Fig. 3

Selection of candidate items. A The predictive factors were determined by cable regression method. B Adjusting the penalty coefficient in the LASSO model using cross validation and minimum criteria. The vertical black line represents the optimal lambda (i.e., the model provides the best fit with the data). C The importance of the candidate items in PG-SGA and QLQ-C30 for individual prognostic evaluation. The relationship between the error rate and the number of classification trees (left). The variables are ranked in terms of importance (right, red rectangle: minimal variables; black rectangle: optimal variables). D The curve of risk score, survival status of the patients and heat map of optimal variables scores were shown