Linking cardiovascular and musical change points for personalised music-based cardiovascular interventions

Music-based interventions provide an under-explored non-pharmacological support for existing cardiovascular therapies with high safety profiles. However, the potential for targeted musical interventions in cardiology is limited by a lack of a precise approach to capture cardiovascular responsiveness to music. Existing approaches predominantly treat entire music pieces as uniform stimuli, failing to account for expressive shifts in music that elicit cardiovascular changes.

Our goal is to use canonical correlation analysis to link expressive change points to physiological changes, serving as biomarkers for the body's reactions to music, which can inform the design of tailored music interventions.

Musical and physiological data were collected from 112 listeners (63 females, 43.9 ± 16.0 years) during a 5-minute baseline and while listening to a playlist of 9 pieces (approx. 40 min). A total of 755 musical change points were identified, automatically from audio/MIDI recordings and through manual expert annotations. Using canonical correlation analysis (CCA) at the individual and group level, musical change points were linked to automatically detected physiological change points in time series extracted from ECG (RR intervals and spectral HRV parameters), continuous BP signals, and respiratory intervals. The model’s canonical loadings were then used to investigate distinctive patterns in different groups of subjects, segregated by baseline sympathovagal balance, blood pressure (BP), and gender. Moreover, we used generalised additive models (GAMs) to analyse how baseline and inter-subject variables are linked with observed patterns.

For the general population, the first two canonical variate pairs showed significant canonical correlations (0.56 and 0.36) compared to surrogate data. Considering the strongest associations (canonical loadings>|+/-0.5|), the results suggest a decrease of the resp. intervals in response to novel melodies and to increases in tempo and loudness. At the same time, we observe a negative loading value for HRV power spectra in the high frequency band (PHF),suggesting parasympathetic (PNS) withdrawal due to an increase in music intensity and/or the occurrence of novelty. The GAMs suggest that baseline PNS index (p=0.04) and female gender (p=0.028) are negatively correlated with general patterns. Subgroup analysis revealed that distinct sympathovagal balance led to opposite RR and PHF reactions to change points in music.

Our research lays the groundwork to use cardiovascular-music linkage patterns as biomarkers for creating targeted micro music-based interventions for personalised or narrow-group level adjustments of cardiovascular variables.

Data Preparation