Staying Alive: Uncensored Survival Analysis with Tabular Foundation Models

2026-06-02 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A novel training-free method for survival regression addresses the central challenge of right-censoring in Survival Analysis (SA) by leveraging Tabular Foundation Models (TFMs). SA, widely used in healthcare and churn prediction, often faces partially observed event times due to right-censoring. While TFMs excel at prediction tasks in a single forward pass without dataset-specific parameter fitting, their application to time-to-event data has been difficult. This new approach uses a TFM to construct an Accelerated Failure Time (AFT) model, requiring no training beyond fitting a single scalar parameter. It then introduces a non-parametric in-context estimator, built on the Buckley-James estimator, for imputing right-censored data. Experiments on standard SA benchmarks demonstrate that this method is competitive with several parametric and semi-parametric survival regression models that typically require extensive training, including Cox regression and parametric AFT models.

Key takeaway

If you are a data scientist working with time-to-event data and struggling with right-censoring, consider this training-free method using Tabular Foundation Models. It offers competitive performance against trained models like Cox regression, potentially streamlining your workflow and reducing model training overhead for survival analysis tasks. You can achieve robust predictions without extensive dataset-specific parameter fitting.

Key insights

Tabular Foundation Models can perform uncensored survival analysis without dataset-specific training.

Principles

• Survival Analysis models time until an event, often facing right-censoring.
• TFMs enable prediction tasks in a single forward pass.
• Right-censoring complicates time-to-event data prediction.

Method

The method constructs an Accelerated Failure Time (AFT) model using a TFM, requiring only a single scalar parameter fit, then applies a non-parametric in-context estimator based on Buckley-James for censored data imputation.

In practice

• Apply to healthcare outcomes prediction.
• Use for customer churn forecasting.

Topics

• Survival Analysis
• Tabular Foundation Models
• Right-Censoring
• Accelerated Failure Time Models
• Buckley-James Estimator
• Time-to-Event Data

Best for: Research Scientist, AI Scientist, Data Scientist

Related on AIssential

• A Survival Analysis Guide with Python: Using Time-To-Event Models to Forecast Customer Lifetime — Survival Analysis predicts time-to-event, handling incomplete data, crucial for understanding durations like customer churn.
• TabH2O: A Unified Foundation Model for Tabular Prediction — TabH2O unifies tabular classification and regression via in-context learning, improving training stability and noise robustness.
• When Tabular Foundation Models Meet Strategic Tabular Data: A Prior Alignment Approach — Strategic manipulation creates a prior mismatch for tabular foundation models, which SPN corrects at inference.
• Bootstrapping-based Regularisation for Reducing Individual Prediction Instability in Clinical Risk Prediction Models — A novel regularisation method improves clinical prediction model stability without sacrificing interpretability, crucial for healthcare adoption.
• Adaptive Experimentation for Censored Survival Outcomes — A new adaptive experimentation framework efficiently estimates causal effects from censored survival data using an optimal allocation policy.
• ReSS: Learning Reasoning Models for Tabular Data Prediction via Symbolic Scaffold — ReSS combines decision trees and LLMs to generate faithful, explainable reasoning for tabular data predictions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.