Project / product name: CKD-xgBoost
Link to the project: https://github.com/KajetanPoliak/hackhealth
Team leader: Kajetan Poliak
Challenge: 6. Are you kidneying?
Problem: Chronical Kidney Disease (CKD) in late stages has huge human and financial cost, including regular dialysis, costly medications and in late stages transplantation (in better case) or death. Therefore any early detection can prevent these costs and enable us to improve life-being of many patients and redirect freed funds to more pressing issues in medical environment or upscale the number of treated patients.
Solution: We have used IKEM data to build a XGBoost model which enables efficient patient targeting using only baseline information and most prevalent test excluding GFR. When following the model recommendations, targeting 20% of the top scoring patients and testing them will identify 99% of CKD cases. In other words model was able to to identify small subgroup of patients which contained 99% of CKD cases in the whole dataset.
Impact: Using the model prediction we can early detect potential patients for CKD treatments using only baseline information like age and BMI and recommend them for more extensive test to prevent further development of disease.
Feasibility & financials: The model would require further development, domain experts for further improvements with expected cost around 10 000 euros.
What is new about your solution?: Compared to existing research we have targeted predictors with minimal costs and also thanks to IKEM the model was trained on one of biggest datasets available for this disease.
What you have built at the hackathon - text explanation + code (e.g. GitHub link): https://github.com/KajetanPoliak/hackhealth description in readmi
What you had before the hackathon, please mention open source as well: nothing
What comes next and what you wish to achieve: We want to tackle problem strategicaly, not only in data realm but also from business perspective.