The Problem the SocialCovid Team is Looking to Solve
A lack of a way of assessing the risk that patients with no symptoms or only mild symptoms may have been infected by their previous journey history starting with the medical sites visited within the national health system.
What Inspires the SocialCovid Team Doctors and medical professionals throughout the world are the selfless front line heroes in the big fight against novel Covid-19 and tragically so many have died themselves. Everyone should be keen to try to help stop the spread of this virus by following guidance whether mandated or common sense and if they can contribute to finding solutions to the multiple detrimental impacts to all areas of our lives. Doctors certainly in the West maintain ethical principles primarily the principle of ‘First Do No Harm’ and Universal precautions already in place in hospitals but these are sometimes hard to enforce for patients and our motivation is to develop an aid to help medical professionals protect themselves ‘come to no harm’ by arming them with decision aiding app. We learn of tracking methods that countries in the Far East are using and initiatives worldwide including in the NHS in England. We intend to develop a system that will provide doctors with an estimated probability for a patient with mild or no symptoms to have been infected with Covid-19 over the past couple of weeks. This awareness will also hopefully enhance compliance with universal precautions for infection control. A big Thank You to all the SocialCovid team members for all their work and our helpful mentors - Dimitar Jetchev, Lukas Mezger, Vishnu Ravi, Ileana Mardare, Arshad Emmambux and Javier Virto for all their advice in this EUvsVirus Hackathon.
What the SocialCovid Solution is to This Problem The App is founded on a ‘privacy by design’ approach to patient contact tracing with personal data never leaving secure data repositories as the responsibility for the data lies with each individual national health system. The data combines patient health data with their medical appointment location journey and it all remains within the data walls of the national health service provider such as the NHS in the UK. In highly enlightening discussions with worldwide mentors in medical, technology, academic and innovation sectors discussions during this Hackathon we have refined the product model as we learnt of many different issues including the various national hospital structures and lack of data availability. We were privileged to get to ask and learn what doctors would find helpful to them and their colleagues and can be sustained. Once this is developed Our App will be fueled by continuous real-time data sourced and managed to an agreed common format (also incorporating detail provided by patient questionnaires). Our App will be powered by a computational engine relying on a combination of social network analysis and machine learning methods and will give medical professionals access to a unified interactive view of patient proximity data and model predictions at the level of individual national health systems. This can in principle be implemented separately in different European countries.
What the SocialCovid App Does The App calculates patient risk factors for doctors where the patient presents with asymptomatic or mild symptoms. The basic algorithm design is health service patient data and structured proximity data for their journeys within national health service locations This will be complemented by patient journey history data collected using questionnaires. The algorithm predicts the probability of them being infected.
The Technical Details and Tools We Used We focused on designing the information interface and the associated computational engine, and on investigating relevant regulatory issues at the EU level, specifically focusing on the requirements around data protection GDPR for this project.
From a Project / Business Development perspective we also explored issues around the possible future deployment in other European health services with a view to building robust and sustainable exploitation pathways. Social graphs can augment / complement this approach.
What We Have Done During the Weekend The Hackathon experience of the SocialCovid team has been a great learning experience and a bringing together of team members of diverse interdisciplinary backgrounds. We have all learnt a lot from each other. It has been extremely interesting and motivational (and fun too) in feeling part of a giant initiative in the EU Hackathon to help find sustainable solutions to this pandemic. We have established the feasibility of the project and the variety of expertise of the mentors - including several clinicians, has guided us in refining the project focus to aid doctors to help prevent them catching Covid-19. We have developed several models showing predictions and also location analysis and are very keen to apply fresh data to these. Our initial models have used the South Korea datasets which contain more individual patient data including their length of stay in hospital than is available in the UK presently. We have lots of other ideas of models to devise and parameters to incorporate and use to extract as much knowledge from the data.If we had had more time we could have devised dummy data for missing parameters and applied these as proxies to other models. From a regulatory perspective we now understand from the mentor (thank you Lukas) about GDPR issues we need to be aware of and ensure we can meet and are pleased that the project can confirm to these regulations and what differences we need to be aware of for future European roll-out. From talking to Clinician mentors we have learnt what the different issues doctors have faced particularly in this crisis in a few other countries. We have learnt that it appears that medical protocols regarding staff wearing appropriate protection has not always been possible to be followed due to lack of protective equipment available to them. Also strict following of Universal medical protocols when dealing with Covid-19 patients such as not using gloves may also have led to increased levels of infection amongst medical personnel. We have learnt what doctors in the health service would find helpful to them and they would like to have to use so that we could tailor the App to to make it more helpful.
How We Built Our Model
The Methodology used is machine learning augmented by social network analysis and checking for Ethics and regulatory issues.
We considered what data would be needed as predictors of morbidity (from Covid19) to provide the risk factor. These include country hospital sites, age, gender,number of days in hospital. As well as geolocality data such as number of schools (kindergarten through to senior) and universities and numbers of seniors homes and seniors as a proportion of the population at different geolocations. We chose the South Korea dataset augmented by other data for our analysis.
The Databiology South Korea dataset was selected as looking to have a selection of data to mine. The data extracted was augmented with Kaggle 'patient routes' data and the resulting dataset was cleaned and relevant features for analysis identified. The data models developed were built using Python/ Jupyter notebooks The models defined were Basic network analysis and graph analysis. A Random Forest model showed a good level of accuracy of prediction at 88% with key predictors identified including age, elderly living alone and elderly percentage in the population.
Some Code is available at: Model Notebook -https://wu9553.lab.dat.bio/notebooks/sk_data_proc.ipynb Model Source code: https://wu9553.lab.dat.bio/tree/source Data: https://wu9553.lab.dat.bio/tree/External%20Data Regression model: https://www.lab.databiology.net/dbe/userlab/show-workunit.html?workunitId=9769&tab=details Graph model: Basic Route Graph - Workunit 9702; Patient and Proximity Graph
The Challenges the Social Covid Team Encountered We found the availability of usable data to run our models for European countries was very limited and did not allow us to produce this in the time available. Also with more time we could construct dummy data for other countries such as England to run simulated models..
The Accomplishments That SocialCovid are Proud of The Hackathon experience of the SocialCovid team shows the results of the project are technically feasible and it can be made compliant for GDPR. We worked very well as a team aided by multi-disciplinary skills and a combined passion for finding a solution. Thus although we haven’t quite got to a prototype yet we have created a POC to show what is possible and present our findings from the modelling done of the available data.
What the SocialCovid Team Learnt We learnt what doctors really would find helpful. We want to explore development of the App to find a way to promote strong adherence to high standards of observance of these protocols, perhaps through linking to medical e-training and working with hospital departments such as those that monitor the disease prevalence.
The SocialCovid Solution’s Impact to the Covid-19 Crisis This will provide a metric to doctors to enable them to keep well and hopefully ensure they adhere to Universal protocols and use all necessary protective equipment to help keep themselves safe. We also think that when lockdown restriction will be lifted, for some time afterwards patients will present asymptomatically or with mild symptoms and the App can be particularly useful then. Adaptability although CV related can be used for any infectious disease so a long term advantage to be invested in.. And also for environmental exposure such as asbestos.
What's Next for SocialCovid The necessities in order to continue our project. Firstly to propose and get support for a Student project to develop a prototype App. Then with early success in the prototype to seek and gain support for the results and then proceed to a pilot study of the suitability and performance of the App to one NHS trust in England. This will incorporate additional research and user experience and requirement surveys. Next steps could include exploring the application of this App to the aid of other infectious diseases and also environmental health issues such as asbestosis or any other conditions requiring proximity for spreading. The SocialCovid team members are keen to set up a consortium to create a supportive environment to aid this research.
Further developments could contribute to the development of the App to use in other European countries where the patient data is kept within each country - the principle of ‘Privacy by Design - fuelled by each country’s health service server's provided patient data.