All Models Are Wrong

Problem: We are challenged by the fact that CKD is a very serious disease, which unfortunately often goes unidentified during the early stages (when medical help is still possible). If left untreated, the patient is referred to a dialysis machine for the rest of her/his life. This could be partially solved by coming up with novelty screening methods with focus on early identification of patients with a higher risk of the disease.

Solution: Mixing the knowledge of our medical expert with statistics and data science, we have been able to devise a Cox model for time varying variables, which is able to grasp the problem at hand in a state-of-the-art way. It provides us with insights on which variables are key for the screening and quantification of proportional risk increase along with confidence intervals. Further, we have developed a dashboard for visualisation of the results and integrated FHIR (which ensures wide use of the tool.

Impact: Based on data from standard blood tests (from a file or FHIR) we are able to determine the proportional increase in patient’s risk of suffering from CKD. This enables us to effortlessly identify patients, which we deem high risk and report them using our dashboard to a medical professional. Owing to that fact, we are able to change many a person’s life for the better. Thanks to the nature of the solution, it is also verifiable by medical professionals.

Feasibility & financials: As of now, the product is in its MVP stage and requires further development (improvement of the statistical model, provision of further data streams, production-level interconnecting of its parts, which are written in three different programming languages…). Aside from that, the future use of the solution itself comes at nearly no costs for anybody, which makes it widely applicable.

What is new about your solution?: Similar problems are usually solved by the means of Machine/Deep Learning, neither of which enable real understanding of the problem at hand. We have decided to build a model using statistics – one that provides sensible results and is simple and scalable in practice. The dashboard prototype provides a neat and tidy interface for medical professionals to identify patients with a higher risk of CKD.

What you have built at the hackathon - text explanation + code (e.g. GitHub link): We have developed a statistical model that is naturally able to identify patients, who possess higher risk of suffering from CKD. For such a model to be useful in practice, visualisation tools are key. Therefore, we have built a dashboard interface that enables the medical professional to quickly get an overview of patients with higher risk of the disease. Furthermore, we have integrated FHIR into the solution. https://github.com/vhotmar/cee_ehh_2022

What you had before the hackathon, please mention open source as well: Know-how on medicine, statistics and programming, everything (including the idea) was built during the hackathon.

What comes next and what you wish to achieve: As mentioned before, the product is currently in its MVP stage and therefore there is still a lot of work to be done. If we are able to push beyond the walls of the hackathon, we would like to work on our idea further as mentioned in the Feasibility & financials section. Further integration with hospital services and education of medical professionals regarding the use of the tool would also be useful to a great extent.

Video: https://www.youtube.com/embed/QZhHS3LdQGI