Project Harmony 2.0: protocol for an updated and expanded individual patient data meta-analysis and virtual clinical trial of treatments for comorbid PTSD and substance use disorders

ABSTRACT

Background: Posttraumatic stress disorder (PTSD) and substance use disorders (SUD) commonly co-occur. Several treatments for co-occurring PTSD and SUD (PTSD + SUD) exist, yet evidence for comparative treatment effectiveness is restricted by a historical emphasis on symptom severity and substance consumption endpoints instead of recovery-oriented outcomes. Moreover, extant syntheses relied on conventional analytic approaches prioritizing standardized mean group differences, instead of clinically significant change (CSC), and included few pharmacological randomized clinical trials (RCTs).

Objective: In an extension of Project Harmony (R01AA025853), this study (R01DA061338) will apply novel quantitative approaches addressing limitations of prior work to elucidate which PTSD + SUD treatments work best overall and at the individual level.

Method: This study will harmonize data from over 60 RCTs (N > 8,000) and use meta-analysis of individual patient data, integrative data analysis, and propensity score weighting to assess the comparative effectiveness of pharmacological, behavioral, and combination treatments for PTSD + SUD. Clinical severity outcomes will be measured as latent PTSD symptom severity, alcohol consumption, and other drug consumption. Recovery outcomes will be estimated from indicators of psychosocial functioning and substance use consequences. State-of-the-art CSC methodologies will be used in analyses of the original Project Harmony dataset (36 RCTs; N = 4,046), which will evaluate the proportion of patients in each treatment class that showed (non)significant improvement or worsening along clinical and recovery outcomes. Analyses of the expanded Project Harmony dataset will assess comparative effectiveness of active PTSD + SUD treatment classes, compared to treatment as usual, on clinical severity and recovery outcomes.

Conclusions: By applying modernized quantitative methodologies to patient-level data from at least 60 RCTs, this study will advance the field’s understanding of which PTSD + SUD treatment approaches are most effective across multiple indicators (i.e. clinical severity and recovery outcomes). This study is registered in the International Prospective Register of Systematic Reviews (PROSPERO identifier CRD420251250349).

HIGHLIGHTS

• Extant evidence syntheses comparing the effectiveness of different treatment approaches for co-occurring posttraumatic stress disorder and substance use disorder (PTSD + SUD) have been limited due to methodological issues and have not examined recovery-oriented treatment outcomes.

• This study will combine data from over 60 randomized controlled trials of PTSD + SUD treatments, equivalent to at least 8,000 patients, and apply advanced quantitative methods to compare the effectiveness of treatment approaches, all compared to treatment as usual.

• This study will clarify which PTSD + SUD treatments work best overall and at the individual patient level across PTSD symptom severity outcomes, alcohol consumption outcomes, other drug consumption outcomes, and recovery-oriented outcomes.

1. Introduction

Substance use disorders (SUDs) are globally prevalent and costly, especially when comorbid with posttraumatic stress disorder (PTSD) (Simpson et al., 2019; Zhang et al., 2024). Compared to patients with either disorder alone, patients with co-occurring PTSD and SUD (PTSD + SUD) have poorer treatment outcomes, greater psychosocial functioning difficulties, and higher risk for self- and other-directed violence (Blakey et al., 2022; Najt et al., 2011; Ouimette et al., 2006; Simpson et al., 2019). To address this public health issue, several treatments have been developed for people with PTSD + SUD, including behavioral interventions, pharmacotherapies, and combined interventions (Hien et al., 2023). Behavioral approaches may separately address PTSD (Hamblen et al., 2019) or substance use (Marlatt & Donovan, 2005), although some therapies are integrated to explicitly target PTSD and substance use simultaneously (Back et al., 2015). PTSD-focused and integrated interventions can be further differentiated (Hien et al., 2023) based on whether they are trauma-focused (i.e. directly addressing trauma-related memories, emotions, and beliefs) or non-trauma-focused (i.e. present-centred treatments) (Najavits, 2002).

Given the numerous available PTSD + SUD treatment options, research has aimed to identify which treatments work best overall. Systematic reviews and meta-analyses of interventions for PTSD + SUD have offered varying conclusions but collectively provide support for trauma-focused behavioral therapies and certain SUD-focused interventions (Hien et al., 2024; Petrakis & Simpson, 2017; Roberts et al., 2015; Roberts et al., 2022; Simpson et al., 2017; Simpson et al., 2021). Pharmacotherapy findings are limited and mixed, however, potentially because extant reviews were based on a small number of pharmacotherapy trials. One likely reason for mixed evidence synthesis findings is that investigative teams did not classify treatment approaches consistently across reviews. Further, most meta-analyses did not estimate direct comparative effect sizes for distinct treatment approaches against a common comparator. Another explanation for equivocal findings is that traditional meta-analytic methods in general can suffer from inconsistency in information for specific subgroups across published reports, publication bias, ‘ecological fallacy’ effects (i.e. study-level predictors that do not mirror effects at the individual-level), and general bias in effect sizes with summary data compared to raw data (Hien et al., 2023; Lubinski & Humphreys, 1996; Patall et al., 2008). Such limitations could lead to faulty recommendations about which treatments for PTSD + SUD work best and under what circumstances. Moreover, measurement bias across measures, populations, and reporters cannot be corrected in conventional meta-analysis (Morgan-López et al., 2022; Saavedra et al., 2021).

To address these limitations, Hien and colleagues (2023) conducted Project Harmony, an evaluation of treatments for PTSD + SUD that used meta-analysis of individual patient data (MIPD) to appropriately manage measurement and selection bias in evidence synthesis. Harmonizing and combining data from 4,046 patients who participated in one of 36 PTSD + SUD randomized controlled trials (RCTs), Project Harmony examined the comparative efficacy of eight different PTSD + SUD treatment classes, all relative to treatment as usual (TAU), on latent PTSD symptom severity, alcohol use, and drug use (Hien et al., 2023; Saavedra et al., 2021). Project Harmony also used integrative data analysis to combat measurement bias in PTSD and substance use severity scale scoring across multiple factors (e.g. studies, populations, time, measures, reporters) (Bauer & Hussong, 2009; Curran et al., 2008; Hussong et al., 2021). Propensity score weighting (PSW) was incorporated to systematically identify and place treatment classes on ‘equal footing,’ necessary when ‘mixing-and-matching’ treatment classes from multiple studies for cross-RCT comparisons against a common control arm (Brincks et al., 2018; Morgan-López et al., 2022).

Through greater power and measurement precision, Project Harmony detected effects that likely could never have emerged using conventional meta-analytic methods (Cooper & Patall, 2009; Lubinski & Humphreys, 1996). Specifically, compared to TAU, the combination of trauma-focused behavioral therapy and pharmacotherapy for alcohol use disorder (AUD) yielded the strongest comparative effects for PTSD severity and alcohol use outcomes (Hien et al., 2023). Other treatment classes with large comparative (vs. TAU) effect sizes included SUD-focused pharmacotherapy alone, trauma-focused behavioral therapy alone, and integrated trauma-focused behavioral therapy. Although drug use severity declined overall across treatment classes, no single class produced statistically significant comparative effect sizes relative to TAU under intention-to-treat, but favourable effects for trauma-focused therapies (whether integrated or not) were observed when accounting for treatment dosage as a post-assignment confounder (Morgan-López et al., 2026).

Project Harmony advanced the knowledge of which treatments for PTSD + SUD worked best. However, certain issues make it difficult to offer treatment recommendations for individual patients with PTSD + SUD that are aligned with many patients’ treatment goals. For example, although reductions in PTSD symptom severity and substance consumption (i.e. clinical severity outcomes) are valuable indicators of improvement, they do not fully capture patients’ treatment motivations and perceived gains. Many patients with PTSD and SUD hold additional goals (i.e. recovery-oriented outcomes), such as improving relationships and social functioning, physical health and wellness, and overall quality of life (Andersson et al., 2021; Gressler et al., 2019; Rabinowitz et al., 2025). However, despite the many relevant dimensions of recovery (Üstün et al., 2010; Vogt et al., 2019), no PTSD + SUD evidence synthesis has examined differential effects on recovery-oriented outcomes (Hien et al., 2023; Petrakis & Simpson, 2017; Roberts et al., 2015; Roberts et al., 2022; Simpson et al., 2017; Simpson et al., 2021).

Another limitation in the field is that existing standards for reporting MIPDs do not provide information about patient-level change. Conventional analysis of both single RCT and conventional MIPD data emphasize average differences in changes over time across intervention arms (i.e. standardized mean differences). However, significant group-level comparison tests can obfuscate a non-trivial proportion of patients who get worse (Jensen & Corralejo, 2017; Morgan-López et al., 2022; Roberts et al., 2022; Westen et al., 2004). Failure to appreciate such realities, as conventional analytic approaches inevitably do, can hinder the specificity with which investigators and interventionists can make PTSD + SUD treatment recommendations on a patient-by-patient basis based on the known proportion of patients who improve, failed to improve, or get worse during treatment (Saavedra et al., 2021). A more patient-centred approach involves tests of clinically significant change (CSC). The CSC framework, widely used in the 1970s, (Ogles et al., 2001) emphasizes treatment outcome inferences at the individual patient level. Modernized CSC methodologies adapted for longitudinal designs that accommodate more than two timepoints, account for measurement error that varies across patients and timepoints, and the use of factor scores (versus sum scores) by McNeish and Wolf (2020) have only been applied to single RCT datasets (Morgan-López et al., 2022; Saavedra et al., 2021; Saavedra et al., 2022) and have yet to be applied to MIPD or any PTSD + SUD evidence synthesis. Pairing CSC methodology with the MIPD framework could offer a new, patient-centred paradigm for evaluating comparative treatment effectiveness based on the proportion of patients who improve, fail to improve, or get worse, as opposed to comparative effect sizes.

A final limitation of extant PTSD + SUD meta-analyses is that only two (Hien et al., 2023; Petrakis & Simpson, 2017) evaluated pharmacotherapies for this clinical population, and even so, the number of medications and treated participants included in these reviews were relatively few due to both the state of the science at the time of review and the fact that most PTSD pharmacotherapy trials excluded patients with SUD. Project Harmony included RCTs completed through 2018; since that time, more pharmacotherapy trials for PTSD + SUD have been completed, including RCTs of certain medications not represented in Project Harmony at all. An updated MIPD incorporating these more recent data could bolster the robustness of the primary Project Harmony findings and improve power to detect potential other pharmacotherapy effects that were not able to be tested by Hien and colleagues (Hien et al., 2023).

In this article, we describe the methodology for Project Harmony 2.0, an updated MIPD and virtual clinical trial of PTSD + SUD RCTs funded by the National Institute on Drug Abuse (NIDA). The specific aims of Project Harmony 2.0 are to (1) evaluate comparative effects of behavioral therapy, pharmacotherapy, and combined treatments on recovery-oriented outcomes, (2) estimate individual-level probabilities of improvement and worsening using CSC metrics, and (3) update and extend comparative effectiveness analyses using an expanded MIPD dataset of over 8,000 participants from at least 60 RCTs. Based on the original Project Harmony and other extant evidence syntheses, (Hien et al., 2024; Petrakis & Simpson, 2017; Roberts et al., 2015; Roberts et al., 2022; Simpson et al., 2017; Simpson et al., 2021), we hypothesize that integrated trauma-focused behavioral therapy will yield greater improvements on recovery-oriented outcomes than non-trauma-focused or pharmacologic monotherapies. This updated evidence synthesis is registered in PROSPERO (CRD420251250349; Clinical trial registry not applicable).

2. Methods

2.2. Overview

This study employs a framework combining MIPD, integrative data analysis (IDA), and PSW to integrate raw data collected from recently completed (unpublished) and published clinical trials to address specific gaps in the literature on PTSD + SUD treatment, many of which cannot typically be addressed within any single RCT or even with conventional meta-analytic methods. MIPD involves the analysis of original raw data collected from multiple treatment trials, rather than aggregate-level data. IDA entails pooling data across multiple datasets and applying advanced scale scoring methods (e.g. Moderated Non-Linear Factor Analysis, Item Response Theory) to create comparability in measures across studies assessing the same construct with at least some variation in item content (Bauer & Hussong, 2009; Curran et al., 2008). Advanced analytic methods are then applied to the combined dataset. More details on these methods may be found in reports of the original Project Harmony (Saavedra et al., 2021).

The planned method and design of Project Harmony 2.0 expansion hinges on (1) secondary analysis of the current Project Harmony dataset and (2) data acquisition and harmonization of at least 24 additional PTSD + SUD RCTs completed since 2018 into the original Project Harmony dataset (for a total of at least 60 RCTs). Comparative effectiveness of treatments for PTSD severity and SUD consumption (i.e. clinical severity outcomes) was the primary focus in the original Project Harmony (R01AA025853; MPIs: Morgan-López, A.A., & Hien, D.A.). Project Harmony 2.0. (R01DA061338; MPIs: Hien, D.A., & Morgan-López, A.A.) will first examine (1) comparative effectiveness of treatments for recovery-oriented outcomes and (2) CSC in both clinical severity and recovery-oriented outcomes in the existing Project Harmony dataset. In parallel, Project Harmony 2.0. will involve acquiring new datasets and preparing them for integration with the current 36-study Project Harmony dataset (i.e. data acquisition, data use agreement execution, item-level semantic harmonization, construct scale score estimation). Lastly, Project Harmony 2.0 will conduct comparative effectiveness analyses using the newly combined and harmonized dataset to (1) update MIPD findings for clinical severity outcomes and (2) examine recovery-oriented outcomes across the full dataset, incorporating behavioral and pharmacological treatments that have emerged since 2018. See Table 1 for a visualization of project scope and timeline.

Table 1.

Scope of Project Harmony 2.0 aims over time.

Research Objective	Original Project Harmony Dataset (36 RCTs; N =   4,046)		Expanded Project Harmony 2.0 Dataset (60+ RCTs; N > 8,000)
	Comparative Effectiveness	Clinically Significant Change	Comparative Effectiveness	Clinically Significant Change
PTSD Severity and Substance Consumption Outcomes	Completed under R01AA025853	Years 1 & 2	Years 3 & 4	Beyond scope (future direction)
Recovery Outcomes	Years 1 & 2	Years 2 & 3	Years 3 & 4	Beyond scope (future direction)

Note. PTSD = Posttraumatic stress disorder; RCT = Randomized controlled trial.

2.3. Inclusion criteria and data sources

Following the same protocol from the original Project Harmony (R01AA025853), we conducted an updated systematic review in accordance with the PRISMA systematic review guidelines using seven electronic databases searching from 1995 to the present: Cochrane Depression, Anxiety and Neurosis Group (CCDAN) Clinical Trials Registers Database, MEDLINE, EMBASE, PubMed, Web of Science, PsycINFO and PTSDpubs. Studies must have met the following criteria: (1) evaluated a behavioral and/or pharmacological intervention; (2) targeting either PTSD symptoms, AUD symptoms, and/or drug use disorder symptoms; (3) including measures of PTSD and AUD/drug use disorder outcomes even if explicitly targeting only one of the focal disorders; (4) in an adult sample defined as individuals (a) aged 18–65, (b) with a current diagnosis of SUD (typically DSM-IV abuse or dependence or DSM-5 AUD/drug use disorder), and (c) a current diagnosis of full or subthreshold PTSD (DSM-IV/5). To cast the broadest net for international trials addressing PTSD + SUD, we did not require an RCT design for inclusion. We then used Covidence systematic literature review software to screen full-text articles. Four hundred and twelve articles were screened at the abstract screening phase by study staff and then reviewed by senior study team members. In addition to the 412 identified articles, the study team used the Clinical Trials Registers Database to identify eligible ongoing and recently completed trials. Using a combination of these approaches, we identified 44 completed trials, 20 primary outcomes papers via updated literature search and screening in Covidence, and 28 clinical trials still in progress that we will continue to monitor.

After retrieval and preliminary review, primary investigators of potentially eligible studies were contacted to confirm eligibility. Thus far, we received consent for datasets from 29 completed trials, which will be added to the original Project Harmony dataset of 36 RCTs (n = 4,046), totalling N > 8,000 for the expanded Project Harmony 2.0 dataset. We received no refusals to participate and are in the process of engaging the investigators of all other identified studies to acquire their datasets.

2.4. Scientific advisory board

The original Project Harmony scientific advisory board (SAB) was a diverse group composed of senior PTSD and SUD investigators. The expanded SAB for Project Harmony 2.0 sought to broaden representation by including multidisciplinary, internationally recognized experts in trauma-exposed populations, PTSD + SUD, trauma-focused interventions, clinical research methodologies, and healthcare implementation. Project Harmony 2.0 SAB members attend videoconference meetings approximately 3 times per year to receive progress updates, advise on clinical and methodological issues, provide feedback on interpretation of findings and make recommendations. Measures are taken to minimize conflicts of interest among board members when making recommendations and evaluating evidence.

2.5. Data organization and measure harmonization

Harmonization procedures for latent PTSD severity and alcohol/drug use severity from the original Project Harmony datasets will be similarly applied to the new/incoming datasets. For example, PTSD severity and SUD consumption measures already have an extensive and established cross-study, cross-measure, and cross-reporter harmonization protocol (Hien et al., 2023; Morgan-López et al., 2022; Saavedra et al., 2021). The psychometric item parameters from the established protocols (see the section on moderated nonlinear factor analysis below) will be used to ‘score’ (i.e. apply existing item parameters to new item-level data to estimate scale scores on the same metric) the PTSD symptom severity and substance consumption measures from the new datasets without the need to re-estimate new item parameters. Specifically, clinical severity outcome item parameters from the original Project Harmony will be retained and applied to new datasets unless tests of model fit or differential item functioning indicate that re-estimation is required. For recovery-oriented outcomes, new psychometric work to estimate and establish item parameters will be conducted in the existing Project Harmony dataset and then these parameters will be used to score recovery-oriented outcomes in the new datasets, with the likely model reflective of recent MNLFA approaches to modelling latent recovery outcomes (Espinosa et al., 2026).

Key measures are organized prior to item harmonization. Logical/semantic harmonization of items involves the assessment and combining of face-valid indicators (i.e. recoding and/or renaming to a common variable standard in the dataset) reflecting similar item content from different measures across multiple studies per recommendations from our own work (Bauer & Hussong, 2009; McDaniel et al., 2023). Although some measures can be harmonized across studies and coded with relative ease (e.g. gender, race, and ethnicity), other factors may need to be reduced to a lowest common denominator (e.g. a yes/no response) to place items on the same metric (Hien et al., 2023). The adequacy of these measurement decisions (i.e. measurement [non]invariance) is tested later under moderated nonlinear factor analysis models, the primary tool in modern IDA studies (Morgan-López et al., 2022; Saavedra et al., 2021). Harmonization of treatment categories, semantic harmonization of key covariates (e.g. demographics, pre-treatment psychiatric status) and outcome measures (i.e. PTSD, SUD) will follow the procedures in Saavedra et al. (2021) and the Supplemental Material for Hien et al. (2023) with brief descriptions below.

To facilitate error reduction and systematic data acquisition, as well as to achieve accessibility, interoperability and re-usability (i.e. FAIRness) (Haering and Van Zuiden, 2026; O’Neil et al., 2025 ; Wilkinson et al., 2016) in our dataset, we have implemented a standardized common data element (CDE) naming framework to ensure consistent variable labels across studies. Naming conventions have been aligned with the established schema from the original Project Harmony.

To facilitate the integration of new datasets to Project Harmony, we are building a transparent and reproducible pipeline using the targets package in RStudio (Landau, 2021; Posit PBC, 2025) which runs on the open-source statistical programming language R. (R Core Team, 2024). The pipeline consists of functions that perform a specialized task using base R and core tidyverse packages (Wickham et al., 2019). The key tasks in the pipeline are summarized below:

1. Load the raw dataset and codebooks.

2. Extract variables that are included in Project Harmony.

3. Restructure data into long format with columns for variables, timepoints, and values.

4. Standardize the variable coding and apply instrument-level harmonization rules.

5. Apply the construct-level harmonization rules and CDE variable names.

Between some of these steps, a variety of outputs are generated that facilitate quality control. For example, tables including the extracted item, its description from the codebook, and the CDE variable names facilitate confirmation that the correct items are being extracted; plots showing the distribution of item values across studies facilitate standardization of values and of missing data indicators. The pipeline is being developed and tested using the studies from the original Project Harmony so that its output can be checked with the previously harmonized the original Project Harmony dataset. Once the pipeline is completed and passes all quality control checks, new datasets can be imported into the pipeline and processed using the same functions. In addition to incorporating new datasets, additional functions designed to extract, restructure, standardize, and harmonize new variables (i.e. those not included in the original Project Harmony) can be integrated into the pipeline.

2.6. Measures

We will harmonize outcome measures from the new studies into commensurate items across datasets for subsequent scoring and/or estimation of latent variable scale scores for PTSD + SUD severity and recovery-related outcome (see IDA/MNLFA Models section below). Covariates (some of which will serve as moderators depending on the analysis) will be included in propensity score models for addressing cross-study ‘mixing-and-matching’ of treatment classes across RCTs with different covariate distributions (despite within-study randomization; see the PTSD and SUD Latent Variable Indicators section below).

(Harmonized) Covariates/Moderators. Harmonized metrics will be generally consistent with previously vetted harmonization decisions made under the original Project Harmony (Hien et al., 2023; Saavedra et al., 2021). Age will be retained in its natural metric. Sex/gender will be recoded to a common indicator (female/woman =   0, male/man ; it is possible that post-2018 studies may have non-binary as a reporting option for participants, but this was not the case for the earlier trials. Race and ethnicity data will be harmonized to dummy variable indicators for White, Black, Hispanic, Asian, and Other (many later trials in the original Project Harmony allowed the reporting of more than one category). Education level will be harmonized to a three-category dummy variable: high school or less, some college, or college degree or higher. Marital status will be harmonized to reflect a common indicator of married (1) or else all other categories (0). Population type will be coded as either civilian, veteran, or incarcerated. Treatment dosage will be harmonized across studies to reflect a proportion of intended dosage as the dosage received (number of behavioral treatment sessions attended and/or number of medication doses taken) divided by the maximum possible intervention dosage. Depression will be harmonized into a binary variable indicating either a formal diagnosis of major depression (e.g. CES-D) or a clinical cutoff for severe depression for the continuous measure of depression that was available (e.g. Beck Depression Inventory). Finally, concomitant medications will be harmonized into a binary variable indicating the use of any non-study psychiatric medication prior to entry into the trial.

Structuring of Chronological Time. Two time variables were/will be created to reflect two piecewise linear trajectories (Roberts et al., 2022): an in-treatment outcome trajectory capturing the model-estimated change in the outcome variable from baseline through the end-of-treatment (Time₁), with many studies in the original Project Harmony having a mid-treatment assessment point, and a post-treatment outcome trajectory capturing the model-estimated change in the outcome variable from the end-of-treatment through 12-month follow-up (Time₂).

Clinical Trial Treatment Information Form. This form includes information about the treatment components implemented in the specific clinical trial. It will include categorization of the treatment approach and a description of specific and non-specific therapeutic components. Information about treatment modality (e.g. individual, group, remote) and pharmacological treatment components, including dosage and duration, are captured here as well.

Coding of Treatment Classifications. Treatment conditions, based on groupings of each within-study treatment arm, have been defined and coded based on whether the treatment was: (1) trauma-focused, (2) integrated PTSD + SUD treatment, (3) a pharmacotherapy that targeted PTSD (e.g. sertraline, paroxetine), (4) a pharmacotherapy that targeted SUD (e.g. methadone, naltrexone), (5) solely a behavioral SUD treatment, (6) a pharmacotherapy, or (7) a non-manualized, community-based treatment (i.e. treatment-as-usual). These groupings are not mutually exclusive and were derived through an expert consensus process, described elsewhere (Hien et al., 2023).

PTSD and SUD Latent Variable Indicators. Prior to MNLFA scale score estimation of underlying PTSD and SUD, items/symptoms for each construct have been/will be recoded and renamed to initially be on the same metric (i.e. item harmonization). As noted by Bauer and Hussong (Bauer & Hussong, 2009), harmonizing items across studies is a necessary but insufficient step to ensure measurement comparability of the construct; this is what is done under the IDA/MNLFA framework described below and as has been described in published psychometric work generated by the original Project Harmony (Morgan-López et al., 2023).

PTSD Items/Symptoms. A total of 42 PTSD indicators (21 symptoms from a clinical interview, 21 self-report symptoms) were/will be harmonized across the studies that have PTSD item-level data and will formulate the indicators of a 42-item latent PTSD construct estimated under MNLFA; 16 of the 28 ‘raw’ data/item-level studies in the original Project Harmony had both a clinical interview measure and a self-report measure. The primary clinical interview measures available across studies were (1) the Clinician Administered PTSD Scale for DSM-IV (CAPS-IV; Blake et al., 1995), (2) the CAPS-5 (Weathers et al., 2013) and (3) the Post-traumatic Stress Scale-Interview (Foa et al., 1993). Each assessment system has a primary method for converting frequency and intensity items into binary indicators of symptom presence/absence. A similar process was/will be undertaken for self-report measures. The self-report measures available across studies were (1) the PTSD Checklist for DSM-IV (PCL-IV; Blake et al., 1995), (2) the PTSD Checklist for DSM-5 (PCL-5; Weathers et al., 2013), (3) the Post-traumatic Stress Scale-Self Report for DSM-IV (Foa et al.,1993), (4) the Impact of Events Scale for DSM-IV (IES; Weiss and Marmar) and the Post-traumatic Diagnostic Scale for DSM-IV (PDS) (Foa et al., 1997). See Hien et al., 2023a and 2023b papers and supplementary materials for specific details on all harmonization procedures. Scale scoring under MNLFA, whether there are differences in the measurement parameters for each symptom across assessment system and/or other factors (i.e. measurement non-invariance/differential item functioning [DIF]), is addressed later in the Approach section on latent variable scale score estimation under MNLFA.

Drug Use Items. In previous analyses of Project Harmony, binary indicators of any use in the past 30 days of the following substances were harmonized across three assessment systems: Timeline Follow-back (TLFB) (Sobell and Sobell, 1992); the Addiction Severity Index (ASI; McLellan et al., 1980) and the Substance Use Inventory (SUI) (Weiss et al., 1995). However, in order to balance both (1) coverage of use beyond any use against (2) skew in using continuous indicators, we propose harmonizing 30-day use measures into ordered categorical indicators with 4 categories: 0 days, 1–10 days, 11–20 days, and 21 + days, as recommended by Burlew et al. (2025). Here, variability in the amount of substance use for each substance is preserved, while handling skew more easily by using estimators that can handle ordered categories (e.g. robust maximum likelihood, weighted least squares) in advanced factor analytic/item response theory models. The following indicators will be used to support a five-indicator latent substance use variable: cocaine, sedatives, heroin and non-heroin opioids, other psychostimulants, and hallucinogens.

Alcohol Use Items. Factor analysis for ordered categorical outcomes (as described in Hien et al., 2023) suggested that alcohol use severity was a separate latent factor from other substance use severity. A latent alcohol use variable consisted of two indicators: number of days of alcohol use in the past 30 days and any alcohol use to intoxication in the past 30 days. Two-indicator latent variables can be supported in factor analysis scoring, so long as the two indicators are highly correlated, Morgan-López et al. (2022), as are 30-day use and use to intoxication; the 30-day alcohol use indicator will be converted to ordered categorical using the procedure described in Burlew et al. (2025).

Recovery-Oriented Outcomes. As part of pre-analytic activities, we will harmonize PTSD + SUD recovery indicators across multiple measures that focus on psychosocial functioning, substance use consequences, and other recovery-related constructs. These focal constructs will be derived from the consequences subdimensions of the Addiction Severity Index (McLellan et al., 1980), which provide valuable data about functioning related to five domains: (1) Employment/Support (4 items); recent MNLFA work on the ASI has shown a transportation sub-dimension that is part of the main employment factor, (Espinosa et al., 2026) (2) Family/Social Relationships (3 items); (3) Medical Consequences (3 items); (4) Legal Consequences (6 items); and (5) Psychiatric Status (11 items). Approximately half of the trials in the original Project Harmony dataset, as well as from the eligible new studies (based on a survey distributed to principal investigators of the post-2018 studies identified in our search), administered the ASI. ASI consequences item parameters/scale scores will be estimated using IDA/MNLFA (see below); in later outcomes analysis, missing consequences scale scores will be estimated via multiple imputation, on the conceptual justification that missing observations on consequences would be predictable by data that are available (e.g. all consumption measures, demographics, pre-treatment psychiatric status).

Additional recovery-oriented measures reported in the survey distributed to principal investigators of the new post-2018 RCTs that are not widely available in the original Project Harmony dataset include the Medical Outcomes Survey 12- and 36-item versions (SF-12/SF-36), which measure functioning related to health, work/daily activities, and relationships, and quality of life assessments (e.g. Quality of Life Inventory, World Health Organization Quality of Life Scale, Quality of Life, Enjoyment, and Satisfaction Questionnaire). With these limited measures, we will conduct smaller-scale harmonization analyses restricted to studies that include the relevant measure, as we previously did with a four-study MIPD/IDA subset examining baseline cannabis use as a treatment moderator (Hill et al., 2024).

2.7. Data analytic plan

Analytic Model Extensions. The general analysis frameworks to be used will be (1) PSW for balancing covariates across treatment classes (which we have used within and outside of Project Harmony; Hien et al., 2023; McDaniel et al., 2025; Morgan-López et al., 2022; Saavedra et al., 2021), (2) IDA under the MNLFA framework for scale score estimation (Morgan-López et al., 2020; Morgan-López et al., 2022; Morgan-López et al., 2024; Morgan-López et al., 2023; Ruglass et al., 2020) in the presence of differential item functioning, and (3) the multilevel modelling framework, used in a conventional manner for treatment outcome analysis (Hien et al., 2023; McDaniel et al., 2025; Morgan-López et al., 2020) and as adapted for analysis of individual-level CSC (Morgan-López et al., 2025; Morgan-López et al., 2023; Saavedra et al., 2021; Saavedra et al., 2022), PSW, and IDA/MNLFA. Conventional multilevel modelling will be used for all Aims, while multilevel models adapted for CSC will be used for Aim 2 analyses.

As noted earlier, application of advanced CSC methodologies has only been to single datasets (Morgan-López et al., 2025; Morgan-López et al., 2023; Saavedra et al., 2021; Saavedra et al., 2022) and will now be applied to the original Project Harmony dataset for latent PTSD, AUD severity, SUD severity and recovery. This will illustrate a new paradigm for understanding meta-analytic effects of any type (conventional meta-analysis or ‘raw data’ meta-analysis) that does not rely on averaged effects across treatment classes. Aim 3 analyses will use an expanded PTSD + SUD RCT database adding at least 24 additional studies to the original Project Harmony dataset (for a minimum of 8,000 patients) to evaluate comparative effects of PTSD + SUD treatment types on recovery.

Propensity Score Weighting. ‘Mixing-and-matching’ of treatment classes across studies may introduce potential confounding within and across treatment classes (Brincks et al., 2018; Morgan-López et al., 2022; Saavedra et al., 2021). To address this, propensity score weighting will be used to balance covariates that influence both the probability of selection into two or more non-experimental groups and intervention outcomes (Harder et al., 2010; Rosenbaum & Rubin, 1983). In a manner similar to the original Project Harmony (Hien et al., 2023), weights for the probabilities of being in each treatment class (compared with TAU as the reference) will be estimated using a random effects multinomial logit model, with the treatment classes as a nominal variable and the covariates as predictors of treatment class assignment, accounting for study-level clustering, using SAS Proc GLIMMIX. Predicted probabilities of treatment assignment will be conditional on the covariates listed in the Measures section. These predicted probabilities are then converted to inverse-probability-of-treatment weights (IPTWs) and added as weights in outcome analyses to produce causal estimates of the intervention effects of interest (Cook et al., 2020; Harder et al., 2010; McCaffrey et al., 2004) so long as covariates are balanced to a standardized mean difference (i.e. d) of <   |.10| (i.e. mimicking a multi-arm, multisite RCT).

For analyses involving the original Project Harmony dataset only, the original PSWs published in Hien et al., 2023; Brincks et al., 2018) will be used for CSC-related consumption outcomes because these weights already have baseline PTSD/AUD/SUD severity as the primary predictor of treatment assignment; baseline levels of the primary outcome (Cook et al., 2020) have been touted as the best singular covariate for PSWs in quasi-experiments (Cook et al., 2008; Cook et al., 2020). However, for recovery-oriented outcomes, new PSWs will be estimated that contain baseline levels of the recovery-oriented outcome scale scores, after scale score estimates and standard errors of measurement are estimated and output from IDA/MNLFA models (see next section). Separate PSWs will be estimated to achieve cross-study covariate balance for all outcomes analysis that involve the newly added studies to the current 36-study Project Harmony dataset.

IDA/MNLFA Models. While the item harmonization steps described above critically enhance harmonization validity (e.g. face validity), these procedures are necessary but not sufficient to guarantee equivalence of the construct across studies (or populations or time points, for that matter); see, for example Beaujean & Osterlind (2008), once items are converted to scale scores. Scale scores estimated under IDA/MNLFA properly weight items within a construct and address the extent to which this weighting varies across studies, measures, time, populations, or reporters (i.e. DIF). MNLFA, unlike other models (e.g. Item Response Theory), can more flexibly handle simultaneous DIF from multiple predictors and DIF for continuous variables (e.g. age), particularly variation in factor loadings. MNLFA models are extensions of the general structural equation model that can be used for scale score estimation with a mix of observed indicator types (e.g. continuous, binary, or unordered categories). The model formulation (for binary items as an example) measured repeatedly over time would be:

$${\mu }_{\mathrm{jti}}={\displaystyle \frac{1}{1+e^{\left[-\left({\nu }_{\mathrm{jti}}+{\lambda }_{\mathrm{jti}}{\eta }_{\mathrm{ti}}\right)\right]}}}$$ (1)

Where μ_jti is the conditional probability of a response of ‘1’ for item j (e.g. presence/absence of PTSD symptom j) at time t for person i; η_ti is the latent trait (e.g. PTSD severity score) that underlies the item responses at time t for person i; ν_jti is a threshold for item j, which is the value of the latent severity score η necessary for a 0.5 probability of endorsing the item (note that, for item j, ν can vary either across time t, person i, or both [based on an item parameter moderator variable] if DIF is present; Bauer, 2017) and λ_jti is a factor loading for item j (also potentially conditional on time or item parameter moderators). MNLFA does not require responses to all items for the estimation of latent trait scores; non-overlapping items can be treated as missing data under the missing-at-random assumption (Shafer & Graham, 2002).

For IDA/MNLFA analyses involving the original Project Harmony dataset only, the MNLFA scale scores in the current dataset for PTSD severity will be used. MNLFA scale scores for AUD and SUD severity will need to be re-estimated with the newly proposed ordered categorical consumption items described in the Measures section (as opposed to the binary consumption items used in previous Project Harmony consumption scale score estimation); the resulting item parameters from this analysis will then be used to score the PTSD and consumption items from the 24 new datasets. Scale scores for the proposed ASI recovery subdimensions will be newly estimated across the original and expanded Project Harmony datasets.

Recovery-oriented outcomes will first be assessed for dimensionality; dimensionality of the substance use related areas of functioning measures will be assessed using the ASI configural structure identified in Alterman et al. (2007) under nonlinear factor analysis using mean-and-variance-adjusted weighted least squares (i.e. Muthén ‘WLSMV’ estimation: Muthén & Muthén, 2017). After dimensionality is established, MNLFA models will be used to estimate and output scale scores for each construct, while incorporating DIF by allowing item parameters to vary (if necessary based on statistical tests of item parameter differences) across all covariates and assessment waves. For final scoring models, MNLFA scores (and MNLFA standard errors of measurement) will be estimated with the scale set to a mean of 0 and variance of 1 at baseline so that changes over time in outcomes are in standard deviation units, as would be the case in a conventional meta-analysis that uses Cohen’s d, then output to an outcomes analysis data set, in a manner consistent with the structure in Espinosa et al., (2026). As a result, changes over time in outcomes can be modeled after we remove bias in measurement across time, studies, and populations. Establishment of the measurement structure for cross-study estimates of PTSD severity has already been established (Hien et al., 2023; Morgan-López et al., 2022). Dimensionality and structure for the harmonized consumption measures will be re-checked, given the change to ordered categorical item structure for past 30 d use from binary (Burlew et al., 2025), but it is anticipated that the general configural structure (i.e. one dimension for SUD items, separate dimension for AUD items) will hold.

2.8. Comparative effectiveness and moderator analyses

2.8.1. Aims 1 and 3

Aim 1 of Project Harmony 2.0 is to conduct new analyses on recovery-oriented outcomes to evaluate comparative effects of behavioral, pharmacological, and combination treatments for PTSD + SUD. Aim 3 of this study is to update and extend MIPD comparative effectiveness analysis of PTSD severity, SUD/AUD consumption, and recovery-oriented outcomes to an expanded (at least 60 trials, N > 8,000) participant Project Harmony dataset. Accordingly, the primary analysis model for comparative effectiveness outcome models (i.e. under Aims 1 and 3) will be an IPTW-weighted, three-level linear mixed model (using SAS Proc MIXED) to account for (1) clustering of repeated observations within participants and (2) participants clustered within studies (see Hien et al., 2023; Saavedra et al., 2021). At the within-individual level, we will estimate three random effects as growth parameters within (a_psi):

$$Y_{\mathrm{sit}}={\pi }_{0\mathrm{si}}+\sum _{p=1}^P{\pi }_{\mathrm{psi}}{a}_{\mathrm{psi}}+e_{\mathrm{sit}}$$ (2)

a random intercept (π_0si; baseline on the outcome), the 1st random slope (π_1si; in-treatment change on the outcome), and a 2nd random slope (π_2si; post-treatment change in the outcome). These values will vary across individual i within study s. The between-individual model is:

$${\pi }_{\mathrm{psi}}={\beta }_{p0s}+\sum _{q=1}^Q{\beta }_{\mathrm{psq}}{X}_{\mathrm{qsi}}+r_{\mathrm{psi}}$$ (3)

β_psq is a matrix of coefficients corresponding to baseline differences (there should be none after IPTW weighting), as well as in- and post-treatment effects of each treatment class on outcomes (compared with TAU as the reference). The between-study model is:

$${\beta }_{\mathrm{pqs}}={\beta }_{00}+\sum _{q=1}^Q{\gamma }_{\mathrm{pq}}{W}_{\mathrm{qs}}+u_{\mathrm{pqs}}$$ (4)

W_qs will be ‘empty’ in that study-level covariates will not be entered directly into the model, because study-level covariate impact will be included in the IPTW weights. Variance components within u_pqs capture (residual) study-level heterogeneity, a requirement in MIPD reporting (Stewart et al., 2015).

Aim 1 comparative effectiveness analyses will focus on recovery-related outcomes only, since consumption outcomes were already completed for the original 36-study Project Harmony dataset; Aim 3 comparative effectiveness analyses will focus on all outcomes with the to-be-expanded dataset.

2.8.2. Sub-aims 1.2 and 3.2

We will test for PTSD severity as a mediator of treatment effects on SUD and recovery-oriented outcomes across treatment models.

Causal Mediation. The potential outcomes framework for mediation analysis (MacKinnon et al., 2020; Miočević et al., 2018) will be used, which we have applied elsewhere (Hien et al., 2025; Saavedra et al., 2023), such that a series of comparison-specific estimates for the effect of changes in the mediator (i.e. PTSD) on changes in the outcome (e.g. SUD severity, recovery-oriented outcomes) will be altered by the treatment classes(s) themselves compared with TAU (i.e. treatment x mediator interaction effects on the outcomes). IPTWs will be included in outcomes analysis to nullify the relation between ‘mixed-and-matched’ treatment classes and measured covariates.

MNLFA scale scores will be modeled in parallel growth processes, with random intercepts and slopes over time for both the mediator and outcomes under the longitudinal mediation framework (Cheong et al., 2003; Jagers et al., 2007). The key effects of interest in general will be the product of the paths from the treatment dummy codes to the random slope for PTSD severity as the common mediator across studies (i.e. the ‘a’ path) and the path from the random slope for mediator to the random slope on the outcome (i.e. the ‘b’ path). However, under the potential outcomes mediation framework, there are two distinct mediation effects: (1) the pure natural indirect effect [PNIE], which, from a counterfactual perspective, is the mediation effect in the control/reference condition (TAU), and (2) the total natural indirect effect [TNIE], which is the mediation effect in the focal treatment condition(s). The differences between the PNIE (i.e. TAU) and the TNIEs (for all other active treatments) will be dependent on whether the treatment × mediator interaction effects on the outcomes are non-zero. PNIE and TNIE estimates and confidence intervals will be estimated using the procedures outlined in Miočević et al. (2018). Aim 1.2 comparative effectiveness analyses will focus on recovery-related outcomes with PTSD as the mediator; Aim 3 comparative effectiveness analyses will focus on both consumption and recovery-oriented outcomes, with PTSD as the mediator.

2.9. Aim 2

Project Harmony 2.0 also seeks to enhance MIPD/IDA methodologies to incorporate individual-level inferences regarding the proportion of patients with clinically significant improvement or worsening across treatment classes. The original methodology for assessing CSC (Jacobson & Truax, 1991) suffers from significant limitations (Brouwer et al., 2013; Jabrayilov et al., 2016; Speer & Greenbaum, 1995) despite its continuing popularity in intervention science. Morgan-López and colleagues (Morgan-López et al., 2022) developed a two-stage, measurement error-corrected form of CSC estimation, combining (1) MNLFA scale scoring (stage 1) and (2) measurement error-corrected multilevel modelling (stage 2). Presuming that MNLFA scale scores have been estimated in Stage 1 (see IDA/MNLFA Models section above), we present the advanced CSC model. The Level-1 model for measurement uncertainty/scale score unreliability is:

$${\theta }_{\mathrm{it}}={\hat{\theta }}_{\mathrm{it}}+s_{\mathrm{it}},$$ (5)

where ${\hat{\theta }}_{\mathrm{it}}$, the unknown true score on the outcome is a function of the estimate in the MNLFA scale score dataset and the s_it measurement error term, with heteroskedastic measurement errors from MNLFA output; Morgan-López et al. (2025) simplifies modelling this heteroskedasticity in scale score uncertainty by fixing the level-1 residual variance to 1 and weighting each observation by the inverse of the square of the standard error of measurement from the MNLFA as is done with uncertainty for effect sizes in meta-analysis (Cheung, 2008; McNeish, 2016). The repeated measures model for, typically estimated at Level-1, is estimated at Level-2:

$${\hat{\theta }}_{\mathrm{it}}={\beta }_{0i}+{\beta }_{1i}\left(\mathrm{Tim}{e}_{\mathrm{it}}\right)+r_{\mathrm{it}}$$ (6)

The between-individual model for the random effects $\beta$_0i and $\beta$_1i is estimated at Level-3:

$${\beta }_{0i}=g_{0i}$$ (7)

$${\beta }_{1i}=g_{1i}$$ (8)

Typically, growth models will have a fixed effect component in Equations 4 and 5 that represent (when no covariates are present) the average intercept and slope ($\beta$₀₀ and $\beta$₁₀), respectively. However, in estimating CSC parameters under the measurement error-corrected MLM model, fixed effects are set to 0 so that g_0i and g_1i are not interpreted as deviations from the mean intercept and slope, but as their ‘raw’ intercept and slope values. The patient-specific tests of significance for g₁ across all i patients test whether each individual patients’ change over time differed significantly from 0 (i.e. an individual-level significance test). On the basis of the direction of each patient’s change, and whether their change is significant at p < .20 (Wise, 2004), each patient will be classified into one of four RCI categories: significant or non-significant improvement or else significant or non-significant worsening. These nominal classifications will then serve as outcomes in a mixed multinomial logistic regression (with study-level random intercepts and propensity score weights) with differences in the percentages of patients demonstrating (non)significant improvement or worsening on across all treatment conditions (compared to TAU); patient type × treatment class interactions will also be included, to better identify which treatments should be advocated or avoided for certain patient types.

Sample Size and Power. Based on this search, our proposed Aim 3 updated IDA/MIPD will more than double what is already considered the largest representation of pharmacological and combined PTSD + SUD treatments to date. The original Project Harmony dataset included 27 behavioral trials (n = 3,206), seven pharmacological trials (n = 360), and four combination trials (n = 352), of which 18.1% included a pharmacological treatment. The newly identified datasets will add at least 24 behavioral trials (n = 3,840), six more pharmacological trials (n = 296), and five more combination trials (n = 645), expanding the number of participants in pharmacological treatments to 1,536 patients across 24 trials.

Statistical power was estimated using Monte Carlo simulations (Muthén & Muthén, 2002) of CSC inferences at the individual-level (separately for PTSD and SUD outcomes) based on 250 simulated samples of 4,046 patients generated based on the fixed and random effect parameter estimates in the Supplemental Materials in Hien et al. (2023). The focus was on power to detect differences in the percentages of patients with statistically significant improvement across all treatment classes (compared to TAU), analyzing the simulated data using the method described in Morgan-López et al., (2022) and Saavedra et al. (2022) For PTSD severity, power was 98.4% or greater for all therapeutic combinations except SUD- and AUD-focused behavioral therapies (power = .34) and integrated/non-trauma-focused therapies (power = .06). For substance use severity (relevant for detecting CSC effects on SUD outcomes), power exceeded 80% (in the desired direction) only for SUD- and AUD-targeting medications, with the next highest effect that was superior to TAU being trauma-focused/integrated therapies (power = .57); power was 1.0 for PTSD medications but for detecting effects showing a lower percentage of patients with significant improvement compared to TAU.

2.10. Data quality assurance and missing data procedures

Missing Data. Models will be estimated under full information ML and supplemented by multilevel multiple imputation for missing data using BLIMP 3.0 (Keller & Enders, 2021). BLIMP 3.0 handles multilevel missingness for up to three-levels of aggregation and uses fully conditional specification for missingness on variables of multiple types (i.e. nominal, ordered categorical, continuous). This will be done for key predictors and outcomes similar to original Project Harmony analyses involving multilevel multiple imputation (Hien et al., 2023; Qu & Lipkovich, 2009). In assessing sensitivity to the missing-at-random assumption (i.e. missing-not-at-random), we may explore latent pattern mixture models which we have used elsewhere (Hien et al., 2012; Morgan-López et al., 2013; Morgan-López et al., 2014) for non-ignorable missingness.

2.11. Analysis and dissemination timeline

Analyses will proceed in a staged manner across the project period, beginning with Year 1 pre-analytic activities (data harmonization, validation, and protocol registration) and initial comparative and CSC analyses using the existing Project Harmony dataset, followed by expanded analyses of moderators, mediators, and recovery-oriented outcomes in Years 2–4 as additional trials are incorporated. Dissemination will occur throughout the project and include submission of multiple peer-reviewed manuscripts, conference presentations at major PTSD and addiction meetings, and deposit of all final publications in PubMed Central.

Harmonized datasets, analytic code, and documentation will be shared through NIH-designated repositories (e.g. NIDA/NIMH Data Archive) with persistent DOIs, and public-facing project materials, protocols, and updates will be archived and maintained through an Open Science Framework (OSF) repository to support transparency and reuse. Efforts to ensure FAIR and FACT responsible data sharing and use will be employed (Hien et al., 2024; O’Neil et al., 2025). Together, these activities are designed to promote reproducibility, facilitate secondary analyses, and accelerate clinical and policy translation of integrated PTSD–SUD treatment evidence.

3. Ethics and consent

The Rutgers University Institutional Review Board (IRB) approved changes to the original Project Harmony IRB request (Pro2018001200; originally approved in 2018) on June 9, 2025. These amendments aligned the IRB protocol with revisions made to the current award, which was submitted to NIDA and awarded to Rutgers University as the Prime awardee. The study is designated as Exempt with appropriate de-identification of 17 categories of participant PHI.

Participating Principal Investigators (PPIs) who contributed datasets from their clinical trial either received approval from their IRB to allow sharing of de-identified individual patient data related to this study or had sufficient justification in the original consents to share de-identified data with IRB exemption. PPIs were informed of the purpose of the study and the data needs required for successful inclusion and execution (e.g. variable and data requests as well as data use agreements). After the project was funded, PPIs were contacted again and given more specific information about the preferred item-level variables along with technical assistance and support for acquiring individual institution IRB approval. When necessary, PPIs were also provided with support in drafting and executing one-way data use agreements. Several webinars were hosted to offer additional support and guidance for PPIs and the needs of their specific studies and data sharing matters. This data science project was conducted in full adherence to the ethical principles and requirements for medical research involving human data sets, as set forth in the World Medical Association Declaration of Helsinki.

4. Discussion

PTSD and SUDs frequently co-occur and are associated with distress and impairment as well as significant public health burden. Several treatment approaches exist for individuals with comorbid PTSD + SUD, yet evidence syntheses have been critiqued for several methodological limitations (Hien et al., 2023; Roberts et al., 2022): study-level measurement bias cannot be corrected in conventional meta-analytic methods, covariates are not always appropriately managed, and few syntheses estimate comparative effect sizes for distinct PTSD + SUD treatment approaches against a common control. Additionally, SUD and PTSD + SUD trials have historically emphasized reduction in clinical severity outcomes (i.e. PTSD symptom severity and substance consumption), with far less attention to recovery-oriented outcomes despite the importance of improvements in relationships, self-efficacy, and quality of life to patients seeking PTSD + SUD treatment. The current study will deliver an updated MIPD of PTSD + SUD RCTs incorporating recovery-oriented outcomes and advanced quantitative methods to inform PTSD + SUD research and clinical decision-making.

Although Project Harmony 2.0 could meaningfully advance the PTSD + SUD treatment literature, future findings will be contextualized by certain limitations. For example, although a strength of MIPDs is that analyses are conducted on a larger, combined sample, MIPD samples are only as representative as the collective set of RCTs included for harmonization. Accordingly, this study may be underpowered to examine many forms of intersectionality and their impact on treatment outcomes. Although Project Harmony 2.0 will include a greater number of pharmacotherapy trials than any PTSD + SUD evidence synthesis to date, analyses will be shaped by the reality that there is a larger body of evidence (and contributing MIPD sample sizes) for behavioral interventions. Although alcohol and drug consumption will be treated as distinct outcomes, this study may be underpowered to isolate SUD outcomes to specific substances other than alcohol. Recovery-oriented outcomes were not uniformly assessed across all trials, and some recovery domains may be represented in only a subset of studies, which may limit precision or power for certain functioning dimensions. Finally, while Project Harmony 2.0 will be the first to incorporate modern CSC methodologies into MIPD by applying these methods to the original (k = 36 RCTs; N = 4,046) dataset, CSC analyses on the larger dataset to be updated as part of Project Harmony 2.0 activities are beyond the scope of this study and will remain an important future research direction.

5. Conclusion

Limitations notwithstanding, Project Harmony 2.0 will address limitations of the PTSD + SUD treatment literature by combining IDA and MNLFA (to adjust for PTSD and SUD measurement error), PSW (to mitigate against confounding that can result when ‘mixing and matching’ treatment arms across multiple trials), and contemporary approaches to quantifying CSC (instead of only examining mean group differences) with MIPD to test the comparative effectiveness of pharmacological, behavioral, and combination treatments for PTSD + SUD on clinical severity and recovery-oriented outcomes. The novel integration of CSC into a patient-centred MIPD will yield results that, instead of comparative effectiveness of treatments on average, will give providers information on optimal patient/treatment matches – through greater precision in the expected probabilities for meaningful improvement or worsening on PTSD severity, SUD severity, and recovery-oriented outcomes. If successful, this project will yield a synthesis of evidence in short order for the benefits of current PTSD + SUD treatments on clinical severity and recovery-oriented outcomes.

Declaration of Generative AI Use

The authors report generative AI was not used in their research or preparation of this manuscript.

Acknowledgements

This manuscript contains recycled material that was included in the funding application to conduct the described clinical trial, in accordance with guidance from the National Science Foundation-funded Text Recycling Project (https://textrecycling.org). We wish to thank our Scientific Advisory Board members, Kathleen Brady, Shelly Greenfield, Jessica Hamblen, Nancy Kassam-Adams, Debra Kaysen, Shannon Kehle-Forbes, Katherine Mills, Ingo Schäfer, Tracy Simpson, and Anka Vujanovic. Consortium on Addictions, Stress and Trauma (CAST) members Sudie Back, Steven Batki, Malcolm Battersby, Matthew Boden, Kathleen Brady, Deborah Brief, Christy Capone, Kathleen Chard, Joan Cook, Thomas Ehring, Paul Emmelkamp, Edna Foa, Linda Frisman, Jessica Hamblen, Moira Haller, Denise Hien, Debra Kaysen, Shannon Kehle-Forbes, Annett Lotzin, Asa Magnusson, Meghan McDevitt-Murphy, Mark McGovern, Katherine Mills, Lisa Najavits, Sonya Norman, David Oslin, Jessica Peirce, Beatrice Perez-Dandieu, Ismene Petrakis, M. Zachary Rosenthal, Lesia Ruglass, Michael Saladin, Claudia Sannibale, Ingo Schäfer, Rebecca Schacht, Tracy Simpson, Cynthia Stappenbeck, Susan Sonne, Anka Vujanovic, Debora Van Dam, and Caron Zlotnick, as well as trial principal investigators who agreed to contribute data to Project Harmony: Michele Bedard-Gilligan, Erica Eaton, Amani Elbarazi, Christine Hahn, Laurien Hakvoort, Ahmed Hassan, Amber Jarnecke, Rebecca Schacht, Kyle Possemato, Katherine Watkins, Laura Watkins, Kelly Peck, Anna Persson, Jordan Davis, Eric Pedersen, Shannon Blakey, Leanne Hides, Annett Lotzin, and Vanessa Somohano. We also thank Santiago Papini, Jordan Gette, Emmanuel Thomas, Steven Curto, and Tasha Bulgin for their contributions to this work.

Funding Statement

Research described in this publication is supported by the National Institute on Drug Abuse of the National Institutes of Health under Award Number R01DA061338, which builds on research supported by the National Institute on Alcohol Abuse and Alcoholism under Award Number R01AA025853. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Disclosure statement

S.M.B. receives book royalties from APA Publishing. T.K.K., S.E.B., and K.M. receive book royalties for the Concurrent Treatment of PTSD and Substance Use Disorder Using Prolonged Exposure therapist manual and patient workbook. I.P. has received in kind (medications) support for research studies from BioXcel Therapeutics and Indivior, has provided consultation for Eli Lilly and Company and Sophrosyne, is Editor in Chief for American Journal of Addictions, and has received textbook royalties from MacGraw/Hill. D.K. receives book royalties and workshop honoraria related to Cognitive Processing Therapy and serves on the Scientific Advisory Board for NEMA Health. All other authors declare that they have no competing interests.