Survival Analysis Using Time-Frequency Analysis of Heart Rate Variability During Exercise

Sammanfattning: Heart rate variability as an indicator of increased morbidity has been established in previous studies. It is defined by different frequency bands, corresponding to different biological mechanisms. This thesis aims to study heart rate variability indices in the time-frequency domain through extraction from the UK Biobank dataset. Time-frequency indices for different time intervals during exercise were extracted. A time-frequency based filter analyzing the presence of high frequency noise was developed as a way of detecting and removing noisy sections of the ECG recordings. This noise reduction algorithm was coupled with time-domain noise reduction methods as a part of pre-processing the data before survival analysis. Intra-individual repeatability of indices was calculated and finally a combination of a random survival forest and Cox proportional hazards regression was used to evaluate said indices' value as indicators of cardiovascular risk. The noise reduction performed by the combination of time and time-frequency domain noise rejection algorithms did not improve intra-patient repeatability. Selected time-frequency indices show high levels of intra-individual repeatability over a three year period. Out of the selected indices, one remained a significant predictor of cardiovascular event when combined with demographic data. The use of time-frequency based noise filtering shows promise for detection of high frequency noise artifacts, and should be further studied and tested on other kinds of recordings. The significant relationship between one time-frequency index of heart rate variability and increased risk of cardiovascular events needs to be further investigated.