As part of the master's thesis, we have established a methodology for analyzing population rhythmicity. We have implemented a set of functions that can be directly used for the analysis of rhythmic longitudinal data. Analyzing such data requires combining rhythmic methods with methods for longitudinal data analysis. The implementation enables the use of three rhythmic methods, namely COSOPT, cosinor, and ARSER. The proposed implementation allows for the use of multi-component methods cosinor and ARSER. For the analysis of longitudinal data, we have implemented three different methods, namely averaging individual models, GEE models, and mixed-effects models. The implementation allows users to test various combinations of rhythmic and longitudinal methods. We tested the implemented methodology on generated synthetic data. The most robust combination proved to be the use of the cosinor method and GEE models. We demonstrated the use of GEE models and mixed-effects models on real data obtained from smartwatch measurements. We tested the combination of the cosinor rhythmic method and GEE model, which proved to be the most robust in the analysis of synthetic data, on concrete experimental data of patients with open sleep apnea.
|
