AI & ATTR-CM: Faster Trial Candidate Identification with Chart Review

A new study highlights a promising application of large language models (LLM) in accelerating clinical research, specifically for identifying potential candidates for trials focused on transthyretin amyloid cardiomyopathy (ATTR-CM). The research, conducted by Cleveland Clinic and Dyania Health, demonstrates that an AI-driven chart review system can accurately pinpoint individuals who might benefit from participation in clinical studies, potentially streamlining the often lengthy and resource-intensive process of patient recruitment.

Understanding ATTR-CM and the Challenge of Trial Recruitment

Transthyretin amyloid cardiomyopathy, or ATTR-CM, is a progressive and often fatal heart condition caused by a buildup of abnormal protein deposits in the heart muscle. It’s a type of heart failure that disproportionately affects older adults, and can be challenging to diagnose due to its varied and often subtle symptoms. Identifying suitable patients for clinical trials is a critical step in developing new treatments, but traditional chart review methods are notoriously sluggish and require significant manpower. Clinicians must sift through extensive electronic medical records (EMRs), both structured data and unstructured notes, to determine if a patient meets the complex inclusion and exclusion criteria for a given trial.

The complexity of modern EMRs adds to this challenge. As noted in a recent article in The Journal of Clinical and Translational Science, clinical research is heavily reliant on manual chart reviews, which are both time-consuming and expensive. Automating this process has been a long-standing goal, and recent advances in LLMs offer a potential solution.

How the AI System Works

The system developed by Dyania Health and deployed at Cleveland Clinic utilizes a medically trained LLM to analyze EMR data. The study, detailed in a news release from Cleveland Clinic, assessed the AI’s performance in pre-screening patients for the DepleTTR-CM Phase 3 trial. The LLM was able to synthesize both structured variables (like lab results and diagnoses) and free-text notes (like physician observations) to operationalize the trial’s complex criteria. In a single week, the AI screened 1476 EMR records – a task that would typically take significantly longer with manual review.

Accuracy and Potential Impact

The study demonstrated a high degree of accuracy in identifying potential trial participants. While specific performance metrics weren’t immediately available in the news release, the findings suggest that AI-driven chart review can significantly reduce the burden on clinical staff and accelerate the pace of research. What we have is particularly important for rare diseases like ATTR-CM, where finding enough eligible patients can be a major hurdle.

Beyond ATTR-CM: Broader Implications for Clinical Research

The success of this AI-driven approach extends beyond ATTR-CM. The underlying technology has the potential to be applied to a wide range of clinical trials, across various disease areas. By automating the initial screening process, researchers can focus their efforts on more in-depth evaluation of promising candidates, ultimately speeding up the development of new therapies. The ability to quickly identify eligible patients could also improve diversity in clinical trials, ensuring that research findings are representative of the broader population.

This isn’t the first time LLMs have shown promise in chart review. Previous studies have demonstrated the effectiveness of LLM-driven screening for trial criteria, but the generalizability of these findings has sometimes been limited. The Cleveland Clinic study adds to the growing body of evidence supporting the use of AI in clinical research, and suggests that these technologies are becoming increasingly reliable and adaptable.

What to Consider: Limitations and Future Directions

While the results are encouraging, it’s important to acknowledge the limitations of this technology. LLMs are only as good as the data they are trained on, and biases in the training data can lead to inaccurate or unfair results. The AI system is not a substitute for human judgment. Clinicians still need to carefully review the AI’s recommendations and craft the final determination of patient eligibility. The study does not address the potential for false positives or false negatives, and further research is needed to fully understand the system’s performance characteristics.

Looking ahead, the researchers plan to continue refining the AI system and expanding its capabilities. Future operate may focus on incorporating additional data sources, such as genomic information and imaging data, to further improve the accuracy of patient identification. The team is also exploring ways to integrate the AI system into existing clinical workflows, making it easier for researchers to access and utilize its capabilities.

Next Steps: Ongoing Evaluation and Refinement

The development and deployment of AI-driven chart review systems are an iterative process. Continued monitoring and evaluation are essential to ensure that these technologies are performing as expected and are not introducing unintended biases. Researchers will need to conduct ongoing studies to assess the long-term impact of AI on clinical trial recruitment and to identify best practices for implementation. Collaboration between clinicians, data scientists, and regulatory agencies will be crucial to realizing the full potential of AI in accelerating medical research.