Ian Ferria, CPO Artificial Intelligence, discusses four ways artificial intelligence can help clinical trials run more efficiently.
Although the COVID-19 pandemic has managed to stop large-scale international spectacles like the Olympics in their tracks, it has also sparked public interest in a type of major event that until now had mostly been hidden from public view: the clinical trial.
In the past, people would eagerly check how their NBA or football team had fared the previous evening. But in 2020, we were all awaiting news on the latest vaccine trials with bated breath. How had Pfizer performed in Phase III? Was Johnson & Johnson performing well enough in Phase II to make it through to the next round? Would Novavax’s efficacy rate stand tall against some tough competition?
Media reports on these clinical trials tended to focus just on their outcomes, because all that most of us were interested in was whether someone would manage to create an effective, safe vaccine. What was left out in coverage of these trials was the vast complexity of organizing and conducting a clinical trial for any medication that requires regulatory approval.
While some of that complexity concerns tangible aspects like organizing the administration of the medication to patients or reserving laboratory and production facilities, perhaps the most complicated part of a clinical trial in the twenty-first century concerns data.
The fact of the matter is that, even in the earliest stages of planning a trial, data—and how effectively it is processed and interpreted—determines what the trial will be able to achieve, or even whether it will be viable at all. Handling all that data can be an enormous challenge for researchers. The smallest-scale clinical trial will produce more data than a team of scientists can make sense of without the help of computing power.
It’s not just a question of giving scientists access to computer hardware with serious processing power. Clinical trials now rely on servers and sources of computing power that may be distributed across the world, especially when the trial is a joint initiative involving research facilities located on different continents. Then there’s the fact that the processing within a trial takes many different forms, depending on whether the research team is, for example, compiling and analyzing data obtained from participants, examining the molecular-level properties of the drug under trial, or feeding the data obtained into predictive models.
The processing of the data across multiple research institutions and for dozens of different purposes means that a trial’s computing-power needs are in constant flux. And amassing and allocating computing power from different sources is an undertaking of exceptional technical complexity.
For all these challenges arising from managing computing power and data, AI is providing solutions that allow clinical trials to be run more efficiently than ever before. Let’s take a look at four ways in which it is doing so.
Replacement of Manual Processes
In terms of workflow, a clinical trial can be broken down into thousands of steps. Although computers have taken over more and more of these as time has passed, scientists are still used to carrying out many of them manually. In a pre-AI world, they had no other choice.
AI has the power to relieve researchers of the burden created by a large number of the tasks that need to be performed. An appropriately designed AI model can, for instance, identify suitable participants for a trial from patient and clinical data at a far higher speed and with a much greater level of consistency than the human mind can manage. It even has the potential to make use of data points that the organizers of the trial may have overlooked.
Clinical trials are an uncertain business, and not just in terms of the efficacy of the drug under trial. Many clinical trials are ultimately not viable, because they turn out to be incapable of producing data that is scientifically robust. Unfortunately, their inviability does not become apparent to the researchers until an advanced stage of the trial, meaning that substantial resources have been wasted.
The predictive power of AI is able to prevent this problem. By analyzing data from previous trials on criteria such as enrollment and by producing models based on the trial design, AI can predict the likelihood of the planned trial being able to yield viable data. The research team is then better able to make sensible decisions about whether to proceed with the trial or whether to modify the study to increase the likelihood that it can be carried through to completion.
In a clinical trial, accuracy is crucial when it comes to interpreting the data on how participants have responded to the therapy or drug under trial. Especially when a trial contains a large number of participants, achieving consistency and accuracy in the interpretation can be difficult. But machine learning can assist the research team here. In recent years, for example, machine learning has successfully been used to detect or classify tumors in medical imaging at a speed that the human mind will never be able to achieve.
As we mentioned, provisioning the compute power required to handle a clinical trial’s data is an enormous challenge. But it’s also a task that a good AI setup is perfectly suited to. Core Scientific’s AI platform, The Cloud for Data Scientists, is able to seamlessly integrate compute power from a wide array of sources—it does not matter whether these are cloud services, the various research institutions’ on-site facilities or our on-premise hardware—and then allocate it with optimum efficiency to the different processing and modeling tasks that need to be performed by the research team at any given time. By unifying all of the compute power in this way, The Cloud for Data Scientists removes processing bottlenecks as well as availability and decision-making delays from clinical trials and ensures that they are always provided with the compute power that they need, when they need it.
The Cloud for Data Scientists makes it easy to get started with using AI in your scientific research. You can start as small as one data scientist with one GPU, and be training your models in under 2 minutes or access the most advanced supercomputing centers for HPC on demand. Sign up for a free POC.