Survival prediction of head and neck squamous cell carcinoma using machine learning models

Head and Neck Squamous Cell Carcinoma (HNSCC) is one of cancer type that is most distressing leading to acute pain, effecting speech and primary survival functions such as swallowing and breathing. The morbidity and mortality of HNSCC patients have not significantly improved even tough there has been advancement in surgical and radiotherapy treatments. The high mortality may be attributed to the complexity and significant changes in the clinical outcomes. Therefore, it is important to increase the accuracy of predicting the outcome of cancer survival. Few cancer survival prediction models of HNSCC have been proposed so far. In this study, genomic data (whole exome sequencing) are integrated with clinical data to improve the performance of prediction model. The somatic mutations of every patient is processed using Multifractal Deterended Fluctuation Analysis (MFDFA) algorithm and the parameter values of Fractal Dimension (Dq) is included along with clinical data for cancer survival prediction. Feature ranking proves that the new engineered feature is one of the important feature in prediction model. In order to improve the performance index of models, hyperparameters were also tuned in all the classifiers considered. 10-Fold cross validation is implemented and XGBoost (98% AUROC, 94% precision, and 93% recall) proves to be best model classifier followed by Random Forest 93% AUROC, 93% precision, and 93% recall), Support Vector Machine (84% AUCROC, 79% precision, and 79% recall) and Logistic Regression (80% AUROC, 77% precision, and 76% recall).