Poster Session 1
Category: Obstetric Quality and Safety
Poster Session 1
Avihu Krieger, N/A (he/him/his)
The Department of Obstetrics and Gynecology, The Chaim Sheba Medical Center, Ramat-Gan, Israel
Ramat Gan, Tel Aviv, Israel
Michal Axelrod, MD, MPH (she/her/hers)
Sheba Medical Center
Sheba Medical Center, HaMerkaz, Israel
Uri Shemesh, MD (he/him/his)
The Department of Obstetrics and Gynecology, The Chaim Sheba Medical Center, Ramat-Gan, Israel
Tel Aviv, Tel Aviv, Israel
Liron Seidman, MD
Lis Hospital for Women’s Health, Tel Aviv Sourasky Medical Center Gray Faculty of Medicine, Tel Aviv University, Israel
Ramat Gan, HaMerkaz, Israel
Roni Plaschkes
Sheba Medical Center
Ramat Gan, HaMerkaz, Israel
Daphna Komem
The Department of Obstetrics and Gynecology, The Chaim Sheba Medical Center, Ramat-Gan, Israel
Ramat Gan, Tel Aviv, Israel
Shlomi Toussia-Cohen, MD (he/him/his)
The Sheba Medical Center
The Sheba Medical Center, HaMerkaz, Israel
Keren Ofir, MD
Sheba Medical Center
Sheba Medical Center, HaMerkaz, Israel
Boaz Weisz, MD
The Department of Obstetrics and Gynecology, The Chaim Sheba Medical Center, Ramat-Gan, Israel
Ramat Gan, Tel Aviv, Israel
Ronen Fluss, PhD
Sheba Medical Center
Ramat Gan, Tel Aviv, Israel
Shalom Mazaki-Tovi, MD
The Department of Obstetrics and Gynecology, The Chaim Sheba Medical Center, Ramat-Gan, Israel
Ramat Gan, HaMerkaz, Israel
Michal Fishel Bartal, MD
Division of Maternal-Fetal Medicine, Department of Obstetrics, Gynecology and Reproductive Sciences, McGovern Medical School at UTHealth Houston
Houston, Texas, United States
Rakefet Yoeli-Ullman, MD
The Department of Obstetrics and Gynecology, The Chaim Sheba Medical Center, Ramat-Gan, Israel
Ramat Gan, Tel Aviv, Israel
Abraham Tsur, MBA, MD
Director of Anterpartum High Risk Beyond, Medical Director ARC
The Sheba Medical Center
The Sheba Medical Center, HaMerkaz, Israel
Electronic health records (EHRs) of individuals in high-risk antepartum units contain valuable clinical data. Large language models (LLMs) offer a promising tool for transforming these notes into patient-facing summaries. Our aim was to identify the optimal prompt for this task via a structured, clinician-led evaluation, and to conduct an initial real-world assessment of patients’ interest in such summaries.
Study Design:
A prospective two-phase study. In phase one, 10 clinicians evaluated EHR summaries generated for 10 patients using a HIPAA-compliant ChatGPT-4.0 platform. Summaries were created using three prompts (Figure 1): Prompt A – minimal instructions; Prompt B – detailed summarization guidance; and Prompt C – Prompt B plus integration of verified external medical knowledge. Summaries were rated using a validated questionnaire assessing correctness, completeness, conciseness, and safety. We analyzed prompt ratings using linear regression, adjusting for physician, patient, and presentation order. In phase two, 11 summaries were generated with the selected prompt, reviewed them for safety, and shared with patients, who rated their interest and provided feedback.
Results:
Prompt A, carrying minimal instructions, received the highest overall score, ranking significantly higher than Prompt B (p=0.024) and trending higher than Prompt C (p=0.053) (Figure 2), and was selected to generate 11 summaries for the next step.
In phase two, we generated summaries for 11 patients based on their EHRs, excluding one for safety concerns after incorrectly implying completion of a genetic workup that was only partial. The remaining 10 summaries (90.9%) were delivered in patients’ native languages (9 Hebrew, 1 Arabic). Overall, 80% of patients were interested in additional summaries, but only 40% reported improved understanding.
Conclusion:
Most high-risk antepartum patients expressed interest in continuing to receive LLM-generated summaries of their EHRs after initial exposure. While this approach shows promise for empowering and engaging our patients, clinician review remains essential to ensure safety.