Background
Effective recruitment of patients remains one of the most significant hurdles to running a successful clinical trial. Issues with patient enrolment and retention are well-known as leading causes behind the termination and delay of trials globally. Further, low levels of diversity amongst trial participants remains an enduring issue in the life sciences space. There are a myriad of reasons behind these struggles with recruitment and retention in clinical trials, including a lack of awareness about clinical trials amongst the general public, issues around the availability, location and quality of trial sites, and strict patient eligibility criteria which fail to engage a variety of potential patient demographics. In the UK these issues are perpetuated by a systemic struggle to prioritise research within the NHS, contributing to lack of conversation about clinical trials between patients and clinicians (particularly where those patients belong to minority groups).
The AI upsides
The introduction of artificial intelligence into the patient recruitment process could help to address some of these issues.
Consolidate patient data
First, large language models can be used to consolidate patient data, which can then be used to identify clinical trials for which patients are eligible. This would streamline the clinical trial recruitment process and improve efficiency within healthcare systems. One such AI algorithm has been developed by the US National Institutes of Health and is currently being assessed for application in real-world settings.
Improve trial eligibility criteria
However, the potential for AI in clinical trial recruitment goes beyond consolidating and scanning patient data, it can also be used to improve trial eligibility criteria. A study by Liu et al (subscription required) found that by using an AI framework to analyse the data found in electronic health records of patients with advanced non-small cell lung cancer, the eligibility criteria for trials could be broadened without a significant impact upon trial hazard ratios. This means that the pool of patients eligible for trials could be expanded significantly, helping to ease struggles with patient enrolment. There is also potential for AI to support the improvement of diversity in clinical trials by using health records to identify prospective patients from different backgrounds.
Synthetic control arms
Within clinical trials themselves, AI can be used to create synthetic control arms to fill the role that would ordinarily be played by placebo groups. These synthetic control arms are created by training AI models on data from historical patient records (meaning that they would be of particular use for trials with historically consistent control group results). This could ease clinical trial recruitment pressures by reducing the number of patients to be recruited, and the reduction in the number of patients would also save on the costs of clinical trials. Synthetic control arms also have the potential to circumvent ethical complaints raised in relation to control groups. Potential participants, understandably, may feel disincentivised to participate in a clinical trial because of the possibility that they will be selected for a control group and will only receive a placebo throughout the trial. Synthetic control arms could remove this apprehension and may indirectly ease the process of clinical trial recruitment.
Possible concern: ethical and diverse AI training data
Underpinning all of the above is an assumption that the relevant AI tools are trained on appropriate data. This is a significant point of concern: if AI is to be used to improve diversity in clinical trials, for example, it will not be of significant use if its training data consists of electronic health records which do not include certain demographics. If historical data introduces bias into AI tools, they may only exacerbate some of the problems they are intended to address. The use of patient data to train these AI tools also raises a number of obvious ethical and privacy concerns. It is worth noting, however, that forthcoming reforms look set to broaden both the definition under data protection law of 'scientific research' and the circumstances under which prior consent to research can be relied upon for new purposes. Research and regulation in this space should progress with a view to addressing these risks.
Conclusion
Recruitment for clinical trials is an especially pertinent issue in the UK, which has seen a drop of around 9% in recruitment for commercial clinical trials from 2022/2023 to 2023/2024. This statistic reflects the stagnation of the life sciences sector in the UK more broadly, notwithstanding its excellent capabilities. In July 2024 Richard Torbett, Chief Executive of the Association of the British Pharmaceutical Industry, highlighted "embedding innovation in healthcare" as one of the biggest hurdles to competing effectively in the sector. The adoption of AI patient recruitment technology could play a part in making the UK a more attractive prospect for running commercial clinical trials.