Trial and Error

Covid pushes AI driven clinical trials and drug manufacturing into the limelight

It is safe to say that never before has so much of the world’s attention been focused on the clinical trials process. Unfortunately, it is a process beset by uncertainty, high costs, long timelines and all too often failure. ICON PLC figures show clinical trials account for 40 percent of pharma research budgets. And these trials are set to become much more costly as our therapeutic modalities become ever more complex. Even when the biology of a disease is straightforward and well known, the economics of the market plus cost of trials often stops the biopharma industry from developing a drug that might otherwise save millions of lives.

There is substantial opportunity for companies who can bring greater efficiency to the process. In fact, Novartis estimates concluded that companies can eliminate 20% of clinical trial costs if they were to deploy technology at scale. Venture capital firm Andreessen Horowitz projects that a 5-10% improvement on efficiency on cost or time—a Moore’s Law for clinical trials—would mean that in seven years, it might cost half as much to run a new therapeutic through clinical trials, and an order of magnitude more diseases would be cured. The industry has responded by introducing more mobile sensors, apps, and software that leverages electronic medical and administrative records. These devices generate a profusion of data and according to Tom O’Leary, CIO of ICON PLC: “When we combine this data with advanced AI algorithms processing, it has the potential to reduce trial costs. This is achieved by increasing patient compliance and retention, and identifying treatment efficacy more efficiently and reliably than we could previously.”


Patch AI is one company that is focused on increasing trial adherence and they do this through the use of machine learning to feed an engagement algorithm that can predict, based on past interactions with patients, the likelihood that they will drop out of a clinical trial. “We have seen that many of the problems that clinical trials have in common is low patient engagement. Dropout and retention rates are high and are directly linked to patient engagement as well as health outcomes. That is why we are trying to transform the data collection process into something interactive, real time and empathetic” said Alessandro Monterosso, Co-Founder and CEO of Patch AI.

“With the upfront startup costs, I think there is a valid use case in terms of site selection, patient recruitment, using the data to tell you where those populations or sites are going to be most effective and focusing on that, rather than the broad net they cast today. I definitely see site selection patient recruitment being higher quality, hopefully shorter phase, and more hits per encounter to get somebody into a trial.”
Don Ragas, CTO, ERT

Simulating Placebo

Another angle a company can take to improve the quality and efficiency of clinical trials is by limiting the placebo arm of the trial. To achieve this without altering the statistical significance of the trial, Unlearn AI uses machine learning approaches to generate a digital twin, which is a concept taken from engineering where you are able to simultaneously build a device while running computer simulations to see how the device would react under a variety of different scenarios. Unlearn is applying this concept to clinical trials by creating digital twins of patients in order to project how they would react if they were to receive a placebo. This allows more patients to take the experimental treatment and when patients are ill and dying it is far more comforting to know that you are at least receiving something that is attempting to combat your illness. The value proposition of Unlearn’s approach is its ability to run trials with fewer patients and in the most ideal circumstance, Founder and CEO Charles Fisher believes: “you can cut the number of patients by half, because you are running a trial in which every single patient who enrolls receives the experimental therapy, and you are using a computer simulation to provide the control group. With half as many patients, a trial is at least twice as fast. In some cases you can get an even a better acceleration. This allows costs to be cut dramatically.”

Predicting the Responders

Let's say you go into clinical trials with a drug that binds a target. Even when the patients express the target, that alone is not enough to identify if they will respond. For example, immuno-oncology drugs are only effective on a percentage of patients and there are adverse events associated with those who do not respond. The reason these drugs do not work for everybody is because some people do not have the molecular phenotype that is amenable to that particular therapy. Consequently, a patient's precious time is wasted and they do not have a lot of time to waste as the cancer spreads. For this reason, Auransa is using its AI engine to predict which drugs will work for the responder patient populations before enrollment. This allows them to very quickly recruit the right cohort of patients. According to company Co-Founder and CEO Pek Lum “Eventually we hope that we are correct often enough in the clinic that we can immediately recruit only the responding patients. A patient should enroll in a trial only if they know they will respond to the treatment. That is the holy grail of clinical trials.”

Covid-19 Impacts Trials

The Covid-19 pandemic has caused hundreds of trials to stop and in many indications medical research has completely ground to a halt. According to Medidata, There was a 65% worldwide average decrease in new patient enrollment year-over-year during March and although the number of enrollees has now improved, numbers were still down 29% year on year as of June 2020. The challenge of recruiting patients to enroll in non Covid trials has been particularly challenging, because people do not want to deal with the inconvenience of travelling to a site or going to a hospital during a pandemic. This is particularly true of those elderly patients that are at high risk of Covid-19. There are also high rates of attrition and as a result, several trials have been stopped.

The challenge of patient recruitment is a big factor in this - 85% of trials fail due to lack of enrollment. The ability to take trials to the patients leads to higher enrollment and retention rates, increasing the success of clinical trials, shortening development timelines, and ultimately reducing costs.

Tom O’Leary, CIO, ICON PLC

In order to salvage trials, companies have had to adapt and increasingly, virtual or decentralized clinical trials are becoming normalized. This presents another opportunity to apply machine learning tools, as it forces data to be digitized and expands the amount of quality data available to be collected. According to ERT CTO Don Ragas:“Ten years ago nobody cared about wearables, they are starting to care about it now, imaging was not capturing at scale because it was too costly many years ago, but storage is pretty cheap now. This allows us to bring in huge amounts of imaging data that previously was not very economical. With cloud storage being cheap, unit costs have come down, enabling greater data collection at lower cost. Additionally, the ML tools are now being plug and play, thus removing friction points that inhibited growth prior to today.”


In addition to having an impact on clinical trials, Covid is also expediting the adoption of AI and machine learning in the drug manufacturing process. Applications of AI in pharmaceutical manufacturing are wide ranging. They can perform quality control, shorten design time, reduce materials waste, improve production reuse and perform predictive maintenance. Also, the challenge of manufacturing vaccines and therapeutics to deploy around the world at scale has illuminated the importance of speed and agility. Many drug manufacturers have had to pivot their operations and supply chains and AI can be instrumental in expediting that process. is using transfer learning to commercialize new drugs through internal supply chains. According to the company’s founder and CEO Rajiv Anand, “Quartic is able to successfully make something in one manufacturing facility, train the machine learning model on that manufacturing facility, let's say in North America, and then when an additional manufacturing facility in Taiwan is opened, the process and product knowledge that was gained is captured into an algorithm and applied to the new process.” With algorithmic knowledge transfer, Quartic is demonstrating that subsequent facilities used for large scale manufacturing are able to come online much faster. Therefore, the cost and time to commercialization are compressed, which in turn benefits patients and increases the manufacturer’s margins.

“The people who can apply AI and ML best are the people who understand the manufacturing process. Much of the focus of tech development has been for people in IT and data science. This creates a big barrier to adoption for chemists, biologists or chemical engineers, it is unsustainable.”

Rajiv Anand, Founder & CEO,


In analyzing clinical trials and drug manufacturing, it is apparent that adoption of AI and machine learning technology holds astonishing potential to improve the healthcare sector. In light of Covid-19, it will be as important as ever to be able to execute clinical trials in a timely and successful manner and manufacture vaccines and therapeutics with increasing precision and efficiency. The underlying technology is here and now it is about executing improvements with that technology.