A lung cancer risk-prediction model based on the open-source algorithm XGBoost outperforms a logistic regression algorithm when it comes to prediction accuracy and robustness, according to a study published January 7 in the Journal of Evidence-Based Medicine.
Making use of lung cancer risk prediction models at the screening stage can improve the accuracy of identifying high-risk individuals eligible for lung cancer screening, wrote a team led by Tiantian Zhang, PhD, of Jinan University and Southern Medical University, both in Guangzhou, China. However, the group noted that there is a lack of research on such models in China, particularly regarding machine learning algorithms.
To address the knowledge gap, Zhang's group conducted a study that used a stratified random sampling method to randomly divide data from 11,708 participants in the Guangzhou Lung-Care Project Program into a training set (70%) and a validation set (30%). The team tracked key variables via least absolute shrinkage and selection operator (LASSO) regression, then used logistic regression and XGBoost to build a lung cancer risk prediction model in the training set and confirm its performance in the validation set.
The investigators reported that the area under the receiver operating curve (AUC) of the logistic regression model in the validation set was 0.647, while the AUC of the XGBoost model based on the machine-learning algorithm in the validation set was 0.658.
In an additional and unexpected finding, the team found a significant effect of childhood exposure to cooking fuels on the risk of lung cancer, it noted.
"The lung cancer risk prediction model constructed based on the XGBoost algorithm is better than the logistic regression algorithm in terms of prediction accuracy and robustness, aiding in the risk assessment of individuals undergoing screening," the group concluded.
Access the full study here.
