Machine learning-based mortality prediction models for smoker COVID-19 patients

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning

Background:

• The COVID-19 pandemic overwhelmed healthcare systems globally.
• Existing mortality prediction models may exhibit bias in specific subpopulations, like smokers.
• Targeted models are needed for high-risk groups such as COVID-19 patients with a history of smoking.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting in-hospital mortality in COVID-19 patients with a history of smoking.
• To identify key predictors of mortality in this specific patient cohort.
• To improve resource allocation and patient management strategies for smoker COVID-19 patients.

Main Methods:

• A retrospective study involving 678 COVID-19 patients with a history of smoking across six medical centers (2020-2022).
• Development of multiple machine learning models using 10-fold cross-validation, incorporating demographics, care levels, vital signs, medications, and comorbidities.
• Creation of two model sets: one for at-admission prediction and another for post-admission prediction, followed by probability calibration.

Main Results:

• The in-hospital mortality rate among smoker COVID-19 patients was 20.1%.
• The best-performing at-admission model (XGBoost) achieved 87.5% accuracy and an 86.2% F1 score.
• The top post-admission model (XGBoost) demonstrated 90.5% accuracy and an 89.9% F1 score, with active smoking identified as a crucial predictor.

Conclusions:

• Machine learning models can effectively predict mortality in COVID-19 patients who smoke.
• These models offer potential for enhanced management and survival prediction in this high-risk group.
• The findings highlight the importance of considering smoking status in COVID-19 mortality prediction.