Scientists develop new way to determine which patients will respond best to bowel cancer treatment

• New AI method predicts which patients with advanced bowel cancer are unlikely to respond to the targeted drug bevacizumab – a drug recently approved for use on the NHS.
• There are limited treatment options for bowel cancer once it spreads. Bevacizumab extends life but only works for a small group of patients.
• The AI method combines large amounts of biological data with clinical outcomes to predict which patients are likely to benefit, meaning some patients could be spared unnecessary treatment and side effects.
• Bowel cancer cases are rising, with nearly 10,000 cases of advanced bowel cancer diagnosed in England each year. Worldwide, this approach could impact around 500,000 patients per year.

Scientists have developed an AI-powered method that could determine which patients with advanced bowel cancer are most likely to respond to a targeted drug used on the NHS – potentially sparing thousands of patients from treatments that won’t work for them.

Nearly 10,000 cases of advanced bowel cancer are diagnosed in England each year, with cases in young adults rising. There are limited options for treating advanced bowel cancer.

The targeted drug, bevacizumab, was approved in December for treating advanced bowel cancer patients on the NHS. It slows the growth of cancer, but it only works for a small group of patients and carries the risk of serious side effects including high blood pressure, gastrointestinal problems and blood clots. 

Now, scientists at The Institute of Cancer Research, London and RCSI University of Medicine and Health Sciences, Dublin, have developed a method to identify the patients most likely to benefit from the drug, and those least likely to respond. In the future, this approach could spare these patients from side effects associated with a treatment that won’t work for them.

By identifying the patterns linked to resistance, the researchers hope this could also lead to new treatments for these patients.

In research published in the journal Scientific Reports, the team studied 117 European patients who had been treated with bevacizumab and chemotherapy.

The work was funded by The EU Horizon 2020, Research Ireland, the Ian Harty Charitable Trust, and The Institute of Cancer Research (ICR).

The team used an artificial intelligence tool developed at the ICR called PhenMap – short for phenotype mapping – to integrate complex data on the genetic make-up of the tumour, with clinical information including gender, age, and which side the tumour was on.

They used this to search for new biological signals – patterns relevant to a patient’s response to bevacizumab.

Until now, scientists have grouped cancers by a small number of subtypes. PhenMap can pick up more complicated patterns and narrow these groups, putting patients on a scale of one to 100, for example.

Based on the patterns from PhenMap, another AI tool then generated a score to indicate the risk of dying after treatment with bevacizumab and chemotherapy.

Each patient was allocated to either ‘high’, ‘moderate’, or ‘low’ risk. The highest 10 per cent of risk scores were placed in ‘high risk’, the lowest 10 per cent in ‘low risk’, and the rest placed in ‘moderate risk’.

Looking at the clinical outcomes, the researchers noted that none of the patients in the ‘high risk’ group responded to the treatment.

The complex pattern of features present within ‘high risk’ patients could be used as a biomarker, for clinicians to identify patients who are unlikely to respond to bevacizumab.

One of the patterns identified by the AI was that patients with a mutation in the BRAF gene were all in the high-risk group and had poor outcomes.

The next stage for the research will be to validate this in more patient samples, and to develop the method into a test that could be used in a prospective clinical trial, to help guide treatment decisions.

The researchers will also explore whether the test can predict response to other targeted therapies, and they believe that the method could be applied to other cancer types.

Professor Anguraj Sadanandam, Professor in Stratification and Precision Medicine at The Institute of Cancer Research, London, said:

“Once bowel cancer spreads to other parts of the body, there are very few treatment options available for patients. It is therefore positive that patients can now access the targeted drug bevacizumab on the NHS. However, we know that the majority of patients won’t benefit from the drug, meaning thousands of people in England could be facing unpleasant side effects unnecessarily. Until now, we haven’t been able to identify these patients.

“Our research uses advanced AI methods to pull together large amounts of complex data, helping us to spot patterns that would otherwise be impossible for a human to see, and to uncover the clues hidden within a patient’s tumour. In our research, we have shown that this allows us to identify the patients least likely to respond to treatment with bevacizumab. While these findings are encouraging, they will need to be validated in a larger cohort, to ensure they are applicable to all patients.

“In future, I hope this approach will lead to a test that can be used by clinicians, to ensure patients receive personalised care that has the highest chance of working against their cancer.”

Professor Kristian Helin, Chief Executive of The Institute of Cancer Research, London, said:

“The approval of new drugs to treat cancers is a significant milestone, but we must recognise that one drug won’t work for everyone – understanding why certain patients won’t benefit from the treatment is crucial to improving outcomes.

“AI has revolutionised cancer research – by enabling us to rapidly analyse large, complex datasets and predict how patients will respond to treatment. This research is a powerful example of how the ICR is leveraging AI to develop smarter, kinder therapies, and deliver them to patients sooner.

“This approach also has the potential to be explored in many cancer types, and it will be interesting to see whether the method can predict responses to other targeted therapies across a range of cancer types.”