Automation, machine learning, artificial intelligence, robots – hearing such terms mentioned in the workplace can strike fear in the everyday employee, that soon the human workforce will be replaced entirely by a ‘digital workforce’. And of course many sci-fi movies only play on this hype. Whilst in certain industries or for specific job roles there may be more than an element of truth to this, in the pharmacovigilance world, using machines to supplement human input is not only a necessity but also a blessing. Or so I learnt at the International Society of Pharmacovigilance (ISOP) Seminar on Intelligent Automation in Pharmacovigilance, held in Boston in December 2017.
The Seminar brought together delegates from the pharmaceutical industry, technology companies, service providers, regulatory agencies and other organisations with an interest in pharmacovigilance. The scene was set by Priya Singhal (Biogen) in the introductory session, who noted that although there has been an explosion in technologies in recent years, the top priority for pharmacovigilance professionals is to focus on the risk-benefit balance for patients, and therefore we should be implementing the best systems to address this. With that in mind, it is certainly not a given that any form of artificial intelligence (AI) will be a 100% solution. April Davis (Accenture) and Andrew Rut (MyMeds&Me) reminded us of the old adage of ‘data out = data in’, meaning that no amount of polishing by a state-of-the-art AI tool will rectify fundamental gaps in information from the source. So whilst complete automation of simple case reports could be successful, more complex cases will still require human input.
In fact this was a recurring theme across the sessions, that human input will always be required for highly skilled tasks, such as for medical review and handling outliers, and potentially for training the machines in the first place. So pharmacovigilance professionals can breathe a sigh of relief. Well to an extent. As although there will still be roles for people in the PV space, over time they will need to up their skill levels and even include a level of tech-savvy, as an over-seer of the work the robots are doing.
So why is a digital workforce a necessity? This can potentially be answered in two words: big data. The FDA has over 14 million reports in FAERS, the WHO ADR database has over 16 million reports, and at Adis Pharmacovigilance we are experiencing a steady increase in the volume of ICSRs published in the literature each year. And that’s not even considering ADR reports in social media. Add to this the increase in the number of drugs, complexity of drugs and level of detail of information about drugs and it is a wonder we are not drowning in data-overload.
During the seminar we heard from several organisations how they are currently using or evaluating automation to assist in the handling of these vast volumes of data including:
- Automated follow-ups of ICSRs at AstraZeneca
- Computational methods, including Natural Language Processing (NLP), to enhance signal detection by the UMC
- Information retrieval, case classification and case summarisation using NLP
- State-of-the art auto narratives for ICSR processing
It was great to hear these real-life examples, as they provided insights into the practicalities of integrating AI technologies into existing workflows as well as the key considerations when selecting which workflows to target and choice of technology and vendor. The presenters also showed that even though implementing such technologies is not without challenges, the anticipated benefits include more consistent processing, faster handling of data, improved quality, and ability to visualise large amounts of data. Therefore, ultimately freeing people from doing menial and repetitive tasks, to focus instead on the bigger picture of risk management and strategy.
The panel discussions and Q&A sessions were lively as there was still a feeling of uncertainty in several aspects, such as
- What is the position of regulatory authorities on using AI technologies?
- What is the ultimate benefit to patients?
- Can AI account for cultural differences in reporting?
- Is the industry actually ready for these technologies, in light of where we are with social media reporting, and the lag time for regulatory changes?
- What are the acceptable levels of quality from AI?
- Who owns the intellectual property and who bears the cost?
However, despite the uncertainty, one thing that was clear from all the panellists, speakers and moderators was that AI technologies are expected to become common place in the industry within the next 3-7 years. I, for one, am looking forward to embracing this inevitable change that will enable the industry to move forward in the face of big data, allowing us to see the wood for the trees. And I also believe that Polanyi’s Paradox, “We can know more than we can tell”, is applicable to AI in pharmacovigilance, because machines can never quite equal the subtleties of human skills, cognitive capabilities and tacit knowledge in this field.
Image credit: Johannes Plenio