AI in Clinical Research: OWL’s Approach & OWLbot
Artificial intelligence is everywhere! At OWL Oncology Research, we're embracing its potential while recognizing that it doesn’t replace good old-fashioned human experience. Knowing when to use AI, and particularly when not to, is vital to getting the most out of this technology.
Although most of us first started using artificial intelligence only a few years ago, today, it’s hard to imagine living without it. AI quickly went from more of a slightly better version of Siri to an everyday business necessity. And on nearly a daily basis, we’re still discovering ways it can help us. Whether it is helping us with trivial tasks such as drafting an email to much more complex requests such as identifying patterns in large datasets, AI has the ability to make us faster, more efficient, and better informed about almost everything that comes across our desk.
In our industry, these capabilities are especially enticing. Clinical trials involve enormous amounts of planning, loads of data, complex regulatory requirements, and depend on coordination among people from diverse backgrounds and locations.
But there's an important distinction between using AI to support clinical research and allowing AI to take on actual decision-making.
At OWL Oncology Research, we depend on AI and consider it an extraordinarily useful assistant. It is not, however, a substitute for expertise, professional judgment, or accountability.
Opportunity: What AI Does Well
Soon after we started experimenting with AI, we realized it can be a lifesaver with all sorts of repetitive tasks. It quickly handles a ton of busy work and can quickly provide summaries or suggestions which would have taken us much longer to develop on our own.
Some particularly useful applications include:
Researching and summarizing complex topics. Since AI can read and understand thousands of documents in an instant, it’s great for helping us get up to speed quickly on the goals and challenges surrounding a new project. AI is excellent at organizing complex information, summarizing available research, and highlighting important topics to help us get started.
Study planning and feasibility. AI is really valuable in helping us plan studies. It’s very capable in projecting enrollment, assessing geographic feasibility, and providing site recommendations.
Document development. AI is also great for creating initial drafts of documents, improving existing documents, and finding mistakes or inconsistencies across different documents.
Resource planning. AI is excellent at taking a birds-eye view of complex projects and figuring out how best to staff the team with the right people.
Instead of replacing experienced people, these tools help them reach an effective starting point where key decisions can be more effectively made.
The Dos and Don'ts of AI in Clinical Research
Clinical research is a highly regulated environment, and getting something wrong can be catastrophic. Knowing how to effectively use AI can help you navigate the complex clinical research environment.
Here are some principles we try to abide by:
DO: AI can help create a solid first draft, but never a finished product.
Don’t automatically trust AI! No matter how convincing it sounds, it often makes mistakes! AI has been known to invent references, completely misunderstand context, or reach conclusions with absolute certainty that are not in line with reality. Before accepting anything created by AI, you must first verify it’s correct.
DO: Protect confidential and sensitive information.
Even though it's widely used and marketed as secure, never assume that an AI platform is secure. Anything confidential (which is a lot in our industry!) must be kept confidential. Patient information, proprietary study data, and confidential study materials need to remain protected at all times. Review the platform's data-handling practices, contractual protections, and your organization's requirements.
DO: Follow sponsor requirements and organizational policies.
At this point, almost all companies have policies regarding the use of AI, including restrictions on specific platforms, data processing, and use in study activities. Get to know your company’s guidelines and remember that AI use that is acceptable for one client may be prohibited for another.
DO: Maintain meaningful human oversight.
Make sure you are carefully reviewing any AI-created material before calling it good. You are ultimately still responsible for anything it produces. Double-check everything, especially if it involves clinical quality, regulatory compliance, and data integrity. And if it involves patient safety, you better triple check it.
DON'T: Confuse confidence with accuracy.
AI excels at producing materials in polished, convincing language. Unfortunately, its ability to sound convincing doesn’t always mean it’s right. Verify everything.
DON'T: Let AI make regulated decisions.
AI is great at finding potential issues and red flags. But once identified, people need to come in and make the final decisions. AI should never be responsible for determining the significance of a clinical finding or whether a protocol deviation needs to be escalated. And when dealing with regulatory agencies, AI can be a great assistant, but qualified human judgment remains essential.
DON'T: Use unapproved AI tools for protected information or regulated activities.
The convenience of AI makes it tempting to use it all the time. However, bear in mind it’s not appropriate to use in all situations. Before using AI with confidential, personal, or regulated information, ensure the tool and proposed use are appropriately reviewed and approved.
DON'T: Assume that using AI eliminates accountability.
Remember that you are ultimately responsible for anything you produce. Whether you’re using AI to help draft a simple email or analyze a complex data set, you are still responsible for those results. Always review anything AI-generated before putting your name on it.
The bottom line: AI can help you do your job. It cannot take responsibility for doing your job correctly.
How OWL Is Putting AI to Work
At OWL, we embrace AI to improve the quality and efficiency of our work. We routinely use a number of AI-assisted approaches to help us with things like strategic planning, enrollment forecasting, site identification and feasibility, risk assessment, and regulatory strategy. We also use AI to support everyday business activities such as automated note taking during meetings, email polishing, and process improvement. It’s quickly become a very important part of our daily life.
However, everyone at OWL is trained on our AI and Automation Use Policy (OWL-POL-005), which establishes rules for responsible use, protection of sensitive information, and the need for human oversight. This policy strictly prohibits the use of autonomous AI for decision-making in GxP activities and the submission of confidential or regulated information to unauthorized AI systems. For us, responsible adoption is about using AI in ways that strengthen, rather than compromise, the quality of our work.
OWLbot: AI-Assisted Resource Intelligence
One of our recent AI initiatives was the development of OWLbot, an internal AI tool which helps our team navigate OWL's clinical research network. In short, OWLbot helps us quickly identify candidates whose experience and expertise best match a project’s staffing needs. Here’s how it works:
Sponsors are staffing a project and need a CRA with a particular therapeutic experience and a history working with a particularly challenging site, a medical writer familiar with a specialized treatment, and a project manager with the right background to run a study. It’s usually a more complicated question than, "Who is available?"
OWLbot helps our team explore those questions by reviewing vast amounts of information from our network directory and professional experience records. It then provides a list of potential candidates, which is a good starting point. We can then narrow the search using any additional relevant factors, including their OWL history, geographic location, and availability to take on new assignments. In the end, we can have a customized list of potential candidates who are likely a strong fit for a new project.
In the end, OWLbot doesn't replace our management team's firsthand experience with the people in our network. It simply helps the team sift through mountains of known, documented information more efficiently. Our team still reviews the OWLbot recommendations and determines which candidates are the best fit for each assignment. However, we can get to this stage much faster than we did in the past.
The Human Advantage
Today, clinical research is highly dependent on technology that helps it remain organized, and to move quickly. The days of three-part NCR forms, paper Trial Master Files, and hand-signed documents are mostly a thing of the past. Electronic data capture, clinical trial management systems, electronic trial master files, and other digital platforms have transformed the way studies are conducted. And AI can play a helpful role in all of these systems, as long as it’s used thoughtfully and in accordance with applicable requirements.
We at OWL truly believe we should neither dismiss AI out of fear, nor embrace it indiscriminately in the interest of efficiency. Our approach is to go in “eyes-wide-open,” understanding AI’s strengths and acknowledging its weaknesses, while establishing appropriate safeguards and ensuring proper training of OWL staff.
At OWL, our greatest asset has always been the experience, judgment, and dedication of our people. AI can help us work more efficiently. But it doesn't replace the relationships, critical thinking, or accountability that our industry demands.
The future of clinical research isn't AI versus human expertise – it’s human expertise strengthened by responsible AI. And that's a future we're excited to help build.
