Ovarian Cancer Screening and AI (AI-OCS-Gyn)

Charité University Medicine Berlin

Status: Not yet enrolling

Conditions: Ovarian Cancer Screening Recommendations by Gynecologists

Treatments: Behavioral: ChatGPT - Control; Behavioral: ChatGPT - Evidence-Based Screening Discussion

Study type: Interventional

Funder types: Other

Identifiers

NCT07503054

2025ChatGPTGyn

Details and patient eligibility

About

Gynecologists frequently overestimate the benefits and safety of ovarian cancer screening. AI-supported discussions may help correct these misperceptions. This study tests whether an AI-guided conversation about the evidence on ovarian cancer screening can improve gynecologists' knowledge and reduce non-evidence-based screening recommendations, compared with a control AI discussion on ovarian cancer prevalence.

Full description

Previous research has demonstrated that gynecologists often substantially overestimate both the effectiveness and safety of ovarian cancer screening, despite robust evidence indicating that such screening does not offer a net clinical benefit. These findings highlight the need for innovative communication strategies to support evidence-based clinical practice and reduce low value care.

AI-based conversational interventions have shown promising results in other fields when aiming to correct misconceptions or encourage engagement with evidence, particularly among individuals who are initially resistant to factual information. Leveraging these insights, this study investigates whether AI-facilitated discussions can effectively improve gynecologists' knowledge of the benefit-harm profile of ovarian cancer screening and subsequently reduce non-evidence-based recommendations.

The study employs a cross-sectional study design in which gynecologists who have previously indicated to regularly recommend ovarian cancer screening with transvaginal ultrasound and potentially with additional CA 125-testing to their asymptomatic, average-risk patients are randomized to one of two conditions:

1. Intervention Condition: Participants engage in an AI-guided conversation in which they explain their reasons for recommending ovarian cancer screening. The AI is instructed to address misconceptions and clarify the lack of evidence supporting a positive benefit-harm ratio.
2. Control Condition: Participants engage in an AI discussion on the prevalence of ovarian cancer, without receiving information or corrective feedback related to screening outcomes.

Before and after the AI-based discussion, all participants are queried on their numerical (X out of 1,000 women) and subjective perception of ovarian cancer screening's benefits and harms and their screening recommendations. Measures are derived from instruments used in prior research.

The primary objective of this study is to assess the change, from before to after the AI-based conversation, in clinicians' understanding of the benefit-harm ratio and their recommendations regarding routine ovarian cancer screening for asymptomatic, average-risk women, within and between study groups.

Enrollment: 350 estimated patients

Sex: All

Ages: 24+ years old

Volunteers: No Healthy Volunteers

Inclusion criteria

• gynecologists in outpatient care who provide ovarian cancer screening to asymptomatic, average-risk women (not guideline consistent)

Exclusion criteria

• gynecologists in inpatient care
• gynecologist in outpatient care who do NOT provide ovarian cancer screening to asymptomatic, average-risk women (guideline consistent)

Trial design

Primary purpose: Screening

Allocation: Randomized

Interventional model: Parallel Assignment

Masking: Single Blind

---

350 participants in 2 patient groups

Control (ChatGPT Control Condition)

Other group

Description:

Participants in this arm engage in a three-turn conversation with ChatGPT. The AI's role is to: * Discuss the participant's perception of how dangerous ovarian cancer is. * Provide factual information on prevalence, lifetime risk, mortality rates, and general epidemiology. * Avoid any mention of screening tests, guideline recommendations, or screening benefits/harms. * Keep responses concise (5-8 sentences per turn). * Begin by reacting to the participant's opening question: "In your mind, how dangerous is ovarian cancer?"

Treatment:

Behavioral: ChatGPT - Control

Experimental (ChatGPT Evidence-Based Screening Discussion)

Experimental group

Description:

Participants in this arm engage in a three-turn conversation with ChatGPT. The AI's role is to: * Ask participants to elaborate on their reasons for recommending ovarian cancer screening. * Provide clear, evidence-based information about benefits and harms of ovarian cancer screening in average-risk women. * Refer to key findings from large trials (e.g., PLCO, UKCTOCS) with absolute numbers (false-positive rates, unnecessary surgeries, complication rates, lack of mortality benefit). * Summarize positions of major U.S. guidelines (e.g., USPSTF, ACOG), including recommendation against routine screening in asymptomatic, average-risk women. * Evaluate the evidence and state whether routine screening is supported based on current data. * Maintain a respectful, non-judgmental tone; critique evidence, not the participant. * Keep responses concise (5-8 sentences per turn). * Begin by responding to the participant's opening question: "Why do you recommend ovarian cancer screening?"

Treatment:

Behavioral: ChatGPT - Evidence-Based Screening Discussion