Early prediction of sepsis using chatGPT-generated summaries and structured data

In this paper, we propose a large language models (LLMs) assisted algorithm that uses ChatGPT to summarize clinical notes and then concatenate these generated summaries with structured data to predict sepsis. We perform a human evaluation of the summaries generated by ChatGPT and evaluate our algorithm using an independent test set. Our algorithm achieves a high prediction AUC of 0.93 (95% CI 0.92-0.93), accuracy of 0.92 (95% CI 0.91-0.92), and specificity of 0.89 (95% CI 0.88-0.90) 4 hours before the onset of sepsis. The ablation study demonstrated a 2% improvement in predicted AUC score when utilizing ChatGPT for clinical notes summarization compared to traditional methods, 4 hours before the sepsis onset. The experiment results in turn revealed the remarkable performance of ChatGPT in the domain of clinical notes summarization.