Integrative Data Analysis of Clinical Trials Network Studies to Examine the Impact of Psychosocial Treatments for Black People who use Cocaine: Study Protocol

Abstract

Background:

Cocaine overdose death rates among Black people are higher than that of any other racial/ethnic group, attributable to synthetic opioids in the cocaine supply. Understanding the most effective psychostimulant use treatment interventions for Black people is a high priority. While some interventions have proven effective for the general population, their comparative effectiveness among Black people remains unknown. To address this gap, our NIDA-funded Clinical Trials Network (CTN) study 0125, will use Integrative Data Analysis (IDA) to examine treatment effectiveness across 9 CTN studies. This manuscript describes the study protocol for CTN-0125.

Methods:

Of the 59 completed randomized clinical trials in the CTN with available datasets, nine met our inclusion criteria: 1) behavioral intervention, 2) targeted cocaine use or use disorder, 3) included sub-samples of participants who self-identified as Black and 4) included outcome measures of cocaine and psychostimulant use and consequences. We aim to 1) estimate scale scores of cocaine use severity while considering study-level measurement non-invariance, 2) compare the effectiveness of psychosocial treatments for psychostimulant use, and 3) explore individual (e.g., concomitant opioid use, age, sex, employment, pre-treatment psychiatric status) and study-level moderators (e.g., attendance/retention) to evaluate subgroup differences in treatment effectiveness.

Conclusion:

The NIDA CTN provides a unique collection of studies that can offer insight into what interventions are most efficacious for Black people. Findings from our CTN-0125 have the potential to substantially inform treatment approaches specifically designed for Black people who use psychostimulants.

Introduction

The increase in overdose death rates among Black adults due to fentanyl contamination of cocaine mandates the need to identify effective treatments for Black people who use psychostimulants.¹ The prevalence of psychostimulant use among Black adults has been estimated at 11.4%, with cocaine at 12.4%.² In 2017, Black adults experienced the highest rate of cocaine-involved deaths compared to other racial/ethnic groups and the largest relative rate change of cocaine-involved deaths compared to previous years. Synthetic opioid use primarily drove these changes.³ Thus, understanding treatment effectiveness for Black people who use cocaine and other psychostimulants is a high priority issue.

Despite strong evidence of the efficacy/effectiveness of psychosocial treatments for substance use disorders (SUDs)⁴, their comparative effectiveness among Black people who use psychostimulants remains unknown. A scoping review of studies from the National Drug Abuse Treatment Clinical Trials Network (CTN) found that of the 5,804 Black participants that have been enrolled in CTN studies as of May 2019, only ten trials reported substance use outcomes specific to Black participants.⁵ Hence, little is known about differences in treatment outcomes (e.g., by treatment type, sociodemographic characteristics) among Black people who use psychostimulants across psychosocial SUD treatments that employ evidence-based approaches.

The literature assessing the effectiveness of psychostimulant use treatment programs focuses on race comparisons (e.g., Black people relative to White people). These between-group comparison studies do not distinguish between treatment differences due to non-equivalence of outcome measures across racial groups ^5–7 and differences explained by moderators such as contextual or socio-demographic factors (e.g., private vs public clinics, retention, treatment adherence, attrition, age, sex, employment), some of which may be uniquely relevant for Black people (e.g., treatment access).^8–12 Further, race comparison studies simply describe differences between racial groups without illuminating within-group differences, which is useful for understanding treatment effectiveness. Studies are needed that focus solely on Black individuals who use psychostimulants to circumvent these issues and identify key factors that inform variations in treatment outcomes among Black people.^13,14 More importantly, focusing on one subgroup is key for identifying treatment effects that would otherwise be obscured in studies that combine data from distinct populations with differential effects.

Prior single CTN RCTs were too small to examine within-group variation in treatment efficacy among Black participants and were statistically underpowered to examine treatment outcomes. Pooling data across RCTs is a viable option for resolving the sample size issue but requires special attention to ensure scale scores for key constructs are commensurate across studies, time, and populations. ^15,16 Integrative data analysis (IDA) is popular for harmonizing data across multiple studies where the measures vary across studies, with one of the noted advantages being increased sample size for analyses with understudied and/or priority populations.^17,18

Using IDA, this study funded by NIDA’s National Drug Abuse Treatment Clinical Trials Network (CTN-0125; MPIs: Burlew, A.K., & Ruglass, L.M.) aims to investigate the comparative effectiveness of psychosocial treatments for cocaine and psychostimulant use among Black adults by combining data from multiple CTN RCTs. This manuscript describes the study protocol that will be used to examine comparative effectiveness and explore other potential moderating factors that may influence treatment effectiveness.

Additionally, convincing evidence suggests that substance use among Black adults is inextricably linked to the social determinants of health (e.g., quality healthcare)¹² and other social contextual factors.¹⁹ These factors are associated with both the development of ^19,20 and recovery from SUD.²¹ Thus, we hypothesize that substance use treatments that also address factors such as social contextual conditions, co-occurring disorders, and self-efficacy (which we identify as social contextual interventions) will have better outcomes among Black adults than treatments which solely focus on substance use behaviors (e.g., contingency management or assessment only approaches).

Methods

Research Aims

CTN-0125 has three specific aims: 1) Estimate scale scores of cocaine use severity while considering study-level measurement non-invariance, 2) Compare the effectiveness of psychosocial treatments (e.g., social contextual vs. contingency management or assessment only) for cocaine use, and 3) Explore individual (e.g., concomitant opioid use, age, sex, employment, pre-treatment psychiatric status) and study-level moderators (e.g., attendance/retention) to evaluate subgroup differences in treatment effectiveness. While the study aims primarily focus on cocaine, data for other psychostimulants will also be explored. This study was determined as non-human subjects research by the Yale Institutional Review Board (IRB; # 2000031552).

Overall Strategy, Methodology, and Analysis

The project will use a combination of three analysis frameworks to integrate raw data across RCTs: random treatment effect multilevel modeling (RTE-MLM), moderated nonlinear factor analysis (MNLFA), and propensity score weighting. The use of RTE-MLM is similar to the analysis framework that is used in meta-analysis of individual patient data (MIPD),^19,22 where cross-study variability in intervention effects is of primary interest. Although our project is not structured for the conduct of a meta-analysis, in that we did not systematically search studies beyond the NIDA CTN, there is still utility in using this framework. MIPD characterizes variation in effects across studies (like conventional meta-analysis), ^22,23 but has specific advantages for raw data analysis, including (1) guaranteeing the same statistical model is used to estimate treatment outcomes across integrated data sets, (2) analyzing subgroups with greater statistical power, and (3) disentangling individual-level moderators of intervention effectiveness from study-level aggregates of individual-level factors (e.g., percentage African American in a given study).²² This latter point allows for addressing concerns about potential “ecological fallacy” effects, where an effect of a variable at one level is not the same at a different level of aggregation.²⁴

In IDA, advanced scale scoring methods within the item response theory (IRT) paradigm, such as MNLFA,^17,25–28 are applied to create comparability in measures across studies that may be assessing the same construct with at least some variation in item content ^17,25 before advanced analytic methods are used to analyze the combined data set. Propensity score weighting, typically used as a statistical method that controls for selection bias in non-experimental studies, ^29,30 will be used because treatment types will be combined across studies, which could compromise the original within-study randomization structures.^16,31

Study Selection

Of the 59 completed randomized clinical trials in the CTN with available datasets in 2021, nine met our inclusion criteria: 1) behavioral intervention, 2) targeting substance use or use disorder, 3) included sub-samples of participants who self-identified as Black; 4) included outcome measures of cocaine and psychostimulant use (biological, self-report, clinician-administered) and consequences. The CTN Protocols that met criteria are: Motivational Enhancement Therapy (MET; CTN0004), Motivational Interviewing (MI; CTN0005), MI in drug free clinics (CTN0006), MI in methadone clinics (CTN0007), MET for pregnant people (CTN0013), STAGE 12 (adapted from Twelve-Step Facilitation Therapy; CTN0031), Therapeutic Education System (TES; CTN0044), Screening, Brief Intervention, and Referral to Treatment (SBIRT; CTN0047) and Seeking Safety Treatment (CTN0015). We will focus on six timepoints: baseline, in-treatment, end-of-treatment, and 3, 6, and 12-month follow-ups. To examine standardization across trials, we examined all protocols to identify measures of fidelity.

Description of Included Studies

Social Contextual Interventions.

Previous research suggests that issues such as social contextual ³² and psychosocial ^33,34 factors are associated with substance use among Black adults. Six of the interventions address social contextual issues: Motivational Interviewing (MI, CTN0005), Motivational Enhancement Therapy (MET; CTN0004), MET for pregnant women (CTN0013), STAGE 12 (CTN0031), Seeking Safety (CTN0015), and the Therapeutic Education System (TES, CTN0044). Together, these interventions address psychosocial, psychological, and contextual factors related to substance use such as trauma, mood, relationship issues, and readiness to change.

Contingency Management/Assessment-Only Interventions.

The other three interventions do not address social contextual issues that Black people who use cocaine and psychostimulants experience: Contingency Management (CM; CTN0006,) for Drug Free Clinics, CM for Methadone Clinics (CTN0007), and Screening, Motivational Assessment, Referral, and Treatment in Emergency Departments (SMART-ED; CTN0047). The focus of these interventions is either on incentivizing abstinence (CM) or assessment/referral to treatment (SMART-ED). Based on evidence that treatments that address social contextual factors may be more effective in reducing substance use, we hypothesize the 6 interventions that address social contextual issues will be associated with better outcomes among Black adults who use psychostimulants relative to the other three interventions. More detail on all nine interventions appears in Table 1.

Table 1:

Description of treatments and treatment as usual (TAU) or control condition procedures across the nine CTNs of the current study

Protocol	Experimental Condition	Control Condition/TAU	Overall Study Outcomes
Social Contextual Interventions
CTN-0004 Motivational Enhancement Therapy (MET)	Based on similar principles as MI, MET also provides the individual with clinically-relevant personal assessment data to further enhance motivation for change. Participants receive three individual session that aim to strengthen and consolidate commitment to change and promoting a sense of self-efficacy while reducing ambivalence for abstinence.	Three weekly sessions of individual counseling as is usually provided by CTP, without any form of MI.	The MET group did not differ from CAU on substance abuse outcomes. However, MET participants had significantly better retention at follow-up than CAU participants.
CTN-0005 Motivational Interviewing	Brief one session, 2 hour, intervention based on client-centered principles. The counseling strategies, reflective listening, summarizing, and paraphrasing, are designed to motivate individuals to identify and reduce the discrepancy between their current and desired states.	2-hour assessment/evaluation session where therapist collects standard data in accordance of their CTP (i.e., current substance use history, treatment history, psychosocial functioning, etc); referral for standard treatment (typically group treatment) at the CTP, without any form of MI.	Participants assigned to MI had significantly better retention – a key component of success --through the 28-day follow-up than those assigned to the standard intervention. There were no significant effects of MI on substance use outcomes at either the 28-day or 84-day follow-up
CTN-0013 Motivational Enhancement Therapy for Pregnant Women	This 3-session intervention is similar to MET but is tailored to be appropriate for pregnant women. Along with discussing the perceived pros and cons of using substances, the sessions also include a discussion of the possible adverse effects on the fetus.	Three individual sessions, 1–2 hours each; no MET offered.	No differences between MET and treatment-as-usual (TAU) participants. Some evidence that the efficacy of MET varied between sites, and that MET might be more beneficial than TAU in decreasing substance use in minority participants
CTN-0015 Seeking Safety (SS)	Based on the assumption that including tools to restore a sense of safety for trauma survivors is essential to reducing substance use, SS in the CTN is an integrated cognitive-behavioral 12 session treatment for women that addresses trauma and substance use simultaneously.	Control group received Women’s Health Education (WHE), a treatment that focuses on topics such as the women’s body, human sexual behavior, and pregnancy. Additionally, control received six weeks of individual and group sessions administered in a variety of orientations and philosophies of addiction treatment.	Seeking Safety Participants did not differ from control group (Women’s Health Education) on substance use outcomes
CTN-0031 Stimulant Abusers Groups to Engage in 12-Step (STAGE-12)	STAGE-12 focuses on addressing the physical, psychological and social factors that impact both the development and the maintenance of drug addiction. The intervention included five 90-minute group sessions plus three complementary individual sessions.	12-step oriented – general approach resembles 12-step, but without some key empirically based practices that were emphasized in experimental treatment.	Compared to TAU, STAGE-12 participants had significantly greater odds of self-reported stimulant abstinence during the active 8-week treatment phase along with higher rates of 12-step meeting attendance and were engaged in more related activities
CTN-0044 Therapeutic Education System (TES)	TES incorporates cognitive behavioral skills training along with skills to improve psychosocial functioning (e.g., family/social relations, managing negative moods, etc.). Two internet individual modules per week replace TAU for 12 weeks.	Combination of group and/or individual counseling typically that is offered at the included CTNs; administer at least twice a week.	Compared to TAU, participants who received Therapeutic Education System had higher levels of abstinence and reduced dropout rates.
Contingency Management or Assessment-Only Interventions
CTN-0006 Motivational Incentives: Drug-Free Clinics	Motivational incentives are based on the behavioral theory of operant conditioning, which posits that behaviors increase or decrease based on type and amount of reinforcement. In this protocol, patients participate in variable ratio abstinence-based incentive procedure in addition to their usual care treatment. The maximum cash value for tangible incentives is a max of $400/participant.	Counseling utilization, or treatment that was offered but not required during 12-week study. Offered in individual (15+ minutes) and group format. No incentives are offered to control group.	Participants in the abstinence-based MI condition produced more drug and alcohol free samples than those in comparison condition.
CTN-0007 Motivational Incentives for Drug-Free Recovery: Methadone Clinics	Motivational incentives are based on the behavioral theory of operant conditioning, which posits that behaviors increase or decrease based on type and amount of reinforcement. In this protocol, patients participate in variable ratio abstinence-based incentive procedure in addition to their usual care treatment. The maximum cash value for tangible incentives is a max of $400/participant.	Counseling utilization, or treatment that was offered but not required during 12-week study. Offered in individual (15+ minutes) and group format. No incentives are offered to control group.	Participants in abstinence-based (MI) condition produced more drug and alcohol free samples than those in comparison during study period; no group differences at 6 month follow-up
CTN-0047 Screening, Motivational Assessment, Referral, and Treatment in Emergency Departments (SMART-ED)	SMART-ED is an integrative approach that combines screening to assess for level of severity of substance use, a brief intervention based on MI principles designed to assess and enhance motivation for change, and referral for additional treatment (depending on level of substance use severity). This protocol tests 3 conditions: Minimal screening only (MSO), Screening and Referral to Treatment (SAR) and Brief Intervention + Booster (BI+B). the BI+B participants receive the same info and referral as SAR, and are provided with a brief manual-led intervention based on MI principles.	The MSO participants are provided with an informational pamphlet with no further intervention. The SAR participants are provided with the same informational pamphlet as the MSO participants and receive referrals if they have an ASSIST score of ≥ 27 (probable dependence).	No group differences in self- reported days using the primary drug, days using any drug, or heavy drinking days at 3, 6, or 12 months

Note. CTN = Clinical Trials Network. MI = motivational interviewing. MET = motivational enhancement therapy. CTP = community treatment program. CAU = counseling as usual . TAU = treatment as usual.

Measures

Outlined below are the predictors (at the individual and study level) and outcomes (cocaine use severity, other substance use consumption, and substance use consequences) that will be used for the current study. See Table 2 as an overview of the available measures for these variables within each study.

Table 2.

Study Details

CTN Study	Black Participants (N)	Substance Use Outcome Measure	Time Points for Data Collection
CTN-0004 Motivational Enhancement Therapy (MET) vs Counseling as Usual (CAU)	194	UDS SUC ASI	UDS and SUC Baseline; Weekly for 4 weeks of treatment; 4 and 7 week follow up ASI-Lite Baseline and 4 week follow up
CTN-0005 Motivational Interviewing (MI)	41	UDS SUC/ASI-lite	UDS, SUC, ASI baseline, 28-days follow up; 84-days follow up
CTN-0006 Motivational Incentives (MI; Drug- Free)	149	UDS	Weekly for 12 weeks; 3,6 month follow up
CTN-0007 Motivational Incentives (Methadone)	194	UDS	Weekly for 12 weeks; 3,6 month follow up
CTN-0013 MET for pregnant substance users	69	UDS SUC ASI	UDS and SUC Baseline, weekly during treatment, 1 month - and 3 months- follow up ASI baseline, end of treatment, 1 month and 3 month follow up
CTN-0015* Women’s Treatment for Trauma and Substance Use Disorders	120	UDS ASI-Lite	UDS Baseline, during treatment, and 1,3,6,12 week follow up ASI-Lite Baseline and 1,3,6,12 week follow up
CTN-0031* Stimulant Abuser Groups to Engage in 12- Steps (STAGE-12)	~171**	UDS ASI SUC	UDS and SUC Baseline, 4th and 8th weeks of treatment, and 3,6 month follow up ASI Baseline, 3,6, month follow up
CTN-0044* Web-Delivery of Evidence-Based Psychosocial Treatment for SUDs	116	UDS SUC	UDS and SUC Baseline, twice weekly for 12 weeks; 3,6 month follow up
CTN-0047 Screening, Motivational Assessment, Referral, and Treatment in Emergency Departments (SMART-ED)	440	SUC	SUC Screen, Baseline, 3,6,9 month follow up

Note. CTN = Clinical Trials Network. UDS = Urine Drug Screen. SUC = Substance Use Calendar. ASI = Addiction Severity Index – Lite.

^*Treatments that address psychosocial factors in addition to substance use.

^**This is approximate because only the percent of the total sample that is African American is provided.

Predictors

Individual-Level Predictors.

At the individual-level, the following will be examined as variables that account for individual-level variation in treatment outcomes: gender, age, education level, treatment dosage (proportion of available sessions attended/doses taken), substance use other than cocaine, baseline major depressive disorder and baseline concomitant psychotropic medication usage. All of these variables will also be aggregated to the study-level and the individual-level analogs will be study-centered to capture separate variation of treatment effects that vary, for example, between studies that have a high versus low proportion of women versus treatment effects that vary for any specific woman within study.¹⁵

Study-Level Predictors and Intervention Arms.

At the study-level, in addition to study-level aggregates of individual-level variables listed above, we list the proposed treatment condition classifications: Social contextual interventions will include 1) Motivational Interviewing/Motivational Enhancement Therapies (CTN0004, 0005, 0013); 2) Seeking Safety (CTN0015); 3) STAGE-12 (12-Step facilitation) (CTN0031); and 4) Therapeutic Education System (CTN0044; TES). Contingency Management/Assessment-Only Interventions will include: 1) Contingency Management (CTN0006 and 0007); and 6) SMART-ED (CTN0047).

Outcomes

Latent Cocaine Use Severity.

Items from the ASI will be used to investigate cocaine use severity. These items will include measures of abstinence from cocaine and reductions in days of use of cocaine.

Latent Substance Use Consumption.

The current plan is to explore latent substance use consumptions using five indicators: a) three indicators of number of days used in the past 30 for cocaine, heroin, and non-heroin opioids from the Addiction Severity Index ([ASI]³⁵ Available in 8 of 9 trials) and b) two binary indicators from the Urine Drug Screen (available in all 9 trials) based on cocaine and opioid metabolites. Consumption of other psychostimulants will also be explored.

Latent Substance Use Consequences.

Items from four consequences sub-domains within the ASI ³⁶ will be used: a) Drug problems b) Family/Social Relationships, c) Legal Consequences, and d) Psychiatric Status.

Data Analysis Plan for Aim 1: Harmonization and IDA

The procedures described in this section are similar to other published IDA protocol papers.^15,16 Aim 1 of CTN-0125 is to estimate scale scores of cocaine use severity while accounting for study-level measurement non-invariance (MNI). The analytical procedures for Aim 1 require the following sequence of steps: (1) Measure harmonization, (2) Dimensionality testing, and (3) Integrative data analysis (IDA).

Measure Harmonization.

Upon obtaining data corresponding to all 9 RCTs, we will combine face-valid indicators reflecting similar item content from different measures across multiple RCTs. This procedure will entail two steps. The first step corresponds to the harmonization of key factors (e.g., demographic characteristics, pretreatment psychiatric status). Although some key factors can be harmonized across studies easily (e.g., sex), others (e.g., pre-treatment psychiatric comorbidity) will be reduced to a lowest common denominator (e.g., a yes/no response) so that items are on the same metric. Cocaine severity measures will include a combination of cocaine use from urinalysis, self-reported use, and consequence measures in a manner similar to the mix of alcohol use and consequences detailed in Bauer and Hussong.¹⁷ As most RCTs will have used the same cocaine severity measures, little semantic harmonization will be needed because item content will be similar.²⁰ However, consistent with other IDA work, ^28,31 analytic harmonization will still be required to ensure commensurate scale scores for latent variables.²⁵

Tests for Dimensionality.

We will then assess the general factor structure of all severity measures to determine if they fall within a single or multiple dimensions. We will reflect this structure in item parameter and scale score estimation in the next step.

Integrative Data Analysis.

Finally, we will employ Integrative Data Analysis (IDA) with Moderated Nonlinear Factor Analysis (MNLFA) to estimate scale scores of cocaine use severity. This procedure will ensure that measures are commensurate across studies (and other predictors listed in the Measures section) in the unbiased estimation of scale scores - a key step highlighted by Bauer and Hussong ¹⁷ and Curran, ²⁵ who juxtaposed classical (e.g., calculating sum scores for items) and modern methods (e.g., MNLFA, IRT ²⁷) of scale score estimation. MNLFA ^17,27 is an extension of the general structural equation model that can be used for scale score estimation with a mix of observed indicators (e.g., continuous, binary, unordered categories, etc.). MNLFA extends the conventional factor analysis model, which includes estimation of factor loadings and item intercepts/thresholds. Unlike conventional factor analysis, MNLFA allows factor loadings and/or item intercepts/thresholds to differ across factors such as time or demographic variables (i.e., DIF). Thus, differences in measurement can be separated out from true differences on the construct of interest as demonstrated in previous work.^6,16,21,37 When used for IDA, ²⁵ MNLFA allows different item sets from different scales to be used across different samples; item content that does not overlap across studies can simply be treated as missing data under the missing-at-random (MAR) assumption. Additional technical detail can be found in Bauer, ²⁷ with examples found in aforementioned articles.^38,39

Data Analysis Plan for Aim 2: Primary Outcome Analysis

Aim 2 is to compare the effectiveness of psychosocial treatments for cocaine use severity. This aim amounts to the estimation of comparative effect sizes for psychosocial treatments that utilize evidence-based approaches relative to treatment-as-usual (TAU) using IDA-estimated cocaine use severity scale scores. Our hypothesis centers largely around anticipated general differences favoring treatments that attend to social contextual conditions, co-occurring disorders, and self-efficacy (MET, CBT, and TSF) compared to contingency management or assessment-only approaches on cocaine use severity.

Nonrandom Cross-Study Grouping: Propensity Weighting.

In analyses under all aims, intervention and comparison conditions will be combined across studies in a manner not intended in the trial design and may compromise within-study randomization. Thus, non-zero correlations between covariates and treatment assignment may be introduced by “mixing-and-matching” treatment arms across studies.¹⁹ This may undercut the benefits of the original randomization within each original RCT, because the factors such as inclusion criteria, demographic makeup of the trial location, etc. were different and may be (inadvertently) related to the treatment classes that were studied in each trial.^16,39,40 Here, propensity score weighting for multiple treatments will be used to weight the data so that the correlations between covariates and the treatment conditions are near 0, in a manner that will mimic randomization across studies. To generate weights for the probabilities of being in each treatment class, we will use the R function ‘mnps’ for multiple treatments ⁴¹, extension of the R package ‘twang’ originally designed for comparisons of two nonrandomized groups. When the propensity score is controlled for, differences on covariates across conditions when mixed across studies should be balanced to a standardized mean difference (i.e., Cohen’s d) of < |.10|.

Propensity Score-Weighted Multilevel Linear Modeling (PSW-MLM).

After scale score estimation under MNLFA but before model fitting, we will assess if there is significant variation across the four (potential) levels of aggregation for each set of MNLFA scale scores for key outcomes: (1) within-individual level (repeated measures), (2) between-individual level, (3) node-level, and (4) study-level; node-level within-study must be examined because many of the CTN studies were multi-node studies. We also will examine the functional form of changes over time in outcomes (e.g., linear, quadratic, piecewise linear). As an example, we present a three-level model (omitting the site-level) with linear time steps linking time to repeated measures, with year being the unit of time. Propensity score weights will be incorporated (see below). For Level 1 (within-individual), we will estimate two random effects as growth over time:

$${\beta }_{pqs\mathrm{ }}=\mathrm{ }{\beta }_{00}+\sum _{q=1}^Q{\gamma }_{pq}{W}_{qs}+u_{pqs}$$ (1)

On the basis of the structure of the time steps within a_psi, we will estimate a random intercept (π0si), which is the estimated (conditional) mean value of the outcome at time = 0 (e.g., baseline) and a random slope (π1si), the estimated per year change in Y. These values will vary across individual i within study s. In Level 2 (between-individual), we have the following structure, where the Level-1 intercept and slopes are outcomes at Level 2:

$${\pi }_{psi\mathrm{ }}=\mathrm{ }{\beta }_{p0s}+\sum _{q=1}^Q{\beta }_{psq}{X}_{qsi}+r_{psi}$$ (2)

βpsq is a matrix of coefficients corresponding to the effect of between-individual predictor q on Level-1 outcome p (intercept, slope) within study s. The key predictors included in Xqsi would be a series of 0/1 dummy indicators indicating whether individual i was in the active intervention condition within study s or the comparison. The key coefficients would be included within β1sq, corresponding to a series of intervention effects comparing changes over time among the reference comparator (e.g., TAU) against the focal intervention, allowing for variations in TAU across study. Study-level predictors will be included at Level 3, as shown in equation 3 below:

$${\beta }_{pqs\mathrm{ }}=\mathrm{ }{\beta }_{00}+\sum _{q=1}^Q{\gamma }_{pq}{W}_{qs}+u_{pqs}$$ (3)

βpq0 is a matrix of conditional means; β110 would capture the average (adjusted) intervention effects across all studies. Wqs contains a series of study-level predictors that account for variation across studies in (among other parameters) intervention effects. The primary study-level predictors (parameters within γpqs) are a series of study-level dummy indicators capturing the study-level covariates (e.g., % Male).

Data Analysis Plan for Aim 3: Causal Moderation Analysis

Aim 3 of CTN-0125 is to explore individual-, and study-level moderators of treatment effects on cocaine use severity outcomes. Examination of moderation will largely be exploratory. In this vein, this study will be among the first to identify differences in treatment effectiveness between subgroups characterized by concomitant opioid use, age, sex, ethnicity, employment, pre-treatment psychiatric status, and retention within Black people who use cocaine. To address this aim, we will use the causal moderation framework of Bansak ⁴², which involves parallel multilevel regressions as proposed for Aim 2 analysis but adding subsets of each treatment class as well as propensity weights.

Propensity Scoring for Causal Moderation.

Because the proposed moderator variables were not randomized, we will incorporate propensity scoring for moderation.⁴² This method, delineated by Bansak ⁴² under the parallel within-treatment regression framework, first stratifies the dataset by treatment condition(s). Next, separate multilevel models are fit by treatment condition for the moderator(s) effects (Z) on the outcome Y (net of covariates). For each of these models, parameters estimates and standard errors of Z’s effect on Y are saved for each treatment condition. The difference between the Z-to-Y effect between two treatment conditions (e.g., TAU and CBT) constitute the test for causal moderation, but under the assumption that there is no relation between treatment conditions and covariates. The addition of inverse probability of treatment weights (IPTW) to the treatment-stratified models would only remove the relation between treatment conditions and measured covariates, which is a general limitation of propensity score models.

Missing Data

The main analyses under all aims will be modeled under full information ML (FIML) and supplemented by multilevel multiple imputation (MI) for missing data (i.e., R package ‘mice’ ^43,45) on key predictors for models under Aims 2 and 3. Both FIML and MI produce accurate estimates and standard errors under the assumption that missingness is predictable by variables that are observed but unrelated to the values that are missing themselves (i.e., missing-at-random or MAR).⁴⁴ For missingness on covariates with regard to propensity score estimation, we will follow the recommendations of Qu and Lipkovich ⁴⁵, who specify a MI approach similar to the typical MI steps where (1) n copies of the data set (e.g., 20 ⁴⁶) are made, with plausible values for missing data on covariates imputed under Markov Chain Monte Carlo estimation; (2) propensity scores are estimated in each of imputed data sets, capturing the probability of each participant’s exposure to each intervention in combination with their status on the moderator (conditional on the covariates); (3) treatment effect models are estimated in each data set, weighted by IPTWs; and (4) results from outcomes analyses across imputed data sets are combined, accounting for both between-imputation and within-imputation variation.

Discussion

Black adults who use cocaine and other psychostimulants are disproportionately impacted by the current opioid epidemic due to the contamination of fentanyl in drug supply, experiencing a 575% increase in cocaine/opioid related deaths from 2007–2019.⁴⁷ Despite strong evidence from the NIDA CTN that certain psychosocial treatments for SUDs (e.g., MI, SBIRT, CM) work, the comparative effectiveness of these treatments for Black people who use substances remains unknown. Furthermore, due to the influence of specific social contextual factors ⁴⁸ and social determinants of health ⁴⁹ that disproportionately impact Black adults who use psychostimulants, a viable hypothesis is that specific treatments that do not address these factors may not be adequate for the treatment of SUDs among this population.

Integrative data analysis overcomes the fact that previous studies are often underpowered and under-resourced to study treatment outcomes among Black adults by combining multiple data sets to provide statistical power to investigate understudied research questions in priority populations. By using multiple key measurement and analysis frameworks, IDA will extend beyond more traditional approaches to cross-study data analysis by creating comparability in measures across studies that may have assessed constructs differently.

The current study has the potential to contribute substantially to both the research field as well as to inform treatment. This novel data analytic approach could serve as a model for future integrative data projects focused on other priority populations. Exploring data across several CTN studies may also uncover a potential need to unify measurement approaches for important variables such as race/ethnicity to allow for increased ease of future cross-study analyses. Importantly, our findings could increase our knowledge on the most appropriate treatments for Black people who use psychostimulants. Explorations of other factors contributing to treatment efficacy could inform future intervention development.

Limitations

As with all studies, this project has several limitations. The lack of attention to variables relevant to the Black experience (e.g., discrimination, ethnic identity) is a limitation of secondary analysis of studies that were not developed with the goal of informing treatment outcomes among Black adults. Also, Integrative Data Analysis has several inherent challenges. The first is that the retrospective nature used in harmonizing variables sometimes results in the loss of measurement variability when the variable has to be reduced to the lowest common denominator (e.g., yes/no). A second is related to hypothesis development. Our initial hypothesis was that interventions which address social contextual conditions (focusing on social determinants of health or other social contextual factors) will have better outcomes among Black adults than interventions which solely focus on substance use (contingency management or assessment only interventions). However, the fact that the interventions within both categories are so broad suggests that more categories may be more appropriate in future research. For example, it is possible that expansion of the definitions of treatment categories based on expert consensus ¹⁹ can lead to treatment conditions that do not have full representation of individuals with certain covariate characteristics.¹⁵ For example, one of the studies in CTN0125 that was recruited because of SUD and PTSD comorbidity was an all-women study (CTN0015 ⁵⁰); thus, baseline PTSD, gender, and cognitive-behavioral treatment approach would be perfectly collinear. Finally, given lack of data availability in this study we will not be able to differentiate between intended and unintended use of opioids and, due to the timing of when these data were collected, we will not have data on fentanyl use.

Conclusion

The increase in opioid-involved overdose deaths for Black people who use substances, largely attributable to the presence of synthetic opioids (such as fentanyl) in the psychostimulant supply, is evidence of the immediate need for greater understanding of substance use treatment effectiveness for Black people. While evidence of the efficacy of behavioral interventions for SUDs does exist, the comparative effectiveness of treatments for Black people remains unknown. The NIDA CTN provides a unique and underexplored collection of studies to address this question. Findings from this study have potential to substantially impact treatment by identifying treatment approaches that are more appropriate for Black people who use cocaine.

Highlights.

• Opioid-involved overdose deaths among Black people who use cocaine are on the rise

• While interventions exist, the comparative effectiveness for Black people is unknown

• Integrative data analysis will be used to combine data across 9 NIDA CTN trials

• A protocol for examining comparative effectiveness is presented