Pharmacovigilance in 2020 – future horizons and efficiencies in data acquisition, evaluation and risk management

To manage the ever increasing volume of information, companies have turned to cost-effective resources. Jeffrey Ho, Principal Consultant, Navitas Life Science, highlights how the advent of technology will contribute to a more sustainable model of pharmacovigilance

There is a continual increase in case volume, expanding sources of information and rising expectations from patients and regulators globally. How do PV functions achieve further efficiencies to continue meeting growing demands, make sense of data as quickly as possible and respond accordingly to ensure that patients safely benefit from medicines?

Over a million adverse events (AEs) were reported to the FDA in 2015, a five-fold increase from the 206,000 reported in 2004. Additionally, industry benchmarking shows an acceleration of the CAGR of adverse events to at least 20 per cent in recent years. If this direction of travel holds, we would expect to see case volumes increasing by another 50 per cent by 2020. A myriad of factors is contributing to this change. These include increasing public awareness of safety; rising expectations from regulators implementing EMA ‘GVP-like’ regulations, but with local adaptations; fast track / conditional approvals; apps and websites for consumers to report AEs; new product launches; and patient support programmes. The bulk of these cases are reported directly to the marketing authorisation holders (MAH), regulators, or partners. Even in light of these increases, there is the long shadow of what is not reported, which by some estimates, is as much as 90 per cent of AEs.

In addition, early indications from social media analysis suggest that these forums may not necessarily surface any unexpected adverse events, but could provide insight into real world data including misuse and methods of abuse. There is also more and more exploration of wearables in clinical trials along with public adoption of these technologies to monitor their own health. Along with this, digital assistants from companies like Amazon and Google are becoming more common in certain markets. It’s not to far fetched to imagine that in the near future, a patient will tell their smartphone that they take a medicine. This information could be sent to his/her primary healthcare provider and the MAH simultaneously near real-time.

To manage this ever-increasing, but also variable volume of information, companies have often turned to vendors for more cost-effective resources. However, this only means that vendors are faced with the very same challenge that the MAHs faced. At the end of the day, if there are 100,000 hours of additional work that needs to be completed by a certain deadline, additional resources will need to be deployed, be it with the MAH or with the vendor. And even if this challenge is addressed in one year, it is likely that the same challenge will be faced again the following year – and the following year. It is also worth noting that the pharmacovigilance resources are not ubiquitous, and following the current path, there are indications that this area will face marked skill shortages.

The convergence of these factors brings pharmacovigilance to a situation that is becoming untenable, and to address this, efficiencies must be identified.

The natural areas to turn to for this are process improvements and more significantly, technology. Process improvements can certainly have a profound effect on the efficiency of data acquisition. For example, implementing clear, consistent global processes; eliminating redundancies; and reducing hand-offs simplifies the flow of information and reduces the likelihood of mistakes. Judicious use of medical resources also enables for more scalability. However, some companies are approaching diminishing returns on these types of improvements because there is still a core of work that will inevitably need to be completed regardless of how optimised a process is.

This is where technology will play a prominent role. At the most basic level, common standards for safety information will reduce the double data-entry, or even triple data-entry often seen in ICSR processes. With the ongoing rollout of the more comprehensive E2B(R3) format, the same information can be consistently shared by MAHs, partners, and regulators without having to supplement the file format as was often the case with E2B(R2). It would be natural to see E2B(R3) as the de facto standard for moving information gathered via apps and other sources of safety information also.

The more significant benefit will be process automation. Some technologies like OCR and translation have been available for a while but have recently reached new maturity levels. It is unlikely that automation in these areas will eliminate the need for human participation, but the role of the human will likely shift to more of a quality one, checking the output of the automation tool and refining the output where necessary. The key is that the number of human hours required in process will be significantly reduced, thus allowing for a more scalable model.

Along these lines, but perhaps a little further into the future, the human will not only serve the role of refining the automation tool’s outputs, but also provide feedback to the tool so that the same mistake is not repeated. Machine learning and AI are already being piloted by some organisations. By 2020, some of the pilots will be completed, and the industry will have a better understanding of how these technologies will contribute to a more sustainable model for pharmacovigilance. The likely areas to benefit will be narrative writing with natural language generation (NLG), triage and medical review. Using a risk-based approach, human time could be focussed on serving as a second set of eyes for more ‘interesting’ information.

As the pieces come together beyond 2020, it can be imagined that new safety-related information is rapidly acquired, processed, and added to the data pool for signal detection. The trick is that signal detection will be like trying to find a needle in a haystack, when the haystack keeps getting bigger and bigger by the hour. This will lead to automated tools that continuously scan this pool of data for anomalies and promptly notify a safety physician or scientist to vet an observation. This will in turn   bring any confirmed signal to a cross-functional team to determine if the benefit-risk for the product has shifted. Once a label change or risk management action is implemented, technology supporting the monitoring of real world data could help gather evidence on the effectiveness of the action, and any refinements could be made accordingly.

Pharmacovigilance must find more sustainable models, and in the interest of ensuring patients can continue to safely benefit from novel medicines. This shift will need to happen soon.