Plasma-transfusion has shown to be a promising therapy to treat severe cases of COVID19. Preliminary studies from USA, China and Europe have delivered promising results, as the treated patients were alleviated from their symptoms within 12 days after the transfusion.  The therapy requires a donor with antibodies against COVID19 to give plasma to a patient in a severe stage of the illness. To enable large-scale clinical trials quickly and effectively, suitable donors have to be identified and connected to medical institutions performing the therapy.  We create a bridge between medical research institutes and COVID19 convalescent patients, in view of clinical trials for plasma transfusion therapy.


  • Use of convalescent plasma therapy in SARS patients in Hong Kong, Cheng Y et al., Eur J Clin Microbiol Infect Disch 2005 Jan;24(1):44-6.
  • The use of convalescent plasma to treat emerging infectious diseases: focus on Ebola virus disease. Winkler AM, Koepsell SA., Curr Opin Hematol. 2015 Nov;22(6):521-6
  • Use of convalescent whole blood or plasma collected from patients recovered from Ebola virus disease,WHO, 2014Sep
  • The effectiveness of convalescent plasma and hyperimmune immunoglobulin for the treatment of severe acute respiratory infections of viral etiology: a systematic review and exploratory meta-analysis. Mair-Jenkins J et al., J Infect Dis. 2015 Jan
  • Treatment of 5 Critically Ill Patients With COVID-19 With Convalescent Plasma. Chenguang Shen et al., JAMA. March 27, 2020


What it does

We started building a web platform that will facilitate the deployment of large scale clinical trials of plasma-transfusion therapy in Switzerland. It is aimed at two groups: the general public and research institutes. We wish to identify both patients who were diagnosed with COVID19 and those who experienced symptoms but didn’t receive diagnosis.

This platform features a quiz that questions users who do not hold a medical diagnosis about the symptoms they experienced. Their answers is the input for a machine learning algorithm that outputs a percentage chance that the user had COVID19. We are training the algorithm based on diagnosis labeling on a range of profiles by medical experts.

All users are then sensitized about the importance of plasma donation and the huge potential it has to save lives during this pandemic. They are encouraged to fill a survey that checks whether they are eligible to donate plasma. We create an extended database of potential donors with an estimation of the probability they feature COVID19 antibodies.

How I built it

Medical content (quizz, donor form & information)

We have several medical and biology experts on the team. They built three pieces of content:

  • COVID19 assessment quiz: It is a self-assessment quiz that evaluates the likelihood of having anti-SARS-Cov-2 antibodies based on symptoms and exposure of a person to the virus. This will allow the scientific community who want to use our platform for clinical trials to be able to call in individuals who were likely to have had COVID-19 but were not tested. They can then do a serological test to these individuals to confirm or infirm the presence of antibodies. This allows to widen the database, increasing the chance of finding as many donors as needed, keeping in mind that if the therapy is validated in the future, a very large number of donors could be required.
  • Plasma donation form: This form assesses whether someone is an eligible plasma donor based on his/her overall medical status. Data collected includes age, weight, health state (do they suffer from pathologies that will exclude them as donors) etc. If a person was likely to have had the virus and meets the requirement to become a possible plasma donor for patients with COVID-19, they can sign up for becoming a donor directly on the form.  We are aware of the necessity of data protection. The data collected for now are simulated cases only. To collect real data, we would first need to go through the ethics committee, and it would be kept safe in a DataCenter approved for this purpose.
  • Plasma transfusion info page: This page of our website gives short explanations on what convalescent plasma transfusion is, how plasma donation is done, and why we believe it is a very promising therapy. The website contains a link to a pdf document  also written by us, which contains evidence-based data, that explains these points more in detail for anyone who wishes to get more information.


The platform was built on Wix, a web platform providing cloud-based web development services. We didn’t have much web-building experience on the team so this was a quick and easy way of putting a website together. The website features:

  • home page with a quiz which outputs a percentage chance that the user had COVID19.  Information on plasma transfusion, incentive for plasma donation and a contact us form can also be found on the same page.
  • literature review giving information about plasma transfusion and its effectiveness.
  • Members area, secured with credentials. This area is aimed at research institutes, they can browse our data center of potential plasma donors. All aspects of the website were built using features from wix, supplement by embedded third party websites: Google drive/forms/maps, bootstrap and flask for data acquisition.

Data acquisition and storage

The quizz part of the website was a custom Flask app embedding a machine learning algorithm that was then hosted on Google Cloud. The training data for the algorithm was acquired directly from healthcare professionals collaborating with the project.

Machine learning algorithm

To assess the suitability of a potential donor in the context of their symptoms and exposure to the virus, a Machine Learning algorithm is used. The algorithm is trained based on randomly generated answers to the quiz mentioned above, which have been rated on a score of 1-4 by healthcare professionals (1 = very unlikely to have had covid, 4 = very likely). The advantage of this approach over a rule-based logic is that it avoids bias due to the subjective opinion of a single expert, and instead allows to generalize the judgement of different professionals. We use a Random Forest algorithm, which takes as input the answers to the 10 questions of the quiz, which are converted into 16 features, and outputs a label between 1-4. The suitability score is a function of the probability of a tested sample to be in the predicted class - if it lies below a threshold (the median of the class probability in the training set), then the predicted class is obtained as the average of the two most likely classes. The algorithm has a test accuracy of 70-75%. The trained Random Forest model is stored using a cloud service, and the predictions are performed in real time.

Challenges I ran into

  • Running our code on a server (Deployment issues on Google Cloud) we had no experience deploying projects on professional solutions like Google Cloud. This was solved after a lot of trial and error and I learned a lot about the handling of such systems.
  • One of the main challenges lies in data protection.  Collection and storage of health-related data and biological samples is governed by the “Loi relative à la recherche sur l'être humain” in Switzerland. We were lucky to have in our team a research associate from CHUV as well as a data protection IT expert who were very active in dealing with this issue. Since our data cannot be totally anonymous, but only coded, since we need the possibility to contact the donors, we would have to go through the Commission cantonale d'éthique de la recherche sur l'être humain, th swiss ethics committee and get their approval to use this data. This can be done without too much trouble, but would take at least a month to get an answer, therefore the data that we have for now cannot be real patient data but only simulated data.
  • Another challenge was getting enough training data for an accurate machine learning model. Indeed, we are in contact with medical professionals, but the short deadline of the project allowed us to get labelling from only few of them. Our quiz could get more accurate with more training data that we could collect with more time. Furthermore, future studies will likely yield datasets that relate symptoms and exposure to positive or negative test results. As such data does not exist yet in a sufficient quality, we have decided to use medical expert advice instead.

Accomplishments that I'm proud of

  • Teamwork and cooperation
  • Coming out of our comfort zone and going through a steep learning curve
  • Effective division of labor (ML people, IT architecture, Industry contacts, Medical people  and so on)
  • Implementing a functional website, machine learning model and subscription form in a very short time.
  • Professional platform for the quizz (Google Cloud)

What I learned

  • Building web platforms, the difference between different languages, how user-friendly they are vs. what they are able to do
  • Protecting sensitive data is challenging
  • Deployment of applications on professional cloud
  • Important medical facts about COVID-19 and the applications of plasma transfusion therapy.

What's next for Plasma-transfusion donation data base

  • Ensure legal aspects of data protection by submitting a request to the ethics committee, and giving maximal protection to the database. Once this is done, we can start sharing the platform and adding donors.
  • Approaching University Hospitals and clinics who are starting plasma transfusion therapy clinical trials to evaluate the need and fine tune the project. .
  • Widen the database by sharing it on social websites. We need to get as many potential donors as possible.

Try It out



data-science, http, machine-learning, skill

Devpost Software Identifier