AI for health, powered by health workers

Machine learning that incorporates health worker expertise enables predictive supply chains for vaccines and builds capacity for data use.

Health supply chains are increasing in complexity and scale. More than 100 new vaccines have been added to the market in the last 5 years and vaccine prices have increased more than 3000% in the last decade [Gavi, the Vaccine Alliance]. Global coverage for basic childhood vaccines has reached a record 86%, but there has been a parallel increase in vaccine wastage –~30% of vaccines are estimated to be wasted – decreasing resource efficiency. Vaccine stock-outs compound the problem by wasting invaluable opportunities for immunization and harming the trust of communities. macro-eyes is working with the Ministry of Health in Tanzania and the Bill & Melinda Gates foundation to design the first predictive supply chain for vaccines to break the link between higher rates of immunization and increased wastage.

Machine learning is not just focused on algorithms; it is also concerned with how machines can learn from humans. Human-in-the-loop (HIL) machine learning can markedly decrease the number of labelled data points necessary for prediction tasks, instead relying on domain expertise. HIL ML focuses on how to use human insight: to efficiently train machine learning models and access difficult-to-acquire information that can increase the accuracy and precision of predictive models.

Machine learning is rarely put to work to tackle problems in global health. The oft-cited reason is the lack of large-scale, high-quality labelled data. Accurate immunization data requires engagement from frontline health workers and trust that data will empower and not merely check a box. Data entry for the sake of data entry is how data collection is often perceived on the ground.

macro-eyes will deploy human-in-the-loop machine learning to engage frontline caregivers and gather valuable information to power the predictive supply chain for vaccines. Frontline health workers are domain experts. They know more than anyone about the delivery of vaccines where they work and how the catchment population perceives immunization and will access facilities. Frontline health workers bear the brunt of data work, but see little reward. They deserve better.

By meaningfully engaging health workers and encoding the provided information for machine learning, routine data entry will be reduced and data quality and actionable insight increased. Direct engagement also creates data champions at the point of care who feel valued. Health worker insight on populations and demand, conveyed via text message [WhatsApp], will programmatically augment the analysis of supply chain and immunization data.

This initiative will introduce health workers to the link between data quality and insight, opening the door for widespread use of machine learning for global health. This will be one of the first deployments of machine learning – specifically human-in-the-loop machine learning – for global health. Our aim is to maximize childhood vaccination coverage and minimize vaccine wastage with precise utilization prediction, translated into vaccine deliveries that anticipate demand. H

macro-eyes AI was designed by Chief AI Officer Suvrit Sra, PhD: MIT professor and expert in large-scale machine learning (ML) and optimization. macro-eyes AI leverages multidimensional similarity; similarity metrics are learned in real time to personalize prediction. macro-eyes AI is the result of refining and deploying ML at Stanford, one of the largest health systems in the US and at federal qualified health centers across the country. The Gates Foundation awarded macro-eyes funding to design the first predictive supply chain for vaccines (PSCV). Human-in-the-loop (HIL) ML informed the work at Stanford. The HIL component of the PSCV will be ground-breaking.

Supply chains for vaccines rush to catch-up (bringing supply too late) or deliver too early. Core macro-eyes AI technology for real-time learning from data is core to implementing the first predictive supply chain for vaccines. To get to scale it is critical to encode information that health workers share as part of the infrastructure for predicting utilization. The regions that most require vaccines often lack in the data to ensure the right vaccines are delivered at the right time and in the right quantity. The problem cannot be solved without technology for learning from data and learning from people.

Over the next 12 months we will (1) identify and engage health worker communities, developing champions and super users, (2) design and deploy human-in-the-loop machine learning in parallel with Gates Foundation funded work to implement the first predictive supply chain for vaccines (3) refine and document the process of iterative data gathering for AI in the real world. 

We will partner with existing systems and infrastructure. Productization will bring a repeatable solution to a problem faced by every nation: supply chains deliver too few or too much of critical health commodities, too late or too early. Human in the loop ML productization also brings a repeatable solution to the data quality issue. AI has not come to global health because of the lack of high quality data. Through partners we will touch hundreds of millions of lives and reduce delivery cost for billions of dollars of vaccines, pharmaceuticals, and other health commodities to those most in need.

Partnerships with existing operators and political and government leaders will be the engine of our growth. Our relationship with the Bill & Melinda Gates Foundation and the DRK Foundation has significantly de-risked macro-eyes technology and our approach. Further deployments will provide greater access to the global health community. We are working with our partners to develop a risk-based licensing model (license fees are paid through achieved cost-savings) and a licensing model that echoes the Gavi approach: software cost is based on ability for health system or government agency to pay (measured by GDP); countries graduate into assuming greater cost.

macro-eyes AI has analyzed 1,000,000 patient medical records and 3,000,000 appointment records in the United States and East Africa.  We have used this data to refine core AI, analyze patient behavior, increase operational efficiencies, and surface the most effective clinical intervention at the point of care.

Within 12 months this technology and process will touch more than 160,000 patients within the Tanzania BID VIMS database. We are currently working with this data base and this new approach would be integrated with our existing technology. 

Within three years we expect to reach 10x this population by expanding into the Zambia BID VIMS database currently under construction along with data bases with existing partners in Zimbabwe and Mozambique.

CEO Benjamin Fels is a DRK Entrepreneur. He led quantitative, market and technology teams for a hedge fund. Built team of experts in healthcare + AI deploying macro-eyes AI at leading academic medical centers and rural FQHCs.

Chief AI Officer Suvrit Sra, PhD is a world-renowned expert in AI and optimization and professor at MIT. 

Chief Business Officer Drew Arenth, MBA is a Worldwide Fistula Fund board member, DRK Entrepreneur, and World Affairs Council Fellow. Drew was a Principal with the strategic leadership team at Providence St. Joseph Health and led global business development for energy supply chains.

Revenue model: standard technology licensing model. We are a diversified machine learning company providing software and services to domestic healthcare organizations, medical research organizations, and global health organizations. Diversification enables sustainable growth, impact, and expansion. 

We work with development partners to improve and de-risk innovative solutions.  We build trust with end users through pilots and proof of concept work.  Our tools all provide positive ROI. Sales rely on demonstrating this impact and enabling the organization to allocate anticipated ROI towards license fees.

We want to be part of the Solve community.  The funding would accelerate an innovation in data collection and data use that enables machine learning for global health at scale and brings the predictive supply chain for vaccines to scale. The Solve community is an environment where we can learn from others and share our own expertise. The support network and the sector visibility will be invaluable as we work to bring new, critical technology to a cautious domain.  The trust in MIT and the Solve community can help us to remove barriers to bringing machine learning to global health. 

Key barriers: government trust and participation and willingness to make supply chains dynamic and responsive to machine learning generated insight. Solve can help by helping to form bridges to government leaders and connecting us to experts in working effectively with ministries of health and political leaders. Most importantly, Solve can work with our technology and our team to refine how we communicate the core science and the impact in terms that are compelling and clear to government officials.