Entry to Rosie, an AI chatbot for maternal well being, confirmed no statistically vital nervousness or despair profit at 12 months in contrast with a month-to-month youngsters’s e book program in a 400-person randomized trial.
The examine, revealed Nov. 4 in JAMA Community Open, examined entry to well being recommendation amongst pregnant and postpartum contributors with usually low beginning symptom scores.
Rosie was designed to reply maternal and toddler well being questions, supply customized steering, and supply reminders and academic assist. A medical social employee reviewed misery alerts, contacted contributors, and supplied sources, so the intervention included human assist alongside the chatbot.
Researchers assigned 200 contributors to every group. The US examine enrolled English-speaking individuals aged 14 or older who have been pregnant or parenting an toddler youthful than six months and recognized as members of racial or ethnic minoritized teams.
Enrollment ran from October 2023 by way of December 2024, with follow-up accomplished in December 2025.
The first measures have been self-reported nervousness signs on the GAD-7 questionnaire and despair signs on the PHQ-9 questionnaire. The 12-month analyses included 361 contributors for nervousness and 362 for despair.
The adjusted distinction between the chatbot and management teams was 0.10 factors for nervousness, with a 95% confidence interval of -0.77 to 0.96. For despair, it was -0.23 factors, with an interval of -1.05 to 0.59.
Each intervals embody zero, leaving uncertainty concerning the dimension and route of any underlying impact. The trial established neither a transparent profit nor equivalence between the approaches.
Learn the 2 main outcomes
Learn the 2 main outcomesAnxiousness · GAD-7Melancholy · PHQ-9
Anxiousness: adjusted imply distinction 0.10 factors. The 95% confidence interval crosses zero.
Anxiousness: adjusted distinction and 95% confidence intervalReported adjusted imply distinction 0.1; 95% confidence interval -0.77 to 0.96. The interval crosses zero.+0.10 points95% CI: -0.77 to 0.96-10+1Chatbot minus management, questionnaire pointsZero = no between-group differenceLower scores imply fewer signs.Line: confidence interval. Dot: estimate.
Anxiousness: adjusted distinction and 95% confidence intervalReported adjusted imply distinction 0.1; 95% confidence interval -0.77 to 0.96. The interval crosses zero.+0.10 points95% CI: -0.77 to 0.96-10+1Chatbot minus management, questionnaire pointsZero = no between-group differenceLower scores imply fewer signs.Line: confidence interval. Dot: estimate.
Melancholy: adjusted imply distinction -0.23 factors. The 95% confidence interval crosses zero.
Melancholy: adjusted distinction and 95% confidence intervalReported adjusted imply distinction -0.23; 95% confidence interval -1.05 to 0.59. The interval crosses zero.-0.23 points95% CI: -1.05 to 0.59-10+1Chatbot minus management, questionnaire pointsZero = no between-group differenceLower scores imply fewer signs.Line: confidence interval. Dot: estimate.
Melancholy: adjusted distinction and 95% confidence intervalReported adjusted imply distinction -0.23; 95% confidence interval -1.05 to 0.59. The interval crosses zero.-0.23 points95% CI: -1.05 to 0.59-10+1Chatbot minus management, questionnaire pointsZero = no between-group differenceLower scores imply fewer signs.Line: confidence interval. Dot: estimate.
Each confidence intervals cross zero. The trial establishes neither a transparent profit nor equivalence.
US, English-speaking pregnant/postpartum contributors from racial or ethnic minoritized teams, with usually low baseline signs. 400 randomized; endpoint samples: nervousness 361, despair 362. Outcomes have been self-reported. Rosie included incentives and human assist; each teams may contact workers.
Human assist and low baseline signs in AI chatbot examine
Contributors obtained small funds for questions and e book rewards for probably the most lively customers. The management group obtained one youngsters’s board e book per 30 days for a 12 months, and each teams may talk with examine workers.
These situations restrict how immediately the findings translate to routine chatbot use.
Contributors usually started with low symptom scores. Amongst these with out there baseline responses, the median nervousness rating was 2, and the median despair rating was 3. About 11.1% and 9.5% of the contributors, respectively, met the examine’s moderate-to-severe threshold.
Self-reported well being outcomes are one other limitation, and the authors urged warning about secondary and exploratory findings.
Rosie, delivered with incentives and human backup to this inhabitants, confirmed no statistically vital 12-month benefit on the 2 main mental-health measures. Whether or not different AI interventions or completely different medical populations would yield completely different outcomes stays open.
