Deepfake perpetrator conversation for adults with sexual abuse-related posttraumatic stress disorder: intervention development and multiple baseline study protocol

Background: A conversation between a victim and a perpetrator of sexual abuse has the potential to reduce posttraumatic stress disorder (PTSD) symptoms in the victim. However, an actual conversation may not always be possible or feasible. Deepfake technology may offer a way to simulate a conversation in a clinical context. Therefore, a feasible, acceptable and potentially effective intervention protocol is required. Objective: We aimed to develop an intervention protocol for a deepfake victim-perpetrator conversation to be used in the treatment of patients with sexual abuse-related PTSD, that would be technologically, legally, ethically, and clinically feasible as well as acceptable. In addition, we aimed to design a single-case experimental design (SCED) multiple baseline study to evaluate its efficacy. Method: The intervention protocol was developed through an iterative multidisciplinary process following established guidelines for complex intervention development. Based on a literature review and interviews with 15 experts and patient representatives, a draft intervention protocol was created and subsequently evaluated, piloted and refined. A consequent study protocol was then designed in accordance with the SPIRIT guidelines. Results: The development process resulted in an intervention protocol for a half-day perpetrator conversation using deepfake technology. The intervention is grounded in cognitive–behavioural therapy and emotional processing theory, and consists of a preparatory session, a deepfake session, and a debriefing session, to be conducted by two therapists. The intervention is tested in a SCED multiple baseline study of 10 patients with sexual abuse-related PTSD who are randomly assigned to a baseline period of 10, 15 or 20 days. Outcomes (changes in negative cognitions, PTSD, guilt and shame, forgiveness, and empowerment) are assessed through 20–30 daily measurements and four main assessments, and analysed using Bayesian statistics. Conclusion: A feasible and acceptable intervention protocol for a deepfake intervention was developed. Its efficacy is tested in a multiple-baseline study.