Here’s a cliché: The future is here. Particularly for pharmacovigilance (PV) solution users. Anybody working with PV software solutions in the next few years is going to experience a major change; working with PV solutions driven by machine learning (ML) algorithms. Whether you’re a veteran of the life sciences or you’re just starting your career in the field, machine learning will likely be part of the PV solution your organization uses very soon, so you will be experiencing this change unless you change your career.
But what does that mean exactly? If you’re not an enthusiast of fields like computer science or robotics, you might not have had occasion to look too deeply into machine learning. Sure, you’ve heard the word tossed around, along with the words “Artificial Intelligence” (AI). And everyone likely has some vague idea informed by common references as to what Artificial Intelligence is. But a lot of stakeholders in clinical research and post-marketing surveillance of drugs never had occasion to examine these concepts very closely. You generally hear that some of the solutions you use at work are powered by AI to some degree and that it’s supposed to improve the execution of certain functions.
At a similar depth, you heard the words “machine learning” being discussed. Those of you who follow discussions for years ahead might have even learned a little more about how ML can assist with pharmacovigilance functions when it’s finally implemented. Because during 2018, ML in PV was not in the implementation phase. Nor was machine learning implemented in 2019. By implementation, we’re talking about PV software solutions in active, widespread, clinical use which actually involve ML.
Up until 2020, the dialogue about machine learning in pharmacovigilance was dominated by theoretical analysis of the possible benefits, interspersed with occasional experimentation. But now, starting from 2021, we are witnessing the first implementation of ML in PV. The United States’ Food and Drug Administration (FDA) launched the FDA Adverse Event Reporting System II (FAERS II), which features ML-driven end-to-end automation of adverse event report processing. So, it might be a cliché, but it’s a cliché because it’s true. The future is here.
Machine Learning & Artificial Intelligence
But what does machine learning mean exactly? Now that it’s going to affect pharmacovigilance more closely, let’s do that quick overview of ML. You know. That overview you’ve been avoiding. You’ll often hear people saying things like “AI and Machine Learning”. Even highly specialized industry commentators. And truthfully, they’re not necessarily wrong. Amongst certain computer science enthusiasts, saying “AI & ML” is understood to mean “AI in general & ML in particular”. Saying “ML & AI” can mean “ML and other AI applications”.
But the truth is, for life sciences’ veterans with no professional affiliation to, or deep personal interest in computer science, reading phrases like that can be highly misleading. It suggests that ML and AI are two separate aspects of computer science. They’re not. Machine learning is, in fact, a part of artificial intelligence. Machine learning grew out of artificial intelligence, and although many would argue it’s grown into a separate field altogether, the predominant arguments are that ML remains a part of AI. Even those who disagree will concede that machine learning came from the artificial intelligence field.
AI as a field is generally concerned with studying “intelligent agents”, which includes any system that can identify enough of its environment to initiate actions that increase the system’s chance of attaining successful results. This means that designing AI technology is the effort to build such intelligent agents. There are multiple approaches to building such intelligent agents. For example, systems can be designed to apply logical steps based on probability, for tasks that require fairly linear reasoning or problem-solving skills.
Machine learning, however, is an advanced approach within artificial intelligence. Machine learning involves the designing of computer algorithms that automatically refine and improve their own parameters by experiencing results in the form of data. The algorithms are primarily designed to build models extracted from sample data, which machine learning pioneers call training data. These models help the algorithms make predictions and decisions they weren’t specifically programmed to make, and as the data collected from the results grows, the models’ parameters are changed and refined.
In this way, the algorithm supposedly “learns” from the data. Machine learning has been applied to a wide variety of industries, having been used to filter emails, recognize speech, and assist with healthcare (although its use in PV is new). The role of machine learning algorithms is to provide the flexibility required to achieve desired results in highly data-variable environments, whenever the conventional algorithms used to accomplish tasks in more data-stable environments cannot be utilized.
Machine Learning Benefits in Pharmacovigilance
So how does machine learning improve upon the predominant approaches to pharmacovigilance before ML was applied? We can look at this from the perspective of the present combined with the immediate future, versus the perspective of the near-but-not-immediate future. In the near-but-not-immediate future, there are any number of new benefits machine learning can bring to pharmacovigilance. For example, there was an experiment published last July, which attempted to mine adverse drug events (ADEs) from social media. This experiment utilized “Deep Learning”. A form of machine learning. And while the results were rather imperfect, they were manageable.
Of course, challenges that can be managed with a little more effort defy the point of technological application, so don’t expect to see all pharmacovigilance professionals eagerly looking through or referencing datasets from Twitter and Facebook in the next few months. But the manageable results of today are the automated and fine-tuned algorithms of tomorrow. In a year or two, who can say? More immediately however, machine learning can already advance pharmacovigilance norms in numerous ways.
One of those ways relates to data aggregation. Machine learning has proven to be particularly valuable in scanning for precise types of information from vast and unstructured data sets and extracting parameter-improving conditions as a consequence, with an end result towards automating Adverse Event Report aggregation more efficiently. Unstructured data can be found in various ways, gathered from emails, physical documents, and incidental mentions in other sources. This grants PV teams greater bandwidth and frees up highly-trained professionals from addressing largely repetitive tasks. This is particularly important for teams struggling with smaller numbers due to a shortage of talent in the regional workforce.
Another way in which ML can advance pharmacovigilance in the immediate future is by analyzing vast data sets relating to confirmed adverse reactions and identifying patterns indicative of new patient safety concerns which have hitherto gone unnoticed. Outside of confirmed adverse reactions, machine learning is also currently capable of automating the follow-ups for unconfirmed adverse reaction cases in order to fill in the gaps for datasets that are not optimized for analysis.
Machine learning has not peaked in advancement yet. Its application is imperfect. Its potential for benefiting pharmacovigilance is enormous, but anybody expecting a radical improvement in pharmacovigilance within the space of a short few months should probably stop and take stock of the bigger picture a little bit. Let’s not forget that machine learning is what some streaming services use to predict the kind of shows you might enjoy, based on previous shows that you’d watched. I don’t know about you, but in my experience, they make some wildly inaccurate predictions sometimes. Machine learning is also used by search engines to predict your entries and by giant online stores to predict the kind of purchases you might want to make.
The main thing to do now is to get ready for the role of machine learning in your PV solutions in the upcoming years, as opposed to the upcoming months. As for improvements in the field, predict them confidently, but be a little cautious in your timeline expectations. Because while we can reasonably hope that the machine learning algorithms for processing advert event reports will follow stricter data parameters than the algorithms designed towards expanding sales or streaming viewership, let’s not forget that the root of the technology being used is the same. With the arrival of machine learning in the widespread use of pharmacovigilance, you can expect almost instant gradual improvements right now, and radical improvements in the general vicinity of “tomorrow”. Just not the immediate tomorrow.
Cloudbyz Safety and Pharmacovigilance (PV) software is a cloud-based solution built natively on the Salesforce platform. It offers 360 degree view across R&D and commercial. It also enables pharma, bio-tech and medical devices companies to make faster and better safety decisions. It helps to optimize global pharmacovigilance compliance along with easy to integrate risk management features. Cloudbyz pharmacovigilance software solution easily integrates the required data over a centralized cloud-based platform for advanced analytics set-up along with data integrity. It empowers the end-user with proactive pharmacovigilance, smart features with data-backed predictability, scalability and cost-effective support.
To know more about Cloudbyz safety & pharmacovigilance contact info@cloudbyz.com