may have exhibited an imbalance between treatment arms. Since causal inference in RCTs completely relies on effective balance of baseline covariate characters, the residual imbalance present in this study may carry over substantial bias in the analysis stage and nullify the results. Therefore, the current study aims at: I. Re-examine the primary and secondary outcomes of interest using propensity score analysis in estimating treatment effects II. Re-examine primary and secondary outcomes of interest using TMLE in estimating treatment effects By employing either propensity score weighting of covariate matrix alone or TMLE doubly-estimation of ATE to the original dataset, the current study reached at the same conclusions as the original findings from Anderson et al. In other words, we confirmed the absence of significant impact of modafinil treatment on the change of weekly cocaine non-use days over the course of the trial. Additionally, we also confirmed that secondary endpoint of interest, i.e. maximum consecutive cocaine abstinence days over the entire trial period, was significantly extended by modafinil 200mg treatment as compared to subjects in the placebo group.

Subject Area


Recommended Citation

Shan, Weiwei, "Utilization of Propensity Score Weighting and Targeted Maximum Likelihood Estimation in the Post Hoc Analysis of a Randomized Controlled Trial for the Treatment of Cocaine Dependence (NIDA-MDS-Modafinil-0001)" (2018). Texas Medical Center Dissertations (via ProQuest). AAI10846306.