Artificial Intelligence as an Aid to Pharmacovigilance

Unless life sciences organizations can find ways to filter the high volumes of online noise, their ability to stay on top of postmarketing safety signals as such signals emerge across the web will be impossible. This means up to 17% of adverse events will continue to be missed. Christopher Rudolf and Adam Sherlock explain how artificial intelligence can transform that challenging scenario through the application of reliable and efficient life sciences Web and social listening as an integral part of safety monitoring.

Good pharmacovigilance practice demands that life sciences manufacturers go further and become more proactive in keeping patients safe. But that’s easier said than done, as potential post marketing feedback channels multiply across the internet. Monitoring all of them is a vast undertaking for even the best-resourced safety teams.

Beyond manufacturers’ web sites, forums, incoming e-mail, and the scientific literature, important safety signals could manifest in Twitter and public Facebook posts, in independent patient forums, through special interest groups; and in blog articles and comments published on platforms like WordPress; and via more-visual channels such as YouTube and Pinterest. Signals could also appear in any language, anywhere in the world.

Monitoring scientific studies and the channels under brands’ control is mandatory, but extending the same vigilance across the whole of the public web is recommended to build a more complete picture-and ensure that no adverse events are being missed. Estimates suggest that 10 to 17% of adverse events go undetected today because companies are not “listening” to social media and other Web channels outside those currently mandated by regulators.

But how can companies reliably and efficiently achieve complete digital vigilance? Until now, the life sciences industry hasn’t found a definitive way to overcome that challenge-despite high levels of concern about it. (It was a hot topic at last March’s DIA EuroMeeting in Glasgow.) To date, the main options have been to buy very expensive proprietary turnkey analytics solutions such as IBM Watson Analytics or to patch something together from a series of tools designed for other purposes, such as for mainstream social listening. But what’s suitable for trawling Twitter is going to differ greatly from tools capable of scanning scientific literature or capturing the sentiment and context of online discussion forums, opinionated blog posts, or photo and video posts, whose potential signals are typically visually based rather than text based.

Unless their monitoring efforts are holistic and deliver something both meaningful and reliable, companies will be wasting their time and their budgets. With so many channels to keep track of, life sciences firms need a more intelligent and focused approach.

Speed and selection require intelligence and learning

This is where artificial intelligence (AI) comes in: by offering to take the strain off human teams so they can focus their time and budgets more productively. As optimized solutions for life sciences become available, AI is beginning to transform what pharmacovigilance and safety teams can do.

Natural-language-processing algorithms and artificial intelligence have made it now possible to sift and clean data, thereby reducing irrelevant or false-positive content by looking for the right signals that match teams’ criteria. The ability to interpret natural human language and semantics means the technology isn’t merely following blind search rules; it can identify mentions in context, and it can read into subtext to determine how relevant the mentions are. The parameters for that ability might vary between those needed to analyse a short mention on Twitter and those required to interrogate longer narratives on WordPress, but the technology is sophisticated enough to recognize those differences and adapt to them.

As teams interact with and classify data, machine-learning algorithms enable the software to observe and adapt to teams’ preferences and then hone the next iterations of findings accordingly. An added benefit is that different teams, with their own individual tasks and interests, can train a system in their own priorities and preferences so that the system supports their own particular purpose based on the same master data.

Filtered, meaningful data gets served to users via the equivalent of an e-mail in-box with options to both share findings with the team and feed important adverse-event findings into regulatory processes for urgent action. Incredibly, all of this can happen in near real time because the latest technology is capable of returning comprehensive but accurately filtered findings from entire global Web and social media searches within just 90 seconds, and red-flag events can be escalated to supervisors just as quickly. That kind of responsiveness to adverse events is unprecedented.

Achieving what humans can’t
 

Although many industries are concerned about AI’s undermining of people’s jobs, in a skilled and resource-pressed environment like the one found in life sciences quality/safety/regulatory affairs, time is of the essence and AI’s role is to free up life sciences teams to focus on what’s important. Given that the web never sleeps, another advantage of AI-based web and social monitoring is that it keeps working continuously-outside of office hours-so there’s much less danger of falling behind. What used to be an insurmountable task for humans has now become viable.

AI-based web and social listening can also significantly improve levels of accuracy and reliability of human-directed monitoring, so as to make sure nothing critical goes under the radar. Systems that have been built specifically for life sciences and that are supplied with preloaded algorithms have levels of accuracy that start high-above 80%-even before the software has been trained in what’s of interest to a particular company and business function. That’s a substantial advantage in winning the big-data war.

Without AI, firms would have to hire more and more people to trawl Google, look through individual Web sites and forums, capture reports, highlight what’s of interest, enter findings into a database manually, and pass anything important on to decision makers. It could take days to find something of value. And in the meantime, more could be missed.

To maximize the impact of AI-based web and social monitoring, life sciences companies must track everything holistically rather than in silos. There also has to be efficient work flow that feeds straight into established systems and processes-that is, existing pharmacovigilance systems and recognized regulatory procedures.

Another critical capability that an AI-based approach brings to the table is the ability to adhere to strict rules designed to protect patient privacy-rules that can differ from one international market to another. The technology can monitor for when relevant social media or web forum posts get subsequently deleted by the poster, for instance, and it can flag that in the pharmacovigilance database so that compliant processes around patients’ data privacy can be applied. Staying on the right side of regulators is essential in managing risk and maintaining public confidence.

Other AI applications

Once companies have seen the potential of AI-based web and social monitoring for the area of safety management, they start to realize its scope and ways it can be used in additional applications. With consent, firms have an opportunity to become more proactive in monitoring patient user groups about diseases those firms are interested in-for example, discussions about hay fever-to get a clearer picture of where they stand in the market and where needs are not being met.

Feedback, too, could help in the design of more-rounded clinical trials, which would supplement clinical data points, biomarkers, and reported outcomes with feedback from patients about how they are feeling and measurements from connected devices that track key health statistics. Once companies can monitor and cross-analyse all of those signals, they can start to predict adverse events before events happen in an individual and can then intervene preemptively.

Even if some of this feels a bit futuristic, intelligent web and social monitoring can add plenty of other value right now, today-for example, by helping companies keep track of regulatory changes and revisions to timelines, which are things that are hard to keep track of globally. Commercially, meanwhile, the ability to monitor how a brand or product is perceived in the market and how its safety profile measures up to the competition offers invaluable insights that could be fed back to product development and marketing teams.

Adoption rates

As things stand, levels of adoption and sophistication vary considerably. Some companies are already fairly advanced in their web vigilance and are now looking to see how they can drive further value. Others feel that they should be doing this and are actively researching the potential. Typically there are two drivers for this. Either they have taken an ethical stance where they want to be responsible to patients and they see this as an extension of that responsibility. Or they see digital vigilance as something modern and progressive, where they’d rather be ahead of rather than behind the curve.

There also companies that have attempted web and social monitoring but have found their teams overwhelmed finding due to all the additional ‘noise’ they are picking up, which is creating an insurmountable workload. (That is where we can help.) Still others are considering the opportunity but only want high-level dashboards they can monitor - without having to do any patient case-handling. The last group are the ones who for now are waiting and watching. It’s hard to estimate what percentages of companies fall into each category, but we currently see the whole spectrum.

Certainly a wealth of executable intelligence is out there-if companies can find efficient ways to extract and harness it. The good news is that all of the challenges companies are trying to meet can be overcome today. AI-based data monitoring and analytics for life sciences are becoming increasingly accessible and affordable too, thanks to the emergence of open-source solutions that can be run in the cloud and operated as a managed service on companies’ behalf, if needed.

Recent news reports have illustrated how AI and machine learning might help speed up patient diagnoses in the future because of the ability to reference and learn from patterns in global data at great speed. That ability has powerful potential in helping clinicians recognize rare conditions so that the conditions can be treated much earlier. And AI’s potential to transform life sciences insights and proactivity is exactly the same.

Christopher Rudolf (crudolf@volv.global) is CEO of Volv (www.volv.global). Adam Sherlock (asherlock@productlife-group.com) is CEO of ProductLife Group (www.productlifegroup.com).