# Creating strong predictive models in oncology

## TL;DR

Many cancer prediction models don't help patients due to bias, weak studies, and lack of real-world impact.

## Contribution

The paper outlines a path to better models by focusing on actionable questions, rigor, and generalizability.

## Key findings

- Current models often fail to improve patient care due to bias and underpowered studies.
- Emphasizing clinically actionable questions can enhance model effectiveness.
- Generalizability and rigor are key to achieving better clinical impact.

## Abstract

Many oncology predictive models fail to improve care. Issues include risks of bias, underpowered radiomics studies, and limited clinical impact. A path forward involves an emphasis on clinically actionable questions, rigor, and generalizability.

Many oncology predictive models fail to improve care. Issues include risks of bias, underpowered radiomics studies, and limited clinical impact. A path forward involves an emphasis on clinically actionable questions, rigor, and generalizability.