Fareed Melhem talks about how artificial intelligence can help to reduce the burden of trials on patients
Artificial Intelligence (AI) is the technology everyone is talking about. How important do you think it will be for clinical research and healthcare?
AI is changing the world around us and clinical trials will be no different; we are already seeing real progress being made. At the heart of clinical trials is the collection, synthesis and analysis of complex data from multiple sources. AI will play a major role here.
As part of clinical trials, we also generate and analyse a large number of documents and images. Again, AI will be key to this.
Finally, and most importantly, clinical trials are about patient care and, even here, AI will play a role, helping us reduce the burden of trials on patients and giving investigators and clinical staff the tools to support better patient care.
With so many complex disease areas out there, it’s exciting to see how AI is helping us to create a holistic picture of patients and disease areas, increasing the chances of being able to meaningfully treat patients and improve health outcomes.
How is AI impacting clinical trial design?
Clinical trials still have a high failure rate and often it’s the choices made in the design of studies that lead to success or failure. However, these choices are often being made with limited information.
By harnessing the available data from previous trials and patients, AI can provide important insights into patient populations, treatment effects and likely outcomes. In addition to the scientific design of studies, we are also using models to minimise the operational burden of trials on sites and patients. Our models show that when you reduce the operational burden of a trial on patients – for example, by minimising visits or painful procedures – enrolment increases and patient retention improves. By putting these pieces together, we can use data and AI to generate better designs, reduce patient burden and optimise trial planning.
We are rapidly moving towards a digital twin of the trial, where we will be able to generate protocols, simulate the entire trial based on the design, test different scenarios and significantly increase the likelihood of success.
As we move forward, we will also see AI increasingly used to identify new forms of evidence, such as digital biomarkers. We have seen an increased use of sensors in clinical trials and this kind of continuous data can provide insights that cannot be seen through a standard visit schedule. AI applied to the data coming in from these sensors will allow us to develop more objective measures of patient health in areas like cardiac effort, gait and activity complexity.
What role do you see for AI in data management?
Data management is becoming increasingly challenging as new technologies and study designs have expanded the types of data collected. Today’s trials collect data from multiple sources, including sites, sensors, wearables, labs, omics, imaging and also directly from patients.
As a result, data managers have a more complex role to play than ever before, as they need to integrate this data, identify anomalies and ensure quality. AI can provide support to data managers in this process, identifying missing information and discrepancies in the data, supporting auto-coding of data to global standards such as MedDRA and WHODrug, and supporting the generation and follow-through on queries to ensure quality standards are met. As a result of AI alleviating the burden on data managers in this new era, they can focus on outputs and delivering important trial insights.
Does AI support other clinical trial solutions, such as synthetic control arms?
Synthetic Control Arms (SCAs) are not an AI solution, but are powered by statistical methods and can act as a ‘virtual twin’, providing a solution where data from past participants is utilised as the control arm in a clinical trial, removing the need to enrol all or part of that population.
SCAs are becoming more accepted by regulators in certain disease areas where it is hard to recruit into a control arm due to the rare nature of the disease, or where it may be unethical. We have successful examples of significantly reducing or eliminating the number of patients recruited into the control arm. Patients joining these trials, especially those with aggressive diseases, are searching for hope. The more that we can help patients get the experimental treatment that they’re really hoping for, the better, and SCAs will increasingly play a vital role in facilitating this.
You mentioned imaging earlier, which plays an important role in clinical research in some therapy areas. To what extent can AI help this process?
AI shows tremendous promise for supporting imaging throughout all phases of clinical research. Many imaging steps in clinical trials, including image quality control and interpretation, are still performed manually, which is time-consuming and error-prone. AI can support automated text detection and protected health information (PHI) removal to reduce PHI in images and flag quality issues, helping to identify deficiencies more consistently and earlier in the process, ensuring the highest quality imaging trial data.
AI can also support radiology assessments, reducing read times and improving quality. Radiologists are one of the most expensive resources on a trial and standardised assessment criteria like RECIST used on oncology trials can leverage AI to help identify and measure lesions, saving the radiologist time and reducing the variability that is often seen in these assessments.
AI clearly has huge potential when it comes to clinical research, but do you have any concerns about the use of this technology?
As we continue along the journey of adopting AI in the healthcare industry, there will certainly be challenges and there are things that we need to keep in mind as we move forward.
We must ensure we are prioritising ethical principles. Issues such as the right to privacy should be taken seriously in the design of AI models and be front of mind throughout its deployment.
Likewise, we have seen that some models can end up perpetuating biases by learning from the wrong inputs. We must use high-quality and representative data to reduce this risk.
Finally, we need to ensure that interpretability and understandability are at the forefront of models so that users – investigators, patients, study managers, data managers and others – are confident in their outputs and can use them to augment their work.
If we can do these things, I am confident that AI will continue to play a key role in delivering better outcomes in clinical trials and healthcare more generally.
Looking ahead, do you have any thoughts on the future of AI in clinical research?
Moving forward, the focus will increasingly turn to finding new and innovative treatments for complex diseases. AI will play a pivotal role in allowing us to do this. As we consider patients moving between clinical trials and the real world, AI will allow us to bring together patient data in a holistic way. Clinical trial data, real-world data, data collected via sensors, genomic data and more can all be combined to give scientists, investigators and doctors a real 360-degree view of a patient.
With this full view of the patient, trial outcomes and real-world outcomes can be optimised to benefit patients. Harnessing AI will hopefully lead to better research and therefore better treatments.
