Flinders University researchers are using Machine Learning – a form of artificial intelligence – to help predict who will benefit from specific anti-cancer drugs during treatment, while also predicting who is more likely to be harmed by the same drugs.

Professor Michael Sorich and his team are using higher dimensional data collected from close to 100,000 cancer patients across 80 clinical trials to learn more about the complex relationships between predictors and outcomes than what is currently possible with traditional statistical approaches.

Using a Flinders Foundation Health Seed Grant, Professor Sorich will focus on patients with urothelial cancer – a common form of bladder cancer with aggressive malignancy.

He’s aiming to develop a prediction model utilising routinely available patient and tumour characteristics to accurately identify patient subgroups with substantially different treatment benefit from the drug ‘atezolizumab’.

“The specific focus of this project is on better informing patients with urothelial cancer with respect to their personalised likelihood of treatment benefits and harms, and thereby allowing them to make better treatment decisions,” Professor Sorich explains.

“Immunotherapy drugs called ‘immune checkpoint inhibitors (ICIs), such as atezolizumab have become an important approach in the treatment of metastatic urothelial cancer.

“Nonetheless, the proportion of patients who achieve adequate responses with ICIs is low, and identifying individuals upfront that are most likely to respond may enable better selection of treatment, and for patients to be better informed about the likely benefits and harms of treatment.”

Over the past decade, Machine Learning has been applied to cancer diagnosis and risk assessment but there have been limited applications of Machine Learning in personalising prediction of anti-cancer treatment benefits and harms.

“Although prediction models of outcomes to cancer medicines have been reported, most models are developed manually using traditional statistical approaches, such as regression modelling, to assess a small number of predictors, and have insufficient evidence they are good enough for clinical use.

“There is also no universal prediction model for locally advanced or metastatic urothelial cancer.”

“This project aims to utilise Machine Learning – a form of artificial intelligence – and clinical trial data to develop robust and accurate clinical-grade prediction models of outcomes in patients with locally advanced or metastatic urothelial cancer.”

Research category: Cancer

Project title: Machine learning for personalised prediction of anti-cancer treatment outcomes

Lead researcher: Professor Michael Sorich