Brian Wellins, senior director, PV, discusses how our company philosophy, built on using best-in-class processes and technology, gives us better agility and supports our investments in innovation.
What specific artificial intelligence and machine learning techniques is PPD’s pharmacovigilance team currently implementing or hoping to implement in the future?
To date, we have invested in a range of innovative artificial intelligence and machine learning techniques, including the following:
- Artificial intelligence and machine learning for tasks that require decision-making from staff during pharmacovigilance case intake and triage
- Use of robotic process automation and optical character recognition in repetitive safety processes
- Use of native language processing for converting large amounts of text data into structured data format
- Use of native language generation to convert structured data into more readable text
- Automated quality control checks for case processing to lower vendor oversight needs
In collaboration with a software vendor, our team tested the ability of advanced data mining software with machine learning capabilities and native language processing abilities to analyze intake forms to streamline processes. We found the machine had the ability to nearly match the quality of output from our team. Our longer-term strategy for implementing this technology is to have our staff work with output generated from the machine in lieu of a traditional data-entry step. Additionally, the machine would make suggestions for the triage staff on certain items, such as case seriousness. From there, we would work within a feedback loop created between the triage team and the artificial intelligence and machine learning programs. This feedback loop would ensure the algorithm would continue to learn and improve performance based on the feedback from our experienced staff.
As with many novel technologies, it will take some time to validate the output and return on investment before we jump to the solution phase. This is where our disciplined approach to process improvement (vetted in lean six sigma methodologies) will ensure we are able to deliver on new technologies and able to offer those benefits to our clients.
We will continue to remain innovative in the industry and pursue opportunities that will advance our service offerings for clients. We have many exciting pilots planned for 2019.
How do these technologies help bend the cost/time curve in drug development?
The technology we are considering implementing will have the biggest impact on overall project cost and quality.
In our industry, there has been a trend toward delegating pharmacovigilance tasks to lower-cost geographical regions to create a more attractive service offering. An alternative model—one that we feel is worth pursuing—is to shift work to a machine-augmented environment, like the one described above. When the machine can assist our staff, it will allow us to scale up the work we perform without adding additional headcount. We will then be able to pass those savings off to our customers.
As for quality, we see artificial intelligence and machine learning as resources or tools that our staff can use to assist their usual processes. The tool on its own—or the staff on its own—would be less effective than the two working in tandem. We have witnessed plenty of instances in our pilots where machines catch certain quality issues differently than their human counterparts and visa versa. When a human and artificial intelligence work side-by-side, we find that our consistency and ability to find errors are both greatly improved. This helps safeguard our approach against potential system errors. We also think it will be a more tolerable approach for both our customers and regulators.
The improvement in efficiency and quality will allow us to help bend the cost and time curve for our clients without compromising the high-level quality we are accustomed to delivering.
Read Brian Wellins’s article “Is artificial intelligence the future of pharmacovigilance?”, originally published by Uppsala Reports, for more on artificial intelligence and machine learning implementation.