Early Triage of Critically Ill Adult Patients With Mushroom Poisoning: Machine Learning Approach

Area of Science:

• Medical Informatics
• Toxicology
• Machine Learning

Background:

• Early triage of mushroom poisoning is critical for effective treatment and reducing fatalities.
• No standardized clinical data-driven triage method currently exists for mushroom poisoning.

Purpose of the Study:

• To develop and evaluate a machine learning-based triage system for mushroom poisoning using clinical indicators.
• To assess the predictive accuracy of various machine learning algorithms for mushroom poisoning triage.

Main Methods:

• Utilized data from 567 patients across five hospitals in China, divided into training (322) and testing (245) cohorts.
• Applied four machine learning algorithms, including extreme gradient boosting (XGBoost), to build the triage model.
• Evaluated model performance using metrics like AUC, decision curves, sensitivity, and specificity, with feature importance assessed via Shapley additive explanations.

Main Results:

• The XGBoost model demonstrated superior discriminative ability with an AUC of 0.90 (95% CI 0.83-0.96) on the test set.
• XGBoost achieved a sensitivity of 0.93 and specificity of 0.79, outperforming physician assessments (sensitivity 0.86, specificity 0.66).
• The 14-factor XGBoost model showed strong predictive performance in identifying critically ill patients.

Conclusions:

• The developed 14-factor XGBoost model offers a rapid and accurate method for early triage of mushroom poisoning.
• This model can aid in selecting appropriate treatments and patient referrals, potentially lowering mortality rates.
• The XGBoost model provides a valuable clinical decision support tool for managing mushroom poisoning cases.