BACKGROUND: Data on the changes of heart rate variability (HRV) and their prognostic significance during ageing are limited. METHODS: HRV analyzes were done from standardized 45-min ECG recordings, which consisted of 15 min recordings in lying down, sitting positions and during walking. We used time domain-, frequency domain- and non-linear methods to estimate HRV. The baseline ECG recordings were done in 1991-1993 (n = 783) and follow up recordings were done in 2013-2014 (n = 466). Endpoints were reviewed in 2021. RESULTS: During a mean follow-up of 22.1 ± 0.7 years, high-, low- and very-low-frequency powers (HF, LF, VLF), standard deviation of RR intervals (SDNN), the short-term fractal-like scaling exponent analyzed by the detrended fluctuation analysis (DFA1) and approximate entropy (ApEn) decreased statistically significantly (p-values from <
0.05 to <
0.002). Larger decrease of VLF(ln) and LF(ln) predicted total and cardiovascular (CV) mortality in the multivariate model (p-values from <
0.05 to <
0.002). Baseline natural logarithm of LF (LF(ln)) dichotomized and DFA1 had the strongest prognostic value on total and CV-mortality in multivariate analysis after adjustments with relevant clinical characteristics. Also, lower values of baseline ApEn retained their predictive power for total mortality and decreased ratio of LF to HF (LF/HF) for CV-mortality after adjustments. CONCLUSIONS: Almost all the HRV parameters decreased during ageing. Larger decrease of VLF(ln) and LF(ln) predicted total and CV-mortality after adjustments indicating that larger attenuation in cardiac autonomic regulation during ageing yields prognostic information. Of the baseline HRV parameters LF(ln) dichotomized and DFA1 had the strongest prognostic value on total and CV-mortality.