Maintaining proper health records improves machine learning predictions for novel 2019-nCoV

Area of Science:

• * Computational epidemiology
• * Health informatics

Background:

• * Novel coronavirus (2019-nCoV) pneumonia outbreak impacts global health.
• * Accurate outcome prediction, especially mortality risk, is crucial for managing the pandemic.

Purpose of the Study:

• * To estimate outcome risk (recovered, isolated, death) for 2019-nCoV using early datasets.
• * To evaluate the impact of patient disease history on prediction accuracy.

Main Methods:

• * Utilized machine learning classifiers (AdaBoost, bagging, extra-trees, decision trees, k-nearest neighbor).
• * Analyzed two 2019-nCoV datasets, incorporating features like country, age, gender, and patient disease history.
• * Assessed prediction performance considering variations in reporting rates.

Main Results:

• * Incorporating patient disease history improved the prediction of death by over sevenfold.
• * Models without disease history exhibited poor predictive performance.

Conclusions:

• * Patient disease history is a valuable feature for enhancing 2019-nCoV outcome prediction using machine learning.
• * Improved predictions can aid in patient treatment and alleviate healthcare system burdens, especially with recurring infections.