Development and proof-of-concept of a treatment target recommendation algorithm in the context of cognitive...

Background: Cognitive Processing Therapy (CPT) is an effective, widely supported treatment for PTSD, but patient response varies considerably. Optimally targeting maladaptive trauma-related beliefs (stuck points) may significantly enhance treatment outcomes. This study evaluated the feasibility of developing a recommendation algorithm that helps clinicians identify specific patient-endorsed stuck points, which, when effectively restructured, could lead to greater reductions in PTSD symptoms.Methods: Data were drawn from 898 veterans and service members participating in a two-week CPT-based accelerated PTSD treatment programme. Measures administered at pre-, mid-, and post-treatment included the Posttraumatic Cognitions Inventory (PTCI), PTSD Checklist for DSM-5 (PCL-5), and Patient Health Questionnaire (PHQ-9). Patients were clustered at pre- and mid-treatment using K-means based on their PCL-5 subscores and PHQ-9 and PTCI item endorsement profiles. Within each cluster, random forest and elastic net regression models identified PTCI items most predictive of PTSD symptom reductions. The algorithm's predictive utility was illustrated through simulated patient cases.Results: Machine learning models identified specific PTCI items within each cluster that appeared most predictive of PTSD symptom reductions. Simulations indicated that targeting these algorithm-selected cognitions could yield an additional 5-7 points improvement on the PCL-5 by mid-treatment, with further improvements of 4-6 points thereafter. Overall, following the algorithm's personalised recommendations could theoretically result in approximately 11 additional points of symptom reduction on the PCL-5 beyond expected standard improvements.Conclusions: This study demonstrates the feasibility of developing a data-driven recommendation algorithm that personalises CPT by identifying the most impactful stuck points to target. Prospective clinical trials are necessary to validate the algorithm's effectiveness, establish clinical utility, and facilitate implementation into routine care.