Bioethical Considerations for Building, Evaluating, and Implementing Artificial Intelligence in Perinatal Mood and Anxiety Disorders

PROJECT SUMMARY Postpartum depression (PPD) is a common, yet treatable illness if detected early, but it can also have deleterious effects to the mother and child if left untreated. Routine screening for PPD is considered best practice but does not consistently occur due to time and resource constraints. As a result, therapeutic interventions are initiated late and many PPD cases go undetected altogether. Artificial intelligence (AI) models can bridge the PPD identification gap and have been shown to proactively, accurately identify women with an elevated risk for PPD. Iris OB Health, digital health startup company, has built a predictive AI model for PPD. The Iris team is developing an interface that presents our AI model to patients and clinicians to facilitate shared decision-making about interventions to decrease risk. Through this work, we are recognizing the need to better understand the bioethical implications of patient-facing AI. Bioethics research in AI has focused on presenting model output and fostering trust in AI among clinicians. However, important ethical questions from the patient perspective remain unanswered. Specifically, it is unclear if patients are informed about or approve of their data being utilized for model building purposes. As patient engagement and shared decision-making continue to rise in importance, it is likely that AI output will also be presented to patients. PPD presents a complex bioethical case for studying patient-facing AI because, in pregnancy, the autonomy, harms, and benefits afforded to the perinatal patient, newborn/ fetus, and partner must be weighed simultaneously. Therefore, this supplement will focus on developing concrete guidance for creating and implementing patient-facing AI while upholding ethical principles to utilize patient data sensitively and equitably, using PPD as a use case. The specific aims are to: 1) triangulate perspectives for transparent, ethical, and equitable use of patient data in AI for PPD from diverse stakeholders through semi-structured interviews, and 2) evaluate knowledge, attitudes, and preferences related to the utilization of AI in PPD among women of child-bearing age via a nation-wide survey. To accomplish these aims, we will leverage our multidisciplinary team from the parent grant and include new investigators who have collective expertise in obstetrics, perinatal psychiatry, AI development, informatics, qualitative research, survey methods, and user-centered design. Given the explosion of bioethical questions related to AI but the lack of attention paid to patient-facing AI, this project fills an important gap in advancing bioethical AI research. This work will both inform future AI work in other mental health domains and be directly incorporated into Phase II development activities by Iris OB Health.PROJECT NARRATIVE Predictive artificial intelligence (AI) models can support more effective shared decision-making for prevention of postpartum depression (PPD), but important bioethical questions pertaining to patient-facing AI remain to be answered. This supplement will focus on developing preliminary guidance for creating and implementing patient-facing AI while upholding ethical principles to utilize patient data sensitively and equitably, using PPD as a use case.