Statistical analysis was performed using SPSS v23 and R v4.20. Continuous variables are presented as the mean and standard deviation, and categorical variables are presented as frequency or percentage. Comparisons of continuous variables between two groups were performed with the t test, and categorical variables were compared using the chi-square test or Fisher’s exact test. p-Value < 0.05, two-tailed, was considered statistically significant. Kaplan–Meier survival plots were generated with the log-rank statistic using the survival package of R.
We first screened the clinicopathological parameters associated with the therapeutic effects of second-line treatment with axitinib using univariate Cox proportional-hazards regression (CPHR) analysis. Because CPHR is not used to analyze multidimensional survival datasets, the least absolute shrinkage and selection operator (LASSO) technique was subsequently performed for variable selection and shrinkage from many clinical variables identified by univariate CPHR, using the glmnet package of R (8 (link), 14 (link)). Finally, we identified and then established a predictive model based on four clinical parameters (albumin, calcium, IMDC grade, and adverse reaction grade) through multivariate CPHR.
To evaluate the predictive accuracy of the model, time-dependent receiver operating characteristic curve (ROC) and area under curve (AUC) at 3-, 6-, and 12-month PFS after axitinib treatment were constructed using the survival package of R. Concordance index (C-index) is used to evaluate predictive accuracy (15 (link)). The consistence between predicted PFS probability and actual PFS probability was confirmed using a calibration curve after 1000 bootstrap resampling. The ROC curve and AUC are not used to make clinical decisions. In clinical practice, decision curve analysis (DCA) was used to estimate the net benefit for patients based on threshold from the predictive model.
We first screened the clinicopathological parameters associated with the therapeutic effects of second-line treatment with axitinib using univariate Cox proportional-hazards regression (CPHR) analysis. Because CPHR is not used to analyze multidimensional survival datasets, the least absolute shrinkage and selection operator (LASSO) technique was subsequently performed for variable selection and shrinkage from many clinical variables identified by univariate CPHR, using the glmnet package of R (8 (link), 14 (link)). Finally, we identified and then established a predictive model based on four clinical parameters (albumin, calcium, IMDC grade, and adverse reaction grade) through multivariate CPHR.
To evaluate the predictive accuracy of the model, time-dependent receiver operating characteristic curve (ROC) and area under curve (AUC) at 3-, 6-, and 12-month PFS after axitinib treatment were constructed using the survival package of R. Concordance index (C-index) is used to evaluate predictive accuracy (15 (link)). The consistence between predicted PFS probability and actual PFS probability was confirmed using a calibration curve after 1000 bootstrap resampling. The ROC curve and AUC are not used to make clinical decisions. In clinical practice, decision curve analysis (DCA) was used to estimate the net benefit for patients based on threshold from the predictive model.
Free full text: Click here
