Federica Santoro 0:15

Like many other health professionals, pharmacovigilance experts had to make huge adjustments to their processes during the pandemic. Caught in an avalanche of data but short on time and resources, they quickly had to rethink how they produced, used, and communicated safety information. And now that the pandemic has relented, we can't help but wonder, what did we learn from it all? My name is Federica Santoro and this is Drug Safety Matters, a podcast by Uppsala Monitoring Centre, where we explore current issues in pharmacovigilance and patient safety. Joining me today is Elena Rocca, Associate Professor of Pharmacy at Oslo Metropolitan University in Norway. Elena works at the boundary between pharmacy and philosophy of science. She formerly led Cause Health Pharmacovigilance, a project financed by UMC to investigate new concepts of causality in drug safety. When COVID hit, she got interested in how pharmacovigilance scientists were dealing with the emergency and how that experience was affecting the science of drug safety. So I called her up in Norway to find out what she learned. Hi Elena and welcome to Drug Safety Matters. You are one of those rare guests who actually volunteer to be on the pod. So thanks for being so brave and reaching out. It's a pleasure to have you here.

SPEAKER_00 1:54

Thank you, Federica. It's a pleasure to be here, of course.

Federica Santoro 1:57

So today we're talking about pharmacovigilance in times of pandemic. And that's because you wrote an article on this with Birgitta Grunmark, former pharmacovigilance expert here at Uppsala Monitoring Center. And based on what I read there, I think we're gonna have a thought-provoking discussion.

SPEAKER_00 2:16

I hope so, yes. I had uh luck of working together with Birgitta and other pharmacovigilant experts during the COVID-19 pandemics. Then I could bridge it to some discussions that are going on in the philosophy of medicine and philosophy of pharmacology, and that's how the article was born.

Federica Santoro 2:37

And I'm looking forward to hearing more. So let's get right into it. In the article, you list three factors that made it especially tricky to conduct pharmacovigilance during the COVID-19 pandemic. So the first one was increased uncertainty of evidence. The disease was new, the vaccines were new, and so were some of the medicines used to treat COVID. How did all that complicate life for pharmacovigilance professionals?

SPEAKER_00 3:08

Yes, it did complicate life, especially if we think about what is nowadays still the cornerstone of pharmacovigilance, which is spontaneous reporting. When we use these databases that collect reports of suspected side effects of medicines, you need to generate a hypothesis that some medicine is causally linked to a symptom. So you can use, you should use assessment of causality, of course, in single cases. So in every single report, you should try to evaluate whether there is, there might be a causal association between the medicine and the symptom. Then you could one should do it at the level of case series. So when you collect a series of cases, then you can use, for instance, a Bradfall Hill criteria to evaluate causality, or at a broader level with statistical methods. And of course, one uses all of these together. But uh when we think about the assessment of causality in single cases, which has been so important in pharmacovigilance, and this is something that makes it different for the rest of medicine, because in medicine single cases at the moment are not that highly evaluated epistemically. Epistemically means at the level of knowledge. So the idea is that to know something about causality, we need to have big data sets. But in pharmacovigilance, we still rely a lot in looking at every single case to try to evaluate causality. And what was the problem with the ascertain of COVID is that in order to try to assess causality in single case, you need to check the presence of confounders. So for instance, it is very difficult to assess whether there might be causality if a patient with arthritis says that he suspects that he gets stiff because of a medicine. Of course, arthritis gives stiffness. So I mean, how can we then assess whether this stiffness is something that is due to the underlying condition or due to other drugs or due to this drug? And in the causality assessment methods that we have now, if you see that there are illnesses or other drugs that could explain equally well or even better the symptom, then causality is not probable. It's either possible or unlikely. So, what was the problem with COVID? Well, it was a partially unknown and very complex pathophysiology. It was due to mainly to respiratory distress syndrome, but um you've got organ failure, thrombosis, multi-organ dysfunction, it was a complexity and an unknown disease. So we it was difficult for uh assessors to know like, is this a symptom due to the medicine or is this a symptom due to the illness that we don't know yet? But not only that, especially in the beginning, the first six months, you've got this uh emergency in which uh severely ill patients were coming to the hospital, and there was no real scheme of how to treat these patients. So they were treated with a huge amount of co-medications of everything that was um available at the moment just to try to save these patients. So you've got also the huge arsenal of drugs used at the same time. So it's very difficult to think, okay, what is what? What is this symptom due to? Then you've got new medicines that had an unknown safety profile, or old medicines that were repurposed for COVID-19. And of course, the safety profile might change with the background of a new syndrome. And also the fact that the risk population for COVID-19 was a particularly weak population with already existing conditions or elderly, and this can also contribute to give some of the symptoms. So it was really difficult to perform this uh causality assessment in the single case.

Federica Santoro 7:14

But even when the evidence is uncertain, regulators still need to make decisions, and in a healthcare emergency, they need to do that even faster, one could argue, so that patients are exposed to as little harm as possible. But that's obviously easier said than done. How did regulators cope in practice with this uncertain evidence during the pandemic?

SPEAKER_00 7:39

Yes, here comes the interesting part. There were actually two main lines of action that also reflect two main ideas of you know how one should collect knowledge or two main ideas of what is causality, if you want. And one line of action was to focus on the population level because we are in lack of a precise single-case causality assessment. So some said, well, it's impossible here to rely on methods that require to evaluate the difference made by each single causal factor. So we shouldn't rely on that. We should then go to the statistical level. So let's forget about the causality evaluation in a single case, and let's look, for instance, at whether a certain combination of drug and symptom is more interesting because, for instance, it's reported more than expected. So, for instance, the drug remdesivir, that was almost a new drug, let's say, it's not completely new, but it was started to be used for COVID-19, so the safety profile was unknown before, was reported a lot together with the renal symptoms, so renal failure. Okay, so pharmacovigilant experts say that they're reported disproportionately, that means more than expected if they were reported by chance. And this might indicate causality. I mean, it's not perfect, but this is what we can rely on. Pity is that other statistical uh designs gave different results. For instance, later on it was possible to make uh cohort studies, so retrospective cohort studies. That means that you take a group of uh patients who were treated with Radiasivir, another group that wasn't treated with Radiusivir, and you look at observationally, it's not an experiment. You just look at what happened with them, and it doesn't seem that there was more renal failure in one group respect to the other. So people who were uh supporting one or the other statistical design were criticizing each other, were like warning against the use for instance of disproportionalities. Another line of action instead, I would say it was the reverse, which was improving clinical quality over the quantity of data. In some countries, there is a kind of decentralized reporting system, for instance, in Norway, in France, and some other European countries, maybe also outside Europe that I don't know about. So in these countries, you have a decentralized pharmacovigilance system. That means that you have clinicians reporting side effects who are out there, for instance, in the hospitals. Experts who look at clinical cases and make the reporting in very big clinical detail. And the point is that this allows you to understand the phenomenon at place in the context. So you've got a deep contextual understanding. Because, of course, no one says that these things should be mutually exclusive. No one in pharmacovigenists would say this is mutually exclusive. So, of course, the best is to have the deep clinical understanding together with the statistical designs, but in times of uh emergency and with limited resources, then you you need to choose where to put your effort. And this is a bit interesting because uh it's another idea of causality, which is a singularist, so that causality happens in the single case, or that the single case is the, they would say in philosophy, is the truth maker of the statistics. So the statistics are there because there are single cases where there is the causality happening, and at least some cases worked. For instance, it has become rather known from the media the cases of certain COVID-19 vaccines that provoked very rare side effects. For instance, this intracranial venous sinus thrombosis together with thrombocytopenia. So it is a very rare side effect, and it was promptly detected through the analysis of few, very few single cases. So there are these two different lines of actions, let's say, that were used.

Federica Santoro 12:03

Thank you. And we'll get back to that: the importance of statistical tools on one hand and clinical expertise on the other a little bit later. But now I wanted to ask you what we learned from the crisis management point of view? Because the COVID-19 pandemic was certainly not the first healthcare emergency the world had to face. And there are plenty of resources out there on effective crisis management. But what would you say the pandemic added to that body of knowledge? Did we learn anything new about dealing with uncertainty?

SPEAKER_00 12:41

Well, I am sure that if you ask this question to different people, they would answer a lot of different things. So I'm sure we learned a lot about dealing with uncertainty. From my point of view, at the side of all the practical advances, there were some conceptual advances and concepts are needed to make practice better. I think that we were reminded of why is, for instance, single-case causality so important in pharmacovigilance respect to the rest of uh medicine. And this is because pharmacovigilance is an explorative practice and it is a practice that should catch the unexpected. There is a whole name for that, which is serendipity, and there is all a range of experts that study how things are discovered in science that care about serendipitous discoveries, which is discovering things not during a normal scientific process. So you start with a hypothesis, you set up a test, you try to corroborate the hypothesis or not, etc. No, there are these other types of discoveries that happen just because things emerge from otherwise different practices. So for instance, in pharmacovigilance, observations emerge from clinical practice. So what do we need for these uh serendipitous discoveries? Well, the tradition was well, we need a chance and a prepared mind. Well, this shows up to be not the whole thing. It's not enough to have the genius who makes an observation and understands the importance. That's just a narrative, a myth. That's not how this type of things happen. What happens is that there is maybe a person, there is the chance, the observation, and there is a network ready to act. Something that is normally transdisciplinary, there's a diversity of expertise, plurality of methodological perspectives, and there's not rigidity. You know, rigid methods are not apt to catch the unexpected. Rigid methods, for instance, study designs, are very good for confirming something. What we learned is that we need to settle the premises for the unexpected to be found. We knew that, but we were reminded the hard way. For instance, having clinical expertise and uh in situ-causal evaluation, it was even more important when uncertainty was high. And uh that's because, in my view, we need to have these responsive networks at place because we have to prompt them when necessary. It's like uh the immune system, you know, it's a responsive network we have at place, and it's prompted just when necessary. And that's the same thing when we want to have this kind of discoveries from the unexpected, for instance, pharmacovigilance uh discoveries. I think that what we learn is that we have to nurture this kind of transdisciplinary networks and that decentralized systems win, and that uh plurality of methodology also wins over rigid uh designs.

Federica Santoro 15:56

Important lessons learned. Now, the second challenge pharmacovigilance experts were faced with during the pandemic was the unprecedented amount of data coming in. And that's not difficult to understand why. When COVID vaccination programs were launched around the world, millions of reports of suspected side effects began to pile up, partly because just a sheer number of people the vaccines were being administered to, and partly because regulators deliberately went out there and encouraged reporting so that they could detect any emerging safety issue quite quickly. So, how did the field cope with this deluge of data?

SPEAKER_00 16:41

It was quite a job. Of course, the system at place with this data mining approach is efficient, but it is also very time consuming because the reports that come in, they need to be, well, if they're not already digital, they need to be transcribed. Many countries have already a digital system at place, but then even if they're digital, they need to be coded because nothing becomes data. Information becomes data when they are apt to go into the CISES. So symptoms and drugs need to be coded following a standardized vocabulary, they need to be structured by the national agencies. This became an enormous amount of work. I talked to very overwhelmed and tired uh experts who were working at the national agencies. And I mean, from January to June 2021, at the time when I wrote this article, June 2021, I checked and I saw that there were 1,100,000 COVID vaccine reports in VGBase that came in that six months. This is unprecedented. So this was also a chance, especially with the manufacturers, so the industry, to implement new technology often based on artificial intelligence. For instance, automatic coding of the adverse drug reactions into standardized medical terminology, or even automatic translation into and from different languages. Because of course, although the terminology, for instance, the major vocabulary is in different languages, still a lot happens in English. And also there was improved meaning of unstructured data. Unstructured data means that place of the questionnaire where you are free to fill in the question. For instance, describe how you feel, describe your symptom. That's unstructured data. That you know you need to somehow put it there in the system to be retrieved. So, how do you do this? You have to try data mining also this kind of data, or even in clinical studies, social media, so this kind of data mining was improved to show how important these type of skills became. EMA, the European Medicines Agency, in 2020, established a big data task force to build technical skills, capacity, and tools to analyze jointly different types of data sources, for instance.

Federica Santoro 19:22

And of course, it's great that the pandemic served as an incentive to invest time and resources in developing all these things: statistical tools, data mining algorithms, artificial intelligence technology, and all that. But as you mentioned earlier, tech solutions alone won't make medicines and vaccines safer. So let's explore this topic a little more. What role do clinical experts play, in your opinion, in the age of big data?

SPEAKER_00 19:52

Yes, let's start from the naive empiricist ideal, which was an ideal from you know the 1930s and no one believes in that anymore, apparently, which is that scientific truths will just emerge out from the data. We don't need to interpret or anything, we just need uh good perfect data, and then they will tell us the truth. The reason why we don't have this is that it's impossible to get perfect data. But if we could, then the truth would emerge out of the data. And this is an idea that has been abandoned because we know that we need to interpret it, etc. But there's been a kind of revival of this idea in the era of big data. Some even asked whether the big data is a death of theory, is it a death of interpretation? So, are we going now to have so many data that what we we need is just to improve the way we handle this data? And that's everything we need. So we need this technical expertise, data science, and then we will have truth emerging. Okay, but you know, there are some philosophers of big data, for instance, Sabina Leonelli, who's uh UK-based, who work hard to show how the computational mining of big data involves a lot of theory commitment. So the contrary, nothing emerges directly from data. For instance, uh she says, in order for information to become data, you need to choose and define keywords used to classify data. As I said before, you've got a symptom, it's classified in a standardized vocabulary. So, how you choose to classify that symptom, and how do you even more you choose to link together different concepts. When you use a database, the advantage is that you can retrieve some groups of data together, but the possibilities you have to retrieve these groups are possibilities that are decided by the curators. So, me as a user of a database, that I can retrieve some groups is because someone thought that somehow. These groups belong together. Why do they belong together? Why do all vaccines, all types of drugs that interact with a certain enzyme, I mean, why do they belong together? For instance, if you think that the vaccine platform has a say in the way vaccine acts and interacts with the body, then it makes sense to group the vaccines based on their platform. And this has been done. So for instance, it's possible to retrieve all life-attenuated vaccines or RNA vaccines. But this kind of thinking, this is not generated by the data. This is not something that jumps out from the data, tells me how to organize this data. This was a collaboration, very nice, between data scientists and people who know about how the body works, how drug works. So clinicians, pharmacologists, this kind of collaboration makes the database better because when you have better retrievers, more ways to retrieve that are more relevant, then you can use it better.

Federica Santoro 23:00

But are we saying that collecting data in a purely unbiased way is impossible? I am saying that.

SPEAKER_00 23:12

So I'm not saying that everyone will agree. But um the ideal of a view from nowhere is an idea that has been abandoned and should remain away from science even in the era of big data. There's a lot of work in biology with Sabina Lonelli and her group in Exeter in the UK, where she shows how much theoretical commitment and human labor there is if you want big data to work and be useful because data have to travel. You know, information needs to become data, and this data have to travel and to be repurposed. So this work that has been done in the realm of biology, I think it's time now to apply it also to pharmacovigilance, which is so data-driven.

Federica Santoro 24:02

Fascinating. And I'm curious to see what our listeners will think about this. So I do encourage them to share their thoughts on social media when we share this episode. And big data doesn't only pose practical challenges like how do we deal with all this information, which we talked about so far, but also some important ethical challenges. For example, how do we share personal data without infringing privacy? Or how do we strengthen all these high-tech solutions for data processing without leaving low and middle-income countries behind?

SPEAKER_00 24:42

Exactly. I don't have any answer on the house. But I I do think that these are big questions that came to the forefront, and it's important to notice that. And I think another bridge that it would be important to create is between pharmacovigilance and this new field, I mean, relatively new field called data ethics, which studies moral problems related to data, algorithms, and all these practices that are connected to them in order to support morally good solutions. And why do we need this? Well, we need it a lot. For instance, because we have this uh false belief, this cultural assumption that if you increase the data set and you perfectionate the algorithms, then you increase the objectivity of the research, and then you are less discriminatory because then the algorithm will do it for you, and you have this inclusive data set, so it you include everyone. It's better than having a single researcher thinking, a single experiment, small data set. Well, that's often not that's often dangerous to think like that because there are systematically inbuilt discriminations that are carried on no matter how big the data set or sophisticated the algorithm. There are many examples. I advise a book called Weapons for Mass Destruction. There, there are so many of these uh inbuilt systematic errors inside the databases and algorithms in many fields. And uh we saw this in uh when the system, of course, rightly enough, implemented a lot of these big data pharmacovigilance, uh, artificial intelligence, data processing. I mean, this was a good thing to do, but we also have to keep in mind that then we represent almost exclusively European countries and the Americas and a handful of other countries globally. So, for instance, 80% of COVID-19-related adverse reaction reports that were shared in VigiBase in 2020 were from Europe and the Americas. Only 1% came from the Africa region. And this difference is always there, but during COVID-19 was even more extreme. So this shows that uh there are global differences in medicine availability, this we knew, but it has become more pronounced during the pandemic. And this is just to say that these are things we should keep in mind when implementing these automatized systems.

Federica Santoro 27:09

Exactly. Important issues we have to keep addressing. As you say, they won't be solved overnight, but we just have to keep them in mind. I've kept you here for a while, so I think I'll start to wrap up, even though this is such an interesting discussion, and I know you could keep talking about this for ages. I can tell you're passionate about the topic. I am. But uh to sum up, we can conclude that the pandemic challenged to the core the way we conduct pharmacovigilance. What needs to change in our current safety monitoring setup to be better equipped for future emergencies?

SPEAKER_00 27:51

Well, I think that both high uncertainty and increased focus on big data require to strengthen interdisciplinarity, interdisciplinary networks between clinicians, pharmacovigilant experts, regulators, data scientists, curator of databases, data ethicists, philosophers of science. So at the moment, of course, wherever you look, there is an increasing uh demand of interdisciplinary practice. I mean, uh, to get the research project finance, you have to show that it's interdisciplinary, etc. But if you look at the reality of what is happening, I think that education, research funding really, scientific journal, regulatory systems maintain a disciplinary focus.

Federica Santoro 28:35

So, how would you go about changing that?

SPEAKER_00 28:37

Yeah, I would start from education, and I am very glad that now I have uh the possibility to do that because uh I will teach pharmacy students. There needs to be this education to interdisciplinarity, very difficult. All universities want to do it, and no one knows how we will do it. So it's a difficult thing to do. Then uh I think the discussion fora platforms, specialized journals, social media should give space also to the big questions, like the big picture. Like, for instance, how do you show causality? How do scientific discoveries happen? What kind of scientific discovery do we have in pharmacologists? These things, they're not very you don't read a lot in these specialized journals. It's like kind of too conceptual for the users or something like that. And then the system should give incentives to be interdisciplinary. It's not easy to be an interdisciplinary researcher, for instance, because it's hard to find a journal who wants to publish your stuff because no one recognizes it. Editors don't recognize that this belongs to their journal, and reviewers don't recognize the methods, so this also needs to change. But then the big question is that who changes it? Who takes the responsibility for this? Everyone thinks it's someone else's responsibility. That is also part of the problem.

Federica Santoro 29:57

And obviously, you're a prime example of this kind of interdisciplinary mindset. You trained in the life sciences and then got increasingly interested in the philosophy of science. So, a final reflection then. What attracted to philosophy of science and what do you think it can ultimately bring to the pharmacovigilance table?

SPEAKER_00 30:21

I think it can bring more expertise on core questions that are core in pharmacovigilance, but are also core of philosophy of science. There is this common interest, for instance, what are causes and how do we find them? I mean, what more than pharmacovigilance is worried with causes of symptoms, and this is a big question in philosophy of science. But also, what is the value of outliers and unexpected observations for scientific understanding? How do scientific discoveries happen? What is an expert? I mean, there is a whole field talking about what is expertise, what is an expert. But also, how do social structures, how do communities help or uh stop the process of discovery? And then there is also all the discourse on big data, data science, data ethics. I think there's really a lot to be gained from this uh marriage. So we hope the marriage uh happens.

Federica Santoro 31:21

And that it will last long, and I am curious to see how it will evolve. Thank you very much for your time, Elena, and for sharing your thoughts on this interesting topic.

SPEAKER_00 31:32

You're very welcome, Federica. Thank you.

Federica Santoro 31:48

If you'd like to know more about Elena's work and pandemic pharmacovigilance, check out the episode show notes for useful links. If you like our podcast, subscribe to it in your favorite player so you won't miss an episode. And spread the word on social media so other listeners can find us. Apart from these in-depth conversations with experts, we host a series called Uppsala Reports Long Reads. A selection of audio stories from UMC's Pharmacovigilance magazine. So do check that out too. Uppsala Monitoring Center is on Facebook, LinkedIn, and Twitter. And we'd love to hear from you. Send us comments or suggestions for the show, or send in questions for our guests next time we open up for that. For Drug Safety Matters, I'm Federica Santoro. I'd like to thank Elena Rocca for her time, Matthew Barwick for post-production support, and of course, you and all the other listeners who tuned in this year. This is the last episode for 2022, but we'll be back in the new year with more stories from the world of drug safety. And I hope you'll be there too. See you next year.