Objective:

To compare the accuracy metrics of three large language models (LLMs) in identifying biometric parameters and supporting toric IOL planning for cataract surgery.

Key Findings:

• The accuracy of LLMs in extracting biometric parameters varied among the models, with implications for their use in clinical settings.
• LLMs demonstrated potential in assisting with toric IOL planning, but performance consistency needs further validation.
• Automated technologies could reduce human error in preoperative assessments, enhancing patient outcomes.

Interpretation:

The study highlights the potential of LLMs to enhance the accuracy and efficiency of preoperative planning in cataract surgery, particularly for toric IOLs.

Limitations:

• The study was conducted at a single center, which may limit generalizability.
• The evaluation was based on retrospective data, which may introduce biases affecting reliability.
• The models were tested without parameter tuning or external tools, potentially affecting performance.

Conclusion:

LLMs show promise in improving the accuracy of biometric data extraction and toric IOL planning, warranting further research for clinical validation in practical settings.