When developing new drugs or managing on-going drug performance, massive amounts of data need to be integrated and reviewed to ensure quality and oversight. However, healthcare data is often unstructured and difficult to access. It is estimated that about 80 percent of biomedical knowledge is buried in text-based documents (including email).
This inability to access the full spectrum of patient data leads to high costs and slow time-to-market for new drugs and impacts patient risk and safety for existing pharmaceutical products. However, as a solution, pharma organizations are deploying AI technologies to identify, process, and analyze valuable patient data that hides in legacy text documents and other unstructured data sources.
The company IQVIA have used their expertise in the life sciences industry to create natural language processing-based artificial intelligence that can break down legacy data silos and make the missing biomedical data accessible.
Digital Journal spoke with Joe Rymsza, vice president, Pharmacovigilance and Regulatory Technology Solutions at IQVIA.
Digital Journal: What are the complexities involved in drug development?
Joe Rymsza: Drug development is by no means a straight line from A to B. One can scarcely think of a process more complex than bringing a new drug treatment to market. Drug development often begins with learning something new about a disease, then progresses to the laborious task of turning those discoveries into safe and effective treatments. Pharmacovigilance (PV) must collect adverse effects (AEs) all along the development path to ensure patient safety.
DJ: What are the typical timelines and stages involved?
Rymsza: Drug development is a process that on average can take a decade and 2.6 billion dollars. It involves many moving parts, including pre-clinical work, clinical trials, the process of bringing a drug to market – with PV throughout the entire product lifecycle.
Each of these stages contains many milestones. First comes pre-clinical work, which is the process of designing new drugs and includes the discovery of new compounds that may influence a biological target. It also includes research into the many ways these new compounds might interact with the human body, as well as identifying dosages and scheduling “first in human” doses.
Then, we have clinical trials, which are typically broken into three phases. In the first phase, the drug developer tests the new treatment on healthy volunteers. The second phase consists of exploring safety and efficacy in a small number of patients who have the disease or condition. Phase three trials involve a large clinical group with more affected patients than phase two. After phase three, the drug goes to a regulatory body, the FDA in the U.S., for review and approval.
PV provides critical oversight along the drug development path by collecting adverse events (AEs) as they arise. Up to 80% of AEs exist in unstructured data formats, like email and hard copies, making a PV practitioner’s job more difficult. Managing and taking a new drug to market can be incredibly challenging from a logistical perspective, with a growing set of datapoints that need to be mined, assessed and evaluated. Monitoring the efficacy and safety of a particular drug once it has gone to market can be equally challenging.
DJ: Why are there many false starts?
Rymsza:The caveat of working in a human centric industry, like pharma, means that proof of concept holds much more weight compared to other sectors. As we’ve seen, the entire drug development cycle is lengthy and complex, which is to some extent unavoidable. A staggering nine out of ten potential treatments fail between phase one trials and regulatory approval. A study from the Tufts Center for the Study of Drug Development between 2000 and 2009 found the most common reasons drugs fail in phase 3 to be lack of efficacy, unmet safety requirements, unexpected AEs, and failure to demonstrate commercial value compared to treatments that already exist.
Pharma organizations are constantly developing new innovations, working diligently to prove efficacy, ensure patient safety and demonstrate value vs. existing therapies. However, one of the prime drivers behind false starts – early on and even later in the process – is the siloing of data across the organization. Data silos hinder collaboration and real-time communication between researchers and developers, as well as between their colleagues working in other divisions of an organization, such as quality and compliance or market access and commercialization. For example, in a large pharma organization there could be nine to ten departments, each with their own collection of data. If those departments can’t easily share their data with each other, it can cause the development process to slow down, or worse, impede the ability to access information that might make the difference in getting a safe and effective treatment to market.
DJ: What is the potential for AI to help tackle data silos?
Rymsza:Tackling data silos is even more difficult in pharma because of the high stakes and immense complexity. As we learn more and more about drug development, the sheer amount of data is almost incomprehensible – to a human. An AI platform can rapidly flag issues to humans for immediate action. With AI assistance, humans can have full visibility, particularly in a PV scenario.
However, in order to avoid costly and inefficient data silos, the data given to the AI must be architected, managed and curated. Creating a data lake, or a single enormous pool of quality data, allows humans to access data across the drug development cycle in real-time, no matter where it resides in the company.
DJ: How can AI help to speed up drug development?
Rymsza:AI can take in documents and emails, scan them for AEs and flag them for follow-up. Additionally, it can help with gathering and reporting these adverse events to the authorities and pharmaceutical manufacturers. The earlier an AE can be identified, the greater the chance of preventing any negative impacts to patient health.
Patient safety must come first. Vigilant detection of AEs during and after the clinical trials process is critical to delivering safe treatments to the market and preventing them from being recalled once they’re commercially available. AI can take in documents and emails, scan them for AEs and flag them for follow-up. Additionally, it can help with gathering and reporting AEs to the authorities and pharmaceutical manufacturers. The earlier an AE can be identified, the greater the chance of preventing any negative impacts to patient health.
These AI-driven solutions can also help manage the most time-consuming, yet necessary, part of drug development – regulatory compliance. AI tools can ensure a product or treatment’s compliance with hundreds of thousands of complex, ever-evolving regulations that differ from country to country, ensuring patient safety worldwide. This, in combination with identifying and reporting on AEs ensures the effectiveness of PV. While all solutions are not created equal, several AI-driven solutions exist in the marketplace today to help pharma companies implement greater PV value, and therefore help bring drugs to market faster.
DJ: Are there any notable case studies where AI has been applied?
Rymsza:In one use case, a biopharmaceutical company developing oncology and rare disease drugs was seeking a way to take scanned image files or PDFs used to report AEs and extract important information. Prior to their adoption of AI, PV experts at the company were forced to manually extract key data elements. These included patient information, dates of disease onset, descriptions of given events, whether prognoses were life-threatening, lab tests, concomitant medications and medical history. Without a sophisticated Software-as-a-Service solution, all of this very specific, structured information would have had to have been pulled manually from unstructured data.
To address this issue, the company created a workflow to process AE forms, using AI to extract relevant patient data. These queries used natural language processing (NLP) to understand the meaning behind the unstructured text in the forms. Humans, with the assistance of AI, were able to identify negations, synonyms for diseases and medications that appear in the medical history section, dates in different formats and drug dosages.
The SaaS solution with AI capabilities also extracted and normalized any AE mentions, structured or unstructured, to MedDRA terms. On top of that, the AI tool delivered data on concomitant medications, date of onset, lab test results and any additional key patient attributes. The company then took that data and uploaded it into their clinical safety database so that research and clinical safety teams could access rapid reporting, speeding up the clinical development and trial phases.
In short, this AI-based SaaS platform helped the company make automated leaps that would otherwise have taken countless hours of human effort. This is a significant step toward speeding the overall drug development process.
With AI assistance, the company has been able to speed up its development of safe, effective treatments. The AI-enhanced solution also monitors patient safety after treatments are approved and in use, searching for AEs that might flag to a pharma company that adjustments are needed in terms of dosage, components, for greater patient safety.
AI plays a significant role in flagging AEs in unstructured data signals. This opens the door to a promising future – one in which drug development can be accelerated, while keeping patient safety front and center.
