OBJECTIVE

This study aims to develop and evaluate a data-driven chatbot to address frequently asked questions by women with infertility, assessing its relevance and usefulness.

MATERIALS AND METHODS

A multi-stage data integration process was employed. A total of 15,457 clinical data from two IVF clinics underwent preprocessing, including cleaning, statistical imputation for missing values, and normalization. The data were then labeled to align with user queries. Community data were gathered from three of the most popular Korean online platforms and semantically aligned using a vector database to ensure relevance to user inquiries, providing context-specific statistics such as hormone levels and treatment outcomes. The distribution of frequently asked questions within the community was analyzed, and topics for detailed evaluation were selected. The chatbot’s accuracy was quantitatively evaluated against a representative set of clinical questions, with statistical measures repeated ten times per question to compute the mean and standard error. Additionally, fifteen experienced reproductive endocrinologists assessed the chatbot’s responses to real community questions.

RESULTS

The quantitative evaluation showed that all correct answers fell within the 98% confidence interval of the chatbot’s responses, highlighting high reliability. Experts’ qualitative feedback rated the system's comprehension (3.8), usefulness (3.33), and relevance (3.26) on a 5-point Likert scale, where 1 is the lowest and 5 is the highest. These scores indicate effective alignment with user needs but potential for increased usefulness.

CONCLUSIONS

The low success rates of fertility treatments can lead to repetitive failures and depression in over 90% of affected women. Many women with infertility seek support in online communities to find peace of mind, yet the fragmented experiences shared in these forums can lead to confusion. The advantages of recent advancements in chatbots, spearheaded by technologies like ChatGPT, include high accessibility and the ability to offer objective consultations based on extensive data. This research demonstrated that a chatbot developed using clinic data and online community resources can effectively serve infertile patients by providing accurate and useful information through both quantitative and qualitative evaluations.

IMPACT STATEMENT

For infertile women who depend on fragmented information and online community activities, the chatbot can provide interactive consultations based on objective data, which may increase access to care and potentially alleviate anxiety and depression.