Heart rate variability suppression in cancer patients versus non-cancer controls: a systematic review and meta-analysis.

BACKGROUND

Cancer-associated autonomic dysfunction may suppress heart rate variability (HRV), yet no meta-analysis has pooled between-group differences in established HRV indices across cancer types. This review quantified HRV differences versus non-cancer controls, HRV-fatigue associations, and longitudinal HRV trajectories.

METHODS

PubMed, MEDLINE, CINAHL, Embase, and the Cochrane Library were searched through March 2026. Adults with physician-confirmed malignancy and ECG- or PPG-based HRV were eligible. Because time-domain indices scale with recording length, the DerSimonian-Laird random-effects pool was restricted a priori to short-term resting recordings (5-10 min) with age-matched non-cancer controls (Part A); long-term ambulatory recordings (12-30 h) were reported separately, and remaining studies synthesized narratively (Part B) or longitudinally (Part C). Certainty was assessed using GRADE (PROSPERO CRD42026130103).

RESULTS

Twenty-eight studies (3232 participants; 9 cancer categories) were included; five met pooling criteria. Cancer patients showed lower SDNN (MD = -14.05 ms [95% CI -18.20, -9.90]), RMSSD (-16.76 ms [-21.23, -12.28]), and pNN50 (-6.05% [-9.36, -2.74]); all p < 0.001, I2 = 22.1-33.7%. HF power was non-significant (p = 0.118). Leave-one-out analyses confirmed robustness, and within-cancer studies consistently showed inverse HRV-fatigue correlations. All pooled outcomes were Very Low certainty.

CONCLUSIONS

Cancer patients show consistently suppressed short-term time-domain HRV, a clinically plausible signal of autonomic impairment. Because it aggregates mechanistically distinct sources (malignancy, cardiotoxicity, perioperative vagal withdrawal, survivorship remodeling), the pooled estimate cannot attribute suppression to any cause, and Very Low certainty precludes recommending HRV as an established clinical biomarker. Treatment-phase-stratified studies with standardized protocols and non-linear indices are needed.