Comparison of nomogram with machine learning techniques for prediction of overall survival in patients with retroperitoneal liposarcoma

Abstract

Abstract Background Retroperitoneal liposarcoma (RP-LPS) is a rare and overlooked tumor type. Because of the rarity and diversity RP-LPS histological subtypes, the diagnosis, treatment and prediction of survival, pose great challenges. This study compared the performance of the 8th edition TNM model, nomogram, and machine learning algorithms in predicting overall survival in patients with RP-LPS to establish a more effective predictive model for clinical use. Establishing relatively reliable survival prediction models has important implications for planning personalized care and patient counseling. Methods The dataset used included clinical data of 2,147 patients with RP-LPS. The machine learning algorithms evaluated included support vector machine, adaptive boosting, decision tree and random forest. These algorithms were evaluated in terms of the area under the receiver-operating characteristic (ROC) curve (AUC) and accuracy values. The performance of the algorithm that produced the optimal results was compared with the 8th edition TNM model and nomogram to better predict overall survival in patients with RP-LPS. Results Comparison of prediction performance indicators of each machine learning algorithm, including accuracy, AUC, F1 score, etc., revealed that the adaptive boosting (AdaBoost) algorithm produced the best prediction effect (accuracy = 69.1%, AUC = 0.70). The performance indicators of AdaBoost were further compared with the traditional TNM model and the nomogram model, and the machine learning algorithm performance was considerably better than other types of models. Conclusions The machine learning algorithm AdaBoost provides more personalized and reliable prognostic information of RP-LPS than the nomogram. However, the level of transparency offered by the nomogram in estimating patient outcomes is higher, which strengthened the principle of shared decision making between the patient and clinician. Therefore, a combination of a nomogram–machine learning (NomoML) predictive model may help to improve care, provide information to patients, and facilitate clinicians in making RP-LPS management-related decisions.