How AI can positively change the clinical trials industry

As AI continues to transform the way we all live, we are seeing that the impact of the technology is equally profound in the healthcare industry, specifically in clinical trials.

Data is a fundamental part of clinical trials and the sheer volume of data being generated is staggering. Trials today collect seven times more data than trials of 20 years ago , and that data comes from a wide variety of sources – data collected at sites, images, sensors, genomics and patient-reported outcomes, among others.

Trials have also become more complex, as therapies that are being developed have become more complex and targeted, making it more challenging to successfully run and complete trials. It’s no wonder that the majority of studies fail to meet enrolment timelines.

With all of this data and complexity, AI has the potential to support those at the front line of trials to make better decisions. However, AI must be deployed correctly, on top of sound data, and in partnership with the people and teams working to improve clinical trials.

Successful clinical trials start with better design, but many design decisions today are made with limited information. With strong data from previous trials and patients, we can deploy AI to uncover new insights in diseases to improve the design of clinical trials and reduce the risk for patients.

The data can be used to simulate trial outcomes and optimise design, identify country and site footprints to maximise enrolment, and support more diverse recruitment in clinical trials. We are rapidly moving to a future where a trial design can be simulated end-to-end and optimised before a patient is ever enrolled.

AI can enable investigators to make better decisions around how they’re treating their patients.

As we move towards an era of precision medicine, where individuals get specialised treatment, researchers must understand patient populations and the likely effects of medical treatments at a more granular level.

AI can combine data from numerous sources including clinical, genomic and laboratory data. This allows clinicians to identify patients who may be right for a clinical trial, which not only helps to enrol patients faster but also recruit populations who are most likely to respond positively to treatment, resulting in better outcomes.

The simulation of outcomes also allows clinicians to identify patients who might be at risk of a serious adverse event in a clinical trial, allowing for early intervention. For example, at Medidata, our teams were able to show differences in common biomarkers that could predict risk for severe cytokine release syndrome (CRS) – a life-threatening complication of CAR-T therapy.

In this case, historical clinical trial data combined with advanced AI predictive modelling is being used to develop safer trials for patients who receive these innovative therapies and increase the success of these trials (see examples from ASCO and ASH).

Data management tools supplemented with AI can also be enormously beneficial. As mentioned, more data from more sources are pouring into clinical trials today than ever before.

Data managers are tasked with integrating this data, identifying anomalies and quality issues, and creating clean patient-level data sets as fast as possible to support internal analysis and regulatory documents. AI can make this job much easier.

AI can be used to identify missing information and discrepancies in the data, to automate the reconciliation of data from different sources, and to auto-code data to defined global standards such as MedDRA and WHODrug.

By using these tools, data managers are given the opportunity to focus on the output and uncover deeper insights rather than spending time on the set up and manual reconciliation of data.

This new data also provides different opportunities for insights. At Medidata we are doing significant work on sensor data processing and the discovery of new digital biomarkers in areas like functional capacity measures using motion-based sensors (e.g., an accelerometer), ambulatory ECG and movement quality.

The algorithms we are developing can compute objective measures of cardiac effort, gait and gait asymmetry, postural sway, and activity complexity.

These types of analytics are being applied to patients with heart failure and pulmonary hypertension, as well as in central nervous system diseases, including Parkinson’s disease, multiple sclerosis and Huntington’s disease bringing a richer and fuller view of the patients to clinical scientists.

These are just a handful of the many uses for AI that we are seeing to improve clinical trials. There is, of course, much more work to be done, and potential challenges ahead. As we deploy more AI as an industry, it is important to ensure we are doing so ethically and thoughtfully.

- We need AI that supports the humans - clinicians, patients, data managers and more – to make better and more informed decisions. To do this, we must make AI as interpretable and understandable as possible and build it into solutions that people will use.

- We need to watch out for models that promote or perpetuate bias. It’s important to use high quality and representative data and to ensure a series of checks are in place to minimise potential bias.

- We need to take core ethical issues, like the right to privacy and security, seriously in the design and deployment of AI and model and build them in from the beginning.

If we take these issues seriously, we can harness the immense power of the new healthcare data that is being created every day. Ultimately, this will deliver better outcomes for clinicians, trials and most importantly patients.

As we continue to research more complex diseases, AI will play a transformational role in connecting data collected across sources. Such abilities can create a 360-degree view of patient and disease – uncovering unique insights in support of better treatments.

Fareed Melham is Senior Vice President, Head of Medidata AI at Medidata Solutions. Go to medidata.com