What happens when you put Artificial Intelligence (AI) technologies to work to improve human safety in Alzheimer’s disease clinical trials?
Major advancements are possible, according to a groundbreaking new study. Led by ATRI Director of Informatics Gustavo A. Jimenez-Maggiora, PhD, MBA, a team of researchers detailed their findings in the paper, “Artificial intelligence-enabled safety monitoring in Alzheimer’s disease clinical trials,” published in The Journal of Prevention in Alzheimer’s Disease last month.
The team tested cutting-edge methods to review and code safety data from clinical trials of Alzheimer’s disease drugs. The goal was to find the most accurate and cost-effective ways to sort through data to recognize adverse events during trials. Trial participants sometimes report injuries like falls, or conditions like headaches. Sorting and manually coding this data has been time and labor intensive, with a potential for human bias.
Alzheimer’s disease continues to be a leading cause of death globally, and demand for treatment is becoming more urgent. The team saw a critical need to modernize clinical trial procedures.
So they set out to test AI methods, comparing them to one another and to human-based methods. Which are most accurate, efficient and reliable when it comes to participant safety? And can the results be applied more generally?
AI is being used more frequently to recognize signs and symptoms of Alzheimer’s disease in patients. Another large-scale research effort is being led by the Keck School of Medicine of USC and USC Viterbi School of Engineering, the Artificial Intelligence for Alzheimer’s Disease Consortium (AI for AD), which spans 12 U.S. sites and is focused on early detection.
A common type of AI called Natural Language Processing is also starting to be used in Alzheimer’s disease research - including in this study. You may not have heard of this type of language processing, but it is already common in our day-to-day lives. Harvard Business Review described Natural Language Processing as the “branch of AI focused on how computers can process language like humans do.” Using it, machines are able to “understand, interpret, and generate human language,” according to Nature. Think of how voice assistants like Apple’s Siri and Amazon’s Alexa listen to what we say and follow our voice commands, or how we use voice dictation on our iPhones. These are forms of Natural Language Processing.
When turned to Alzheimer’s disease research, language processing helps recognize the disease using speech and language detection methods.
In this study, software programmed to use Natural Language Processing methods looked for patterns and relationships in the data from clinical trials. The adverse event data was then sorted and classified. These methods used a combination of longer-standing and newer Natural Language Processing methods, as described in more detail in the paper.
But the study didn’t cut humans out of the process. Expert clinicians were included in each method tested, to varying degrees. The authors noted that the role of clinicians in these processes will remain essential.
The study broke down the performance of each method based on accuracy and costs. AI approaches achieved higher accuracy (∼20% increase in accuracy) and were more cost-effective (∼80% cost reduction) than traditional clinician coding. A hybrid model that combined classic and newer approaches outperformed both when it came to accuracy.
“These methods are really focused on bringing efficiencies to how we conduct clinical research and clinical trials,” Jimenez-Maggiora said. “They’re still in the early stages of development, but the results suggest that an AI-based approach that complements a human expert process is going to see a productivity gain in terms of data quality and accuracy.”
Part of what is driving interest in these improvements is the pressing need for effective, safe Alzheimer’s disease treatment. The demand has never been more urgent, with the number of cases of Alzheimer’s disease expected to soar in coming years. The number of clinical trials exploring Alzheimer’s disease drug development are picking up pace, increasing from 172 in 2022 to 187 in 2023.
Many more trial participants will be needed, along with modernized trial processes and procedures, the authors noted. Participant safety is essential, and current processes are “ripe for transformation,” the authors wrote.
As ATRI's Medical Director, Michael Rafii MD, PhD, further noted, "Our recent paper demonstrates how AI will transform safety monitoring in clinical trials by enabling the automated, real-time detection and analysis of adverse events. AI can identify patterns and anomalies quickly and accurately, ensuring that potential safety issues are addressed promptly, thereby enhancing patient safety and overall trial integrity."
This isn’t about choosing either AI or humans, the authors found. Expert clinicians must continue to play a central role. But the study’s results showed that AI technologies hold great promise for increasing the accuracy and reliability of participant safety monitoring while saving resources. The team is continuing to seek ways to apply these findings to safety monitoring more broadly. Freed up resources will be needed elsewhere in the growing fight against Alzheimer’s disease.
Key Takeaways |
| What was this study about? A team from USC’s Alzheimer’s Therapeutic Research Institute published a paper showing great potential for increasing accuracy and saving resources when using AI to analyze participant safety data from Alzheimer’s disease clinical trials. The team compared approaches combining AI and human expertise, and found winning combinations. |
| What was the goal? The study evaluated new technologies used in AI. The goal was to see if the time-consuming work of reporting and coding health issues like falls could be streamlined. The team is looking to apply lessons found here more broadly. |
| What was found? Using sample data of 1,000 adverse events, the AI-based methods were about 20 percent more accurate and 80 percent less costly compared to human-only methods. |
| Where did the study take place? The study was conducted by the University of Southern California Alzheimer’s Therapeutic Research Institute, based in San Diego, CA. |