Personalized treatment selection and adaptation for depression – the T-SAD study

The central aim of this project is to increase the effectiveness of existing treatments for major depressive disorder (MDD). To this end, we will develop and implement an easy-to-use web-based prediction tool, based on innovative statistical techniques, which will be prospectively tested in routine daily care. This is relevant because, although evidence-based treatments for MDD are available, individual responses to these treatments vary widely and are unpredictable. Clinically, this results in a trial-and-error approach in which different consecutive treatments are being offered to find the optimal regimen resulting in chronicity, long treatment trajectories, and high societal costs. To overcome these problems, prediction models have been studied to develop personalized treatment selection for MDD. However, most of these initiatives were carried out retrospectively as post hoc analyses, and suffer from inconsistent results and power issues. In addition, studies have successfully enriched ongoing treatments with frequent outcome monitoring and feedback using personalized adaptations. However, these personalized treatment predictions and treatment adaptations have been mainly investigated separately as different fields of research. Additionally, previous studies have limited their scope to an examination of clinical response, thereby ignoring important other outcomes as quality of life, well-being, and treatment dropout. To maximize the potential of personalized medicine, the current project will combine selection and adaptation strategies into one comprehensive approach with the integration of different outcome domains. This project is a joint research initiative of three academic centers, four large nation-wide mental health organizations, and a foundation for client empowerment and participation. It aims to develop, test, and implement a tool for personalized treatment recommendations and adaptations to improve quality of depression treatment. We define this as (i) a larger decrease of depressive symptomatology, (ii) an increase of quality of life and well-being, and (iii) reduction of treatment dropout in comparison to treatment as usual (TAU). As core activities, we will i) combine existing datasets and expert opinion to develop innovative Bayesian statistical models that predict these outcomes prior and during treatment with continued incorporation of client and therapist information, ii) test this model prospectively in comparison with TAU in routine care, and iii) integrate this model into a cost-effective and easy-to-use web-based tool for daily clinical use. To this end, the project is divided in four phases: 1. In phase 1, we will build innovative statistical Bayesian models. Advantages of Bayesian inference are (i) incorporation of prior knowledge (empirical, and client and clinician information ensuring expert by experience participation) leading to powerful predictions that enable the use of relatively small samples and (ii) the possibility for constant model updating with new empirical data (information of ongoing treatment). The latter results in real time prediction instead of retrospective monitoring of outcomes and serves as the basis for potential treatment adaptations. 2. In phase 2, an easy to use web-based tool will be developed in which these models are incorporated. Tool development will be a result of close collaboration with experts in tool design, software developers, and end-users (client’s, their relatives, and clinicians). In this phase, we will also make an implantation plan and set up a data-collection infrastructure to prospectively test and implement the newly developed tool. 3. In phase 3, we will test the tool as developed in phases 1 and 2 and compare this to TAU in a multicenter randomized trial. Participants will be randomized to: - TAU (based on regular shared decision-making). - model-informed personalized treatment selection (shared decision-making with a model-informed recommendation for treatment selection). - model-informed personalized treatment selection and adaptation (shared decision-making with a model-informed recommendation for treatment selection, and model-informed recommendations for treatment adaptations). We hypothesize that our tool will result in significant larger effect sizes on the defined outcomes in comparison to treatment as usual. Additionally, we hypothesize that adding treatment adaptation advice to treatment selection will lead to a significant larger effect size than model-informed treatment selection only. 4. In phase 4 we will (i) assess cost-effectiveness of our intervention in comparison to TAU, (ii) evaluate tool and implementation plan, (iii) evaluate and fine-tune our models, and (iv) disseminate and further implement the application of the tool. This project has great potential to improve (cost) effectiveness of MDD treatment treatment by preventing long treatment trajectories that are prone to chronicity and delayed recovery.