Prediction of survival with alternative modeling techniques using pseudo values

Area of Science:

• Biostatistics
• Machine Learning in Medicine
• Survival Analysis

Background:

• Censoring in survival data complicates alternative modeling techniques.
• Pseudo values offer a solution for statistically appropriate survival outcome analysis.
• This study validates pseudo values across seven distinct modeling techniques.

Purpose of the Study:

• To demonstrate the utility of pseudo values in alternative survival modeling.
• To compare the performance of various models using pseudo values.
• To assess predictive accuracy for Head and Neck Squamous Cell Carcinoma (HNSCC) survival.

Main Methods:

• Analyzed survival data from 1282 HNSCC patients.
• Calculated pseudo values to represent individual survival patterns.
• Compared Kaplan-Meier, Cox regression, RPART, NNET, LR, GLM, and SVM models using AUC and RMSE.

Main Results:

• Logistic Regression (LR) and Support Vector Machine (SVM) models showed high predictive accuracy for 60-month survival.
• General Linear Models (GLM) and SVM demonstrated strong performance for continuous survival predictions.
• Predictor importance varied significantly based on the survival aspect and modeling technique.

Conclusions:

• Pseudo values facilitate the application of diverse modeling techniques to survival data.
• This approach enables robust comparison of model performance.
• Pseudo values open avenues for exploring novel survival analysis methods.