New concordance loss improves deep survival prediction models

Researchers have developed a new concordance loss function, called SCL (Sigmoid Concordance Loss), for deep survival prediction models. Traditional models often use likelihood objectives that do not reliably correlate with the concordance index (C-index), a key metric for evaluating survival predictions. SCL ensures that a decreasing loss value directly corresponds to an increasing C-index, providing a more stable and reliable optimization objective. Experiments across eighteen datasets demonstrated that SCL achieves comparable discrimination to standard likelihood losses while maintaining a strong correlation between loss and C-index during training. AI

IMPACT Provides a more reliable optimization objective for deep survival models, potentially improving their performance and training efficiency.

RANK_REASON Academic paper detailing a new loss function for deep survival prediction models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• Cox Communications
• DeepHIT: a deep learning framework for prediction of hERG-induced cardiotoxicity
• Sculptor
• support vector machine

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →