HealthECCO

The vast majority of a rapidly expanding corpus of health knowledge is hidden in data silos -  undiscovered, unconnected and underused. This reduces our ability to respond to health crises because decision makers do not have the information they need. 

HealthECCO is building an open-source platform to combine heterogeneous, unstructured data sources and activate hidden knowledge through connections in the datasets. Our solution can integrate and enrich data from biomedical experiments, publications, patents, clinical trials and electronic health records and more. 

We see our solution positively changing lives by helping:

• Last-mile health workers access to established information sources in combination with case reports and grey literature delivers contextualized insights to improve treatment decisions.

• Policy makers create robust, harmonised health care policies (such as treatment guidelines) faster while accommodating local variations

• Researchers conduct fast, effective research that can speed up the time-to-market of new, life-changing treatments.

The vast majority of a rapidly expanding corpus of health knowledge is hidden in data silos -  undiscovered, unconnected and underused. This reduces our ability to respond to health crises because decision makers do not have the information they need. Data in the healthcare sector is complex and highly connected. The data is available at different places and neither interconnectable with other data sources nor searchable.

This affects various health care stakeholders: Public health policy makers do not have access to the information they need to quickly and efficiently create treatment guidelines. Frontline health workers have to rely on personal experience or hearsay to make the best treatment decision for their patients and health researchers are locked into their domain and cannot translate their knowledge to the emerging health situation. All of these problems are felt more keenly in the developing world where skills and information access may be limited

It also impacts the time-to-market of new treatments for patients worldwide and may slow down the dissemination of accurate guidelines that can help save lives.

We are building a unique solution to combine, annotate and organize the world’s health knowledge and put it into the hands of the right people.

HealthECCO is building an open-source platform to combine heterogeneous, unstructured data sources and activate hidden knowledge through connections in the datasets. Our solution can integrate and enrich data from biomedical experiments, publications, patents, clinical trials and electronic health records and more.  

The solution is based on a knowledge graph that can handle the complexity of the underlying data and stores data and relationships in a network that is both easy to adjust and easy to understand. This helps non-technical stakeholders to work close to the data and facilitates clear communication & understanding of the data. 

We combine unstructured text based data with structured knowledge in a repeatable way. Once the source data is integrated in a central place, we can enrich & annotate it with text mining and natural language processing. The data in the central knowledge graph can be accessed from different, use-case specific end-user applications (either directly or via an API) to extract relevant information needed for a task.

The key advantage of our solution is that we can integrate and annotate the world’s health knowledge to get it into the right hands quickly, efficiently and with the additional benefit of advanced visualisations.

While their daily problems are different, they all need access to similar data. Policy makers need contextualized knowledge to draft guidelines, health workers have to find the right piece of evidence for the specific patient, and researchers need to combine their specialty with the world’s knowledge to quickly respond to health emergencies. We have a growing community of developers, researchers, health workers and policy makers and explore use cases with the community. We are actively establishing collaborations with organisations active across these groups, for example the Clinton Health Access Initiative, private sector organisations and Universities around the world. 

Reviewers in previous applications noted that our solution was not focused. We would like to address that concern directly. The technology and structure of our platform is enormously flexible with a vast number of potential use-cases, many of which are not mutually exclusive. e.g. developing a system that supports Researchers by default supports Policy Makers as many of their basic needs are the same. 

Policy Makers

Around the world there are a number of government and non-governmental organisations responsible for the timely generation of guidelines that need to be accurate and appropriate for a given audience and possibly locality. Such organisations, particularly in the developing world, are often stretched for resources and do not always have the capacity or skills to properly interpret or develop robust, harmonised guidelines.

Our solution is well placed to serve as the basis for an application that will help such organisations to conduct basic literature research and to understand the wider scope providing information access where it may previously have been limited. 

In addition, by exploring the ongoing integration of relevant data sources combined with effective Natural Language Processing, our knowledge graph can present a consolidated view of peer reviewed literature alongside emerging grey literature like donor reports and government policies which can be critical in responding to rapidly evolving situations like the Pandemic. 

Health Workers

For Health Workers, particularly those on the front line, decision making is a vital element of their role and better decisions can equate directly to improved outcomes for their patients. Treatment guidelines may refer to underlying studies but rarely provide enough data for a Health Worker to assess whether the evidence applies to their specific patient or situation. 

We aim to streamline access to this evidence making it easier for Health Workers to quickly understand the underlying bases of treatment guidelines. We will work with Health Workers to build an application that can fit into their workflows without slowing them down. 

This use case naturally complements the Policy Maker use case and there is significant overlap. A tool that can help Health Workers assess the evidence can also help Policy Makers explore the existing landscape. 

Researchers

Data in the healthcare sector is complex and in reality highly connected, but this does not translate into the literature where health information becomes isolated in data silos: undiscovered, unconnected and underused. Combine this with the ever growing mountain of research data and the task of trying to find useful information about any given topic becomes seemingly insurmountable. 

CovidGraph was built to solve these problems by systematically connecting information points not only within the literature, but also to related data sets like biomedical concepts, clinical trials and patents. This process allows our knowledge graph to present a more realistic data model to Researchers. Connections that are inherent and obvious in the real world are created in the knowledge graph. Adding a layer of Natural Language Processing enhances the quality and number of connections we are able to make between all the elements contained in the knowledge graph regardless of their source. 

This means that Researchers are able to quickly find information that relates to, say a gene of interest even if it originates from a completely different speciality or from a database that they would not previously have considered exploring. New avenues to explore are much easier for Researchers to uncover and in the process they can discover other published scientists or institutions working on similar areas. 

In addition the time it takes to uncover such information is orders of magnitude faster using our knowledge graph than it is using traditional methods. 

While the use cases above detail the direct positive impact the HealthECCO ecosystem has on those three groups of stakeholders, the transitive impact should not be forgotten. There is a butterfly effect that magnifies the benefits that HealthECCO has on the wider population. This is especially true in the case of Researchers and Policy Makers whose work can affect the lives of millions.

The user groups outlined above need access to a broad range of up-to-date information to create a meaningful response to health crises: Researchers aim to understand mechanisms and treatments, health workers translate those insights into patient care and policy makers create the operational framework to harmonise and disseminate guidelines.

All groups trade with the same currency: Knowledge about a disease, the mechanisms of treatment and the efficacy of measures taken. 

Our solution enables faster, more efficient access to knowledge to speed up the research process and it increases the quality of evidence by delivering better data with more context.

HealthECCO is the evolution of the CovidGraph project that started in 2020. We have had contributions from more than 30 developers worldwide and built a knowledge graph focused on Covid-19. We have developed two end-user applications that are actively used by researchers. CovidGraph is the prototype for the more generic solution that we want to build.

We have built an active community with more than 100 members, hold regular webinars and presentations, and monthly community calls. Our public outreach has generated positive feedback from researchers, data scientists and health workers. We consider the growth of our community and the number of press mentions plus detailed feedback calls to be indicators of success.

The team that is now setting up HealthECCO has a cross-sector background with members from academic research, IT firms and health companies. We are uniquely positioned to scale up development of our solution with appropriate funding.

Combining cutting edge technologies like Neo4j, Docker, NLP and emerging front end tools allows us to build a flexible single source of truth that is easy to develop, explore and adapt to fast changing needs.

Traditional data analysis pipelines operate on a static input and produce a result in a serial fashion. We store both the input data as well as the results of our data enrichment in the same, integrated knowledge graph. The data storage allows us to combine cutting edge analyses (NLP, text mining, deep learning) in a way that was not possible before. Data integration, enrichment, analysis and visualization are not separate steps anymore but contribute to an organically growing knowledge graph which enables policy makers, health workers and researchers to keep up with the fast growth of data volume and complexity, a huge shortcoming in existing data integration projects.

We apply graph based deep learning methods to predict new relationships in the data. We are able to add context to the results and our API allows for rapid development of new use-cases. This is a huge step to overcome one of the major shortcomings of deep learning methods: Lack of interpretability and trust in the results.

Our approach has attracted interest from Trinity Challenge members who are interested in exploring how it might be applied in their particular problem areas so we see our solution encouraging the evolution of open healthcare solutions using graph technologies.

We understand that a direct, postive impact on patients' lives is the ultimate goal. We aim to do this by making workflows more efficient for researchers, health workers and policy-makers which in turn improves the lives of patients. A tool that helps government agencies create better treatment guidelines has the potential to impact the entire population. We currently have less than 20 active users but have a strategy for growth.

Plan for one year:

Together with one of our collaborating partners, the Heidelberg Institute of Global Health (HIGH) we will test the feasibility of our approach in the low-resource context of Burkina Faso. We will investigate the outputs of the graph database focusing on policy-makers and health workers. HIGH has been a longstanding partner with the Centre de Recherche en Santé in Nouna, Burkina Faso, and has a strong expertise in conducting health research in Burkina Faso. This targets a population of 21 million people.

Together with the Clinton Health Access Initiative, which operates in over 35 countries around the world, we plan to expand with similar pilot projects. 

Plan for five years:

After establishing pilot projects with our partners, we aim at growing the impact of our solution by opening it up to low income countries worldwide. We hope to (indirectly) serve a population of at least 1 billion people.

We have further detailed the model of our impact in a blog post: https://healthecco.org/healthe...

Our approach is to focus primarily on the three use-cases where we can see HealthECCO having the most direct impact, especially in disadvantaged communities and the developing world. The key performance indicators are users within these groups and the size of the patient population that each user affects. The number of main users can be measured directly with user analytics. The number of patients reached by the individual users will be estimated in close collaboration with our users and with our partners (e.g. CHAI). Also, the value created by the knowledge we deliver will be assessed in close collaboration with the users. Here, important KPIs are the time required to perform a patient-relevant activity and the quality of the output of that activity. 

To promote global awareness of our project we plan to run a number of community and public outreach campaigns including surveys and outreach campaigns in traditional media, print, and social media. 

KPIs for our knowledge graph are the number of data sources, the ratio of connected to unconnected data points, the number of evidence extracted with NLP and the quality of predicted relationships. The frequency of updates and respective changes of KPIs are indicative of our ability to keep up with data growth.

HealthECCO core team:

• 7 part-time members

Community leads

• 4 part-time members 

Scientific Collaboration

• 3 active collaborations 

The HealthECCO core team will manage the project. Depending on funding we can draw resources from our community and network of developers. Our aim is to grow our team as our solution gains momentum.

The team behind CovidGraph (and now HealthECCO) has a cross-sector background with domain experts, IT specialists and leaders in the graph community. We understand the needs of our users and possess the capabilities to deliver robust and scalable solutions. The organisations behind HealthECCO as well as our collaboration partners (both through Trinity Challenge and our existing network) are committed to supporting us throughout the project. All elements of our technology pipeline are supported by either a member of HealthECCO or a collaboration partner (e.g. Dr. Evidence, Neo4j and yWorks). Use-case oriented partners such as CHAI and the academic partners enable us to reach out to our users and hence understand their needs. In addition HealthECCO’s engagement strategy will seek to attract new technical and user-oriented partners.

A key capability of our team is managing international, multi-stakeholder, data-centric IT projects. We have long standing experience both in development of the conceptual background as well as implementation of end-user applications covering development, deployment, maintenance and security. With appropriate funding we are able to scale our solution globally.

Right from the start our founders reached out across disciplines and functions to form the core of what is now HealthECCO. The knowledge that maximum benefit derives from including as many different people as possible to start such a project stays at the heart of what we do. Consequently, our existing community network spans researchers at universities and medical institutions to PhD students to IT experts and software developers across Europe and America, but we are reaching out to a wider community across the globe, especially in Africa via our collaboration partners. 

Our community is built on the core values of doing good things, making good things and being good to one another. We are an inclusive and welcoming community that fosters learning and support. We do not accept or tolerate discrimination in any form and are currently drafting a code of conduct.

We are looking to expand our team – from backend development to medical research – and we welcome everybody to join our existing team.

We have an Matrix Chat open to anybody that wants to get in touch with us and is looking to participate in any way. Weekly meetings are by video call with the core team so that no matter where people are based they are able to participate. We conduct monthly community sessions where we share the progress of the project and encourage open exchange and discourse.

All of this is reflected in the knowledge graph (it’s the relationships that matter) that is HealthECCO.

The cross-sector nature of HealthECCO makes it difficult to acquire funding through grants aimed at academic research or innovation in the private sector, respectively. Basic technology development often does not fit into the project based funding logic of research grants. Hence, the MIT SOLVE challenge is a unique opportunity for our organization to professionalize and scale software development and dissemination.

Our solution started life as a direct response to the current pandemic and is well suited to helping improve future responses so it fits well with the MIT Solve objectives. 

The impact of our solution depends on users who can multiply the effect to patients. We hope to create new collaborations through the MIT SOLVE network. 

Being involved with the MIT SOLVE challenge increases awareness of our project, adds to our credibility and paves the way for future collaborations regardless of the outcome of our application.

Another barrier for our solution is global impact in multilingual settings. With the reach of the MIT SOLVE network we can acquire expertise in managing global projects and implement technologies for multilingual NLP. 

Public relations

Our solution is used by organisations to deliver the impact to patients. We have already acquired partners (CHAI and HIGH) that support us building our brand and increasing turnover in our communication channels with our target organisations. With support from MIT SOLVE we hope to further improve our capabilities to target health policy makers globally. 

Product & Service Distribution

Solution development that works for both high and low resource settings poses many challenges. We hope to get support in building robust technologies that work well with slow internet connections and limited computing resources as well as adapting for location and language. 

Technology

Our team has a strong background in technology development. There are still key areas where we seek support and partners, e.g. improving our natural language processing pipeline to adapt to more languages. Aside from anything else, access to additional development resources will help us to improve our platform and expand the applications we can offer end users in a much shorter time frame.

There are several organizations among the MIT Solve Members that we would like to partner with. The Bill & Melinda Gates Foundation could provide additional resources or funding and help us get our solution into the hands of public health organizations and government authorities worldwide.

The Novartis Foundation and Pfizer could act as mediators to bring our solution to drug discovery and medical research.

We will partner with the Clinton Health Access Initiative to reach policy makers, particularly in the developing world, and scale our impact globally. The Heidelberg Institute for Global Health will act as a partner for pilot projects and has a long standing engagement in western Africa.

Academic collaborations: We have already initiated collaborations with renowned academic researchers, including Prof Karl Morten (Oxford University), Dr Shamith Samarajiwa (Cambridge University), Prof Dagmar Waltemath (Greifswald University) and Dr Davide Mottin (Aarhus University). We are also in discussions with a research institute in Spain affiliated with the University of Valencia. We aim at a) setting up a scientific advisory board, b) build partnerships for graph-based machine learning and c) co-apply for research grants.