By Mark Brown and Lucas Glass
The business of finding new therapies is fraught with risk. The process is slow and expensive. It takes an average of 10 years to get a product approved. Patients pay the price in the time lost getting much needed therapies to market. Enter big data, artificial intelligence and machine learning. How we curate, synthesize, and apply data to our decision-making is a game-changer, and it is poised to fundamentally transform clinical development as we know it. However, to successfully change the healthcare space, we must keep our eye on the clinicians and medical experts and on how technology can support them.
Reducing risk and saving cost
One of the most expensive elements of clinical development is the length of time it takes to find and enroll patients in studies — 80 percent of studies miss timelines. Under the traditional model, a panel of disease experts and clinical trial gurus convene and try to predict trial length. Based on data from IQVIA’s Study Optimizer software, sponsors of clinical trials predict study timelines with an average 55 percent error rate.
There is a wide range of data and information that can support the estimation of enrollment rates. However, the human brain cannot analyze the vast amount of data available today. Machine learning techniques and augmented intelligence can significantly improve the prediction of study timelines. For example, we at IQVIA™ leveraged large nonidentified public and proprietary datasets and reduced the error in timeline prediction from 55 percent to less than 30 percent, and this is just a start. As the models continue to adapt, learn, and improve their accuracy, the goal is to achieve an error rate lower than 20 percent.
Additionally, these algorithms help sift through the massive amounts of information to identify the key data points that are most likely to influence the enrollment rate. For example, a recent publication on a novel mechanism of action might have substantial impact on enrollment rates. AI can monitor for activities that would have such an impact, change the estimates of the enrollment rate, and highlight factors affecting enrollment, such as a journal article, for medics building the strategy. In the long term, arming clinical trial feasibility experts with this type of augmented intelligence will help sponsors limit their financial risk.
Identifying patients sooner
What if we had the capability to prevent diseases before they happened? Many diseases have specific symptoms preceding diagnoses and treatments that can help us identify patients at high risk. Using deep learning and AI technology, it’s possible to identify sites and investigators with a high number of patients at risk and enroll them in appropriate trials. These disease detection algorithms can significantly improve enrollment success. For example, in a retrospective analysis of a difficult-to-enroll Alzheimer’s study, deep learning phenotyping improved screening precision from 20 percent to 79 percent.
Machine learning technology can identify the right patients for clinical trials. This technology not only leads to improved clinical trial success, the technology has the ability to integrate clinical research with clinical care, augmenting the physicians’ toolkit to better treat their patients. By partnering with hospitals and healthcare organizations, solutions for disease detection and patient referral could be installed on-site. Algorithms could quickly analyze complex phenotypes in a patient’s record, match the patient to the optimal trial, and refer the patient to the doctor in real time.
While there are some existing solutions that can suggest diseases from a patient profile, proper integration into the clinical care setting is essential and challenging. A more robust solution with referral capabilities — one that focuses on supporting the clinician and one that understands how clinicians operate — ultimately benefits the patient, the physician, and the trial sponsor.
Speeding patient recruitment with precise site recommendation and AI
A significant portion of trials miss timelines because of the difficulty in recruiting optimal investigators. Identifying the right investigator is critical for clinical trial performance. When it comes to selecting the right investigator partner, we want to know if the trial is appropriate for him/her and if he/she will successfully enroll patients. It is also helpful to select an investigator with a background of strong collaboration with other doctors to help enhance patient recruitment through patient referral. Again, the amount of data and information available to support this decision-making has gotten too big. Machine learning can synthesize the massive data and help generate the optimal investigator list. More importantly, matching the right physicians with the right clinical trials puts the clinicians in a position to succeed in the research setting and thereby further the aim of clinical research as a care option.
Why AI has finally arrived in healthcare
Pharmaceutical companies spend billions of dollars every year on research and development. A significant portion of that investment is in clinical development. The ability to more accurately forecast trials, and to increase the speed, efficiency, and precision of patient recruitment has the potential to bring enormous financial relief to the industry and, most importantly, bring much-needed therapies to market sooner.
The old way of doing things is not sustainable, and the industry is undergoing a vast transformation. This transformation and innovation will be powered by computer scientists working closely with clinicians through the application of artificial intelligence and machine learning to improve and accelerate development of new therapies.
Mark Brown is VP, Global Patient and Site Solutions, IQVIA and Lucas Glass is Global Head, Analytics Center of Excellence, IQVIA.