Adding respiratory-rate features to heart-rate metrics provides a modest amount of incremental information about exercise modality in wearable sport recordings, according to an observational secondary analysis of a public exercise dataset. Researchers evaluated whether active-exercise descriptors derived from respiratory rate could enhance the characterisation of exercise contexts beyond what is captured by cardiac signals alone.
The study analysed data drawn from 126 initial wearable source recordings. Following the application of rule-based signal and annotation criteria, the researchers included 124 recordings gathered from 79 modality-qualified participant identifiers across 10 distinct exercise modalities. In this experimental framework, exercise modality functioned strictly as an analytical context label rather than as a clinically targeted classification endpoint.
The core feature set evaluated in the investigation consisted of four heart-rate descriptors and four respiratory-rate descriptors, each calculated across aligned active-exercise recording segments. To control for potential confounding individual characteristics, all primary models incorporated a baseline covariate block designated as block B, which accounted for participant age, sex, body mass index, and total active-exercise duration.
Model performance was evaluated using a fixed L2-penalised multinomial logistic regression architecture subjected to 50 repeated modality-stratified, identifier-grouped two-fold partitions. Primary statistical inference relied on paired model differences, an identifier-level paired permutation test, and a modality-stratified identifier-level bootstrap procedure applied to out-of-fold model predictions.
Across the repeated grouped partitions, the mean identifier-level macro-F1 score reached 0.379 for the baseline covariate block combined with heart rate, 0.400 for baseline covariates combined with respiratory rate, and 0.475 when baseline covariates were combined with both physiological signals simultaneously. Incorporating respiratory-rate features alongside heart-rate features increased the macro-F1 score by +0.096, yielding a 95% bootstrap percentile interval from +0.061 to +0.136 and a paired permutation significance value of p = 0.0045. Conversely, adding heart-rate descriptors to respiratory-rate descriptors produced a macro-F1 gain of +0.075, with a 95% bootstrap percentile interval spanning +0.031 to +0.124.
The incremental predictive value of respiratory metrics persisted when the heart-rate reference baseline was expanded from four descriptors to eight descriptors, retaining a ridge-regression respiratory-rate increment of +0.108 with a Holm-adjusted significance value of p = 0.0040. However, evidence across alternative modelling frameworks remained limited. Although an alternative random-forest model yielded a positive point estimate of +0.050, its 95% bootstrap percentile interval of +0.020 to +0.085 and an accompanying paired permutation test result of p = 0.093 presented discordant uncertainty evidence.
The authors concluded that while respiratory-rate metrics contribute modest modality-related information beyond matched heart-rate metrics within the primary ridge regression structure, the findings characterise context-sensitive wearable signal variation rather than practical intervention targets. The dataset and analysis do not establish test-retest reproducibility, measurement validity, training-load validity, external generalisability, or clinical utility.
Subject of Research: Incremental exercise modality information from respiratory rate beyond heart rate in wearable sensor recordings
Article Title: Respiratory-rate features add modest modality-related information beyond heart-rate features in heterogeneous wearable sport recordings: a secondary analysis
Article References: Li, H., Yang, Q., & Wang, B. (2026). Respiratory-rate features add modest modality-related information beyond heart-rate features in heterogeneous wearable sport recordings: a secondary analysis. BMC Sports Science, Medicine and Rehabilitation. https://doi.org/10.1186/s13102-026-02163-0
Image Credits: AI Generated
DOI: 10.1186/s13102-026-02163-0
Keywords: respiratory rate, heart rate, wearable sensors, exercise modality, secondary analysis, sports medicine, multinomial logistic regression, L2 regularisation, macro-F1 score, active-exercise duration, body mass index, out-of-fold prediction, paired permutation test, bootstrap percentile interval, random forest, wearable sport recordings
News Source: Ophelia Keating. (November 12, 2026). Respiratory Rate Features Add Modest Exercise Context Beyond Wearable Heart Rate Metrics. Scienmag.



