Periodic Reporting for period 2 - RE-SHIFT (Dismantling, REdialing, personalizing, and implementing task SHIFTing psychosocial interventions to treat and prevent common mental disorders in low-resource settings)
Berichtszeitraum: 2024-10-01 bis 2025-09-30
RE-SHIFT addresses one of the leading causes of disability worldwide: the ever-increasing prevalence of anxiety and depression. It is known that psychological and social interventions are effective for the treatment of these disorders. In low-resource settings, though, for feasibility reasons, these interventions are not delivered by mental health professionals but by trained lay counselors, so that a wider audience can be reached at a lower cost. However, such interventions are compilations of different, possibly interacting, treatment components, usually delivered in different combinations. And it is not yet known which of these components or combinations thereof are most effective. So how to disentangle the efficacy of individual components and understand which of these works best for whom? This is the research question I am trying to address. To do this, RE-SHIFT implemented a sophisticated research design that identified the impact of participant-level prognostic factors and effect modifiers on intervention outcomes. Now that results have been automated into software, it is be possible to tailor interventions to the needs and preferences of people with depression and anxiety.
Dr. Papola conducted a systematic review of randomized controlled trials (RCTs) evaluating the efficacy of task-shared psychosocial interventions using the PICO framework as follows: P (population) = adults diagnosed with common mental disorders (CMD) or experiencing subthreshold psychological distress, seeking first-level/primary care or identified within the community; I (intervention) = any psychosocial intervention provided by non-specialist providers (NSPs), including mentoring programs, cognitive-behavioral therapy and its derivatives, interpersonal therapy, psychoeducation, problem-solving therapy, and others; C (comparison) = no treatment, treatment as usual, waiting list, or another active psychosocial intervention; O (main outcome) = reduction in symptomatology; S (setting) = any context. The systematic literature search, conducted in line with the above PICO criteria, yielded 13,320 references from four bibliometric databases (PubMed, Embase, PsycINFO, CENTRAL). After a rigorous screening process and full-text review, 34 studies were deemed eligible for inclusion. This phase adhered to Cochrane standards to ensure the reliability and quality of the process.
Dr. Papola then contacted the authors of the 34 selected RCTs to obtain (a) the manuals of the psychosocial interventions used in the experimental arms of their trials so that he could qualitatively analyze the manual and create the taxonomy of components (WP1, WP2), and (b) the individual participant data (IPD) from the trials, with the aim of tailoring the results based on participant characteristics (WP3). To analyze the intervention manuals and extract active treatment-specific components from their text, Dr. Papola and his research collaborators conducted multiple rounds of coding, both collaboratively and independently. Dr. Papola and collaborators created an initial list of eligible elements and reviewed it for duplication and redundancy. Each element was then operationalized and assessed by the research team. This process included several pilot testing rounds, during which four researchers independently coded two or more eligible trials to test and refine the existing taxonomy. Interrater reliability (IRR) was assessed using intraclass correlation (ICC). After each round of coding, the team discussed similarities and discrepancies among the coders and adjusted the taxonomy as needed. By the final round, the coding showed a high level of agreement among the researchers (ICC = 0.85–0.96). The outcome of this process was a three-tiered, tree-shaped taxonomy of treatment-specific components. Notably, although other taxonomies of psychological intervention components have been developed previously, this is the first time such an instrument has been designed to support both qualitative and quantitative analyses. To enhance its utility for quantitative purposes, the taxonomy was structured to include three levels of depth and specificity: the 'family' level, the 'component' level, and the 'technique' level. This hierarchical structure allows the level of component analysis to be tailored to the available data. For instance, more granular analyses at the 'technique' level can be conducted when large datasets are available, while broader but still informative analyses can be performed at the 'family' level when data is more limited. An online survey has been conducted with the authors of the trials to assess their agreement with the classification of the intervention elements.
Parallel to this, thanks to the collaborative spirit of the trial authors Dr. Papola engaged with, he successfully gathered individual participant data (IPDs) from 30 of the 34 trials in the systematic review. This achievement represents the collection of 10,618 IPDs out of a possible 11,969, covering 89% of the total available data. This compilation forms the most extensive IPD database assembled to date in the field of global mental health. We fitted a cNMA additive model integrating both IPDs and AD (depending on data availability) to evaluate the efficacy of each component at the individual level. We implemented the IPDcNMA approach within a hierarchical Bayesian framework, which enables us to estimate main component effects and moderation by individual participant characteristics while accounting for heterogeneity at both the study and participant levels.
Dr. Papola then contacted the authors of the 34 selected RCTs to obtain (a) the manuals of the psychosocial interventions used in the experimental arms of their trials so that he could qualitatively analyze the manual and create the taxonomy of components (WP1, WP2), and (b) the individual participant data (IPD) from the trials, with the aim of tailoring the results based on participant characteristics (WP3). To analyze the intervention manuals and extract active treatment-specific components from their text, Dr. Papola and his research collaborators conducted multiple rounds of coding, both collaboratively and independently. Dr. Papola and collaborators created an initial list of eligible elements and reviewed it for duplication and redundancy. Each element was then operationalized and assessed by the research team. This process included several pilot testing rounds, during which four researchers independently coded two or more eligible trials to test and refine the existing taxonomy. Interrater reliability (IRR) was assessed using intraclass correlation (ICC). After each round of coding, the team discussed similarities and discrepancies among the coders and adjusted the taxonomy as needed. By the final round, the coding showed a high level of agreement among the researchers (ICC = 0.85–0.96). The outcome of this process was a three-tiered, tree-shaped taxonomy of treatment-specific components. Notably, although other taxonomies of psychological intervention components have been developed previously, this is the first time such an instrument has been designed to support both qualitative and quantitative analyses. To enhance its utility for quantitative purposes, the taxonomy was structured to include three levels of depth and specificity: the 'family' level, the 'component' level, and the 'technique' level. This hierarchical structure allows the level of component analysis to be tailored to the available data. For instance, more granular analyses at the 'technique' level can be conducted when large datasets are available, while broader but still informative analyses can be performed at the 'family' level when data is more limited. An online survey has been conducted with the authors of the trials to assess their agreement with the classification of the intervention elements.
Parallel to this, thanks to the collaborative spirit of the trial authors Dr. Papola engaged with, he successfully gathered individual participant data (IPDs) from 30 of the 34 trials in the systematic review. This achievement represents the collection of 10,618 IPDs out of a possible 11,969, covering 89% of the total available data. This compilation forms the most extensive IPD database assembled to date in the field of global mental health. We fitted a cNMA additive model integrating both IPDs and AD (depending on data availability) to evaluate the efficacy of each component at the individual level. We implemented the IPDcNMA approach within a hierarchical Bayesian framework, which enables us to estimate main component effects and moderation by individual participant characteristics while accounting for heterogeneity at both the study and participant levels.
The efficacy of task-sharing psychosocial interventions for CMDs was analyzed to identify the most effective components and their combinations through a two-step procedure conducted by Dr. Papola.
Step 1: Network Meta-Analysis (NMA)
Dr. Papola initially pooled aggregate-level trial endpoint data using a NMA technique, which allows for the simultaneous comparison of multiple interventions by utilizing both direct (head-to-head trials) and indirect (trials with a common comparator) evidence. This approach ranks treatments to identify the most effective options across a study network. PM+ intervention manual was most frequently evaluated, with 9 RCTs comparing it to enhanced treatment as usual (ETAU) and 2 RCTs against treatment as usual (TAU). Most other intervention manuals were tested only once, and none were directly compared head-to-head. This led to a network with limited connectivity. Network coherence, representing the agreement between direct and indirect evidence, was assessed within each closed loop and using the “side-splitting” method for individual comparisons. Statistical heterogeneity was evaluated using the heterogeneity variance parameter (T2 = 0.12) and the I2 statistic (I2 = 87.4%; 95% CI: 81.3%–91.5%). The Cochrane Risk of Bias tool 2 was used to assess the risk of bias, and transitivity was ensured by comparing effect modifiers across studies grouped by interventions.
Step 2: Component Network Meta-Analysis (cNMA)
In the second step of the project (last year of the fellowship), the interventions in the form of treatment-specific components were analyzed at the IPD level to determine which components have the best efficacy and for which subpopulation, and to identify the impact of participant-level prognostic factors and effect modifiers on intervention outcomes, thus achieving the ultimate goal of the project. The result of the statistical analysis suggest that there is strong evidence that strengthening social support is beneficial, but moderate evidence that the relaxation component is detrimental. Moderate evidence suggests that problem management and behavior activation are also helpful, while weaker evidence suggests that cognitive reframing may diminish the efficacy of interventions. Dr. Papola repeated the one-step IPDcNMA for the acceptability outcome. The obtained estimates for the components were all close to the null effect and thus not clinically informative. However, it is possible that psychoeducation is related to an increase in all-cause trial discontinuation.
Based on these results is now possible to estimate the relative effects of any combination of components on a patient-specific basis. For instance, it is possible to predict the effect of a psychosocial intervention that includes treatment as usual, social support strengthening, behavioral activation, and problem management, as opposed to treatment as usual alone, for a 36-year-old married woman with at least a high school diploma who is employed and has a baseline severity of 62% (corresponding to a score of 15 points on the K6 scale). The predicted effect is -17.7% (95% confidence interval [95% CI], -23.4 to -12). This corresponds to a net benefit of 4.25 points on the K6 scale (95% CI: -5.62 to -2.88). Posterior samples of key parameters and variance-covariance matrices have been automated into a web app that provides individualized estimates for any combination of components. As foreseen in the grant proposal, the app is freely available at the following link: https://www.metapsy.org/tools/cnma-lmic(öffnet in neuem Fenster)
Step 1: Network Meta-Analysis (NMA)
Dr. Papola initially pooled aggregate-level trial endpoint data using a NMA technique, which allows for the simultaneous comparison of multiple interventions by utilizing both direct (head-to-head trials) and indirect (trials with a common comparator) evidence. This approach ranks treatments to identify the most effective options across a study network. PM+ intervention manual was most frequently evaluated, with 9 RCTs comparing it to enhanced treatment as usual (ETAU) and 2 RCTs against treatment as usual (TAU). Most other intervention manuals were tested only once, and none were directly compared head-to-head. This led to a network with limited connectivity. Network coherence, representing the agreement between direct and indirect evidence, was assessed within each closed loop and using the “side-splitting” method for individual comparisons. Statistical heterogeneity was evaluated using the heterogeneity variance parameter (T2 = 0.12) and the I2 statistic (I2 = 87.4%; 95% CI: 81.3%–91.5%). The Cochrane Risk of Bias tool 2 was used to assess the risk of bias, and transitivity was ensured by comparing effect modifiers across studies grouped by interventions.
Step 2: Component Network Meta-Analysis (cNMA)
In the second step of the project (last year of the fellowship), the interventions in the form of treatment-specific components were analyzed at the IPD level to determine which components have the best efficacy and for which subpopulation, and to identify the impact of participant-level prognostic factors and effect modifiers on intervention outcomes, thus achieving the ultimate goal of the project. The result of the statistical analysis suggest that there is strong evidence that strengthening social support is beneficial, but moderate evidence that the relaxation component is detrimental. Moderate evidence suggests that problem management and behavior activation are also helpful, while weaker evidence suggests that cognitive reframing may diminish the efficacy of interventions. Dr. Papola repeated the one-step IPDcNMA for the acceptability outcome. The obtained estimates for the components were all close to the null effect and thus not clinically informative. However, it is possible that psychoeducation is related to an increase in all-cause trial discontinuation.
Based on these results is now possible to estimate the relative effects of any combination of components on a patient-specific basis. For instance, it is possible to predict the effect of a psychosocial intervention that includes treatment as usual, social support strengthening, behavioral activation, and problem management, as opposed to treatment as usual alone, for a 36-year-old married woman with at least a high school diploma who is employed and has a baseline severity of 62% (corresponding to a score of 15 points on the K6 scale). The predicted effect is -17.7% (95% confidence interval [95% CI], -23.4 to -12). This corresponds to a net benefit of 4.25 points on the K6 scale (95% CI: -5.62 to -2.88). Posterior samples of key parameters and variance-covariance matrices have been automated into a web app that provides individualized estimates for any combination of components. As foreseen in the grant proposal, the app is freely available at the following link: https://www.metapsy.org/tools/cnma-lmic(öffnet in neuem Fenster)