May 2025 • PharmaTimes Magazine • 22-23
// AI //
Enemy of the state
What does AI mean for the future of cancer care?
If you ask AI about the future of cancer care, it replies that it will ‘make cancer care more precise, proactive and patient-centred, leading to better outcomes and a higher quality of life’. But is it telling the truth?
A look at the research and implementation of AI in the cancer space certainly points towards this potential. Oncology, with its wealth of data, is primed to utilise AI. Indeed, there are many examples of AI already enhancing cancer detection, care and treatment.
It has also been suggested that the status of AI in this field is a strong indication of its utility in disease areas that have smaller patient data sets to work from.
Globocan estimated that there were almost 20 million new cases of cancer in 2022. That number is predicted to rise to 35 million by 2050.
One reason why we see multiple applications of AI in oncology – including in imaging, genomic profiling, and the digitalisation and storage of medical records systems – is that the prevalence of cancer, and the size of cancer data sets that have been generated, is huge.
While cancer is complex, with many subtypes and variations, this is beneficial in terms of AI, particularly if it is being trained to identify patterns and make predictions.
Right now, AI is being harnessed to detect cancer earlier. It is supplementing the accuracy of diagnostic test reviews, supporting an overstretched workforce, and enabling greater risk stratification.
For example, the use of AI in mammography for breast cancer detection has transitioned from research on retrospective samples to clinical integration in some settings. Recently, Malaysia published a proposed approach to facilitate screening using AI-assisted chest radiography to detect lung cancer earlier.
We have also seen the potential of AI to address workforce capacity challenges. It is being trialled as a ‘first reader’ in both lung and breast cancer screening, ruling out negative cases without substantially increasing the risk of missed referrals.
Efficient and accurate
There is growing evidence about the use of AI in cancer diagnosis. It has been used to inform the grading, staging and classification of images.
Deep learning has also gained attention due to its potential to enhance image analysis and molecular profiling. It has been used, for example, to classify invasive and non-invasive breast cancer subtypes, which can increase the speed and accuracy of diagnosis. We may therefore expect to increasingly use AI to gain a more comprehensive understanding of a person’s cancer – as well as an understanding of cancer more broadly – via analysis of genomic, image and clinical data.
Large language models (LLMs) are also being explored as a way to improve cancer diagnosis. These tools may help clinicians rule out cancer, thus avoiding unnecessary tests.
LLMs may also provide more accessible information to patients. However, biases in LLMs’ medical decision-making have been observed, and there are concerns that these biases could exacerbate health disparities.
Informing personalised treatments
The fast-evolving nature of cancer treatments – with increasingly personalised treatments via genomic testing and profiling – means that the pool of data for training AI becomes smaller and more specific.
‘While cancer is complex this is beneficial in terms of AI, particularly if it is being trained to make predictions’
However, given AI’s ability to analyse individual patient characteristics, it could inform more personalised treatment plans, increasing the likelihood that treatments will be effective and have minimal side effects. The potential of AI to predict immunotherapy biomarkers, treatment response and prognosis also looks positive.
Furthermore, AI is being used to support treatment decision-making, optimise and streamline administrative workflows, and strengthen patient engagement.
One example is its use in radiotherapy – automatically contouring areas of risk – which can reduce procedure time and interobserver variability.
Regulation and guidance
Alongside the excitement around the application of AI in the cancer space, the need for regulation and guidance on its use must be recognised at the governmental level.
With the fast pace of technological development, it is vital that this guidance be regularly reviewed to reflect the realities and challenges of AI regulation and governance.
In Europe, the use of AI is primarily regulated by the EU AI Act, which is designed to be updated in line with innovations.
The US takes a less centralised approach with a combination of federal guidelines, state laws and sector-specific regulations, although these have the potential to change dramatically depending on who is in power.
Given the global nature of some AI solutions, it will be imperative to address international regulatory disparities.
Uncoordinated investment
The use of AI in cancer risks uncoordinated and unbalanced investment. There are many cases of financial commitment to support the expansion of AI capability, but also cases where these intentions fall foul of budget cuts.
The recent decision by the UK government to withdraw funding for AI cancer technology for radiotherapy in England has been widely criticised. Radiotherapy UK has calculated that it will add 500,000 extra days to already extensive NHS waiting lists.
Without sustained and considered investment, we are likely to see the development of AI solutions limited to the private sector, potentially restricting opportunities for governments to strategically harness AI to address country-specific challenges.
Importance of representative data
It is important to understand the limitations of the data being used to inform AI, particularly if traditionally underrepresented groups are excluded from data sets and bias is exacerbated by AI algorithms.
For example, if an AI data training set contains predominantly light-skinned patients with melanoma, it is likely to be less accurate at diagnosing melanoma in dark-skinned people. LLMs can also ‘hallucinate’, producing outputs that are statistically likely but not factually correct, which could be extremely problematic in a health setting.
Alongside this, questions and debate often arise about the use of patient data. People have varying levels of comfort about the use of AI, and patients must be treated as key stakeholders in its research and deployment – especially as it is their data that enable its successful use.
Some patient groups, including ‘use MY data’ and ‘Data Saves Lives’, actively engage on this topic. Collaboration with such groups could encourage patients, carers and their families to become more deeply involved in informing decisions about the appropriate use of patient information.
Work in progress
With appropriate integration, AI could play a key role in revolutionising cancer care. But for this powerful tool to have maximum impact, governments and healthcare providers must prioritise the development of unbiased data sets, establish robust yet flexible regulatory frameworks and commit to empowering patients to become active participants in how data is used.
The wealth of research and real-world implementation of AI demonstrate its potential to transform cancer care, and it looks particularly promising in the imaging and ‘big data analytics’ spaces.
But there is certainly still a lot of work required before we consider AI a realistic replacement for a healthcare professional.
Eleanor Wheeler is Director of Consulting, Oncology at The Health Policy Partnership
PharmaTimes
PharmaTimes is the UK’s leading pharmaceutical magazine, tracking the trends and issues that affect the industry here in the UK, in Europe and across the world. We deliver independent, authoritative and trustworthy content through impartial, balanced and informed opinion and analysis.