In our recent webinar, “Improve Medical Product Information Sharing With Virtual Agents,” Prabha Ranganathan, Kari Blaho-Owens, and Nico Frantzen discussed the ways that implementing virtual agents in life sciences companies’ call center operations can help to promote pharmacovigilance and more effectively service customers.
Below are some of the questions we answered during the webinar and some that we didn’t have a chance to answer live.
1. What are the challenges with speech-to-text capabilities?
“Speech-to-text is all around us nowadays — in our phones, at home with your Alexa or Google Home devices, etc. It is a part of AI that has made tremendous progress, thanks to Deep Learning models’ abilities to process complex signals like the human voice. The challenge still remains in specialized domains or applications where uncommon or complex vocabulary, or jargon is used. In these domains, specialized and custom training is often required.” – Nico Frantzen, Director, Artificial Intelligence
2. What do you see as the challenges with different languages and dialects?
“There are still some significant challenges around models adapting to particular dialects and accents. Therefore, vendors and data science teams must continuously monitor, test, and validate their models against various large datasets that have to be regularly augmented with new sample audio.” – Nico Frantzen, Director, Artificial Intelligence
3. Does it make sense to integrate a virtual agent with WHODrug/MedDRA/VedDRA dictionary?
“Absolutely. The FDA has come out with guidance about AI/NLP/ML parsing data and coding AEs with MedDRA or VedDRA. It lends harmonization to the organization but does require revisiting with dictionary updates. The bonus is the downstream in the PV workflow; this task is already done and only requires review.” – Kari-Blaho-Owens, Ph.D., Director, Safety, Pharmacovigilance, and Regulatory Affairs
4. What types of things can be done to improve call drops? Are there any solutions for reaching back out to call drops? How do you ensure call continuation across channels?
“There are a number of techniques that can be used and implemented; for example, programs can be designed to detect a call drop and send an automated SMS to follow-up with the user to resume their call, or continue the conversation via chat or SMS. This is why choosing the right omni-channel platform is just as important as the choice of a conversational AI platform. The ability to switch seamlessly between channels is a very important capability to enable in your solution.” – Nico Frantzen, Director, Artificial Intelligence
5. We have a large volume of scientific resource documents. How do we make conversations friendly?
“The scientific resource documents need to be parsed and stored in other formats (like knowledge graphs) so that it is easier for AI/ML models to search and retrieve the right information for the interactions. The format and structure of the existing scientific resource documents should be examined to see how they can be converted to graph format, and this step will be automated to the fullest extent possible so that a large number of resource documents can be converted with less manual intervention.” – Prabha Ranganathan, Life Sciences Strategist
6. What serves as your source documents?
“The AI/NLP/ML can take voice and convert to text, or any other avenue of communication and produce an accurate source document that can be incorporated into an E2B XML and uploaded directly into the AE database.” – Kari-Blaho-Owens, Ph.D., Director, Safety, Pharmacovigilance, and Regulatory Affairs
And how do you make sure that AEs with source documents get into your safety database?
“As the case is updated after follow-up, the solution records outbound and inbound communication, source documents, etc.”
7. How do you maintain MI knowledge into Virtual Agents?
“A combination of scientific resource documents in the format of knowledge graph and training of AI/ML models to understand the conversation will enable the virtual agents to search/retrieve/maintain MI knowledge. A business rules engine combined with product knowledge from scientific resource documents will enable the Virtual Agent to have an intelligent human-like conversation with the customer. The primary source of MI knowledge will be the knowledge repository.” – Prabha Ranganathan, Life Sciences Strategist
8. Are we aware of validation requirements – described in multiple FDA guidances?
Yes, most of the validation requirements are identical to the guidances for computerized systems validation. There are other guidances published by global regulatory authorities that set the precedence for expectations for the validation of solutions that utilize AI, NLP, and ML. – Kari-Blaho-Owens, Ph.D., Director, Safety, Pharmacovigilance, and Regulatory Affairs
9. Do disaster recovery (DR) and business continuity (BC) plans come standard with such a solution (since this is cloud-hosted), or do we interface with the company’s DR?
“Perficient is capable of hosting this solution and maintaining it for our clients; we have a DR/BC plan in place to minimize the impact of an unlikely production outage. This is standard in our hosting environment. We encourage all of our partners to have a separate DR environment because it limits potential data loss and loss of communication.” – Kari-Blaho-Owens, Ph.D., Director, Safety, Pharmacovigilance, and Regulatory Affairs
10. Would Perficient provide a list of SOPs to go with the solution (as there will be a need for new SOPs), or is it up to the customer?
“As part of developing a customized roadmap and gathering requirements, Perficient is happy to author SOPs. Standard with any solution, we offer training, support after go-live, and user guides. Perficient will also partner with you in the event of a regulatory audit.” – Kari-Blaho-Owens, Ph.D., Director, Safety, Pharmacovigilance, and Regulatory Affairs
Curious to learn more? Watch the webinar recording here or below.
