Oropouche, Drug Resistance, and AI in Outbreak Surveillance
Q&A with David Hamer, MD (Part 2)
Davidson Hamer, MD is Professor of Global Health and Medicine at the Boston University Schools of Public Health and Medicine, co-lead of the climate change and emerging infectious diseases research core at the BU Center on Emerging Infectious Diseases, and an attending physician in infectious diseases and Director of the Travel Clinic at Boston Medical Center. Dr. Hamer is a board-certified infectious disease specialist and medical epidemiologist with particular interest in emerging arboviral diseases, tropical medicine, travel medicine, infection control, and antimicrobial resistance. Dr. Hamer has been involved in travel medicine for thirty years. From 2014 to 2021, Dr. Hamer served as principal investigator and, since September 2021, as Surveillance Lead, of GeoSentinel, a global surveillance network of 68 sites in 32 countries using returning travelers, immigrants, and refugees as sentinels of disease emergence and transmission patterns throughout the world. He also has been actively involved in enhanced screening for and treatment of patients with Chagas disease at Boston Medical Center. Dr. Hamer is currently the Scientific Program Chair for the American Society of Tropical Medicine and Hygiene and Section Editor of the American Journal of Tropical Medicine and Hygiene (global health and Chagas disease).
Oropouche Virus and Emerging Arboviral Threats
Tom Mahoney: Let's start with an emerging infectious threat in which you're certifiably one of the world's experts, namely Oropouche virus. GeoSentinel issued an Oropouche alert in mid-2024 as cases surged across the Amazon basin and spread to previously unaffected areas. At what point did your network first recognize this as a signal worth elevating - and what did the traveler case data tell you that domestic surveillance in source countries was not yet showing?
David Hamer: Probably the first inkling we had about this disease was a report from our site in Atlanta of a patient who had been in the Ecuadorian Amazon who appeared to have Oropouche virus. Interestingly, that case ended up not being Oropouche once it was sequenced further. PCR can cross-react with Madre de Dios virus and Iquitos virus that are very similar, and so we must look at different segments of the viral genome to be able to specifically identify them. There is evidence that major reassortments of the genome of Oropouche virus have led to the emergence of novel strains that appear to be more pathogenic and for which past immunity is not protective. In this case the patient turned out to have Iquitos virus. This case appeared when Oropouche was exploding in the Brazilian Amazon, but then an introduction of the virus to Cuba eventually was detected by our site in Brazil. We were able to sequence one of the viruses and show that temporally, it was associated with the outbreak in Brazil, so probably a Brazilian traveler introduced it to Cuba. We were able to characterize a subsequent big Cuba outbreak somewhat, but importantly, our site had the foresight to learn more about the virus to see how long shedding occurs, and sequenced the virus, leading to the second confirmation that it probably came from Brazil. But they also looked at one patient who showed shedding of the virus in semen. Nobody had looked at that mode of transmission before, but our site was able to show not only were semen samples PCR positive, which could be due to viral particles and not live virus, but they were able to culture and grow the virus and see it at day 16. This suggested the potential for sexual transmission of the virus, which was completely novel, and the tricky part was sharing that information with CDC and public health authorities. Our team was concerned initially that they might get scooped on this novel finding, and were hesitant about sharing the data, but we convinced them that for public health purposes, CDC would keep it confidential, and they did. CDC then put out an alert saying there's a potential risk for sexual transmission of Oropouche virus and made some recommendations online very similar to what had been used for Zika virus. Eventually, that warning was published and subsequently more widely disseminated. As of today, there has been no confirmed case of sexual transmission of Oropouche virus, but definitely there's a possibility that could happen.
Mahoney: The 2024 Oropouche outbreak also raised alarming signals around vertical transmission and deaths in previously healthy adults - features not prominently described in prior outbreaks. How is GeoSentinel positioned to help characterize whether these represent true viral evolution versus improved detection?
Hamer: That's hard to say. I think that in the direction we're going, doing more sequencing, we're going to be able to say that the virus has changed. The Brazilians have done a fantastic job with Oropouche virus by demonstrating multiple cases of mother-child transmission, often with either a miscarriage or a severe congenital syndrome. GeoSentinel has not yet been able to make this kind of identification, but I think we'll be able to in due course as we did with Zika. I think sequencing this virus and being able to compare how it has evolved is something we're going to be doing more of. The Brazilians also showed that the Oropouche virus that caused a big outbreak in 2024 is a reassortment, very different from the variant that had been circulating a decade earlier, so people that had immunity from the prior version were no longer protected. That was a very important finding, and I hope we'll be able to do more work like that in the future.
Mahoney: More broadly, arboviral threats - Oropouche, dengue, chikungunya, and emerging flaviviruses - seem to be expanding their geographic range, likely driven by climate change and vector dynamics. Is GeoSentinel's traveler-sentinel model well suited to tracking that range expansion in near real-time?
Hamer: Absolutely. In near real time is a bit of a challenge, but through travelers, and sometimes immigrants, we are able to identify arboviral infections, whether chikungunya or, say, Mayaro, another alphavirus that can cause an arthritogenic syndrome and we're really worried is going to be one of the next big epidemic explosions in the Western Hemisphere. Fortunately, our ability to detect emerging viruses in returning travelers is high. At many of our sites, there are potential delays in entering the data in the system, and we've been trying to address this challenge by increasing incentives. We give very small incentives to sites to enter records. We've changed that system a bit, but given the new limits on CDC's funding, we have spent sometimes as much as a third of our budget on incentives, and yet these aren’t fully adequate, in my opinion. We've tried to get our network to focus on arboviruses, because identifying them early can be useful for public health authorities such as CDC to issue a traveler's health alert, saying for example there's chikungunya in this place where it has not occurred before, or not for many years, and you should consider being vaccinated. Chikungunya is manageable because there's a vaccine. Dengue, unfortunately, has no vaccine available in the United States, but most European countries have TAK-003 available. So, there are responses that can be useful for travelers and for practitioners, and if we share those data in-country, it can help, with either vector control or enhanced surveillance.
Antimicrobial Resistance (AMR)
Mahoney: Travelers have long been recognized as vectors of antibiotic-resistant organisms. What is GeoSentinel's unique contribution to the global AMR picture - what does the network see that hospital-based or community surveillance systems miss?
Hamer: I think geographic variability is one key item that gets missed. There may be surveillance systems for, say, S. typhi and paratyphi within countries in South Asia, Southeast Asia, to a lesser extent Sub-Saharan Africa, but we're getting a global picture, seeing people returning from pretty much all continents, and almost every country and territory worldwide. I've even seen a couple of cases from cruise ships near Antarctica. But then doing analysis longitudinally, looking at regional differences in antibiotic susceptibility, and how this is changing over time, is a unique contribution we’re able to make. We did one analysis, up to, I think, 2007 or so of S. typhi and paratyphoid enteric fever, a common cause of febrile illness in travelers. We now have a lot more data and are preparing to look at the last nine years to see what the evolution is for resistance to common antibiotics. For example, we've done a very large-scale assessment of diarrheal pathogens revealing very high levels of fluoroquinolone resistance in South and Southeast Asia and rising in Africa as well. In many parts of the world there has been growing multidrug resistance, especially in parts of Asia, because people are apparently not using ampicillin or cotrimoxazole as much anymore, and susceptibility is returning for some of these drugs. This kind of region-specific and country-specific data can be very useful for practitioners and public health authorities.
Mahoney: GeoSentinel has an active AMR Enhanced Surveillance Project, prospectively collecting antibiotic susceptibility data from returning travelers across nine WHO Global Laboratory Surveillance System target pathogens. How does that alignment with WHO's epidemiological monitoring framework translate into GeoSentinel's data ultimately influencing global treatment guidelines?
Hamer: It hasn't yet. We're not directly contributing data to WHO, but they do obviously have access to anything we've published. We have tried to harmonize our reporting with their focus on key pathogens for it to potentially be useful for WHO, but we need to work more towards better ways to share those data with WHO and other multinational agencies like the European CDC and PAHO, for example.
Machine Learning, AI-Supported Detection and Prospective Research
Mahoney: Your 2024 JTM paper on AI-supported outbreak detection describes a move from "manual analysis techniques" toward AI-driven anomaly detection. Along with your collaborators at Eindhoven University of Technology in the Netherlands, you laid out a roadmap for AI-supported outbreak detection using GeoSentinel's 400,000+ record database, highlighting the value of reliability, scalability, dimensionality, and timeliness as GeoSentinel data advantages for machine learning. Where is GeoSentinel on that journey right now - is the AI-supported detection model deployed operationally, or still in development/validation?
Hamer: I'd say the model is still in development and validation. There is an outbreak detector software that our collaborators at Eindhoven have developed, but mostly we've applied it in recent years retrospectively to show that it can in fact detect outbreaks. With the surge in dengue that took place in 2024, for example, where Brazil had a massive amount of cases in one year, our large language model machine learning approach established a typical baseline for Brazil. With just one case, or one or two sites reporting, the case incidence was below that baseline but once we combined the sites, we went above the threshold. So, the power of having multiple sites reporting can allow us to identify an outbreak sooner. We're still trying to figure out the best ways to apply these tools. For example, looking retrospectively at some of our syndromic data, we found, unsurprisingly, in January and February 2020, there was a big spike in viral and respiratory syndromes with no concomitant etiology. At the time, we weren't looking at those kinds of data routinely, so we missed the signal for the beginning of COVID, and we want to avoid doing that in future. Syndromic data can be very valuable, but you need to have a way to look at them. The crude approach we've taken for years is to get a line listing of all the major diseases that have been reported from the day before and several of us look through those, and say, wait a minute, there's dengue in Angola that shouldn't be there. Then we investigate it. Those data do not however include all the syndromic data, because there's so much of them, so we need a more systematic way to review those data. So, I'd say our machine learning data analytics are still a work in progress.
Mahoney: Staying on that landmark paper, it highlights a key tension: AI can flag signals daily, but epidemiologists must set thresholds, and there's a real cost to both under-detection and false alarms. How do you think about calibrating that system - and who ultimately makes the call when a potential outbreak signal appears?
Hamer: I think it depends on the disease, but certainly if we have more retrospective data, we're better able to have a baseline, i.e. a threshold above which we become concerned. But if we surpass that threshold, we need to be aware of it and get a team of epidemiologists to look at our own data, and then also do outreach to sites and contact other organizations to ask if anybody else is seeing a signal like this, and do we need to worry or might this be a false alarm. For example, we first saw chikungunya in the Seychelles in 2005-2007. The Seychelles to that point had almost never had chikungunya cases. When we see a new disease in a place where there's a risk of arboviral infections, the signal becomes a lot easier to detect, because the baseline is zero. Then you have one case, and suddenly you might have an outbreak.
Mahoney: The paper mentions using "labelled datasets" - historical outbreak events tagged as confirmed outbreaks - to train supervised learning models. How complete and reliable is GeoSentinel's historical labelling, and what's the biggest data quality challenge for the machine learning team?
Hamer: I think some of the historical data for certain diseases within GeoSentinel are pretty good, because there have been a lot of cases over time, but for diseases that are rare, or harder to diagnose, there may be more gaps in the data. We've become better at diagnosing arboviruses, for example. There's been an increase globally in dengue and bigger outbreaks of chikungunya. Zika is still smoldering along in a few places, but with no major outbreaks. But in general, I think our ability to detect a surge in arboviral diseases is better than, for example, an increase in shigellosis in some parts of the world. This also relates to where patients are seen. Many of the sites in Europe are very big tropical medicine centers, so if you're coming home, and you're sick, and in Zurich, you go to the University of Zurich, because you know they have the capacity to diagnose your illness. If I come home to Boston, I might go to Mass General, Boston Medical Center, the ER, or my primary care provider, so there may be a delay in our getting those data, or not having them at all, which presents an additional challenge. Another challenge is that once somebody's seen, depending on what's going on at a GeoSentinel site, there may be a delay in reporting. So, these are some of the challenges in our being able to systematically identify outbreaks early and respond.
Mahoney: As GeoSentinel moves from predominantly retrospective analysis to prospective, hypothesis-driven research, what are one or two prospective studies currently underway or in planning that could be genuinely field-changing for the travel and tropical medicine community?
Hamer: One study we did that was fruitful in this regard looked at chikungunya, dengue, Zika, and malaria, longitudinally across all four of these mosquito-borne diseases to see how long symptoms persist, and what is the impact on quality of life. We had problems with the study implementation, because there wasn't a lot of chikungunya or Zika circulating, so we saw a lot more dengue. Malaria is more of a control. Some 60% of the malaria patients had fatigue out to 3 months post infection and then it was gone. But if you look at dengue, there are a lot of people still feeling quite sick, with multiple complaints at 3 months, some of them out to 6 months and even 18 months, and that sort of post-viral syndrome for dengue has not really been described well. There's more evidence being generated on it now. Zika manifests similarly, but the numbers were very small. We already know with chikungunya that people can have arthritis and other persistent symptoms. But in our analysis, we also showed a negative impact on quality of life. In another example we've been conducting a study that was called PRESEVERE, trying to characterize the etiology of fever in travelers, but in particular, trying to work towards pathogen identification in the subset that had fever and symptoms, but for which no diagnosis was made. That work led to a couple new diagnoses. Although we haven't really followed through on this study in depth, now we have new funding for a process called BioThreat Radar. We may be able to dig some samples out of the freezer and retrospectively identify, using pathogen agnostic testing, what diseases the patients who submitted the samples had. We're now also working on designing a prospective approach to try and apply this to both febrile syndromes and respiratory syndromes, and that's going to be a major project over the next year and a half.
Mahoney: Finally, a big-picture question: in a world where metagenomic sequencing is becoming faster and cheaper, wastewater surveillance is expanding, and tools like HealthMap aggregate open-source signals in real time - what is the enduring, irreplaceable value of a clinician-anchored epidemiological event reporting network like GeoSentinel?
Hamer: That's a good question. CDC has been doing wastewater testing, for example, with samples taken from airplanes, but all you can say is that somebody on this flight had this disease. You can't really say who it was. By contrast, GeoSentinel can characterize a specific disease in a specific patient with a lot of clinical and travel-related metadata that allow us to track back to where they were, a little bit about what they were doing, and the kind of travel they had, and therefore where they acquired the infection, and maybe what some of the risk factors are. Tying the clinical and epidemiologic data to the diagnosis in this way really adds a lot. We are now layering on molecular epidemiology, sequencing viruses we routinely now collect for 20 or so sites which are part of SUMMIT, a project to systematically sequence all malaria samples. This data can give us insights into, for example, where is artemisinin resistance arising, or where are we seeing pfhrp2 deletions that are going to interfere with rapid diagnostic testing? Studies like these provide a very exciting application of molecular surveillance to practical public health questions, and I think we'll be doing more of that in the future. Having all the clinical data and detailed epidemiologic - including travel – data in one place is a strength of GeoSentinel that databases like HealthMap and various other similar projects do not possess.
Closing Perspective
Mahoney: If GeoSentinel did not exist and you were designing a global infectious disease early-warning system from scratch today, would you build something that looks like GeoSentinel - or would you design it differently?
Hamer: That's a hard question. I think I would probably want to have multiple different, complementary surveillance systems, certainly starting with something like GeoSentinel, but then probably incorporating some level of airport-based surveillance for returning travelers that could be targeted towards certain destinations, because part of the challenge is that GeoSentinel only evaluates symptomatic patients. We're not catching people who are asymptomatic and could be viremic. For a lot of arboviruses like dengue and Zika, for example, there's a subset of people that have them but have no symptoms, but if they're viremic, and they return home, and you've got the right vector in your neighborhood, you could have an outbreak. So, I think having some form of airport level incoming traveler detection system would be beneficial. Airport wastewater surveillance is helpful just to get an idea whether a disease is circulating in a community or within a region of the world, but it doesn't have the granularity to detect where the disease exactly is coming from. Thus, this data is an adjunct to our work but is still inadequate. Creating a multifaceted system for surveillance could be very helpful.
Mahoney: Finally, what keeps you up at night - which pathogen or epidemiological trend do you think the world is currently underestimating, and that GeoSentinel data is beginning to hint at?
Hamer: One of the big things I'm worried about is the rise in arboviral infections we’ve been seeing over the last 10 to 15 years, with dengue certainly, but also increasing chikungunya outbreaks. Zika might come back, and now we have Mayaro. Oropouche transmitted by midges has disturbing potential. But my real worry is that we're going to see more travel-associated arboviral introductions into the United States, and even warmer parts of Canada during the summer, and definitely in Europe. This is already happening. If you look at the last 8 years of European CDC data, there are introductions and local spread of chikungunya and dengue every summer now in parts of France and Italy. We're going to be seeing more of that because the vectors are endemic. With climate change, we're going to be seeing a lot more arboviral infections in places where they haven't occurred before. So that's one set of things that has me worried. Fortunately, with chikungunya, we've got a vaccine that may help prevent disease. For dengue, we do also, although not available in the United States. We need more tools like these for control and prevention. The other major concern is antibiotic resistance, which appears in common diarrheal pathogens like Salmonella, Shigella and Campylobacter, where fluoroquinolone resistance is so high in most parts of the world, we can't use ciprofloxacin or levofloxacin anymore, and we're limited to azithromycin, but if we start to lose that macrolide, we could be in trouble. But also for enteric fever, there are areas of Pakistan where XDR typhoid is resistant to almost every known agent, except for maybe carbapenems. If this pathogen spreads, we're going to have a big problem on our hands.
About the Author:
Tom Mahoney, a 2024 Senior Fellow at the Harvard Advanced Leadership Initiative, is focused on global venture philanthropy initiatives to catalyze investment in development of breakthrough vaccines, therapeutics and diagnostics for infectious diseases. A career investment banker, technology entrepreneur and asset management senior executive, Tom is an Associate in Immunology and Infectious Diseases at the Harvard T.H. Chan School of Public Health; a member of the Advisory Board of EdJen BioTech, LLC, a developer of novel vaccines, and Virufy, a respiratory disease diagnostics platform; a Founding Sponsor of the Harvard Alumni Entrepreneurs Accelerator; and a member of the Venture Board of the Harvard HealthLab Accelerators, the Massachusetts Consortium on Pathogen Readiness, and the Council on Foreign Relations.
This Q&A has been edited for length and clarity.