Myocardial damage during mountain bike race - an analysis of data from Nordsjørittet 2014 (NEEDED study)

Abstract

During Nordsjørittet 2014, blood samples, blood pressure and ECG were systematically gathered from over 1000 contestants. With this gathering of data, there is an opportunity to study the effects an endurance bike race has on presumably healthy individuals. In addition to this data, there are complete data from sports watches from a third of these contestants. When the medical analysis was done, 25 contestants were diagnosed with myocardial damage. Eight of these were from those with sports watches.

The motivation for this thesis is to test if the data logged with the sports watches can be used to predict myocardial damage. Or whether this data can be used to separate the contestants with high or low troponin I values.

To achieve this, a classification system was designed to test if there are patterns in the sports watch data that can be connected to myocardial damage or troponin I values.

As a consequence of variable sampling, the sports watch data were interpolated and smoothed. After this pre-processing, several features were extracted from the sports watches, from a predefined segment in the race. Two different methods, exhaused search and sequential forward selection, were used to select the feature subset maximizing the prediction rate. Furthermore, a Naive Bayes classifier was trained and validated using leave-one-out cross-validation. To measure reliability of the results in the experiments, Matthews correlation coefficient were used.

With an achieved prediction rate of 86% and a correlation coefficient of 0.58, the results clearly indicates that the data logged in the sports watches can predict myocardial damage.

When testing if the sports watch data can separate the contestants with high or low Tropinin I, the results were inconclusive. Therefore, it is uncertain if this is achievable with the current data. Fjern valgte