A regression-based machine learning approach for the prediction of lung function decline

Pulmonary fibrosis is a progressive disease of the lungs which usually gets worse over time. Once this disease damages the lungs, it cannot be cured totally, but early detection and proper diagnosis can help to keep the disease in control. The Kaggle competition entitled “OSIC Pulmonary Fibrosis Progression Predict lung function decline” ran from July to September 2020 with the goal of early detection of the disease. Our approach achieved a Laplace Log Likelihood score of −6.8590 which was within the bronze medal band. The Kaggle dataset contained CT scans and anonymized demographic and clinical data from multiple patient visits, such as spirometry forced vital capacity (FVC), for 176 unique patients. In our method we predict FVC and a confidence measure using a sigmoid equation. This equation is extracted via a novel transformation using only three of the given parameters. In this way we created a simple but accurate model for the prediction of lung function decline.