Home

Goals

Risk analysis	Dataset	TFM	Setup
single-risk	Metabrics	TabPFN	Neural FG single (emb)
			Cox (emb)
			XGBoost-Survival (emb)
			Survival Stacking
			Neural FG single (data)
			Cox (data)
			XGBoost-Survival (data)
		TARTE	Neural FG single (emb)
			Cox (emb)
			XGBoost-Survival (emb)
			Survival Stacking
			Neural FG single (data)
			Cox (data)
			XGBoost-Survival (data)
	Metabrics vs. SUPPORT vs. PBC vs. Combination of datasets	TabPFN	Best 1
			Best 2
			Neural FG single (data)
			Cox (data)
			XGBoost-Survival (data)
		TARTE	Best 1
			Best 2
			Neural FG single (data)
			Cox (data)
			XGBoost-Survival (data)
Competing risks	Framingham + PBC	TabPFN	Best 1
			Best 2
			Neural FG (data)
			Cox (data)
			RSF (data)
		TARTE	Best 1
			Best 2
			Neural FG (data)
			Cox (data)
			RSF (data)

Baselines

Classical Models

• Cox Proportional Hazards Model
  • [Single-risk setting] https://www.jstor.org/stable/2985181?seq=1
  • [Competing risks setting] https://www.jstor.org/stable/2335290?seq=1

Neural Network Models

• Neural Fine-Grained (Neural FG)
  • [Single / Competing risks] https://arxiv.org/pdf/2305.06703

Tree-Based Models

• XGBoost Survival

  • [Single-risk setting] https://tdhock.github.io/assets/papers/Barnwal2022jcgs.pdf

• Random Survival Forest (RSF)

  • [Competing risks setting] https://ishwaran.org/papers/IGKMGL.biostatistics2014.pdf

Metrics

brier score, integrated brier score, time-dependent AUC, concordance