A Study on Survival Analysis Methods Using Neural Network to Prevent Cancers.

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Cancer remains a significant global health challenge.
• Early, personalized cancer risk prediction is vital for at-risk populations.
• This research addresses the need for advanced predictive tools.

Purpose of the Study:

• To introduce a novel cancer prediction model using recurrent survival deep learning.
• To evaluate the model's predictive performance across ten different cancer sites.
• To compare the novel model against established survival analysis methods.

Main Methods:

• Utilized a large cohort of 160,407 participants from the Korea Cancer Prevention Research-II Biobank.
• Employed advanced recurrent survival deep learning algorithms (nDeep).
• Compared predictive performance using the concordance index (c-index) against Cox PH regression, DeepSurv, and DeepHit.

Main Results:

• The novel deep learning models achieved a concordance index (c-index) exceeding 0.8 for all ten cancer sites.
• A peak c-index of 0.8922 was observed for lung cancer prediction.
• The proposed models demonstrated superior predictive accuracy compared to Cox PH regression and other deep learning survival models.

Conclusions:

• This study presents a state-of-the-art survival deep learning model for cancer prediction.
• The model exhibits the highest predictive performance on censored health data to date.
• Future work will explore causal relationships to further reduce cancer incidence and mortality.