Utu Care

Utu Care is tackling a significant challenge for the diabetic community: there is currently no way to evaluate, in real time, the risk of developing a serious health complication or get personalized insights on actions that can be taken to address a complication. That is, of course, unless you live at your doctor’s office. 

Of the 442 million people globally living with diabetes, only about 50% of them are able to control their condition. As a result, diabetic populations have high rates of morbidity and mortality due to complications, both acute and chronic. They spend more time in hospitals than non-diabetic populations, and they have more than twice the average medical costs compared to people without the condition. 

Data from emergency departments in the US (2018) showed that 242,000 adults with diabetes presented at the hospital with hypoglycemia, and that 248,000 adults presented with some type of hyperglycemic crisis (9.9 per 1,000 adults with diabetes, majority were experiencing diabetic ketoacidosis). Additionally, almost 40% of diabetic adults have chronic kidney disease, adults with diabetes often have co-morbid cardiovascular disease, and diabetes is the primary cause of new cases of blindness in adults. 

These statistics don’t even begin to scratch the surface globally, and misrepresent the true burden of diabetes-related complications for minority and low-income patients who typically have less access to medical services. 

Real human lives are continuously turned into statistics because:

• The current diabetes care model is flawed. In the best case, patients see their endocrinologist every 6 months for treatment and management updates (more realistically, some patients see their doctor every 12-18 months). But 6 months is too long if a patient is progressing quickly towards kidney failure. Instead, patients end up in the hospital, and doctors end up providing sick care as opposed to preventative and wellness care.

• Doctors are making treatment and management decisions based on a slice in time. In between visits, massive amounts of health information go undocumented and therefore cannot be used to support decision making. Patients are expected to self-manage their condition even though their health needs are constantly changing. 

• Access to high-quality care is disproportionate. Particularly for low-income and minority patients, access to specialists or hardware like continuous glucose monitors is limited. This makes it difficult for them to benefit from continuous data collection and insights. 

• Current digital health solutions are limited in scope. Tools on the market today are great at tracking and charting blood sugar over time. Some also offer logging meals, exercise, and insulin injections, which is useful for looking back and seeing how trends have changed. What’s missing, however, is an all-in-one digital platform that delivers the value of a diabetic care team with targeted, data-driven precision care - this is Utu Care. 

It’s clear that predictive analysis, personalized early intervention, and continuous high-quality care can revolutionize diabetic care and provide individuals at risk for complications with the support they need to capture their health information and take action.

Utu Care is an AI-driven digital health platform that provides precision self-care and personalized complications management to improve health outcomes for diabetic patients.

The platform has 3 major components; Capture, Decipher and Action.

It allows patients to: 

1. Capture: Seamlessly capture all of their health information - from glucose readings to food and exercise, to symptoms, lab investigation results and physiological changes. Data capture can be done through the web, mobile or desktop apps. We are also developing integrations with Apple and Google Health platforms. This multi-platform approach makes it easy to record information continuously and from anywhere.

2. Decipher: All captured information is combined and analyzed in the context of the patient’s history to determine the exact physiological progression of the patient. Combined with selective insights from other similar patients, health state and progression can be estimated with high accuracy and clear uncertainties.

3. Action: Patients can take action through personalized recommendations and summarized ahead-of-time insights that follow approved guidelines for patient self-care. These include the state of existing complications, the risk of developing a complication, the current and next states of the body’s physiology, and the possible effects of any interventions taken (hydration, laying down, exercise, etc.).

Additionally, patients can share and sync information with their healthcare providers, allowing for care to be continuous and not just at a slice of time. This creates an opportunity for integrating into clinical practices by providing clinicians with a one-stop shop for patient management.

Powering the platform is a combination of technologies that solve smaller parts of the problem:

1. Online Reinforcement Learning powers our personalized recommendations algorithm by learning the expected responses per patient with respect to their context in real time. We do this using a combination of bandit algorithms, privacy preserving federated learning algorithms, and gaussian processes.

2. Mechanistic Agent Based Models are used to represent patient physiology through simulating organ and system states (pancreas beta & alpha cell functions, blood glucose levels, peripheral cell insulin sensitivity, etc). Accurate physiological representation of patient states provides the basis for all predictions and is by default explainable, interpretable, and auditable.

3. Distributed Patient Graph (Matrix) that connects each patient with other patients to “fill in the gaps” and provide additional needed data. This approach to crowd-learning makes it possible to use data from a patient with a continuous glucose monitor (CGM) to understand the responses of an IDENTICAL patient without a CGM.

The technologies underlying the platform are simple and battle tested across different industries and benefit from decades of research and refinement. Coupled with the deep clinical expertise and current health research, their application to our problem area brings new potential for seemingly unsolvable problems in health care.

Utu Care benefits three populations: 

1. Diabetic patients
2. Diabetic care teams
3. Health payers 

Diabetic Patients

Utu Care benefits patients who want to easily manage and prevent complications, as well as get access to high-quality care between visits with their care team at an affordable cost. Individuals with diabetes who are at high-risk for complications are significantly underserved by the current model of diabetic care. They have higher-touch needs and, as a result, spend significantly more money on healthcare services than low-risk diabetics. Additionally, while there are many digital health tools on the market for diabetic populations, existing tools do not provide integrated complications management, personalized care, or comprehensive risk analyses. 

We are specifically committed to serving minority and low-income patients - both in the United States and globally - who do not have access to traditional care services (i.e. regular visits, lab tests) or tools they need to continuously monitor their health (i.e. CGMs\). Utu Care benefits these populations by leveraging collaborative learning algorithms to estimate an individual’s physiology regardless of how much information they are able to share. 

Providing personalized, high-quality decision support and insights about disease progression and complication management for diabetic patients will: 

• Decrease complication-related morbidity and mortality;

• Improve glucose control and give individuals a deeper understanding of how their food and daily activities (interventions) impact their blood sugar;

• Reduce the cost of managing, tracking, and treating diabetes-related complications;

• Increase the number of patients who have access to specialist-level health decisions.

Diabetic Care Teams

Utu Care benefits entire diabetic care teams - endocrinologists, clinicians, dieticians, and family/caretakers - who want insights into their patients’ behaviors and health progress in between clinic visits. Currently, there are no streamlined ways to get real-time insights outside of the clinic, and patient behavior is difficult to control and predict. Additionally, adjusting patient treatment plans between visits is nearly impossible. Utu Care is able to integrate our real-time platform into the care workflow in order to bridge the gap between patient visits and allow healthcare providers to make more data-driven decisions. 

Health Payers

Our platform benefits health payers who want to reduce their spend on chronic illness complications. Treatment and management for diabetes-related complications costs insurance companies and governments billions of US dollars per year. This is entirely unsustainable as more and more individuals are being diagnosed with diabetes. We work with health payers who offer the platform to their customers as a preventative measure. As a result, they are able to optimize resource use at a significantly lower cost. 

Overall, Utu Care’s patient-centric approach addresses the diverse, unmet needs of diabetic populations and reduces burden and cost across the entire care system.

Our team started working on Utu Care because we saw a massive gap in high-quality and continuous care for marginalized and low-income patients living with diabetes. Megan Allen and Ally Salim met while working together in East Africa on AI solutions for public health systems. While there, we collaboratively built solutions for low-income populations and came to learn about the significant burden of diabetes on our community. We watched as close family friends suffered from leg amputations, blindness, and renal failure. This gave us deep insight into the challenges and opportunities in the care system. 

We moved our operations to the US, recognizing that while the challenges were different, poor health outcomes persisted. Our team is built around a love for technology and a passion for human health. We are multidisciplinary and diverse - composed of young innovators with the expertise, relationships, and grit to address the burden of diabetes at a global scale. 

Across the team, we bring a set of unique skills in artificial intelligence, medicine, software engineering, monitoring and evaluation, public health research, digital health implementation, and ethics. We have been building AI tools for the past decade, and have experience implementing digital health technologies in resource-limited settings. We have an optimistic view of technology's potential to improve our world.

Community & Stakeholder Engagement

To ensure that our solution meets the evolving needs of the communities we serve, as well as involve them in the development of the solution, we:

1. Build in the open: Our team is committed to transparent and open science. We’re taking every effort to build Utu Care in collaboration with our communities and stakeholders by making it easy to provide feedback, track the progress of our roadmap, and critically evaluate our models. 

2. Conduct human centered design activities: We co-develop all of our solutions with target populations and seek regular input to ensure we are developing solutions that take into account varied cultural and social perspectives. We are currently working with diabetic individuals and clinicians to solicit feedback on early versions of our platform. 

3. Do continuous discovery with users: We regularly conduct customer interviews and seek feedback from our users throughout development cycles. 

4. Stakeholder Relationship Development: We work closely with our partners to learn, build, and grow. We have developed strong partnerships with medical experts, researchers, and health stakeholders across the United States and Sub-Saharan Africa, which positions us to effectively develop and scale Utu Care.

The key innovations of our approach are:

1. Crowd-learning of diabetes models from individual patient data and leveraging the learnt model to provide insights to new patients.

2. Agent Based Models (ABMs) to simulate patient internal health states.

By combining these two approaches with the massive distribution channels of software, we are effectively extending the reach of the health system for near-24-hour personal health monitoring.

Crowd-learning makes it possible to transfer learnings across all participants to benefit individual participants and their care teams. Learnings from one patient can be used to support other identical patients in novel situations through similarity matching (using various distance metrics). This approach is potentially the only useful thing to come out of the online advertising industry.

Our use of Agent Based Models takes model personalization to new heights. Traditional statistical/machine learning starts with fitting lines to curves using massive data sources collected from individual patients. ABMs start from what we know and have compiled over decades of diabetes research. They are *supposed* to overfit individual patients to provide personalized insights. ABMs provide us with a unique interactions interface that supports open ended interventions that *do not* need to be captured in training data, as the physiology is already well understood. We can finally ask interesting questions that are not possible otherwise. For example, through ABMs, we do not need to collect data on patients with sickle cell anemia, we just need to describe the role of Red Blood Cells in the body and parameterize it with the features of sickle cell anemia to observe emergent behavior.

This approach doesn’t stop with diabetes and can be applied to new and novel types of care, by simply describing the new agents and the rules of interaction between them. Next on our list are other chronic illnesses of neural degenerative origin.

Utu Care specifically supports progress towards two of the UN Sustainable Development Goals sub-targets for Goal 3: 

• 3.4 reduce mortality from non-communicable diseases

• 3.8 achieve universal health coverage

By focusing specifically on diabetes-related complications (as opposed to only glucose and meal tracking), Utu Care is able to target a critical moment in a patient’s disease progression. Diabetes-related complications are the sole contributor to morbidity and low quality of life for diabetic patients, and significantly contribute to increased mortality. We’re able to address complications before they happen, identify interventions that will positively impact health outcomes, and support patients on their path to wellness. As a result, health expenditures will also be reduced, directly contributing to sub-target 3.8. 

We anticipate being able to evaluate Utu’s impact on patient mortality after 5 years.

The two crucial components of any artificially intelligent system are the data and the algorithms. Our approach innovates on models and algorithms to support leveraging heterogeneous data collected from different sources and individuals.

Our team is in conversations with local indigenous health systems, as well as researchers from Arizona State University and Valleywise Health to leverage their data for the fine-tuning and validation of our technology. We are also pursuing a delivery channel through these same partners.

From a decade of health AI development and deployment, we have learned that it is very difficult to have a single end-to-end model. We are taking a more modular approach and solving individual problems with specific AI techniques:

1. Online reinforcement learning: A combination of Bayesian contextual bandits, variations of Hidden Markov Models, and Gaussian processes to power recommendations and understanding the relationship between different causes and effects in real time. This allows for adaptive recommendations that change as the patient’s condition progresses

2. Graph learning from data: Given the global distribution of patients, we leverage information and graph theory to create a global matrix of similarities that we can perform inference on. For simplicity, think about a large Matrix (a tensor really), and applying K-nearest neighbors on it. This is how we achieve collaboration between patient models across space and time.

3. Agent Based Models: With ABMs, we model and simulate each relevant organ/organ system and observe the interactions between them to understand the patient’s physiology by comparing with the observations. This allows for greater generalizability to patients of all backgrounds and contexts. It also allows care teams to simulate outcomes of different interventions and drugs with a sample size of 1. This approach to personalized care is our contribution to science.

A tradeoff we make that other machine learning approaches don’t typically make is to treat the problem as one of “understanding” and not “prediction/forecasting”. The downside is the computational expense we have to pay, but the upside is the generalizability of the outcome and the explainability/auditability of the overall platform.

The ethical and responsible use of AI has been a core value of the team for the past eight years. We follow at our core the medical adage “Do No Harm” and have robust policies in place that ensure privacy, safety, and accountability in all our work and technology development. 

From our experience, we build the following guards into all our products and services to mitigate risk before any development even starts:

1. Accountability thorough audit trails: Thought processes taken by the models are constantly logged for scrutiny and evaluation at a later time. This auditability, together with the nature of ABMs, allows for “replaying” the exact conditions that happened and debug any issues that arose.

2. Transparency through explainability: The models we use are white or gray box, giving us and third party auditors a simple way to understand the outputs and see the relationships between the inputs and outputs. We also try to stick to Bayesian methods that provide uncertainty distributions, which allow us to communicate ignorance in situations where the model is uncertain.

3. Overridable by a human-in-the-loop: Users of our models can interact with them and can change parameters on the fly and even “converse” with the model through simulating counterfactuals. Furthermore, we allow human users to make the final decision.

4. Differential privacy whenever possible: Instead of sharing the patient data, we have differential privacy algorithms that create realistic data from the real data that can be shared to third parties like researchers. This approach is superior to anonymization because it is not possible to reverse engineer the data to identify the owner.

5. Compliance with GDPR regulations: For patients living in many parts of Europe, data is stored in compliance with territorial laws.

6. Strong encryption of sensitive data: Patient sensitive data is encrypted at rest and in transfer.

7. Robustness against data and model drift: The nature of ABMs and online learning algorithms results in significantly reduced risk of model/concept shift and data shift. However, even though the risk is very low, we still periodically check for these two metrics and keep a close eye on them.

We can also support an offline-only strategy where the patient data never leaves the patient’s device. Patients can still leverage all the collaborative learning models we use, but with slower and slightly reduced performance for some actions.

Utu Care’s long term objective is to be the world's only self-care platform for diabetic-related complications. Our impact goals are: 

One Year: Our primary impact goal over the next one year is to release the Utu Care platform to the public, expand the reach of the platform, and validate the technology that powers the solution at scale. We aim to onboard 50,000 patients by the end of 2024, which we will achieve by collaborating with key partners (health systems) to increase adoption, marketing efforts targeting ideal users, and continuously improving our platform to ensure it meets user needs. We also aim to maintain an average of 4/5 star rating by our users across platforms and distribution channels, which will drive more traffic to our platform. 

Five Years: We plan to achieve our long term objective in five years. Our impact goals are to: 

• Reach 5.5 million patients globally
• Achieve a 20% reduction in morbidity for active users of the platform, as measured by complication rate, amputations, QALY, and delayed onset of complications
• Lower costs for users by 50% compared to similar patients not using the platform

We are specifically focused on impacting low-income and minority populations. To do this, we will work closely with health systems across the US (Native health providers, county hospitals, etc.) to ensure Utu Care is accessible and usable. We are operating under the assumption that patients in our target population have both access to smartphones and the desire to proactively manage their condition. Research shows that 76% of low-income households in the US have smartphones. 

Our strategy to accomplish these goals lies in our commitment to developing state-of-the-art artificial intelligence, deeply understanding the community we serve, and executing on our operational plans. We’re also excited about contributing to the open science and technology movement; we believe this will disrupt the market in positive ways for patients.

Full time: 3

Part time: 2

Contractors: ~8 (as needed)

16 months

Incorporating diversity, equity, and inclusivity (DEI) into our work is a fundamental and unwavering commitment. We recognize that DEI is a source of strength and innovation. We are dedicated to building a diverse team that reflects the varied backgrounds, experiences, and perspectives of the communities we serve. We actively seek out individuals from underrepresented groups and provide an inclusive and equitable workplace where all team members can thrive.

The core Utu Care team is diverse across a range of features (gender, experience, background, and nationality), which allows us to explore a variety of solutions during development and implementation, and ensures that all ideas are considered. Our team includes: 

• Ally Salim, a Tanzanian national with more than a decade of experience in artificial intelligence, software engineering, and mathematics. His work focuses on leveraging emerging technologies to improve health systems and reduce the burden of disease in limited resource settings. He understands deeply how to implement technical solutions in places where both technology and doctors do not exist. Ally regularly contributes to the WHO/ITU Focus Group on AI for Health. 

• Megan Allen, is an American national who received her MSc in Biology from Arizona State University. She is a public health practitioner and researcher with experience designing, implementing, and evaluating digital health interventions in Sub-Saharan Africa and the United States. 

• Dr. Kelvin Mariki is a Tanzanian licensed medical doctor who leads our team’s clinical work. He brings needed clinical skills and knowledge of physiology to our team of technologists, and has experience developing and using AI models for clinical decision support. 

• Sarah Mure, MPH brings expertise in planning, managing, and evaluating public health interventions in low-resource settings. She has worked in bilateral organizations (USAID) and brings valuable evaluation experience from her previous role as the Country Director for an international non-governmental organization in Tanzania.

Our commitment to incorporating DEI into our work is multi-faceted and holistic. We encourage open dialogue, respect for different viewpoints, and actively involve team members in decisions to ensure a range of perspectives are considered. We have a dedicated team member who is responsible for ensuring that both our team and our projects represent varied perspectives. Additionally, we actively engage with the communities we serve to ensure that their voices are heard and their needs are met. This includes seeking input and feedback, partnering with local organizations, and ensuring that our work is culturally responsive.

Our approach to incorporating diversity, equity, and inclusivity into our work is embedded in our core values and is a continuous journey. We are dedicated to fostering an environment that celebrates diversity, promotes equity, and includes everyone's perspectives and voices. We believe that by embracing DEI principles, we not only enhance the quality of our work but also contribute to a more just and inclusive society.

Team Organization:

• Utu Care is a multidisciplinary and experienced team of technologists and healthcare providers. Together, we have worked on a number of AI for health projects that have delivered impact to tens of thousands of patients.
• We regularly bring on consultants to fill specific roles. These individuals are often medical doctors who support algorithm development and validation. 
• The team works in a highly collaborative way. While the team has their defined roles based on their skills, we recognize that an “all hands on deck” approach is sometimes needed to achieve our impact goals. 
• Moving forward, we see the need to bring on an additional software engineer and someone with expertise in medical software sales.

User & Stakeholder Engagement:

• We actively engage with patients and healthcare provider users through continuous discovery and feedback sessions. This is a critical part of our development process. 
• Health providers who benefit from Utu Care’s platform serve as critical entry points for new patients. We work with them to ensure the solution provides value so that they continue to recommend it to patients. 
• We work closely with health service providers and organizations to validate Utu Care’s models and get access to data. In the future we intend to work with these partners to conduct an impact evaluation of the platform.

Tools & Resources: 

• The Utu Care team has access to needed technical resources (compute, data storage, etc.) for the first phase of the project. As we scale, we will seek additional technical resources from partners, as well as scale up our internal capacity as needed. 
• The primary resource needed right now is funding. The core team has bootstrapped the development of Utu Care since its inception.

Implementation Plan: Utu Care has a multi-phased implementation and go-to-market strategy.

• Phase 1 (2023- 2024): This phase consists largely of R&D for our proprietary software and algorithms. We have already started testing early versions of the platform in private beta, and plan to release v1 to the public in early of 2024. We anticipate continuously updating the platform based on user feedback. We will reach our impact goal of 50,000 patients during this phase by investing heavily in top- and middle-of-the-funnel marketing.
• Phase 2 (2025-2026): This phase is focused on growth and continued validation in order to establish a burden of proof for the technology. We plan to initiate an impact evaluation/ randomized control trial to demonstrate that Utu has a statistically significant impact on patient outcomes compared to the standard of care. Throughout this phase, we will continuously improve our solution based on customer feedback. 
• Phase 3 (2027 beyond): Growth and scale to achieve our five year impact goals. 

We plan to become financially sustainable during Phase 2 of our implementation. Our approach relies on a diversified revenue model:

Patient D2C Subscription Service: Although we are still experimenting with pricing strategies, we plan to offer subscriptions for diabetic patients to access the platform. We actively seek new customers through top- and middle-of-the-funnel marketing activities, including paid advertising, case studies, social media, and content. We are also actively establishing partnerships with health organizations and research institutes to expand our reach and generate revenue. 

Healthcare Provider & Payers B2B: We plan to offer a subscription service to providers that allows them to access Utu Care’s insights, as well as seamlessly connect to their patients outside of the clinic. As we evaluate the clinical efficacy of the platform, we will leverage published results in peer-reviewed journals as evidence for providers and administrators. We also plan to work with large health plans in the US, who are able to charge a per-member fee to reduce complication rates (and thus reduce the costs that the health plan incurs). 

Grants and Non-Dilutive Funding: We intend to fund our R&D through grants as a way to catalyze growth and retain ownership. We are currently seeking SBIR and other federal funding to continue development and validation on our proprietary algorithms. 

Investment: We are not currently raising a round of funding. We may seek investment from impact-focused venture capitalists in the next 18-24 months in order to validate our technologies. 

Our revenue streams are projected to cover our expected expenses in the long run. This is critical for our financial sustainability as we expand our reach and improve our platform.

The founding team has bootstrapped the initial development of Utu Care. 

Current Operating Costs: $2,000 - 2,500 per month
The team’s current operating costs are minimal. We have very low human capital costs as the founding team is not paying themselves. Since we are a remote team, we also do not have office rent and related overhead expenses. 

Current expenses primarily include: 

• Technology, including servers, cloud compute, hosting platforms, etc.
• Expert consultation, which primarily covers clinical consultants who support the validation of the algorithms. This varies month over month. 

Projected Operating Costs for Next Year: ~$12,000 - $15,000 per month
We anticipate an increase in our operating costs over the next year. In order to meet our implementation timeline and planned impact goals, we need: 

• Human Capital / Salaries: $130,000 (inclusive of very modest salaries for the core team, as well as an additional software engineer)
• Consultants: ~$18,000 (for clinical validation) 
• Data acquisition: $5,000
• Software infrastructure development and computational capacity: $10,000 
• Overhead and legal: $10,000

We are requesting $100,000 in funding from Cure to cover the remainder of our Phase 1 implementation (platform and algorithm R&D, validation of algorithms, and customer engagement and marketing). 

Utu Care is at a critical inflection point. Our team has been able to do a lot on a very modest budget; we have validated our market and need funding to continue developing and evaluating our solution. The amount requested has been carefully selected based on the needs of our team for the upcoming year, as well as what is required in order to meet our planned impact goals. 

The requested funding will be used directly for programmatic impact and will significantly contribute to our team’s success. It will be used for: 

• Salary for Team Lead: $35K
• Salary for part-time software engineer: $20K 
• Salary for part-time medical doctor: $15K 
• Clinical consultants for model validation: $12K 
• Data acquisition / primary data collection: $5K
• Software infrastructure and computational capacity: $4K
• Marketing: $5K
• SG&A (legal, accounting, offset operating expenses associated with proposed work, etc.): $4K

We are deeply committed to improving global healthcare, and we feel confident in our ability to deliver the technical components of our solution. Right now, our team’s biggest needs are capital, warm intros to relevant stakeholders in health systems and health plans, and mentorship/ guidance on how to significantly grow our reach through D2C and B2B strategies. We’re also looking to better understand the best approaches to scaling AI models in healthcare and navigating regulatory requirements along the way. 

Participating in the Cure Residency program would be an invaluable opportunity for our small team and would provide crucial support in several ways:

1. Seed Funding: As a small company with a limited budget, the financial support provided by the Cure Residency will be a significant boost. This funding will enable us to advance our R&D efforts and bring Utu Care closer to market readiness. It will help bridge the resource gap and enable us to take our work to the next level.

2. Mentorship: One of the most exciting aspects of the Cure Residency is the mentorship component. Our limited access to high-profile mentors has been a significant challenge for us. Being paired with experienced mentors in the field will provide us with guidance, expertise, and industry insights that are otherwise difficult to access. This mentorship will help us refine our strategies, make informed decisions, and navigate the complexities of the healthcare industry effectively.

3. Educational Programming: The educational programming offered as part of the Cure Residency is another exciting element. It will provide us with opportunities to expand our knowledge and skills, staying up-to-date with the latest advancements in AI for health. This continuous learning will be instrumental in refining our projects and ensuring they remain relevant and cutting-edge.

4. Networking Opportunities: Being part of the Cure Residency will expose us to a vibrant community of professionals, researchers, and entrepreneurs. This expanded network will open doors to potential collaborators, investors, and partnerships, which are essential for our growth and sustainability.