Chatbots and virtual assistants: How might they be used in pharmaceuticals?

When browsing the web and using smartphone applications, users increasingly encounter chatbots and virtual assistants (some more sophisticated than others) that interact with them and generate conversations. This is in part due to the development and implementation of artificial intelligence systems.

Numerous websites and apps, from e-commerce sites to sharing platforms, have already implemented programs capable of simulating conversations with users through chats in pop-up windows. Often these support users by answering frequently asked questions about services (a more evolved version of the traditional written FAQ) or provide initial assistance for purchasing or using a service.

Many of these programs interact using preset question/answer sets and cannot go beyond the boundaries of the areas for which they have been programmed. They are unable to adapt a conversation in response to the individual user, as they do not access real-time data sources.

In some cases, however, users may interact with virtual assistants with broader function that can provide complex answers to general questions not on the basis of preset information, but by drawing on real-time data and adapting answers in response to an individual user’s previously saved preferences, relying upon generative artificial intelligence (AI) and machine learning (ML) systems. Google’s Alexa and Apple’s Siri are two examples that have been on the market for a relatively long period of time.

These systems are expected to become increasingly common on websites and applications that offer goods and services to users.

Recently, Amazon was reported to be testing a new virtual assistant for shoppers named Rufus.^[1] Rufus uses generative AI for new forms of product research. For example, a user might ask a general question, such as what kind of product they need to perform a certain activity (from gardening to skiing), what features they should consider when buying a product for a certain use (technology products such as headphones for home use), or even what the difference is between product 1 and product 2 and then receive personalized answers from the software based in part on their purchasing preferences.

Similarly, the well-known e-commerce platform Zalando is in the process of launching a generative AI-based shopping assistant that will rely upon ChatGPT.^[2]

The pharmaceutical sector is unlikely to remain immune to this paradigm shift in user/patient relations, and, in fact, the first moves in that direction are already taking place in the sector.

AI and pharmaceutical promotions

The Italian regulatory framework is particularly restrictive with regard to advertising medicines to the public. Advertising for prescription drugs is prohibited and advertising for OTC and SOP drugs is heavily regulated.

Therefore, it is difficult to imagine virtual assistants being used for promotional purposes, such as on sites for specific products like drugs or medical devices.

Indeed, the authorization procedure and promotional messaging outlined in the Drug Code and related ministerial guidelines clash with promotional interactions created and personalized in real time by generative AI systems, including by means of virtual assistants.^[3]

Prior authorization of a promotional message by the Ministry of Health and the consequent and necessary “static nature” of the message, as the authorization states that the message must remain unchanged after receiving the ministerial green light, appear completely irreconcilable with the use of generative AI systems. Also, it is inconceivable at this point that a virtual assistant used for promotional purposes, capable of generating varied content, could be authorized, since the specific message is the focus of the authorization procedure.

There might be limited leeway for using this tool to answer unsolicited questions from third parties. That could be considered non-promotional under the definition provided in the Drug Code and therefore not subject to the relevant advertising regulations. However, even in this case, problematic issues would arise related to the type of information provided to patients. Society (and the medical field in particular) must be able to exercise absolute control over that information to ensure that it is high-quality and scientific.

What about e-commerce sites?

The issues arising from potential promotional content in messages produced by a chatbot are also a concern with regard to e-commerce, so it is crucial to ensure that users can request information and interact with pharmacists. This includes obtaining information about products and purchasing advice—just as if the customer were in a physical pharmacy. The question is to what extent an automated tool such as a chatbot can replace a human pharmacist providing such services.

A tool might provide answers about things such as placing orders, prices, and transportation, but it must tread cautiously in answering questions about medications or offering medical advice. Such tools cannot replace the presence—even the remote presence—of healthcare professionals, be they doctors or pharmacists.

Additionally, when users are free to provide content through digital channels under the oversight of a pharmaceutical company, that company is subject to pharmacovigilance regulation obligations. The Good Pharmacovigilance Practices (GVP) guidelines explicitly place the onus on drug marketing authorization holders to perform regular audits and oversight of channels under their management or responsibility to look for potential adverse reaction reports that must be transmitted to an authority.

AI and patient support

Virtual assistants and AI might also provide services and support for users and patients, without any direct or indirect promotion of pharmaceutical products.

For example, this could occur in Patient Support Programs (PSPs) sponsored by pharmaceutical companies and delivered through third-party providers to patients being treated with their products, under the Farmindustria Code of Ethics, or a “digital caregiver” service might support patients during treatment. This type of scenario would give rise to the same issues discussed above; oversight of interactions with patients would need to be in place to prevent the virtual assistant from making diagnoses or providing advice that may only be dispensed by health professionals and to ensure that pharmacovigilance obligations are met.

Chatbots and medical devices

Lastly, the more sophisticated AI systems become, the greater the risk that related software will fall under the definition of a medical device and therefore will need to obtain the relevant certification and comply with the relevant European and national standards, starting with Regulation (EU) 2017/745 (MDR) and the numerous guidelines issued by the Medical Device Coordination Group (MDCG).^[4]

If a virtual assistant can analyze and reprocess data provided by users/patients (uploaded reports, information derived from questionnaires) to provide a user with feedback in the form of diagnosis, prevention, monitoring, treatment, or mitigation of a pathology or physiological state, it qualifies as a medical device under the definition provided by Article 2 MDR.

Upon approval and full implementation of the European Artificial Intelligence Regulation (AI Act), software equipped with AI systems and qualified as medical devices will be classified as high risk for the purposes of the AI Act, which means they will be subject to additional compliance assessments under the AI Act.^[5]

The difficulties and obstacles involved in employing these tools in the pharmaceutical sector are clear, but equally clear are the opportunities for improvement and personalization of patient care—so we can expect them to develop rapidly in the years to come.

[1] https://www.aboutamazon.com/news/retail/amazon-rufus.

[2] https://corporate.zalando.com/en/technology/zalando-launch-fashion-assistant-powered-chatgpt.

[3] Legislative Decree No. 219/2006.

[4] On the subject of qualifying software as a medical device, particularly MDCG Guidelines 2019-11, see https://health.ec.europa.eu/document/download/b45335c5-1679-4c71-a91c-fc7a4d37f12b_en?filename=md_mdcg_2019_11_guidance_qualification_classification_software_en.pdf.

[5] https://portolano.it/news/ai-act-quale-impatto-sullindustria-dei-medical-devices.