I have the pleasure of speaking with James Maitland, an expert in robotics and IoT applications in medicine. Today, we will discuss COTA Healthcare’s recent breakthrough in using generative AI for real-world oncology data (RWD) abstraction.
How would you describe the major breakthrough that COTA Healthcare has achieved in real-world oncology data?
COTA Healthcare’s major breakthrough is the development of a generative AI platform that facilitates the large-scale curation of cancer data with high accuracy and profitability. This marks a significant shift from previous industry efforts which relied on less reliable technology, making RWD abstraction labor-intensive and not profitable.
What is the core function of the generative AI platform in this achievement?
The core function of the generative AI platform is to automate initial data interpretation, which significantly reduces the time and cost associated with manual data abstraction. This platform allows for real-time data exploration and has a search-engine-like interface that simplifies data queries, democratizing access to data insights across broader R&D teams.
Can you elaborate on why previous industry efforts to automate RWD abstraction processes were not successful?
Previous industry efforts struggled because the technology available at the time was not sufficiently reliable for expert abstractors to trust. These earlier systems often failed to meet the required accuracy and reliability that this complex task demands.
What were the main challenges COTA faced in its earlier attempts to automate data abstraction?
COTA faced challenges such as inadequate technology that couldn’t handle the complexity of medical data abstraction. They went through multiple cycles of re-engineering their processes, which eventually fell short in delivering the reliability and efficiency needed.
How did the emergence of large language models (LLMs) in 2023 change COTA’s approach to data abstraction?
The emergence of large language models in 2023 provided COTA with the necessary technological foundation to experiment with automating core abstraction tasks and front-end queries. These models offered the reliability and accuracy needed to shift from a purely manual process to a more automated one.
Can you explain the hybrid model where AI and human experts work together?
In the hybrid model, advanced AI performs the initial data interpretation, while human oncologists and data scientists step in to validate the results for accuracy. This combination leverages the speed of AI and the expertise of human professionals to ensure high-quality data.
What roles do human oncologists and data scientists play in this new model?
Human oncologists and data scientists primarily validate the data interpreted by the AI. Their expertise ensures that the data is accurate and reliable, which is essential for informing drug development and clinical decisions in oncology.
How does this model improve the accuracy of data interpretation?
This model improves accuracy by combining the strengths of AI and human expertise. The AI handles large volumes of data quickly, but human experts are crucial for validating complex cases, ensuring that the data is both accurate and trustworthy.
What are the key operational benefits of this new AI-driven approach compared to traditional methods?
The key operational benefits include a significant reduction in costs and turnaround time for data abstraction. The approach also maintains or exceeds the accuracy benchmarks set by traditional methods, making the process more efficient and reliable.
How has the cost per record changed with this new approach?
The cost per record has been reduced by 23% year over year, which is a substantial decrease. This reduction is expected to drive positive EBITDA and cash flow while maintaining a high level of accuracy.
Can you provide details on the turnaround time improvements for data abstraction?
The turnaround time for data abstraction has improved dramatically, with tasks that used to take days or weeks now being completed in minutes or seconds. This rapid processing allows for near-instant queries and faster access to critical insights.
What are some of the key findings from COTA’s research on GenAI-assisted data abstraction?
COTA’s research indicates that GenAI-assisted data abstraction can achieve nearly instant queries with accuracy rates matching or exceeding the 93% benchmark set by human abstractors. This finding underscores the efficiency and reliability of the new approach.
How significant is it for academic and industry researchers to achieve near-instant queries?
Achieving near-instant queries is highly significant for researchers as it saves substantial time and allows quicker access to critical data insights. This capability can accelerate research and development processes, leading to faster clinical decisions and advancements.
In what ways does your platform democratize access to real-world oncology data insights?
The platform democratizes access by allowing users without formal analytics training to perform complex data queries easily. This broader accessibility empowers a wider range of professionals to gain valuable insights from the data.
What are the notable business metrics that indicate the success of this transition to an LLM-driven model?
Notable metrics include a 25% decrease in cost per record, a 32% reduction in abstraction handling time, and a 350% increase in data inventory. These figures highlight the efficiency and scalability achieved through the transition.
How has the reduction in abstraction handling time impacted biopharma operations?
The reduction in abstraction handling time allows biopharma operations to identify patient cohorts more quickly and obtain insights on treatment patterns and biomarker expression almost instantly. This speed enhances the efficiency and effectiveness of biopharma activities.
What are the implications of faster and more precise identification of patient cohorts for clinical research?
Faster and more precise identification of patient cohorts enables quicker initiation of clinical trials, more targeted research efforts, and timely discoveries in treatment outcomes. This efficiency can potentially accelerate the entire drug development process.
How does the platform, named CAILIN, work for users without formal analytics training?
CAILIN operates with a user-friendly, search-engine-like interface where users simply type their queries. The platform then processes these queries to provide accurate and relevant results, making it accessible to those without specialized training.
Can you describe how the search-engine-like interface of CAILIN simplifies data queries?
The search-engine-like interface simplifies data queries by allowing users to enter questions in natural language. The AI processes these inputs and delivers quick, precise answers, eliminating the need for complex coding skills traditionally required for such tasks.
What kinds of questions can researchers now answer in seconds with the CAILIN platform?
Researchers can now answer a wide range of questions, such as identifying the number of patients on a particular therapy line or specific cancer subtypes. This rapid querying capability greatly enhances the speed of research and decision-making processes.
What are COTA’s goals for the next few years in terms of continuing to improve this technology?
COTA aims to further refine its AI capabilities, increase the speed and accuracy of data abstraction, and expand the accessibility of its platform to more users. Continuous improvement and innovation will be key to maintaining their leadership in real-world oncology data insights.
