Adaptive AI-driven Interventions for improved global health

It is increasingly recognised that health data has a significant role to play in improving global health outcomes, by identifying health risks, improving decision making, and evaluating interventions. 

Over the last decade an abundance of data has been created from varied sources including electronic health records, disease surveillance systems and this is increasing due to the proliferation of mobile technology. However this data is fragmented, creating difficulties for accurate decision making and meaningful insights and an increased security risk. 

The current health data is often stored in different formats, scattered across various databases, or even located in different geographic regions. This makes it challenging to consolidate and obtain a comprehensive view of the information. Organizations have to spend a considerable amount of time and effort in integrating the data, cleaning it, and ensuring data quality before they can analyze it. This is if they are even able to support a data science team or single data science professional.

Disconnected health-related data can result in fragmented care, as healthcare providers may not have access to a patient's complete medical history or current health status. This can lead to misdiagnosis, inappropriate treatment, and poor health outcomes. Additionally, disconnected data can create administrative burdens, as healthcare providers may spend significant time manually searching for and consolidating patient information.

Compounding the problem is the shortage of healthcare professionals globally, especially in LMIC and fragile environments. According to the Global Health Observatory (WHO) it is estimated that 18 million additional  health workers are needed in order to achieve Universal Healthcare by 2030. The current shortage is driven by underinvestment in education and training of health workers and complicated by the difficulties in health workers accessing rural areas. 

The shortage of health professionals makes it difficult for patients to access the care they need and leads to long wait times. Health workers will have difficulty prioritizing patients, leading to unnecessary illness and death. 

We need to do more with fewer resources. We need to leverage the tools that we have and the data that they are collecting in smarter ways.

We believe that the next innovation in healthcare will not come from a new vaccine or medication but from behavioral change, resulting from increased engagement in healthcare for both patients and providers. 

Causal Foundry was founded to organize existing data and use the latest AI and machine learning innovations to create adaptive interventions to suggest changes to patient and provider behavior. By driving increased engagement in healthcare tools we can then enable behavioral change, and significantly improve health outcomes in fragile health systems in low resource settings to help bridge the gap in both primary and secondary caregivers.  

Both clinical and patient actions are essential to high-quality, low-cost, effective healthcare. In medical practice, clinicians can be gently nudged to improve decision-making by providing a decision architecture in which optimal default clinical actions are suggested. Outside controlled clinical environments (e.g., intensive care units), patient actions (e.g., lifestyle choices and treatment adherence) primarily determine health outcomes. 

We research and deploy algorithms to improve personalization for health digital services. Our ML products support the work of medical care teams, help patients self-manage, and become more engaged. Our software integrates into existing digital tools, organizing, labeling, and standardizing the data to make sense of it, visualizing and predicting behaviors, and allowing intervention at scale. We developed a data-centric ML platform that allows adaptive interventions, “sending the right intervention to the right person at the right time.”

Our all-in-one platform tracks and organizes provider and patient data to use real-time and predicted behavior to deliver adaptive interventions. It allows massive iterative experimentation with rapid cycles of intervention deployment and optimization.

Our solution fosters patient-centric solutions by supporting healthcare providers and funders in global health to generate more flexible, adaptable and personalized patient journeys.

We work in partnership with governments, not for profit organizations, and healthcare organizations. The Causal Foundry platform integrates with existing digital tools and mobile applications or builds end-to-end, natively integrated digital solutions serving patients directly and through pharmacists, community health workers and health care providers at all levels of the health systems.  

Causal Foundry supports the daily work of healthcare providers by offering, for example, recommendations for drug and treatment adherence, diagnostic tests, patient referrals and the management of medical supplies, based on historical behavior and information gathered in real time. It can also provide a triage service to identify those most at risk and escalate their cases to the secondary care providers to prioritize those who need special attention. 

In addition, the platform supports an iterative cycle of improvements in digital tools for health workers. Program managers can monitor how apps and tablets are being used by the healthcare providers in the field, assessing the most valuable parts of the platforms, and those that are rarely used or used incorrectly. This creates a feedback loop to constantly improve the relevance and usefulness of digital tools for community health workers. 

The platform can also serve funders, both philanthropic and governmental, to transparently monitor the impact and success of health programs and ensure that they are reaching objectives without overly burdening program management teams with additional reporting requirements. We aim to develop the capacity to measure the number of lives that are saved or improved by an intervention, rather than the numbers of people who are involved in a program. 

Causal Foundry's founding team consists of an unusual combination of machine learning engineers, software engineers, and data scientists with extensive experience building machine learning products for personalization. In addition, we also benefit from the support of experienced specialists in Global Health on our advisory board. 

The core team has been working together for more than 10 years;

In 2015, as founders of Yokozuna Data, a start-up in Japan created the first and most powerful ML platform in video games, capable of predicting players' behavior at the individual level and nudging them towards different goals (e.g., playing more, playing better, purchasing more). 

Subsequently, the same team built Zara Brain, the in-house ML platform for Inditex (largest fashion retailer), and the system is still the key ML tool at Inditex. In 2020, the team joined forces as benshi.ai with the Bill & Melinda Gates Foundation to ensure that the benefits of artificial intelligence, personalization, adaptive interventions, and data-centric technologies reach the most under-served communities worldwide. 

Our team has a broad knowledge of data science and ML techniques, bringing extensive expertise (some members up to 20 years) around the cornerstones of our research and development, focusing on Reinforcement Learning, Time-varying and Dynamic Prediction Modelling, Deep and Ensemble Survival Analysis and Synthetic Data Generation technologies. 

AI engineers of Causal Foundry have won (as first authors) top Machine Learning competitions (including the RL competition of NeurIPS in 2021, which is the most renowned AI competition worldwide). 

The leadership team also merges their technology industry expertise with experience in the fields of AI, personalization, and healthcare. 

The advisory board members include impact leaders in Global Health:

Susan Murphy is an American statistician known for her work applying statistical methods to digital health for chronic and relapsing medical conditions. She is a professor at Harvard University and the foremost expert in personalized adaptive interventions in healthcare. 

Pedro Alonso is also part of our advisory board. Pedro is a physician, epidemiologist, and researcher in diseases that affect vulnerable populations. His work focuses mainly on malaria and he served as the Director of the Global Malaria Programme at the World Health Organization between 2014 and 2022. He only advises BioNTech and Causal Foundry.

The Causal Foundry solution is already accessible by the 250,000 pharmacy professionals in the SwipeRX network in South East Asia and their 750,000 clients (WHO). 

We have a number of partners in the development or testing phase on the platform. They reach the following people:

• Medtronic LABS who screened over 1 million patients and trained 3125 health workers in 2021 and serve both private and public sector health facilities

• Appy Saude pharmacy customers in Angola

• Aide Chemists, with 360,000 customers in Ghana

• Chekkit product authentication services in Nigeria

• Drugstoc who currently serve 14 million people with pharmaceutical supplies in Nigeria

• LifeBank -upplying critical health supplies to 1,700 hospitals in Nigeria, Kenya and Ethiopia

As an organization that designs applied science, we are excited at the opportunity of working closely with MIT, an institution that excels at stimulating business innovations in the area of applied science.

We are applying for the MIT Solve program not only because of the wealth of opportunities for mentoring and networking that the program can offer start-ups, but also because it is the right time for Causal Foundry to take best advantage of this opportunity. 

To date we have been focused on developing the product and integrating the partners. We have been very fortunate to have the Bill and Melinda Gates Foundation as our lead investor, they have provided us with great mentoring and introductions to their network. Our relationship with BMGF is strong and we will continue to work with them however, as we move into the next phase in our growth, new challenges are arising, specifically related to business and we recognise that to be part of a cohort of impact based peers, supported by expert mentors, will enable us to learn, network, and grow to be more effective. 

Our business model is based on partnering with the right organizations, thus the exposure possible through MIT solve will be invaluable for reaching the organizations that are best fit for this critical juncture in our development. 

As a company that is trying to stretch our limited resources, we also appreciate any support for software licenses and legal services available.

No single platform performs Reinforcement Learning advanced algorithms to democratize adaptive and personalized interventions in the health sector. The only reinforcement learning platforms that are operational are in-house of big tech or specific companies such as Spotify, Netflix, or JD.com. 

The AI we have seen in the sector is usually a means to organize data and at most a means of gaining insights or making predictions from data. Delivering adaptive interventions from data is our unique added value.

The application of AI in this field is essential because the resources in healthcare are scarce everywhere and will be even more due to the aging of the population. It is also fundamental because there is a transition in medicine, moving from general one-size-fits-all interventions to personalized medicine by optimizing the pathway for diagnosis, therapeutic intervention, and prognosis using large multidimensional biological datasets that capture individual variability in genes, function, and environment. We aim to reconcile multiple data sources to personalize diagnosis and treatment, improve prognosis, and open the door to a new era of evidence-based, individualized clinical practice.

The healthcare sector is transforming digitally. Engagement with patients and providers is becoming easier through mobile health, but little to no thought has been made in the development of these tools of the last 15-20 years as to enable actionable insights on demand like we are offering with the Causal Foundry suite of AI solutions.

Our impact goals over the next five years are based on an iterative cycle of improvement. With each intervention we get access to better data, creating more effective interventions, creating better data. 

As with the theory of change, the impact goals can be divided into three main use cases, pharmacy, primary healthcare and community healthcare support:

• Pharmacy:

  • Next year: 

    • Improved supply chain management 

  • Five years: 

    • Elimination of drug stock outs

    • Working with the public sector health services to predict and prevent disease outbreak 

• Primary health care:

  • Next year:

    • Monitoring and insight on workflows

    • Improved workflows and engagement in digital tools

  • Five years:

    • Improved patient journey

    • Improved management of non communicable diseases

    • Reduced burden on primary healthcare

• Community healthcare worker support

  • Next year: 

    • Development Increased engagement in digital tools 

  • Five years

    • Increased diagnosis of disease and non communicable disease

    • Reduction in maternal, neonatal and infant mortality

Each use case is initially developed with strategic partners to solve specific problems within the global health sector. The AI products subsequently generated can then be fine tuned to get the right market fit and attract additional investment funding.

We are outcome-driven. Measuring impact is a fundamental component of our solutions.

The nature of the data centric Causal Foundry platform is to continuously track behaviors and healthcare records of patients and providers which can be correlated with meaningful KPIs . We can analyze variations over time and the impact that a community health worker has on their different types of patients. We will predict adherence to treatments and engagement of users and the quality of care.

We have the capability to track core KPIs, including the number of health workers who make a patient intervention as the result of an adaptive nudge, the distribution of successful health interventions broken down by population of interest, and engagement with health recommendations. We also conduct time series analyses of changing population behavior over long periods, which serve as proxy measurements for adverse health outcomes prevented.

Causal Foundry believes that the next revolution in healthcare will not be a drug or a vaccine, but rather improving the engagement of patients and providers through personalized interventions. We will nudge patient and provider behaviors towards better health outcomes.

Our mission is to improve healthcare outcomes by making existing digital health tools work harder for all stakeholders in the healthcare ecosystem. 

We will achieve this through three main use cases; Pharmacy, primacy healthcare and community healthcare worker support.

To arrive at this theory of change, we have worked closely with our partners including through qualitative and quantitative research, and with the additional collaboration of the BMGF and our advisory board. 

Inputs: 

Causal Foundry collects, analyzes and organizes data from partners who specialize in three main use cases:

• Pharmacy and medical supply chain

• Primary healthcare

• Community Healthcare worker support

Outputs:

The Causal Foundry machine learning system provides analysis, insight, adaptive interventions,  recommendations and predictions based on an iterative process. 

• Pharmacy: supply and demand analysis, recommendation to re-stock on essential medication and supplies 

• Primary healthcare: recommendations for patient follow-up, increase engagement in digital tools

• Community healthcare worker support: recommendations for relevant information and news, connection with colleagues 

Short / term outcomes:

• Pharmacy and medical supply chain

  • Efficiencies in supply chain saving wasted medication and cost

  • Prediction of demand, especially in response to outbreaks of disease

  • Improved traceability of medication

• Primacy healthcare

  • Monitoring of healthcare worker activity and engagement with digital tools leading to iterative improvements

  • Identification of patients at risk of a chronic illness by monitoring their health data

  • Triage / prioritization of patients 

• Community healthcare worker support

  • Connection to senior nursing and medical staff and secondary caregivers for additional information in complex cases, coaching and mentorship

  • Educational support and advice through a Chat GPT or LLM based chatbot (pretrained with internal information) using natural language to support CHW with the right information, quickly

  • Connection to patients to monitor symptoms and offer ongoing support and advice through whatsapp or other suitable, widely used chat function

Medium term outcomes

• Pharmacy:

  • Identification of disease outbreaks faster, leading to a reduction in disease related illness and death

  • Elimination of drug stock outs and improved access for essential medical supplies

• Primary healthcare

  • Reduction in deaths related to chronic non communicable diseases

  • Improved quality of life for patients with chronic diseases

• Community health care workers

  • Reduction in maternal and infant mortality through improved access to information

Long term outcomes:

• Reduction in inequality in healthcare in the global health 

• Overall improved health outcomes

Causal Foundry researches and builds AI products. Our algorithms are in production both in the machine learning platform where data, experiments, machine learning models, and interventions are governed, and in the mobile application, where the user receives the interventions and is connected to the platform, as well as being able to access information about the status of their disease, nutrition, and exercise. Hence, there is a track to the reaction of every intervention.

An experimentation engine for digital interventions is part of the Causal Foundry software into which the app developed will be integrated. It allows us to easily create experiments for selected cohorts, enabling rapid cycles of experimentation and deployment to continually improve the interventions delivered. The platform permits randomized and adaptive experimental designs and inter- and intra-subject treatment assignments. It includes near real-time monitoring of the experiments, with statistical analyses of the impact and heterogeneous effects in all tracked metrics.

We will compare cohorts using fundamental and advanced features to assess the app's effectiveness and evaluate its impact. The full capacities of the CF platform for experimentation will be available out of the box, and all relevant metrics will be tracked. We will run a series of experiments including different versions of the adherence enhancement adaptive features available, using adaptive designs also in the experiments to maximize statistical efficiency and minimize risk. Impact on patient engagement with the program and treatment will be measured by tracking how they commit to their follow-up schedule, responding to calls and visits, and adhering to the recommended treatment. Additionally, clinical outcome metrics (gravity of symptoms reported) will be tracked and scanned for impact, along with the users' performance metrics described above and additional standard app-usage metrics to control for any unexpected effects of the interventions.

The mission of Causal Foundry is to democratize access to the latest innovations in AI technology to improve global health outcomes especially for marginalized populations. These values of equality of access, diversity and inclusivity are reflected in the way we have built the team, selected the partners that we work with and approach data.

We recognise that there are gaps in the collection of health data globally and that this has a particular impact on women, children and marginalized populations. It is through attempting to identify, and correct or compensate for the missing data that we can start to address inequality in healthcare systems. 

Causal Foundry is a female led startup, a female CEO and 66% female management team. When recruiting team members, we have actively recruited from a diverse pool of talent and the team of 11 represents 8 countries, from Uganda to Pakistan to Russia. This diversity of experience and insight ensures that the platform reduces algorithmic bias, essential in Machine Learning. 

When looking for the right partners to work with to deliver this technology to patients and providers, we seek partners who are locally-established organizations, working in the field and who can provide the local insight and expertise to ensure that we are building solutions that are designed with the needs of the patients and health workers front of mind.

Causal Foundry works with healthcare providers, NGOs and government agencies in the global south to organize, analyze data and create adaptive interventions to improve health outcomes. 

We support these partners to measure and improve their performance against their key objectives and KPIs and to demonstrate the impact of their work. This in turn enables these  NGOs and other organizations to demonstrate value to their funders and when applying for additional funding to continue programs. 

We support our work through different streams:

• Core funding is currently secured through the Bill and Melinda Gates Foundation

• Deployment funding for a selection of partners has been secured through additional grant funding

• Large scale partnerships with industry, specifically pharmaceutical companies 

• Individual partnerships may incur any of a range of fees to cover our overheads:

  • Consultancy fees for data analysis and insight

  • Integration fees for initial set up on the platform

  • Fixed fee for access to the SDK and Machine Learning Platform

  • Variable fees based on performance against KPIs e.g.: increase in activity within a cohort

• Development of digital tools such as android applications

The service is delivered through a software development kit (SDK), connecting the patient and healthcare provider applications with our machine learning platform.

Our aim is to make our products accessible for all healthcare sectors and regions. We work with our partners closely to find the right service and access fees, supplementing with deployment grants when necessary. For our long-term growth we see potential in large scale investments, as well as SaaS and consulting service fees from high-income areas.

Over the last three years, the team behind Causal Foundry has collectively received $10,5M from the Bill and Melinda Gates Foundation (BMGF). This has given the team the opportunity to research into the field and develop an initial version of the product. 

With a 1.5M grant from the BMGF, Causal Foundry has launched its products for its non-profit partners and is currently expanding its revenue-generating client and investor rosters

In addition to the initial funding from the BMGF, Causal Foundry is also shortlisted for two other grant applications for specific partner projects which we should have confirmed within the next month. 

One grant is to enhance our work with one of our existing partners to improve supply chain management and authentication of pharmaceutical products and medical supplies in Africa.

The second grant is to integrate the Causal Foundry Machine Learning platform with a pharmaceutical authenticator and distributor in Pakistan. 

Causal Foundry has also successfully agreed financial terms for a partner to pay a fixed monthly fee for their access to the Machine Learning Platform and SDK. We are in the latter stages of negotiations with additional partners for them to start paying for the access to the service. 

We envisage that the direct costs related to partner hosting on the platform will be covered within 12 months either through chargeable fees or grant funding. 

The business plan has a target of break-even within five years.