Elsa Health

High-quality healthcare care is simply not accessible by a majority of the population across Africa, meaning that people die unnecessarily of preventable causes or have decreased quality of life. We’re committed to improving patient outcomes and reducing misdiagnoses by bringing specialist-level decision support into the hands of healthcare providers in lower-level facilities.

Elsa Health is a clinical decision support platform for healthcare providers across East Africa that optimizes health decisions, increases provider confidence, and improves health outcomes for patients. Elsa Health combines expert knowledge and artificial intelligence to assist providers in identifying the cause of a patient's illness, provide actionable next steps, and identify disease clusters for more precise diagnostics. Our solution is used by healthcare providers across Tanzania, and we are currently working with the government to integrate with the national electronic health system. 

Our goal is to power the health decisions for 10 million patients annually within five years.

High quality care is simply not accessible by a majority of the population across the African continent, meaning that people die unnecessarily of preventable causes or have a decreased quality of life as a result of prolonged illness. Oftentimes, these poor health outcomes are as a result of misdiagnoses at the first point of care; it has been estimated by the World Bank that 4 in every 10 patients in Tanzania at the primary care level are misdiagnosed. For patients with rare diseases, this number is often significantly higher. 

Misdiagnoses in LMICs often result from a lack of specialized care providers, as well as a lack of resources and tools to support diagnostics and treatment. Since rare diseases are complex conditions commonly of genetic origin with nondescript symptoms, lower-level healthcare providers are unable to identify them. As a result, patients who are misdiagnosed experience higher mortality, receive incorrect medications, and lose trust in health institutions - ultimately impacting the development of a nation’s productive workforce. Some estimates indicate patients with rare diseases may wait nearly five years and see an average of seven different health care providers before their condition is accurately diagnosed. 

We believe this should change.

Elsa Health is a clinical decision support platform that optimizes health decisions, increases provider confidence, and improves health outcomes for patients. Powered by causal artificial intelligence models, data, and expert knowledge, Elsa Health supports providers in identifying the cause of a patient's illness, predicting medication adherence, providing actionable next steps based on national guidelines, and identifying disease clusters for better diagnostics and management. 

Elsa Health is delivered as a point-of-care solution to healthcare providers through custom software (mobile and desktop), as well as through integration with existing health infrastructure like information management systems and national reporting tools. Providers can currently interact with the platform in both English and Swahili. 

Elsa is being used by 53 healthcare providers across Tanzania and supports decisions for 5,000 patients a month. We leverage the network of facilities using the platform to constantly improve our clinical algorithms and effectively report to health stakeholders. 

The AI system that powers Elsa is built on data and scientific literature with support from field experts and medical specialists; it is interpretable and capable of expressing uncertainty. The platform has also been co-developed with users  to ensure easy integration and uptake even in the most remote health facilities. The result is low-cadre healthcare providers leveraging powerful and specialist-level decision support in a cost-effective, adaptable way.

Elsa Health serves three populations:

1. Patients (beneficiaries) 

2. Healthcare providers (users)

3. National Ministries of Health and other Health Impact Organizations (customers)

Patients - Beneficiaries

We believe everyone has the right to access high quality healthcare, regardless of where they live or how much money they make. Elsa Health is focused on improving health outcomes for patients by reducing misdiagnoses, increasing time to appropriate treatment, and decreasing unnecessary suffering. Particularly for patients with rare diseases - who suffer from recurring misdiagnoses and high misuse of medications - our technology provides an augmented and optimized health decision across all levels of care. 

Elsa is capable of bringing the decision expertise of high-level specialists to other levels of care providers, strengthening the entire healthcare delivery system and ensuring improved outcomes for patients. 

Healthcare Providers - Users

Healthcare providers benefit from data-powered and evidence-based insights into the condition causing a patient’s symptoms and signs. Our current users report that they like using Elsa because it helps them understand what is wrong with their patient. 

We’ve also heard that Elsa increases the confidence of healthcare providers at lower-level facilities, and helps solidify their roles as effective caregivers in their communities. 

Additionally, since the identification of rare diseases is often complex and requires additional training and diagnostic tools, Elsa helps healthcare providers by flagging possible conditions for further exploration. This reduces burden and cognitive load, allowing them to focus on providing empathetic care for their patients (aka the “human touch”). 

National Ministries of Health & Health Impact Organizations - Customers

Ministries, organizations, and health stakeholders benefit from Elsa through the use of data powered insights that allow them to better understand the health of their citizens and the communities they serve. Elsa’s technology improves their ability to provide care, identify disease hotspots, and allocate resources. For countries where the majority of healthcare is provided through the public system (funded and managed by governments and NGOs), this is a significant benefit. The improved outcomes they see also enables them to achieve national and international goals for healthcare (such as the Tanzania Call for Action for Rare Diseases), as well as solicit additional funding and support as a result of evidence-based insights.

Elsa Health addresses the challenges of misdiagnoses and lack of high-quality care at low level (primary) care facilities by leveraging AI and digital health tools to strengthen health decisions and improve outcomes for patients. Given that a majority of healthcare in low and middle income countries is delivered to communities through community health workers or low-skilled providers, there is a significant need to leverage data and new technologies in order to optimize the identification of rare diseases. In places where there is no oncologist, genetic testing, or specialist care, this is a game changer. Elsa Health aligns with the challenge by leveraging advances in technology to improve the lives of patients with rare diseases and ensure appropriate treatment for better patient outcomes.

We are currently piloting and evaluating the Elsa Health platform in the following areas: 

• With 25 clinical officers in health dispensaries in Northern Tanzania

• With 18 clinicians at Care and Treatment Centers in Northern Tanzania

• With 10 drug dispensers at private drug shops near Dar es Salaam (Eastern Tanzania)

Elsa Health is an innovative approach to empowering healthcare providers and supporting clinical decisions. We have worked closely with healthcare providers and users to develop powerful tools that are easy to use and intuitively aligned with the healthcare provider’s decision making process. Additionally, we have been able to effectively tackle challenges related to low quality data, limited connectivity, and low technology literacy. Our approach of using causal artificial intelligence allows us to develop models that are not reliant on large amounts of high quality data. We’ve built our models (and tools) to work both offline and online, meaning healthcare providers don’t have to have reliable access to the internet in order to leverage the tool’s insights. Since we’ve created tools in collaboration with our end users, we have increased the chance that they will be easy to learn, even if providers have had limited interaction with technology.

Our goal is to make emerging technologies accessible, even in remote areas with limited connectivity. We use a variety of technologies to make this possible: 

• Structural causal models power our AI decision support tools to be able to identify the causal relationships between diseases, symptoms, risk factors, and epidemiological information. Our causal models are built with expert knowledge and literature, and are evaluated against case studies and data for accuracy and performance. Through counterfactual inference we ask the causal models for the best explanations for the observed realities [1].

• All data is stored in a decentralized peer-to-peer network so that sensitive clinical data is available at the main facility and other stakeholders can request for the data from this facility.

• Given that data is stored with the owning health facility, we use federated learning techniques to learn “offline” (on premises) and only synchronize the learnt models instead of real patient data.

[1] Pearl, J. (2019). Causal and Counterfactual Inference. Forthcoming section in The Handbook of Rationality. Retrieved from: https://ftp.cs.ucla.edu/pub/stat_ser/r485.pdf

Digital tools that are capable of supporting health decision making, particularly around disease identification, are becoming increasingly popular. In a health system where a majority of health decisions are made by low-cadre clinicians that work in facilities with sub-optimal diagnostics, these tools are useful to ensure positive patient outcomes. Emerging technologies such as artificial intelligence are capable of augmenting the decision making of providers, and go beyond the current tools such as paper-based algorithms and cumbersome guidelines. These tools learn from experts, literature, and data to arrive at the most (statistically and causally) accurate disease models that improve decision recommendations for the final decision makers.

The recent increase in efficacy of these tools and their role in supporting decisions and improving health outcomes has been impressive. A few examples, among many: learning-based algorithms have shown to accurately forecast the onset of septic shock [1], ML-based pattern recognition methods classified skin lesions with the same level of accuracy as a dermatologist [2], and ML-driven triaging tools have improved outcome differentiation beyond the emergency severity index [3]. 

Many of these tools use a combination of traditional machine and statistical learning methods such as Bayesian Networks, Naive Bayes, and Deep Neural Networks to process patient presentation and symptoms in order to output a list of the most likely underlying causes.

One approach - which we leverage in our work - uses Structural Causal Models (SCMs) or Causal Models (CMs) to process patient presentation and produce the likely underlying causes. The use of Causal Models for decision making [4] is a big improvement to the application of Artificial Intelligence in healthcare [5].

[1] doi:10.1126/scitranslmed.aab3719.

[2] doi: 10.1038/nature21056.

[3] doi: 10.1016/j.annemergmed.2017.08.005.

[4] doi: 10.1093/oxfordhb/9780199399550.013.27

[5] https://doi-org.ezproxyberklee.flo.org/10.1038/s41467...

We recognize that leveraging new types of technology introduces the possibility of multiple risks. The primary risks we have identified are: 

1. Over-reliance on the technology, which could result in the misidentification of a patient’s illness. 

2. Safety and privacy related to patient data, given that we are working in the health field with health information. 

For the first risk, we understand the increased liability that comes with supporting health decisions. We want to ensure that we are effectively supporting the decisions of healthcare providers without introducing an additional cause for concern. We address this risk by ensuring that the healthcare provider is still the one making the final decision, as well as training the healthcare providers on utilizing Elsa in a way that actually supports their decisions. We also take a “safety first” approach to our recommendations, which lets us focus on providing the safest recommendation when we are unsure about a decision. Our models use uncertainty to explain how “sure” they are, and the platform is able to say “I don’t know” if we aren’t sure what might be causing the problem.

Regarding the risks of privacy and data safety, we have an internal data policy, as well as ethical clearance from the Tanzanian National Institute of Medical Research to conduct research related to our technology and collect data. We work closely with data owners and facilitators in order to ensure patient information is protected; where possible, we don’t collect or store personally identifying information. 

Our tools are delivered to healthcare providers across Tanzania on a mobile device in both English and Swahili. We currently have 53 healthcare providers utilizing Elsa’s decision support tools across multiple facility types and in four districts in Tanzania. We are currently supporting 5,000 patient decisions per month, and quickly on track to support 10,000 over the next three months. These healthcare providers all work in the public health sector, primarily in government-owned facilities. 

Over the next 12 months, we expect to scale up massively through high-value contracts with international non-governmental organizations and government partnerships. In one year, we will be supporting 500,000 decisions per month in Tanzania and Kenya. In 5 years, our goal is to power 80% of health decisions in Tanzania, and be present in the healthcare system across 5 additional countries. We expect to scale up massively through high-value contracts with international non-governmental organizations and government partnerships.

In five years from now, we expect to be providing smart medical care services through our platform to 10 million people annually. In addition, we expect to be reaching an additional 20 million patients through our network of institutions. Our impact goals are to: 

• Improve patient health outcomes by decreasing the number of misdiagnoses of patients, particularly for those with rare diseases

• Decrease patient time to care and treatment, particularly for those with rare diseases

• Improve appropriate medication prescription to individuals and reduce the use of unnecessary antibiotics

• Reduce death and mortality (measured in DALYs) and improve patient satisfaction with their healthcare experience.

• Power the decisions of 80% of the healthcare providers in Tanzania

• Increase number of diseases covered by the platform to account for additional rare diseases and hard-to-diagnose conditions

• Increase countries/ locations where we are providing our services

We regularly monitor and evaluate the impact of our solution against the above metrics, and always strive to align with national and international regulations and Calls for Action. In order to achieve these impact goals, we will work closely with key public sector partners, government Ministries of Health, health researchers, and catalytic stakeholders to scale our solution, support healthcare providers, and improve care and treatment for individuals with rare diseases. 

Throughout all of our technology deployments and partnerships, we work closely with the users and beneficiaries to measure our indicators of success. We are already using these indicators to better understand how we are performing. Over the next five years, we will continue to track:

• Number of misdiagnoses of patients (with a focus on those with rare diseases), accuracy of initial diagnosis

• Patient time to care, patient time to treatment (focus on rare diseases)

• Appropriate medication prescription, accuracy of initial medication prescribed, percentage of antibiotic use

• DALYs saved and mortality rates, particularly for pregnant women and children under the age of 5 

• Percentage of decisions that the Elsa Health platform is supporting & effectiveness of the Elsa solution compared with the next best alternative

• Performance and accuracy of our AI models

We have 7 team members (4 full time and 3 part time) working on our team. Four of those members work solely on Elsa Health. 

Contributing to our team and work, we also have 7 interns (undergrad and master's students) who contribute expertise in software development and data science.

Our team is passionate about using technology for social good and we were drawn together by a passion for creating impactful change through technologies that are both powerful and accessible. Our current team has vast experience in front-end and back-end development, system design and architecture, machine learning, and public health research. 

We strongly believe that we can improve health outcomes related to rare diseases by using technology to support decision making and reduce the number of misdiagnoses for patients with rare diseases. We’re excited about the potential partnerships we can learn from and collaborate with through applying to The Horizon Prize. We recognize that working with a globally minded organization that we can make a bigger impact not only in Tanzania but also in the world.

We are excited about partnering with anyone who is working towards improving health outcomes and ensuring that patients with rare diseases receive effective care. We would love to work with any MIT faculty or Solve members, as well as Horizon Therapeutics to see our solution reach people around the world. 

In Tanzania and East Africa, we are excited about future partnerships with: 

• Ali Kimara Rare Disease Foundation (AKRDF), whose mission is to be the voice of an unknown number of people living with rare diseases and to support patients to speak out about their conditions. 

• Tanzania Society of Human Genetics, a non-profit organization coordinating human genetics research and activities in Tanzania in order to generate knowledge and recommendation for the diagnosis and treatment of genetic diseases. 

• Governments and MInistries of Health from around East Africa to share knowledge, influence policy, and support resource allocation.