Harnessing Digital Technologies to Advance Global Precision Health and Development
![Harnessing Digital Technologies to Advance Global Precision Health and Development Harnessing Digital Technologies to Advance Global Precision Health and Development](/pic/harnessing_digital_technologies_to_advance_global_precision_health_and_development/RmpCX0NHZEd3NVk_.jpeg)
Good morning and welcome to the second session in our frontiers in precision population health and health equity. Our inaugural session focused on how we could use precision population health approach to in California for studying COVID-19 in California. In this, in the vaccination campaign in the fall, we plan to host sessions focused on AI and cancer. And today's session takes us in a more global direction. We'll look at how this approach can be used to advance global health and development. Our hope in this series is to stimulate transdisciplinary discussion around precision population health and we are excited to co-host this event with the King Center at Stanford and the Department of Epidemiology and Population Health. Today's discussion is being recorded and the recordings will be available on the Population health sciences website as well as the King centers early next week. And it's
my honor to introduce my friend and my colleague Steve Luby, who will be moderating this session. Steve is a professor of medicine and infectious disease is at Stanford Senior Fellow at the Stanford Woods Institute and the Freeman Spogli Institute. And he also by courtesy, is a member of our department in epidemiology and population health. Before joining Stanford, he worked in Pakistan for five years and in Bangladesh for eight years where he continues to do extensive research in Bangladesh. He has led research teams and have that have advanced our global scientific understanding of hepatitis C transmission and safe injections, water sanitation and hand washing, NEPA virus prevention, typhoid fever, epidemiology and prevention. I'd like
to hand this session over to Steve who will be our moderator and I'd like to thank him and our, our speakers for today and I'm really excited to have be for him to be the host and for us to have continuing these sessions in population precision health. Thank you for your attendance and we look forward to lots of discussion and would really appreciate your participation in this in this seminar. Thank you Steve. I'm handing it over to you. Thanks a lot Melissa and welcome everybody to this really exciting seminar. I have the privilege of introducing our remarkable panel. So we have three panelists today. Manisha Bhinge joined the Rockefeller Foundation in October, 2016 and serves as a managing director for health Manisha leads program strategy and manages the portfolio of global partners for the Rockefeller Foundation's Precision Public Health Initiative and more recently pandemic response and prevention efforts. She has over a decade and a half of experience in social innovation and implementation science in global health. Prior to this role, she created and led the strategic partnership office
at the Tata Trusts India's oldest philanthropic organization and launched the India Health Fund, a leverage fund in collaboration with the global fund. In addition, Manisha was vice president at BRAC, the world's largest NGO for over six years, where she developed and managed programs that promote access to health, education, economic empowerment, and social justice. She's worked extensively across Asian, across Africa and South Asia on community-based service delivery and women's health and empowerment. Manisha's experience includes management of field operations, program strategy and design and partnership for stealing intervention. She spent her early career in the private sector working in technology consulting and then at the un our second panelist is Giulio De Leo. He is a disease ecologist in interested in investing, investigating factors and processes, driving the dynamics of coupled natural and human systems and in using this knowledge to identify levers for health and conservation that is ecological interventions that can improve human wellbeing and the health of the environment that underpins it. In the last 10 years, he's been particularly interested in investigating
how the development of water management infrastructures to support agricultural expansion and intensification may increase the risk of the transmission of schistosomiasis, one of the most important of the so-called neglected tropical diseases. Giulio does not neglect it. Along with Dr. Sokolow, he co-founded the Upstream Alliance partners in schistosomiasis reduction and the Stanford program for disease Ecology Health and the environment with the goal of developing ecological solutions to control infectious diseases with an important environmental component in their transmission cycle. And then Brigitte Gosselink who leads Google leads Google dot org's work to leverage emerging technologies and Google's expertise to address global challenges. She's currently focused on how artificial intelligence can be used for social impact through efforts such as the $25 million Google AI impact challenge with a particular focus on crisis response and sustainability. She previously created programs focused on how technology can improve global education and innovation for people with disabilities. Prior to google.org, Brigitte was a strategy consultant for nonprofits and
foundations at the Bridge Group and worked for the US Agency for International Development and IRD focusing on innovative approaches in post-conflict transitions. She has an MBA from the Yale School of Management and a BS in Systems Engineering from the University of Virginia. So over to Manisha. Excellent. Thank you Steve. And again, thanks everyone for your patience. It's
ironic. Digital technologies practice sessions still have little of challenge, but coming back to Rockefeller Foundation, we started a precision public health initiative in 2018 and the rubric that we used there and the philosophy was how do we ensure that the right interventions read the reach the right people at the right time? And that was pretty, that was the underlying premise of, you know, the, the rationale for engaging in this initiative. There was precedence to this work. So when we started in 2018, I think we'll hear from other participants. There was a wealth of evidence and a wealth of work that we built on. And what we realized very quickly is there the technology that drove
what we call pre precision public health, and I'll get to it in the next few slides, largely settle arrested on three critique, three categories of technology. We had the frontline technologies and the example that I wanna show you here is the personalized frontline tools that were, that are fairly common in most of the world with the community health workers largely that allow them to understand who they are clients are, what are key challenges that are, are operating within the catchment area, as well as allow them to do more periphery analysis and understand workflow modalities in a more, in a more accelerated and effective manner. The second body of technologies that we encountered were what we call visualization technologies. So there's one thing about collecting data and having data there is an entirely different skillset around visualizing and interpreting that information in a way that aligns with decision making and brings forward key interventions and needs that stakeholders have. And again, we found a number of examples around that. And the third is also a source of data, but analyzing, accessing data through remote sensing means, again satellite technologies, G S I technologies and alternative sources of information beyond health information systems, beyond what can be gathered through the public health interface were key areas of opportunity that we identified and where we have technologies on one one end and where are applications of those technologies.
And we found there were wide ranging pretty much the entire gamut of primary healthcare delivery is a ripe for engagement and action and precision at large for the simple reason is populations are heterogeneous, challenges and health behaviors are heterogeneous and the needs and concerns within communities also significant differentiation. So how does one ensure that delivery of health services are not only targeted but also streamlined and personalized if you will, towards specific communities and individuals. And how does one leverage technology for that purpose? At the Rockefeller Foundation we started, and this is again in 2018 with maternal and child, maternal child and reproductive health as so the signature focus of our efforts, again looking very closely at community health workers and the interface between the public health system and households and what are the technologies that work in that interface not only to gather data off individuals, households, context communities, but also support community health workers, which we know are who who we know are paraprofessionals to upskill their capabilities and provide more targeted and precise. Again, I'm using that term services towards those community based on that data. A second level of engagement that we work towards is infectious diseases or communicable diseases and that's largely in the space of tb, malaria and hiv aids to start with. And as we
have learned over the past year, COVID has also been, you know, the remit has increased to COVID. So what does, what do we mean by precision public health at the foundation? We've again very much started with, you know, the three factors of data that we feel that drive precision. First is availability. Do we have enough of the right data? Is it timely and is it granular enough to make inferences that drive health outcomes or inform health delivery When we have those three factors, again availability, acce availability, timeliness and granularity. And I think the underlying piece is quality and accuracy being sort of the fourth
overarching component. We have the opportunity to solve some of the world's greatest problems. We can drive scale in a cost effective manner. Again, remember we talked about the heterogeneity of health behavior services and disparities. How does one drive the solutions to scale in a way that still is applicable? Can we address what were previously insurmountable barriers to progress? And an example that we, you know, aim to tackle was just the access to health services writ large in a world that is deeply constrained of skilled workforce. Can technology augment paraprofessionals and or resource constrained settings to bring services to those communities and can be used data analytics to understand what those are and ensure that those are more targeted in a meaningful manner and broadly address the health equity divide. And I think knowing who your customers are is an important factor in delivering effective services. And if the data on those
customers and those clients and those communities are absent largely because they're off grid, they don't, we we not building services for those needs. So again, how does one use the first suite of technologies that we have available to understand where those hidden populations are, what we call cold spots, spots if you will, off the grid, off the healthcare provision to bring services and and to them or bring them to the services, whichever is the more appropriate mechanism. So again, the, the tools and technologies are important but the purpose and the outcomes that we want to drive is paramount in term, especially when we have the foundation set out to develop this strategy. So what was our approach? There were three areas that we wanted to focus on. Again, we were looking at how does one get, how does one ensure availability, timeliness, granularity of data driving equity and accuracy across the board. And three critical components came to the fore for us to invest and lean into both from a scale as well as an innovation standpoint. The first is the data exists, it's not as though we don't have data in countries,
it is questionable data. And again I'm looking at health information systems right large across lms. Is there, how does one ensure interoperability within these data sets so that they're not silo malaria data sitting in one place, primary healthcare data setting in a separate silo. Infectious diseases are in a third silo, yet it is the same communities that are recipients of services largely from same frontline staff and primary care facilities. How does one create an integrated view of that community, of that household and even of that district to ensure that we are using data and evidence to drive resource planning, to drive service delivery and design services that are more customer and patient focused as opposed to, you know, traditional modeling efforts while those took us as far as they have, what's the next generation to do that in a robust manner, we need to have digital tools at scale. And again that the foundation, we leaned into frontline digital tools because again we felt that the interface between the primary interface between households and the health system is the richest interface where you can gather robust data on specific needs, households, behaviors and context.
And finally, what is the processing that's needed? We have the data, we have it's interoperable yet to make it actionable, it needs to be processed and to address critical use cases and challenges that decision makers need to target. Whether it is resource planning, whether it is response to outbreaks and and disease, disease outbreaks or understanding rationale for a surge in mortality, mobile morbidity in a specific community and what we need to do to drive those. So, so that is where our approach leaned into over the past two years, and this is largely sort of pre C and early days COVID, we worked very closely across India and sub-Saharan Africa and the focus there was again looking at primary use cases in India we looked at malaria and what does active surveillance for malaria look like? How does one inform outreach? How does one gather the data and how does one ensure that at the district level we are understand we are processing and driving analytics to inform how resources can be deployed in near real time. So not at an annual basis but what can we do at a monthly or quarterly basis
that drives more effectiveness and efficiency in those spaces. We're right at time now so we really have to move on. Can you close quickly please? Yes I do. I can. So very quickly we did similar programs in Uganda looking at essential health coverage during the pandemic and similarly in several parts of Africa and this is in Ethiopia, look using g I s services to find where there are graphs very quickly summarizing where we have challenges and I won't let read them out, but fundamentally data is still paper-based, countries have different maturity levels and capabilities are limited. So there's a lot of
work to be done to reach the goal that and the aspirations that we have in front of us. So with that I'll stop and thank you for your patience while we fixed the technology issues. Thank you. Thank you very much Manisha. Our next speaker will be Giulio Deo Giulio. Over to you. Thank You very much to much Nisha for your introduction. I want to follow up with a specific case study that is about integration ecology withdrawn in satellite imagery to the, for the control of neglected tropical diseases. While we are ramping now with the vaccination campaign and we start to have some hope in ending the devastating and tragic COVID-19 pandemics that just today has exceeded 180 million certified ca cases worldwide. Well there is a silent epidemic that has been
going on day after day with no relief for decades actually for centuries affecting nearly 40% of the world population. 3 billion people typically living in tropical andSo tropical countries with limited access to clean water, to healthcare and to Sanitation. And I'm talking about a family of diseases caused by pathogen and parasitic wars that share free features. They have an
environmental component in their transmission cycle, whether freely stage of a parasites or an animal reservoir for this disease contingent occurs through infected bites for instance from a vector such a mosquitoes or a tick or simply by stepping into contaminated soil or water. They are a disease of poverty affecting disproportionately population with limited access to resources in the healthcare and they generally do not trigger long-lasting immunity. So most importantly, for a lot of these disease we do not have vaccines and sick people can be treated which is great, but treatment does not protect from brain infection. Therefore
targeting also the environmental reservoir or preventing exposures are two key approaches for five these diseases. And the government might talk to guys is to show how digital technologies can help to address this problem. And I want to show you that by using one of the most important of theor tropical disease as a reference example, I'm talking about histo disease, rehabilitating parasitic disease of poverty. They in four continents, 74 country in tropical, tropical regions affecting more than 200 million people worldwide. The vast majority in sub-Saharan Africa, the vast majority school aged children. It is caused by parasites of the blood vessel which is represented here in the lower left panel. And this parasites as a complex lifecycle. It required two host, the human host that you can
see here and a limited set of fresh waters, nails the obligate intermediate host of the parasite. These nails are actually really tiny, they looks harmless but they are hijacked by the parasite. They transport them in a factory of infectious stages called saria. They are released in the
water for several weeks after infection. So people get infected by stepping into parasites, contaminated water. Now there is a drug to treat infected people quantal, it works well, it's cheap, but after removing the parasites from people, it does not anything to target the environmental reservoir of the disease. Since they continue to shed infectious stages
for weeks after drug treatment and like on a treadmill, this lead to an endless cycles of treatment and reinfection to interrupt this cycle, the what the organization has recognized that OSI can be controlled only by complementing medical treatment with environmental intervention and cause we always operated in condition of limited resources is to, it is crucial to identify and target snails, asna habitat to fight that EAs to show how we can do that. I want to bring you in a quick journey to my field size in Senegal here. So we are in Western Africa in the lower Beijing of the Senegal River up here at the border with ma, with ma Lithuania. And since 2017 I've been fortunate to lead seven field missions at Senegal. We've been conducting survey of nearly 1500 school aged children. We put a a lot of effort to try to monitor the infection
dynamics in this hyper endemic area of in the world. And this to this end, we've spent countless hours collecting nails from the river in the lake where people congregate for fishing, for bathing and other daily tasks. We have collected actually nearly 50,000 nails and Metis three recorded data about all of them from the sub special location where they came from, what type of vegetation they were founded to precise miser of their shell and of course which parasites live inside them. And this is what we found first we found very patchy distribution, very ephemeral with snails aggregated in clustered ear and they are kind of difficult to find. But we also saw that there was a strong association of snails with certain type of aquatic vegetation like phylo very easily to identify visually like for instance from this picture.
We also actually ended up finding some unexpected mismatch in some villages that were no near shore's nails but high infection levels in people, which was something really odd because we know that the life cycle requires nail and parasite. And so we basically realized that the regular sample scheme was failing to capture this nail where they are. We needed to change perspective. And in order to do that in January, 2017, we bought a simple, cheap recreational drone, A D J ui, phantom four with a simple air GB camera. We started to take shot of our field site and what we found a lot of more suitable habitat for this nail that we could see from the shore. And specifically we even realized that the area that we could safely sample by stepping water up to one meter, one meter and a half at most was just the marginal areas of these large distribution of floating vegetation. And this vegetation could be easily identified from our high definition subter imagery, drone imagery and classified in different habitat of different quality forces. So we
decided basically to move from these occasional observation to a systematic flight plan of the series of overlapping drone photos that we took a every water access point about 50 to 80 meters above the ground level. I will spare you with some of the technicalities and the details, but we took hundreds of images weed to identify our much vegetation. There was different areas and this basically what we found. So the extent of aquatic vegetation
matters, the more aquatic vegetation, the more snail and snail cluster we found. We confirm the fact that these days are extremely patchy and theoral. They fluctuate in time lot and in fact it's much more efficient to understand that the potential risk of transmission to track treatable habitat, which we can spot from the drone imagery that samplings nails as we did side from side. And the relationship between alu vegetation and prevalence of infection is robust enough to understand just by looking basically a drone imagery, whether the site is a high transmission site or a low transmission site. So this has been great cause then we know what we can do in a single village. But the problem as Manisha was saying
before is what happened when we scale up, when we do map the large special scale, you know, we have to understand where do we want to prioritize intervention tailor, what do we want to basically deploy and distribute product quantum force do environmental intervention. And this scale we're talking about something like 20,000 square kilometer. There is no way we can fly a drone over such a big area. And so we rely and we moved to satellite imagery. The quality is not as good as drone imagery but as improved dramatically in the last five years. And this allowed us to detect natives avatar from satellite imagery. There is no way we could parse manually such a big area and we need to automatize the process and thus we rely basically on some of the techniques of machine learning, what we can see by as we can train and machine learning algorithm to do for us. And so we basically use the satellite imagery and
the drone imagery to ground truth in the satellite imagery and ultimately we apply a specific type of convolutional layer of network that has become widely used in biomedical application. And we use this on a really limited number of imagery, surprisingly limited. And the training is a training dataset and the the results were excellent. So despite the limited dataset, we are able to identify 95% of the floating vegetation, the treatable habitat for ISNA by using this type of algorithm. So what we see here is that we can integrate the field ecology with drone and satellite imagery to identify potential hotspot of transmission. We can use this
information to find tune, imp, epidemiological mathematical model of disease dynamics where people movement for instance is calibrated by using mobile phone data and we can explore different scenarios and have a sort of a civilian system where every year we can span this area, identify where the hotspot are and which are the best intervention to control the disease. To conclude, I want to stress and to emphasize how likewise precision medicine there is an untapped potential precision mapping of transmission risk for environmentally transmitted disease. And we need to integrate the old fisher disease ecology, which is the foundation of what we do with the superpower of the digital technology. And with there I stop my talk,
I want to thank you all for your attention and thank also the broad set of collaborators that provided an invaluable support, crucial support to develop this research. Thank you very much. So thank you very much Giulio. And so we've now had two excellent setup presentations and we're getting ready to now move to our keynote speaker, Clinicians in remote hospitals, counselors for crisis, L G b, youth smallholder cotton farmers, online fact checkers. These aren't necessarily the people that you, you think are working at the forefront of cutting edge AI applications, but why not? Why shouldn't those working on the front lines of our most challenging societal problems be using technologies', most advanced tools to enable their work? And we have an opportunity, can you please put the slides back to the That's fine, thank you. Have an opportunity to ensure that really everyone can benefit from ai, advanced data analytics and digital technologies. and@google.org we work to ensure that that's possible. I'm really pleased to be here today to share more of that work with you.
Next slide. Google's is google.org is Google's philanthropic arm. We work to bring the best of Google's resources to address societal challenges. We do that in the form of funding, so every year directing 1% of Google's profits to support nonprofits. We also think about how we can use our technical expertise mostly in the form of volunteers who spend anywhere from an hour to a full-time secondment with organizations that we're partnering with, bringing their expertise alongside that of frontline organizations who are working on these problems day in, day out. And then of course thinking about how we
can bring advanced technology, connecting with expertise and really bringing the, the tools and we use to power much of our work to organizations that we think have the opportunity to change the game for some society's biggest challenges. Today I'd like to share a couple of examples of those organizations and then I wanna talk also about how we're working to enable more organizations around the world to use machine learning in particular and where we see future opportunities. Next slide. The first example that I wanna share is out of Uganda. So air pollution, as many may know,
contributes to a significant number of deaths every year. Some estimates up to 4 million deaths in Africa. And recent studies even show that that impact is greater than malnutrition water and Sanitation sort of problems that I think we, we often hear more frequently about in continuous air pollution. Monitoring is a critical tool for combating and addressing
that problem, but sensors are quite expensive, typical sensors, and they have to be installed across many specific point locations in most of the models that we might think of being used in some higher income countries. The researchers that we are working with and proud to support at Macri University have developed air quo, which is a project where they have built a low-cost sensor. You can see one here on the page that is uses solar power and a lot of local components and they're deploying them across the city, not just in point locations, although importantly they are, but also on motorcycle taxis, on boat bodas that drive throughout the city of Kampala and are able to provide an additional layer of data without necessarily having to have the quantity of, of sensors that you might need in a point system. They're then applying AI and machine learning on top of that data to both interpolate between the data points that they might not be able to, to get and also to do forecasts and predictions that citizens can use through an app that's now available throughout the country and that governments are starting to use to inform policy and also organizations like schools and others being able to take action on days perhaps when the air quality is bad, not having students play outside. We think that this tool and, and the project that they're working on has tremendous potential to change the game for air pollution throughout the continent and perhaps around the world. And they're looking right now to think about how already it can be scaled to Nairobi. We've supported them obviously with funding and, and also AI expertise. But I think that one of the most interesting pieces of this
project is actually that the implementation itself is really where the innovation lies. The low cost sensors, the installation of those sensors on taxis, and then they're also working to engage the community around particularly high risk spots of the, of the city. And also in places where the data has not really shown to be predictive, they're actually asking folks to, to provide crowdsource information. So on days when they might, you know, see a fire burning or
notice a lot of pollution adding that data in. And I think that community engagement approach is something that is really kind of a hallmark for a lot of the work that we're working through and something that we wouldn't necessarily, I think have seen or they had not even necessarily proposed in the initial part of this project. And so innovation in that implementation part of the way that we deploy digital technology in addition to of course the cutting edge tools like AI machine learning that they're applying is really advanced the, the opportunity here. Next slide. And on this page you can see that app that I was mentioning, you can see the points
there zooming in on them and then also an ability to see a forward looking forecast and prediction. Next slide. The next example that I'd like to share is, is focused on the problem of antimicrobial and antibacterial resistance. Obviously bacteria, viruses are always changing
and some have adapted to ensure that medical treatments that we typically use are not necessarily able to be used. And some reports suggest that the number of deaths attributal to Inca microbial resistance by 2050 could be as high as 10 million if we don't take action. And these are particularly challenging cases in in war torn countries where there are common injuries and bacterial infection can actually cause significant loss of life. This is obviously a lot of the countries where M s f al frontier doctors about borders does their work. And this project is a really interesting origin story. So the, the challenge of dealing with antimicrobial resistance
requires prescribing the right antibiotic for the bacterial infection that's being presented. And to get that information, we typically use expensive lab equipment and very highly trained microbiologists to interpret essentially the Petri dish that you see here, a susceptibility test that shows how the vi the bacteria is reacting to various, various antibiotics. Those tests are hard to interpret as you can see here. Then perhaps we hard for, for someone on an untrained eye to to use this tool or to to use this petri dish and sort of understand what to do. And it's very hard to get both expensive lab equipment and
trained microbiologists especially into perhaps a country like Yemen. And that was the case for M S F A couple of years ago. One of their lead microbiologists was, had been recently there and she was interpreting these images being sent to her by an email overnight in Paris and their home office. She connected with a friend who was a computer scientist and said, you know, we've seen all these advances in computer vision, isn't there a better way we might be able to do this? And this project was born, they have now developed a free mobile app that you're seeing the demo here on the, on the left. It supports those sort of non-trained technicians to measure and interpret these important tests so that clinicians can then prescribe the right antibiotics. It works by taking a picture of the Petri dish and then measuring and, and looking
at those inhibition zones. So that's supported by computer vision. There are then interpretive questions that help guide the clinician to interpret the results, resizing the area, for example, around this particular dot, and allows them importantly to layer in some of what they may have seen in terms of patterns or ask questions as they're interpreting. And then they're able to see a recommendation that they can then share with the clinician and also send in for peer review as they've been trying to make sure this tool continues to, you know, deliver results. As noted here on this page, the the way that the studies initially that they've done in in this tool is just in trials at the moment, I should say, is showing a really high level of agreement against both the standard auto automated tools and human interpreted results, which I think is really telling and, and shows the opportunity that we have to layer in these tools to clinical settings. You know, as Manisha was mentioning, where there might not be some of the technical expertise available on the ground. And it's also a really great example of a human in the loop system. So
we talk a lot about AI and, and concern about how it might interpret results that may or may not be accurate. And here you see AI augmenting really the most error prone and challenging portion of the task, this kind of initial assessment, but ensuring there's multiple points throughout this system at which people can layer in their own interpretation and do the necessary checks. Obviously when we think about machine learning in particular, prediction is sort of the ultimate goal for many, for many cases. And if we can get information to help us act in advance of adverse events, that I think is something that we've all been striving for for many years in the social sector. I wanna share this final quick example,
which is work that's done actually not necessarily only by google.org and our third party partners, but also through work that our crisis response team has been leading internally. So this is an effort to really get better flood forecasting predictions, floods are one of the most devastating economic and harmful events in the world, and they are notoriously challenging to predict, especially during heavy monsoon seasons in places like South Asia. We've been able to combine advanced map mapping capabilities using some of our mapping tools and AI to really predict a path of a flood and generate detailed maps that then can alert through our online systems.
And those alerts are also provided to government who often alert through their own systems. But I think this is a really good example of combining digital technology with offline solutions because as you can imagine, many of those who are at greatest risk for impact from floods are not necessarily the folks who are getting a, a smartphone alert with a map showing them where the flood susceptibility is highest. And so we've partnered with the International Red Cross in in India where this has started and expanding now into Bangladesh, into other countries to really build out a community alerting system alongside this. So how do we take the information that these models are outputting,
which is critical information and try to get it into as many hands as possible. And obviously, you know, digital tools in all forms, you know, s m s and WhatsApp groups being formed, but also just word of mouth and using a lot of the, the ways in which community information flows already. Some communities use flags to raise down the river to say, hey, a flood might be coming. Some communities use literally musical instruments to try to, to try to get more information out. And so partnering with this network of people alongside the technology has been a really critical input. And as noted here on this page, we are seeing through some
studies done with, with folks at Yale, a two day improvement in rural alert De dissemination during the pilot that we ran alongside our flood alerting. And that two days, as you can imagine, could be the difference between loss of a home, loss of livelihood or, and you know, worst case loss of life. So we're really excited to see how this might be able to scale to many more places, working closely with governments and working closely with our N G O partners.
So I wanna zoom out and talk a little bit about how do we ensure that more and more problems might benefit from machine learning and artificial intelligence in the advances that we're seeing. And I think Lacuna Fund is a great example of that. So we, with Rockefeller actually some of manisha's colleagues alongside the German and Canadian governments have started a collaborative effort to fund labeled data sets that address problems that are underrepresented in currently available data. There's three areas that that work has started agriculture. So thinking about things like crop disease or soil moisture, not the, the data that we're using, you know, here in the US where I'm standing, but importantly represented across Sub-Saharan Africa.
Language is another area. How do we ensure greater representation of language? And then we just announced actually earlier this week, an ex first round of expression of interest for health data sets. So really looking at how more representative data, both in the US and in low and middle income countries could drive more e equitable outcomes. And we realized that this is a critical input and I think really call on everyone to be thinking about how we can continue to, to close this gap and ensure that we have the right data that we need to drive equitable outcomes and to create new innovation across the board. Finally, I'll just leave with these three areas we've recently published with ID insight as of an article which you can find in S S I R that really points out, you know, was a framework for how do we think about more areas where AI and machine learning may be most impactful. And there are three that I'll leave you with as I think we can all continue to point our attention to do more. So
the first is, of course, point of care diagnostic tools, really thinking about how with especially better data, we can see that differential impact as we just talked through in the M S F example, communication tools that support marginalized communities that really lean into language. So thinking about human translation, translators being as augmented or with those new languages as we mentioned coming online, how we can ensure that people are getting the information that they need. And the final area that I think we feel there's a lot of opportunity is agriculture yield prediction, obviously with a significant portion of the world relying on agriculture for their livelihood, for their health, thinking about where we can see more tools that actually increase productivity, forecast yield, help inform harvest, help inform decisions that are being made. I hope this has been a, a tour through where we see some of the, the fascinating work that's happening with AI machine learning. Again, really in in partnership hand in hand with those
who have worked on these issues for decades. And we'll work on them for many to come. Thank you. Thank you very much, Brigitte. It's really good to hear about Google's work in this area and, and, and how this connects with broader public health efforts. We've had a number of interesting questions come in and I wanna start with a first question to Manisha. I think that we have all watched with dread as the surge of COVID-19 ran through India recently contributing to as much as a, to a couple of million deaths. So first I'd like to
ask Manisha her thoughts on how digital technologies might be used to help manage the crisis. And after we hear from Manisha, perhaps Brigitte could also talk about the perspective that Google might take in terms of where contribution might be. So first Manisha, Thank you Steve. Yeah, I mean it was a crisis and then, and you know, unfortunately a tragic one, one of the challenges with COVID-19 at large India and the world broadly is that we were flying blind and there was limited information in terms of transmission, community transmission, emerging variants and availability of supplies and critical essential supplies. So I think there were failures, information failures across multiple fronts. I think that the key lesson here is two, two things. You know, there's certain things that worked in certain things that didn't. India's
crisis is enormous because its scale is enormous. You have 1.3 or close to 1.3 billion people packed in highly dense geographies. So I think elements like social distancing and some of the, the, the, so public health in interventions that we have Western in countries are harder to implement.
You know, every event, every step out of the house, it's a, it's a potential super spreader. Having said that, there were certain things that did work. The AR app, which was rolled out as, as a contact tracing app was, is fairly well implemented. I mean, I I'm saying fairly well with the, the caveats alo in terms of what that actually means, but in terms of the number of people that went on it were Monitoring, it was much more than expected. So there were certain things that worked. I think we need to acknowledge the role that digital technology can play and acknowledge the fears and behaviors that surround those technologies. So I'll, I'll sort of break it out into three things. First is, you know, contact
tracing individual applications that monitor people their behaviors and their surroundings that has challenges in terms of uptake largely. So I think it's not just the r o GTO app, I think even the other efforts that Facebook and Google put in place, you know, had challenges. The second is scientific information. And this is around, and this is the work that foundation has
really is leaning into going forward is what is, how does one implement and support a cost effective scale surveillance platform around, you know, genomic sequencing. So again, driving precision in terms of what are the pathogens that are circulating, what are the variants of concern and what is the bioinformatics capabilities that we want to build in the country to actually pulled together an early warning system. So again, what have we learned, what are we doing? And again, you know, we were late, India was late in terms of identifying the Delta variant, which was circulating months before, not because we, the, the infrastructure wasn't there, but the Coordination and the ability to pool, aggregate and analyze data was limited in the early days. So again, what is the system that we can put in there? So you've really highlighted that there were true information gaps Yes. That really contributed to our suboptimal response to this crisis That is across the board agnostic of country.
And it's interesting to me because this is actually some of the same things that Giulio was talking about saying, I don't have the information I need to recognize where the snails are. So maybe just Brigitte, could you talk a little bit about how google.org sees this? Are you completely agnostic on where you're working or do you see like with this big health crisis of COVID-19 and in India that there are particular areas where AI and Google strengths really can lean in and contribute? Yeah, we've been doing a lot of COVID related work over the past year. One, one piece of that, which unfortunately to Manisha's point I think didn't quite work for India yet, but is hopefully in place for the future is, is work that we've done to support a global specific case line. It's called line list data in epidemiology world that provides case information across 140 countries in those early days, you know, to to some of what Manisha was sharing, getting that kind of information to say, hey, we're starting to see something that looks a little bit different and we actually don't mean aggregate data. We need to know specifically what were the cases, where were they, that dataset is being run through a university consortium now under the umbrella of global health, also their U R L with India in particular though i, I just a hundred percent agree with so much of what Manisha said. And I would say one, two other things to add. One is that I think we also saw the
need for information. At the end of the day, you sort of end up with frontline workers sort of stuck holding this challenge themselves and, and very poorly equipped to do so because we have been flying blind, having the right protocols, having the right information, even just having the right information to share with families who in, in many cases we're actually providing that care in their homes for their loved ones. It's really challenging. So I think that information gap also extended down to that level and, and just points again to the need to strengthen the infrastructure of those systems in times of not crisis. And we've also, we've been working with an N G O called Armand who supports a lot of government community health workers. They luckily had already stood up a, a mobile academy, which they had been running for, you know, the last year or so in partnership with government, which provided a vehicle to get, you know, just creating a new set of content through the same channels. If you have to also be establishing all of those channels at the same time that you're creating content, it's just impossible. So I think, you know,
and that that even was probably inadequate for where it would've ideally been, but I think again to this point of how do we invest in the channels infrastructure that we need now, how might we use this as a motivator, right? To do some of that work is, is just been critical And this, this is this no notion of where do we invest. We have some capacity that we think can help out, but there's been a lot of concern raised by questioners here too about how can we guarantee that we can actually get this into the hands of people who need it. That if we think of the most marginalized communities, the type of communities where Giulio for example, is working with and what RA and, and, and, and where the Rockefeller Foundation is, is trying to reach out. There are severe shortages of personnel, there's shortages of IT equipment and, and how can we adapt these, these potential tools in the setting of such resource scarcity. So maybe, maybe I'll ask Giulio to answer first, but I would be really interested in others' responses as well. Yeah, absolutely Steve, thanks a lot. I'll give you of course an academic perspective.
So thinking about what is in the real of capacity of the work that I can do and, and I know that Manisha and JIT have, you know, a different scale of operation in action. From my perspective co-design, the research that we're doing with the local community and the specifically the academic communities there that are operating locally is one of the key element. And, and so is there involvement in our specific project we're working with two no-profit organization as Spark for Lacent has been operating in the RF for 25 years with the university Teton Berger in San Luisi and the African Institute for mathematical biosciences.
And so we working with them, we can be sure that part of the knowledge that we so passionately want to develop will be co-developed with the new generation of leaders of these countries. Thanks to Sparko Lasante specifically our no-profit who's working there are, we are also trying to work elbow to elbow with the local community and will chiefs and with the people who are there on the ground to be sure that they need are met. And so this is basically our, we operating, we do also invest in, we will do in the future in training launching undergraduate courses, for instance in drone mapping, which extend from our specific goal of controlling the greater tropical diseases, but expand also to agriculture mapping because this is an area where there is potential for jobs in the future and we want to share these basically technology with them. Last comment is that when we brought drones the first time in Senegal, we were concerned of bringing something that was potentially scary or you know, sounds that completely out of the context, but it's been on the contrary, a fantastic way to engage with all the villagers. I think everybody, you know,
as included looking where we live from the bird eye view, you know, is a, you know, it's a new experience and so we've been able to interact with the people there in a way we didn't even expect. Thanks Giulio, Manisha, I'm interested in your perspective both in terms of how we reach out and also is this a role where NGOs work really well, but does that also give us limits in terms of long-term engagement by governments and use of these technologies and all? And, and Rockefeller thinks a lot about skills, so I'm interested in your perspective. Yeah, no, I think that's a really good question. And you know, for a lot of these solutions to work and deliver on the impact that they promise they need to operate at scale and they need to be institutionalized. So it's, it's, I would argue it's not an either or. It is an, and there's a role that NGOs and social enterprises can play, especially in the leading edge of demonstrating what the innovation or the technology is, how is it applied and really unpacking the implementation science behind the application of that technology to solve a specific set of problems or problems.
The role of the government is in terms of the enabling environment and institutionalizing those that implemented that, that that model of implementation. You know, we talk about access to technologies in rural and remote areas. One of the fundamental areas we leaned in a couple of years ago is access to frontline mobile tools in remote and, and, and rural rural areas.
And for that to function, we need a couple of things. We need network, right? And that is the role of the private sector and government understanding how does one create the Economics for network providers to engage in say, rural Sierra Leone, how do we work with a ranch, how do we work with mtn, et cetera. The second is cost of data. I mean, we don't think about this in the us but fundamentally, yeah, who's paying for data in these countries and how much data are you using? So that fundamentally constrains what you can do and who you can't do. So there is a multiple strand me, you know, approach here where we are trying to use relevant technology to the context as it stands today, but also influence and and advocate for the enabling environment to move in a direction that allows for smart technology, allows for better bandwidth that could deliver on those, those analytics both at the, the facility level as well as the handheld level with community health workers. So I mean the, the cost model changes substantially and if you don't address the cost
models, then scale becomes a fundamental constraints. So you'll have a set of very well-designed, beautiful, highly impact small projects that basically end when the financing ends. And we have probably, you know, millions of those over the last several years in that space. Yeah, I mean, Manisha you make a really good point and something that we often don't bring to the forefront that with focused effort and some funding, non-government actors can come in and collaborate and make really remarkable interventions, but taking those interventions to scale so that they can impact the trajectory of a pandemic is, is another task and, and is a difficult task. And I think part of the reason why the public health community is so attracted to Google and to the tech companies is that they do have a funding model that allows them to put some money in to generate data, yet at the same time, obviously they're a profit making company and trying to figure out how this all plays in ways that can generate data at scale that can matter for public health. And so Brigitte would just welcome your thoughts on, on how, on what you see as the contributions there, the limits, what steps we might take to make this a more productive and enduring partnership.
Yeah, I think that scale is often something that we, we talk about as though it's like not something we really understand how to do when actually we have seen many things go to scale and picking them apart, you know, a little bit to say how did that work and getting the right ingredients from the start with projects would go a long way. I think we, you know, gov we know government isn't in needs to be an involved actor. We've heard it throughout the examples that everyone's discussed here today. And yet too often we do something in a corner and then ask government to look at it without having had them there in the beginning. It's not a pilot if no one's watching, right? We're not actually demonstrating anything to anyone. And I think that is actually something that tech companies have learned, right? We don't build products in isolation, we build products for users, we build them, we know quickly check them and sort of run that process forward.
And I do think there's, there's value in that thinking and that mindset that that, that is under, under invested in I think sometimes in the projects that we collectively support. And I think a lot of that ingenuity and is, and clarity is actually there in the organizations that we're working with. They're just not necessarily, they're just not being offered the opportunity to put the peel puzzle pieces together in the best way that they would want to because they're hemmed in by funding constraints. They're hemmed in by, you know,
this particular Ministry or or agency that, you know, says that they can't do the thing or, or no one's just willing to try. And I think those are things that actually we can, we can do a little bit better job of. I wanna on the, on the role of, of kind of Google and tech and on the data front in particular, you know, I think the pandemic has been an interesting opportunity for us to try some new approaches to releasing more data than we have previously. Obviously we of course take privacy very seriously as a company and so it's quite challenging to think about how we can release information that would maybe be beneficial while still of course preserving privacy in the way that we want to and need to. We did release this year community mobility reports
for the first time offering aggregate data in a fairly small and a sort of county level equivalent geographic area that showed sort of the, the movement based on Android phone data that we were able to connect for those who had turned location sharing on. And we found that those have been quite helpful for a lot of epidemiologists in modeling. So we're asking ourselves questions, you know, about how we continue to do that kind of work going forward. Similarly symptoms, search data sets, so looking at search and saying what are people searching for? How might that be correlated with, you know, upcoming outbreaks and another data set that's now out there that is available to researchers. And I think these are actually helpful having that
forcing function of, hey, this is a big global emergency, how might we be able to contribute? Then brings back the question of what does that look like in an ongoing way. But I do also think sometimes there's a challenge of, of like tech companies and the data will just solve it all when actually so much of the other components are, are critical as, as everyone's been speaking to, right? That that will be one important input as will other data sets out there. But at the end of the day, if governments and other actors aren't there to act and we haven't sort of brought them into that process, then we likely won't achieve the outcome that we wanna have. I think thanks a lot for the comment. I think my, I think my favorite piece of, of what you were saying there was that a pilot is not a pilot unless somebody is watching. And this whole notion that as enamored as we are with big tech and ai,
we still need to work with the broader political and social ecology of public health if we are going to work towards solution, and this is, this is a potential tool, but solving some of these very difficult issues like revenue model and Coordination and figuring out how we're gonna manage privacy are actually really tough nuts to crack. And in fact, when we think about precision health, usually what is thought about with precision health is more precision medicine. That is when you have one patient coming in, how can you get all of the data and all of the genetics and all of the exposure to optimize for that one patient. So, so we've really been optimizing this notion around precision medicine and using precision public health has been, I would say, a less common framing. Cer
2023-10-15 21:03