The Role of Large Language Models in Hormonal Contraception Consultation

Area of Science:

• Medical Informatics
• Reproductive Health
• Artificial Intelligence

Background:

• Large Language Models (LLMs) show potential in healthcare but require evaluation for specialized applications like hormonal contraception counseling.
• The efficacy of LLMs in providing accurate and comprehensive contraceptive advice is not well-established.

Purpose of the Study:

• To assess the accuracy and completeness of LLM-generated advice for hormonal contraception.
• To compare LLM responses against established evidence-based clinical guidelines.

Main Methods:

• Ten simulated clinical scenarios covering common contraceptive counseling needs were used.
• Three LLMs (Chat-GPT, Google Gemini, OpenEvidence) were tested on medical eligibility screening, contraindications, drug interactions, side effects, and emergency contraception.
• Responses were independently reviewed by two board-certified obstetrician-gynecologists against international guidelines.

Main Results:

• All LLMs demonstrated high accuracy in identifying contraindications based on Medical Eligibility Criteria (MEC) and suggested suitable alternative contraceptives.
• LLMs effectively managed drug interactions, including the lamotrigine-estrogen interaction, and offered evidence-based strategies for side effect management.
• Communication styles varied: Chat-GPT focused on structured consultation, Gemini on actionable advice, and OpenEvidence on concise summaries.

Conclusions:

• LLMs generally provided advice consistent with guidelines in standardized scenarios.
• Limitations in individualized assessment and health literacy optimization necessitate cautious LLM deployment.
• LLMs are best utilized as supplementary educational resources, not replacements for professional contraceptive counseling.