Digitizing Protocol Design and Deploying AI to Save Hundreds (of Hours) and Millions (of Dollars)
Kimberly Tableman

wapping digital for paper has proven to be the springboard for operational transformation across dozens of industries from financial services to manufacturing. Yet protocol design—the building block for clinical research—remains rooted in paper, belaboring an already long, slow process.

Paper-based protocol design is not only inefficient but also vulnerable to error, leading to delays, protocol deviations, and potentially trial failure altogether. Considering, for example, that the median revenue of oncology drugs is about $1.6 billion (with the median cost to develop oncology drugs at $648 million), there’s a lot at stake. And not just money: poor protocol design can delay the delivery of life-altering medicines. Unfortunately, it has been difficult to digitize clinical trial protocols due to a lack of standardization, document variability, and significant increase in trial complexity.

For instance, since the 1980s, nearly all protocol design changes have been additive; adding refinements to a process often introduces more documentation and complexity as an unintended consequence. The growth in personalized medicine has also compounded this complexity by incorporating strict inclusion/exclusion criteria on patient enrollment, while biomarker-driven trials—which add many complications to trial operations—are becoming mainstream with nearly 20% of all trials using biomarkers today.

Research shows that as protocol designs become more complex, trial performance worsens. Protocols with a higher relative number of endpoints, eligibility criteria, and procedures are associated with lower physician referral rates; increased procedure administration burden; diminished participation; lower patient recruitment and retention rates; lower dose adherence; and a higher incidence of protocol deviations and substantial amendments. Ultimately, these outcomes contribute to higher failure rates, longer clinical trial cycle times, and increased costs.

Despite all this complexity, modern technology—including artificial intelligence (AI)—can simplify protocol design and automate basic elements to save hundreds of hours and millions of dollars per trial. Historically, one leading global biotechnology company with a large portfolio of lifesaving medicines took an average of 333 days to complete the protocol design. However, when it used a digital tool, the company completed a protocol of similar length and complexity 67 days faster—a ~25% improvement, translating into savings of $438,904. In addition, the company was able to generate an additional $15 million in revenue by getting the product to market faster.

Making Dollars and Sense of Protocol Design

A simple concept has ruled the day: Time saved equals more revenue earned. The cost of a single day of delay is now approximately $1.3 million in lost prescription sales, so the opportunity to capitalize on operational efficiencies is high. Advanced, digital, or electronic protocols (eProtocols) can have a profound impact on clinical trial success. There are three areas where digital protocols offer the greatest return on investment:

  1. Enabling Digital Data Flow
  2. Reducing Audit Findings
  3. Creating a Foundation for Machine Learning and Generative AI

Enable Digital Data Flow

The primary objective of providing electronic protocol data to downstream systems and business processes is to increase operational efficiency. Historically, sponsors have had to program systems manually to move protocol data into the Interactive Response Technology (IRT), Electronic Data Capture (EDC), electronic Clinical Outcome Assessment (eCOA), eConsent, and other downstream clinical trial systems. Literally, humans had to physically enter information from a hard-copy paper document protocol manually into each system, one system at a time.

By capturing all the protocol data electronically at the point of design, information is automatically available to other important systems via an Application Programming Interface (API). The downstream systems have the ability to consume the data via RESTful APIs, as well. Historically, it has taken three to four months to implement an eCOA solution, but with automation this timeline has been reduced by half to six weeks or less. This is a massive accelerator.

Another eProtocol benefit is the ability to easily use the electronically captured information in downstream documents such as the Statistical Analysis Plan, Informed Consent Document, Investigator Brochure, and others. The technology pulls the data “auto-magically” in one click from the protocol into templates to eliminate extensive copy-and-paste time and risk of manual error.

The risk is significant when one considers that sites need to have easy access to the most current information and it needs to be crystal-clear, especially related to the visit schedule, procedures, and dosing. The ability to update this information in one click and ensure that the updates move to downstream systems and documents automatically is critical.

Reduce Audit Findings

According to a 2022 study that analyzed the most common themes in audit findings, most involved missing documentation. In addition to operational challenges with missing information is the risk of providing misinformation to trial sites. It is critical that sites have all the accurate study-related information available at their fingertips. The importance of how medical instruments are calibrated, appropriate timing, and dosing, for instance, all impact the study results.

To prevent the effects of missing information or misinformation, all stakeholders need centralized access to the same data on the same platform at the same time. With eProtocols, all protocol information is digitized and in one place so there’s no more duplicate data entry or the perils of copy and paste. Digitizing and centralizing the protocol helps ensure alignment across all the documentation specific to a clinical trial.

Create a Foundation for AI

AI requires clean, high-quality data in electronic form to train models to identify specific patterns. This scrubbed, electronic data then fuels sophisticated AI algorithms to quickly detect anomalies or trends that inform protocol design much faster and more accurately than humans can alone. AI-based—also known as “smart”—eProtocols can leverage millions of historical datapoints to not only accelerate development of the first protocol draft but also to ensure that the first draft is optimized based on data.

For example, AI can identify and alert the study team on which endpoints in a therapeutic area had successful outcomes and provide insights into the sites and countries that have been successful at recruiting patients. Given that patient recruitment is a major problem in clinical research, optimization of the appropriate country and site strategy will have a significant impact on the speed and success of finding the right patients for the study.

In addition, AI can analyze large datasets at the macro level to provide input at the therapeutic area/disease state level and utilize algorithms to provide input on actual gaps in the research—for instance, optimizing targets in the body, and the mechanisms of action. This approach has yet to be implemented because it requires a robust information architecture and a standardized digital data flow beginning with the protocol all the way through the execution of the clinical trial. The industry is moving in that direction, but the first step is to ensure that protocol data is stored in an electronic format and aligned to ICH M11 guidelines (as recommended by the FDA). In this way, sponsors can “kill two birds with one stone”: Create the backbone for digital data flow and increase the speed of the submission to the FDA.

There is equal excitement relative to using data to predict disease progression within a virtual control arm. For example, sponsors can use synthetic data to mimic the data typically collected from actual patients receiving placebo treatment to reduce patient and site burden, as well as the cost of a trial.

Someday, AI will be able to optimize study design based on real-life data—i.e., providing recommendations on whether to use a pragmatic design, adaptive approach utilizing on-site visits, remote visits, or a hybrid strategy and quickly identify the smallest number of participants necessary to provide statistically significant safety and efficacy results.

There are many opportunities to utilize AI in clinical research, but the first step is to capture data electronically, ensure it is of high quality, and format it to make it easy to train the machine and build complex algorithms. This is the AI foundation to truly transform and improve operations, gain deeper insights, and create smart protocols.