Deep Neural Networks for Survival Analysis Using Pseudo Values.

Area of Science:

• Medical Research
• Biostatistics
• Machine Learning

Background:

• Survival data analysis is crucial in medical research.
• Traditional methods often struggle with censored data.
• Deep learning offers potential but requires specialized cost functions for survival data.

Purpose of the Study:

• To propose a novel, simplified deep learning method for survival data modeling.
• To reduce the complexity of survival analysis to a standard regression problem.
• To enhance the flexibility and practical applicability of deep learning in survival prediction.

Main Methods:

• A two-step approach is introduced for survival data modeling.
• Survival times are transformed into jackknife pseudo conditional survival probabilities.
• These probabilities are used as response variables in a deep neural network.

Main Results:

• The proposed method simplifies complex survival analysis into a standard regression problem.
• Deep neural network construction is significantly simplified.
• The approach offers a flexible and practical solution for risk prediction in survival data.

Conclusions:

• The novel two-step deep learning method effectively models survival data.
• This approach simplifies neural network construction and enhances flexibility.
• The method provides a practical and appealing solution for survival risk prediction.