The clinical development process is incredibly time consuming and only yields a success rate of about 10%. One way to reduce the time, cost and success of the clinical development process is to introduce artificial intelligence (AI) into clinical trials and implement AI-driven process automation.
The dream of AI in clinical trial execution is for trial functions to run autonomously, like a self-driving car. A self-driving car knows how fast it is going, where it needs to go, and the current status of all major systems. It also has algorithms that can detect cars ahead, detect when an obstacle veers off course, and the capabilities to automatically readjust the path of the car.
Similarly, the dream of AI in clinical trials is to have trial functions run autonomously using robust algorithms, analyze insights and workflow processes, and then execute on optimal next steps. Much like autonomous vehicles, an AI-powered clinical trial will know how fast it is going, the current status of all major activities, a view to potential challenges and the ability to automatically adjust based on this knowledge to execute most efficiently.
AI in autonomous cars not only tells us where to go — it drives us there — making all the turns and avoiding risks along the way. In the context of a clinical trial, this would work the same way. Rather than guiding every step of the trial, researchers would plug in directions and parameters and let the trial function run itself. But we’re not there yet.
Where Are We Today?
Today, clinical trials are more like Apple Maps than autonomous vehicles. Data and AI solutions can tell us where we are, where we are going, offer ideas about where to go next, and suggest the best way to get there. Today’s solutions can even predict obstacles based on current and historical data or alert users to problems and recommend alternative actions. However, human input and action is needed to reach the desired outcomes.
Apple Maps does not replace the driver but it makes the driver much better. Similarly, AI does not replace clinical operations resources but improves the outcomes. This advancement has already been instrumental in making trials more efficient than ever before but it’s still an early stepping stone toward the greater dream.
Addressing Challenges
One reason the dream does not meet reality is due to expectations. The self-driving clinical trial is being sold, but it is not available. What is holding our industry back? And, given these constraints, what can we do today?
Access to Data
While substantial data is available, it can be difficult to access due to availability, complexity, size and organizational silos. Organizations do not always have the teams and the infrastructure required to easily store, aggregate, and access the data. In other cases, there are areas that simply do not have enough data captured in a consistent way to draw needed insights. For example, if a sponsor wants to predict enrollment for a specific trial, it may want to use AI to gauge the dropout rate of a previous study with a similar design. However, if the previous study’s dataset was too small, the AI will likely not provide accurate predictions.
Ethics, Compliance, & Privacy Requirements
There are high standards for any technology implemented as part of clinical trial execution, including good practice (GxP) quality guidelines, regulations, and validation requirements that not only ensure the outcome is valid but that the process is predictable and repeatable.
For example, there are ethics and laws that must be factored into the design of trials. Since laws can vary by geography, it is not possible to list them all here. The laws tend to align to ethical principles. The World Health Organization published the following Key Ethical Principles for AI in Health:
– Protect autonomy
– Promote human well-being, human safety, and the public interest
– Ensure transparency, explainability, and intelligibility
– Foster responsibility and accountability
– Ensure inclusiveness and equity
– Promote artificial intelligence that is responsive and sustainable
A key tenant that underlies these ethical principles is to mitigate potential bias. For example, AI algorithms based on electronic medical record (EMR) data may be biased toward the EMR’s workflow, billing procedures, or that of an individual institution. In another case, AI might recommend treatments based on what insurance is most likely to cover, regardless of whether it’s actually the best treatment for the issue.
Lack of Process Digitalization and Automation
Insights are formed by connecting data sets together and running algorithms, but intelligence occurs when you drive the right action to the user. We are slowly creating the insights to make better decisions, but we have limited pathways to share those insights with the decision-maker. In order to pave the way for self-run trials, researchers and data scientists have to digitize and automate processes in order to ensure the AI prediction is used in the workflow.
Improving Clinical Trials Today
While the dream of self-running clinical trials isn’t yet accessible, we can improve clinical trial execution by increasing the focus not on AI development but insight deployment: How the AI-driven insight is automatically and seamlessly introduced in the clinical trial workflow to provide value. To do this, organizations should focus on the following:
Digitize the Process
In order to ensure clinical trials are successful and future-proofed, researchers need to digitize the process or focus their AI/ML implementation in areas where the process is already digitalized. The digital process itself allows for generation of data with which future recommendations can be made.
Focus on User Experience
Ensure the user (e.g., a Clinical Research Associate, Data Manager, Project Leader) understands a recommendation. This means not just communicating what the recommendation is, but also why and how the conclusion was drawn to suggest it. This may require providing data used to generate the insight in plain view of the user for confirmation. Users also need simple interfaces that enable them to quickly execute or override a recommendation.
Capture the Feedback
Allow the user to override the recommendation and capture the feedback. For example, GPS can tell a driver to turn left, but if they see a roadblock, they can then choose to go right. Similarly, if an algorithm recommends a certain investigator for a trial but that investigator is no longer practicing, that’s critical information to capture to ensure that investigator is not recommended in the future.
Manage Insights Centrally
It is very easy for an algorithm to be developed that can then be coded into an application. However, if you tie the algorithm too closely to an application, you lose the ability to re-use that insight in other solutions. For that reason, it is critical to manage these insights centrally and deploy them via APIs to digital workflows. This also offers better visibility into the data and insights collected.
Realizing the Dream
Ultimately, we want the dream of AI in clinical trials where insights are not only created but also implemented to generate action results in better time, cost, and quality. The big reason that the promise hasn’t yet been achieved is that there is not enough focus on the workflows that will benefit from the insights. We need to deploy that insight to the right person at the right time to drive the right decision. Otherwise, we’ll continue to put excess focus on AI without installing the systems that will make the algorithms valuable.
Doing this, will allow us to get closer to that dream.
About Timothy Riely
Tim has 20 years of experience delivering data management, analytic solutions and business intelligence in leadership and consulting roles. Tim currently leads IQIVA’s Clinical Analytics Suite (CDAS), providing both SaaS solutions for sponsors as well as IQVIA’s internal CRO needs. As head of CDAS, Tim is responsible for full lifecycle delivery of R&D data and analytics solutions for clinical operations, clinical data management and intelligent applications. Tim’s background includes a combination of payer, provider, and life sciences experience.
