Objective To investigate the impact of large language models on medical record coding,providing insights for the medical record management industry and professionals to better understand,familiarize with,and utilize large language models.Methods The study compared the time consumption,completion rate,and accuracy rate of coding 93 rare diseases u-sing ICD-10 codes between manual search and multiple large language models,elucidating the influence of large language models on medical record coding.Results In terms of coding time consumption,Model A and Model B required the least time,comple-ting all coding in 8 minutes,which is 90 times faster than manual search.Regarding completion rate,all models except Model C(91.4％)achieved 100％.In terms of accuracy rate,Model A was the highest(87.1％),surpassing manual search coding(84.9％).Model B and Model C had similar accuracy rates,47.3％and 43.5％respectively,while Model D had the lowest(0％).Conclusion There is a significant difference in coding accuracy among different large language models,but the accura-cy of Model A has already surpassed that of manual search coding.This demonstrates the powerful capabilities and potential of large language models in medical record coding.In the future,AI based on large language models may replace much of the manu-al work in disease coding.

Objective To investigate the impact of large language models on medical record coding,providing insights for the medical record management industry and professionals to better understand,familiarize with,and utilize large language models.Methods The study compared the time consumption,completion rate,and accuracy rate of coding 93 rare diseases u-sing ICD-10 codes between manual search and multiple large language models,elucidating the influence of large language models on medical record coding.Results In terms of coding time consumption,Model A and Model B required the least time,comple-ting all coding in 8 minutes,which is 90 times faster than manual search.Regarding completion rate,all models except Model C(91.4％)achieved 100％.In terms of accuracy rate,Model A was the highest(87.1％),surpassing manual search coding(84.9％).Model B and Model C had similar accuracy rates,47.3％and 43.5％respectively,while Model D had the lowest(0％).Conclusion There is a significant difference in coding accuracy among different large language models,but the accura-cy of Model A has already surpassed that of manual search coding.This demonstrates the powerful capabilities and potential of large language models in medical record coding.In the future,AI based on large language models may replace much of the manu-al work in disease coding.