Approximately 30% of patients with gastrointestinal cancer undergoing 5-Fluorouracil (5-FU)-based chemotherapy experience severe toxicity. Presently, there is a dearth of effective tools for identifying individuals at risk within this context. This study aims to fill this gap by constructing a predictive model using a Bayesian network, a robust probabilistic graphical model known for its interpretable predictions. Employing a dataset encompassing 267 gastrointestinal cancer patients, the data underwent preprocessing and was partitioned into TRAIN and TEST sets in an 80%:20% ratio. Variable importance was assessed using the RandomForest algorithm, employing the MeanDecreaseGini coefficient. The Bayesian network model was designed using the bnlearn R library, utilizing a 10-fold cross-validation on the TRAIN set, and optimizing the network structure with the aic-cg method. Model performance was evaluated through accuracy, sensitivity, and specificity, employing cross-validation on the TRAIN set and independent validation on the TEST set. The model displayed favorable performance, achieving an average accuracy of 0.85 (±0.05) and 0.80 on the TRAIN and TEST datasets, respectively. Sensitivity and specificity were 0.82 (±0.14) and 0.87 (±0.07) for the TRAIN dataset, and 0.71 and 0.83 for the TEST dataset. A user-friendly tool was developed for clinical deployment. Despite some limitations, our Bayesian network model exhibited a strong capacity to predict the likelihood of severe hematological toxicity in gastrointestinal cancer patients undergoing 5-FU-based chemotherapy. Future investigations should concentrate on validating the model using larger patient cohorts and in diverse clinical scenarios.
HTML PDFShare this article
Journal of Forensic Medicine received 165 citations as per Google Scholar report
