Evolving AI for Safety and Multivigilance

Today, as changes in safety regulations worldwide generate a significant increase in the number of adverse event (AE) cases that need to be processed by life science companies, the variety of big data sources that can be mined for safety signals is also rising. These trends translate into huge, new pressures on safety organizations as they continue to carry out their mission of multivigilance (the umbrella term for pharmacovigilance or drug safety, vaccine safety, and medical device safety). The multivigilance process covers the entire lifecycle of a medical product, from clinical trials through post-marketing surveillance.  Currently, Oracle Health Sciences is working with artificial intelligence (AI), machine learning (ML), natural language processing (NLP), and Deep Learning technologies to address the trends of increased case volume and additional signal sources.

Have you ever heard the phrase, everything old is new again?   Most likely, it’s because it reminds you of the Peter Allen song by the same name from the movie, All That Jazz.

The strategy of that phase is truly universal because Oracle Health Sciences, working hand in hand with Oracle Labs, has translated it into new approaches for next-generation multivigilance. Oracle is accelerating cognitive computing concepts and applying them to both safety case management and safety signal management. Today, these technology innovations already can play a major role in creating a cheaper, faster, and more efficient case management process, as well as making medicines safer for patients by detecting risks earlier.

Artificial Intelligence through a Safety Lens

Oracle is applying AI methodologies such as NLP, ML, and Deep Learning to the areas of safety and multivigilance. Two such examples are the extraction of AE information from unstructured data to automate manual processes, and the identification of AE signals in diverse data sources. Oracle’s subject matter experts have many years of experience in both safety and AI. With the ultimate goal of embedding AI across the end-to-end multivigilance workflow, for both case management and signal management, subject matter expertise is key.

In the 1950s, modern work on AI began in earnest. The computer was trained to learn patterns and relationships in data, instead of humans pre-programming the rules for its algorithms. Initially, NLP systems offered a very basic interpretation of word sequences and contexts. So, results were not highly accurate.

Neural networks are part of a wider class of machine learning techniques inspired by the function of neurons in the brain. Today, with advanced computing power, Oracle uses deep learning, which is based on large architectures of neural networks designed for learning exceedingly more complex data representations. This, in turn, enables improved accuracy in machine learning prediction and classification. 

Deep learning is based on several, different, architectural approaches that are typically organized as layers of neurons, as shown in Image 1, and which are used to simulate very simple versions of brain functionality.



Above in Image 1, each small circle is a node representing a neuron (as in Image 2) that essentially acts as a filter or gate that allows the passing of signals from one neuron to another.  It takes in as input a numeric value, processes that value, and outputs another numeric value. When the output value is high, the neuron is said to fire, sending information on to the next level neurons.  Generally, including more neurons and more layers in a neural network, allows for higher order learning. 

In mathematical theory, Deep Learning can achieve 100% accuracy. In practice, for certain tasks, and with a suitable neural architecture, deep learning can attain close to perfect accuracy. This is one of the main reasons for its growing application.

Before creating an automated system that can recognize AEs in unstructured free text, researchers have to teach the algorithms to recognize what an AE is, using large sets of training data. Here is where the critical element of domain expertise comes in, to refine the program’s understanding of language and context.

The Role of Domain Expertise in Safety AI

There are three areas where safety domain expertise, in addition to AI expertise, is essential in building a solution that works well. These are: training data preparation, model design, and lexical resource creation.

  • The quality of the solution depends on the quality of the training data.  In the example of AE case intake, information can come from healthcare professional (HCP) reports, consumer reports, literature articles, health authorities, clinical trials, patient support programs, social media posts, and more. But until now, humans have had to read these narratives manually, pull out the AE information, and enter it into database fields. Today, domain experts can curate training data that are needed by AI applications to extract the AE information automatically.
  • Domain expertise is also an important element in the design of classification or prediction algorithms (models). For example, determining which data elements should be used by the algorithm, determining which data transformations are necessary, and in NLP, determining what context of nearby words the algorithm needs to examine in order to distinguish between true AEs and other medical terms that are not AEs (such as indications and historical conditions). Domain experts are also key to stitching together a complete model from a hierarchy of individual algorithms that each do one, particular task well, such as, detecting which check boxes on a form are ticked or identifying a patient’s birth date.
  • Lexical resources (such as dictionaries of medical products, medical conditions, HCP occupations, English names, consumer health vocabulary, etc.) are used to enhance the performance of learning algorithms---a mechanism for incorporating existing domain knowledge into the learning process. Domain expertise is needed for the creation of such lexical resources and for their integration with the learning algorithm.

By including safety domain expertise, researchers can achieve much higher quality and much greater accuracy for AI applications in the multivigilance space.

Oracle and AI in Multivigilance

Oracle is currently combining multi-layered, deep learning, cloud architecture with its extensive safety domain expertise to explore how these AI methodologies can automate AE case processing intake and detect signals earlier. As the only software developer with decades of experience in both multivigilance and AI, Oracle provides its customers with a very competitive solution.

The exploratory project, Oracle Safety One Intake, aims to address issues including: the time-consuming, inefficient, and error-prone manual processing of incoming AE reports; the problem of multiple intake methods creating multiple work queues; the disparate kinds of data privacy protection offered via different methods; duplicate detection issues; and the manual workarounds for the staging of follow-up information.

In addition to case processing, other safety AI projects under consideration include call center log AE flagging and multi-modal signal detection.

The Past as Prologue

By building on existing AI technologies, advancing them, and applying them to the domain of safety, Oracle enables the vision of the past to become a prologue to the future for multivigilance. Through advanced computing, Oracle is evolving the original approaches to machine learning and neural networks into the next generation of safety technology. And, with Oracle’s domain expertise in both safety and AI, real value can be achieved today for customers.

For immediate AI benefits, Oracle Health Sciences Consulting can deliver safety solutions today that use the core Oracle AI engine.

Contact Oracle if you would like to schedule a demonstration of the Oracle Safety One Intake prototype, participate in the beta test of Oracle Safety One Intake, or find out how Oracle Health Sciences Consulting can implement AI in your organization right now.

View a webcast recording about AI for safety here. Learn more about the Oracle Health Sciences Safety Cloud here and here.

Safe Harbor Statement

The preceding is intended to outline our general product direction. It is intended for information purposes only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied upon in making purchasing decisions. The development, release, and timing of any features or functionality described for Oracle’s products remains at the sole discretion of Oracle.


Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.