AI/ML for Outbreak Prediction & Response

Given the unique challenges faced by LMICs in the event of a pandemic, as well as the lack of epidemiologists and pandemic specialists in these locations, machine learning and data science techniques can be leveraged to prevent, mitigate, and respond to future disease outbreaks. Our solution - Elsa for Outbreaks - uses machine learning and digital tools to predict, monitor, and respond to pandemics. Elsa visualizes disease information across a country or region through an online portal and allows health stakeholders to model various interventions to see how they can better deploy resources, such as healthcare providers. If a new disease does emerge, Elsa immediately begins tracking that disease and incorporates new clinical information into its algorithms to augment detection of the disease by front line healthcare workers. Tools are deployed to patients and healthcare providers that offer quick symptom assessment, virtual follow ups, and digital referrals to appropriate facilities.

Despite the experience that low and middle income countries (LMICs) have in responding to novel diseases, they often perform poorly on evaluations for pandemic preparedness. For example, a majority of African countries ranked “none” or “limited” for preparedness capacities related to the COVID-19 pandemic. Health professionals in Tanzania have mentioned that they are in “uncharted territory” and face significant barriers to ensuring that they are able to slow the spread of disease. 

Although the East African Community released a comprehensive plan for COVID-19 response that heavily focused on strengthened communications, health provider up-skilling, and resource mobilization, not all East African countries were capable of implementing this response plan. Key stakeholders need specific data about their communities, and need a way to see specifically how various interventions make a difference. Health providers at the frontlines of care need immediate clinical decision support for unknown epidemics. 

In addition, health data is often collected for data’s sake and is highly fragmented, meaning that they don’t leverage it for useful decision making. Elsa for Outbreaks changes that.

Elsa for Outbreaks is a digital health and machine learning solution for predicting and responding to disease outbreaks. Our tools augment the capacity of healthcare providers and health stakeholders, particularly during a pandemic. 

Outbreak Prediction/ Detection - Elsa allows health stakeholders to visualize disease distributions across the country and see predictions for when and where outbreaks are likely to happen. In addition, Elsa suggests and highlights key decisions for outbreak response and lets stakeholders model various response activities to see their impact on the pandemic. Elsa is trained on disease data from country-level datasets and uses machine learning methods for projection and predictions of disease outbreaks. 

Outbreak Response - During a pandemic, Elsa is deployed to patients and at the clinic through mobile and web applications to track the spread of an emerging outbreak and conduct symptom assessment for triaging patients. As a pandemic unfolds, aggregated data collected from the field gets made available for monitoring, as well as gets integrated with transmission data to inform our prediction models. 

Elsa for Outbreaks benefits key health stakeholders and decision makers, which typically include members of the Ministry of Health, influential non-governmental organizations, top medical professionals, and members of an emergency outbreak task force. The Outbreak Portal allows stakeholders to see where outbreaks are likely to occur and what interventions should be implemented in order to slow the spread of disease or mitigate the impact. Through the portal, stakeholders receive highlighted suggestions for pandemic response based on previous suggestions or best practices, and can test alternatives to various decisions. For example, they can see what will happen if they deploy more mosquito nets to one area and deploy more community health workers in another. We have heard directly from stakeholders that this would significantly support pandemic response, and would enhance monitoring capabilities. 

Our solution also supports frontline healthcare providers and the general public by deploying symptom assessment tools for evidence-based decision making. With these tools, healthcare providers are also able to receive up-to-date and real-time information about the ongoing pandemic, something they traditionally receive through printed materials or formal training. 

Elsa for Outbreaks supports both the response to an ongoing or emerging pandemic, as well as strengthens preparedness for unknown future pandemics. We are able to use data science and digital health solutions at multiple touch points to augment decision making capacity and ensure that a country is capable of appropriately and effectively responding. The Challenge seeks solutions that can do just that. The COVID-19 pandemic has shown the strengths and weaknesses in the global pandemic response. 

Our solution empowers health stakeholders and health providers to take effective action and improve their response to new and emerging pandemics.

Our outbreak prediction tools utilize hidden markov models In addition, our solution and database integrate with DHIS2 in order to get access to current disease trends. We have custom built a web portal to view visualizations of outbreaks and test various implementation options. 

Our outbreak response tools utilize bespoke machine learning models for clinical symptom assessments and next steps recommendations. We have custom build data collection tools and software interfaces to support healthcare providers and ensure that accurate information is both collected and used for monitoring. 

With these tools, patients and providers can directly send data to one another for pre-clinic triage, appointment setting for high risk patients, and manual follow up and interventions when needed. Data from both healthcare providers and patients is constantly synchronised and stored in the database through the Elsa API.

There is a lot of research that supports the use of hidden markov models for epidemiology and disease prediction. The following are some papers we have used to guide our model development. 

Rath et al., 2003: https://link.springer.com/chapter/10.1007/978-3-540-45231-7_48

Watkins et al., 2009: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2735038/

Zhang & Berhane, 2013: https://europepmc.org/article/pmc/pmc4221103

We utilize Bayesian methods for our symptom assessment models. They are developed with support from specialist medical providers and are trained using clinical patient records from Tanzania. There is precedence for using these methods for disease prediction. 

We currently have 5 team members who work on our solution. 

Our team has been at the forefront of developing machine learning solutions for healthcare in East Africa and we believe that this pandemic creates both an imperative to act and an opportunity to learn. We have experience in machine learning, artificial intelligence, software development, and systems design, health research, and digital health solutions. We are deeply connected in the healthcare ecosystem in Tanzania, and have knowledge of the challenges faced by all health stakeholders, from community health providers in villages to the Ministry of Health. 

We have built and deployed a number of solutions that utilize artificial intelligence and machine learning technologies to provide decision support for healthcare providers and patients. We are currently working with PEPFAR and local collaborating partners to develop intelligent tools that will support the HIV/AIDS epidemic in Tanzania, such as risk assessment, prediction of antiretroviral dropout, and prediction of HIV drug resistance. In addition, we are a contributing member of the WHO/ITU Focus Group on Artificial Intelligence for Health, which has been tasked with developing benchmarks for standardization of AI technologies in healthcare. 

We hope to support African governments - and governments around the world - in responding to emerging health crises by utilizing innovative and integrative technology.

Our business model is to license the data-powered insight tools to health institutions, the Ministry of Health, and international non-profit organizations for pandemic response and preparedness. Institutions will use the tools alongside their own healthcare providers to better understand and diagnose patients, as well as use data aggregations to make better decisions about their resources.  

There is also an opportunity to commercialize a patient-facing application with additional symptom assessment, automatic telemedicine, and chronic care management features.

We believe that Solve provides a unique opportunity to connect with mentors and innovators who are developing health solutions aimed at positively impacting the lives of individuals around the world. We hope to learn from others and understand new ways to implement our solutions for high impact. We believe this will help increase impact to the community and ensure that people have access to technology-powered healthcare services.

We would love to partner with any organizations who are interested in developing AI tools for healthcare. This includes organizations such as Welcomme Trust, PATH, and universities focused on this research or those who have access to large datasets that can be used for training. In addition, we would love MIT faculty or Solve members as partners to help develop and commercialize our solution.