Introduction

Functional capacity is an important objective health metric for older adults1,2. Its preservation is also of high propriety from older adults’ perspective3. Unfortunately, functional capacity becomes more susceptible to decline as an individual ages1,2. Evidence shows that a declined functional capacity in physical domains, by itself, is associated with a higher risk of multiple adverse health events and high care needs4,5,6,7. As the population of older adults grows, preserving functional capacity as an individual ages is becoming a major goal of care and represents an unmet healthcare need.

Despite this, detailed mechanisms contributing to the decline of functional capacity in late life remain incompletely understood. As an aggregate health measure that typically declines with age, an increasing amount of evidence suggests that the functional capacity relies on the maintenance of physiological homeostasis by multiple organ systems, including cardiovascular, metabolic, and immune systems, in the face of stress and its decline as an individual age may result from a failure of this hemostasis maintenance2,5,8,9,10,11,12,13,14,15.

The autonomic nervous system plays unique and pivotal roles in maintaining physiological homeostasis16,17,18,19,20,21,22,23,24. These roles are mainly exerted through their effects on the function of multiple organ systems. Aside from its well-known effects on cardiovascular and metabolic systems16,17, recent experimental research even showed the previously unexpected close connections of autonomic nervous system activity with inflammation, immune responses, and skeletal muscle physiology18,19. Consistent with its critical role in maintaining homeostasis, in addition to neurological and psychiatric disorders25,26,27, aberrant autonomic nervous system function has increasingly been identified in clinical and even the earlier subclinical states of multiple diseases, including cardiovascular, metabolic, and inflammatory diseases20,21,22,23,24,28,29. Emerging evidence indicates that autonomic nervous system function may alter during aging, for instance, in response to cellular and molecular-level stressors30,31,32. These alterations are postulated to contribute to the development of geriatric syndromes and, consequently, the accelerated health decline in late life though controversies still exist25,32,33,34,35.

Here, we conducted a longitudinal study with repeated measurements of heart rate variability (HRV), a measure of autonomic nervous system function, and functional capacity. We aimed to examine the longitudinal association of heart rate variability and its change with changes in functional capacity over time in older adults. We hypothesized an association of autonomic nervous system function and its change with functional capacity decline.

Methods

Study design and participants

Study participants were from the Healthy Aging Longitudinal Study in Taiwan (HALST), a prospective population-based cohort study that began in 2009 with the third wave follow-up examination still underway36. From 2021 to 2024, HALST participants aged 55 years and older were screened for eligibility to enter a study that examined the applicability, diagnostic, and prognostic values of digital health technologies in community-dwelling older adults (Supplementary Methods S1 in Supplementary Materials). After being recruited, each participant underwent a structured baseline examination assessing sociodemographic status, diseases, health status, and major lifestyle risk factors. Afterward, each participant underwent three follow-up examinations, which took place one month, three months, and six months after the baseline examination. A total of 233 participants entered this analysis (Supplementary Fig. S1).

An additional 322 HALST participants without arrhythmia or pacemakers underwent heart rate variability measurement during the HALST examinations between September 2012 and April 2015. A total of 309 individuals were also included in this analysis (Supplementary Fig. S1).

The study was conducted according to the guidelines of the Declaration of Helsinki. All participants provided written informed consent. This study was approved by the institutional review boards of the National Health Research Institutes.

Measurement of autonomic nervous system function

The main exposure in this study was autonomic nervous system function, which was assessed by measuring heart rate variability during both the baseline and follow-up examinations according to standardized procedures37. Participants received a 5-minute resting single lead (lead I) electrocardiogram while breathing spontaneously and normally in a room that was maintained at a comfortable temperature with dim light after a 15-minute rest. The electrocardiogram signals were then processed by an HRV monitor (8Z05, Wegene Technology Inc., Taiwan)38,39,40. Both time-___domain (standard deviation of all NN intervals [SDNN] [in milliseconds])) and frequency-___domain (powers [in milliseconds squared] in low-frequency [LF] [0.04–0.15 Hz], high-frequency [HF] [0.15–0.4 Hz] ranges, LF/HF ratio and the total power [in milliseconds squared] [TP] ( < = 0.4 Hz)) measures were used for analysis37. Resting heart rate (HR) was also calculated from the electrocardiogram signals.

Measurement of functional capacity

The main outcome of this study was functional capacity, which was assessed by conducting chair-rising tests during both the baseline and follow-up examinations according to standardized procedures. During the chair-rising tests, participants were asked to stand up (from a straight-backed chair) and sit down as fast as possible five times while keeping their arms across the chest4. The time taken to complete this task, the chair rise time, was measured by a trained examiner using a handheld stopwatch that measured to the nearest hundredth of a second (s) and was recorded. This objective physical performance test represents a composite measure of muscle strength, muscle power, and balance2,41,42.

Measurement of covariates

We measured sociodemographic status (age; sex), education, smoking status, physical activity, body mass index (BMI), chronic diseases (hypertension, diabetes mellitus, cardiovascular disease, stroke, lung disease, cancer, arthritis, chronic kidney disease, and Parkinson’s disease), and resting heart rate during the baseline examinations. (Supplementary Methods S1 in Supplementary Materials)

Statistical analyses

Participants’ characteristics were summarized using descriptive statistics. Categorical variables were summarized using numbers (percentage). Continuous variables were summarized using mean (standard deviation) if they were normally distributed and median (interquartile range) if they were nonnormally distributed.

To conduct subsequent analysis, all heart rate variability measures were natural log-transformed because of nonnormal distribution. Also, in order to facilitate effect size comparisons among different heart rate variability measures, we further standardized the natural log-transformed heart rate variability measures.

To investigate the longitudinal association between the exposure (HRV measures) and the outcome (chair rise time), both on continuous scales, we applied linear mixed models with participant-level random intercept and random slope to the longitudinal data. An unstructured covariance structure was assumed. With this model, the correlations between continuous outcomes measured in the same individual over time were properly accounted for. Time was defined as years (t) since the baseline (t0) examination. Assumptions of the linear mixed model, including the normality and functional form assumptions, were checked by examining the residual distributions.

We hypothesized an effect of baseline and time-varying HRV measures on chair rise time. For the baseline HRV measure k, we fitted a linear mixed model that included t, t2, and the term for the baseline HRV measure k (model 1). We included the t2 term in the model to allow for non-linear change with time. The coefficient estimates of the term for the baseline HRV measure k quantified the association of the differences in baseline HRV measure k with the differences in initial chair rise time (the intercept of the average true trajectory of change in chair rise time). Then, we added covariates that might confound the association between the baseline HRV measure k and chair rise time to the model (model 2). To quantify the association of the baseline HRV measure k with the rate of change of the chair rise time, we tested the interactions of baseline HRV measure k with t adjusting for covariates (model 3), which allow for different rates of change in chair rise time between those with differences in the baseline HRV measure k. In model 4, we tested the interactions of baseline HRV measure k with t2 adjusting for covariates, which allow for different change rates of rates of change in chair rise time between those with differences in the baseline HRV measure k. In models 3 and 4, we retained interaction terms that were statistically significant.

For the time-varying HRV measure k, we likewise fitted the following linear mixed models: model 1 included t, t2, and the term for the time-varying HRV measure k; model 2 additionally included covariates that might confound the association between the time-varying HRV measure k and chair rise time; model 3 tested the interactions of time-varying HRV measure k with t adjusting for covariates; and model 4 alternatively tested the interactions of time-varying HRV measure k with t2 adjusting for covariates. The coefficient estimates of the term for the time-varying HRV measure k in Model 2 quantified the association of the differences in HRV measure k at a time during follow-up with the differences in chair rise time at that time. The coefficient estimates of the time-varying HRV measure k x t interaction term in Model 3 and the time-varying HRV measure k x t2 interaction term in Model 4 quantified the association of the differences in HRV measure k with the differences in the rate of change in chair rise time and the differences in the change rates of rate of change in chair rise time, respectively.

To further isolate and test the effects of within-individual change in HRV measure k, we decomposed the time-varying HRV measure k at time t of subject i (HRVikt) into two parts: the HRV measure k at t0 (HRVikt0); and ΔHRVikt, where ΔHRVikt = HRVikt - HRVikt0. We then refitted the linear mixed models that included HRVkt0 and ΔHRVk. The coefficient estimates of the term ΔHRVk in Model 2 then quantified the association of the differences in within-individual change in HRV measure k from baseline to a time during follow-up with the differences in chair rise time at that time, controlling for baseline level of HRV measure k. The coefficient estimates of the ΔHRVk x t interaction term in Model 3 and the ΔHRVk x t2 interaction term in Model 4 quantified the association of the differences in within-individual change in HRV measure k with the differences in the rate of change in chair rise time and the differences in the change rates of rate of change in chair rise time.

In sensitivity analysis, we specifically examined the effects of hypertension on our findings by fitting the models that did and did not include hypertension. We also examined the possible impacts of medications by fitting models that additionally included β-blockers.

A two-sided P value of less than 0.05 was considered statistical significance. Analyses were performed using SAS, version 9.4 (SAS Institute, Cary, NC, USA).

Results

Characteristics of the study population, consisting of 542 adults, are shown in Table 1. The mean (SD) age at baseline was 70.1 (7.0) years, and 46.5% were women. These 542 adults contributed to a mean of 2.3 functional capacity observations per person.

Table 1 Baseline characteristics of study participants.
Full size table

Baseline HRV measures and changes in chair rise time

There were statistically significant interactions of the baseline LF term with time (Model 3 in Table 2) and time2 (Model 4 in Table 2), where time represents years since baseline examination, in adjusted models. The point estimates of the coefficients were − 0.11 and − 0.02, respectively, meaning that one SD lower baseline LF was associated with a 0.11 s/year faster increase in chair rise time and a 0.02 s/year/year faster increase in the rate of change in chair rise time during the subsequent follow-up.

Table 2 Association of baseline heart rate variability measures with change in chair rise time.
Full size table

Unlike the slope, differences in any baseline HRV measure were not significantly associated with differences in the average true initial (baseline) status of chair rise time (Table 2).

Time-varying HRV measures and changes in chair rise time

A higher HF at a time during follow-up was significantly associated with a longer chair rise time at that time in the adjusted model (Models 2, 3 and 4 in Table 3). The point estimate of the coefficient was 0.18, meaning that one SD higher HF was associated with a 0.18 s longer chair rise time. We also observed a marginally significant association of a higher SDNN at a time during follow-up with a longer chair rise time at that time (Models 2, 3, and 4 in Table 3). Levels of LF, LF/HF ratio, and TP at a time during follow-up were not significantly associated with chair rise time at that time (Table 3).

Table 3 Association of time-varying heart rate variability measures with change in chair rise time.
Full size table

In the analysis of within-individual change in HRV measures, we noted that a within-individual increase in HF from baseline to a time during follow-up was associated with a longer chair rise time at that time, controlling for baseline HF level and covariates (Models 2, 3 and 4 in Table 4). The point estimate of the coefficient reached 0.22, meaning that one SD increase in HF was associated with a 0.22s longer chair rise time. We also noted that a within-individual decrease in LF/HF ratio from baseline to a time during follow-up was associated with a longer chair rise time at that time (Table 4). The point estimate of the coefficient reached − 0.17, meaning that one SD decrease in LF/HF ratio was associated with a 0.17s longer chair rise time. Within-individual change in other HRV measures, including SDNN, LF, and TP, were not significantly associated with chair rise time (Table 4).

Table 4 Association of within-individual change in heart rate variability measures with change in chair rise time.
Full size table

For each of the HRV measures, we failed to reject the null hypothesis that the population value of the time-varying HRV measure x time and the time-varying HRV measure x time2 interaction terms’ coefficients were 0 in both unadjusted (Supplementary Table S2) and adjusted analysis (Models 3 and 4 in Table 3). Likewise, we failed to reject the null hypothesis that the population value of the ΔHRVk x time and the ΔHRVk x time2 interaction terms’ coefficients were 0 in both unadjusted (Supplementary Table S3) and adjusted analysis (Models 3 and 4 in Table 4).

In sensitivity analysis, we noted that the coefficient estimates that quantify the associations of HRV measures with chair rise time in models that included hypertension were comparable to those in models that did not include hypertension (Supplementary Tables S4, S5 and S6).

Discussion

In this longitudinal study with repeated measurements of heart rate variability measures and functional capacity, we showed that heart rate variability was significantly associated with changes in functional capacity. First, a lower baseline LF was associated with a higher rate of gradual increase in chair rise time. Second, an increase in HF and a decrease in LF/HF ratio were each associated with an acute increase in chair rise time. These findings offer unique and valuable information about the relationship between autonomic nervous system activity and functional capacity.

The autonomic nervous system has two major components, namely the sympathetic and parasympathetic nervous systems. Its activity directly modulates the beat-to-beat variability in heart rate, with the sympathetic and parasympathetic nervous systems modulating the variability differently. Whereas the parasympathetic nervous system modulates the HF range variance (power), the sympathetic nervous system modulates the LF range variance (power) in heart rate. As such, autonomic nervous system function is frequently assessed non-invasively by measuring the heart rate variability in time or frequency domains. Whereas HF primarily reflexes the parasympathetic nervous system activity, the LF reflexes the sympathetic and parasympathetic nervous system activity. The LF/HF ratio indicates the sympathovagal balance and the sympathetic nervous system activity37. SDNN and TP reflex overall autonomic nervous system activity.

The association of heart rate variability with late-life health, though possible, remains largely unclear, and controversies still exist25,35. Our data suggested that autonomic nervous system function is related to functional capacity declines, either gradually over a longer term or acutely. For the gradual decline in functional capacity, we demonstrated that a lower LF range variance in heart rate at baseline was associated with an accelerated rate of gradual decline in the chair-rising tests performance. This association seems to be unmasked after covariate adjustment (Supplementary Tables S1 and Table S2; Table 2), which may have controlled for negative confounding43. A lower SDNN reflects a lower overall autonomic nervous system activity. We observed a marginally insignificant interactions of the baseline SDNN term with time2 (Model 4 in Table 2). The point estimate of the coefficient was − 0.01, with 95% CI reaching − 0.03, which were comparable to those of the baseline LF. Our findings are consistent with the results of previous investigations. Koopman et al.44 observed that, among 822 adults aged 50 years and older, a lower SDNN tended to be cross-sectionally associated with lower grip strength, an objective measure of muscle strength. Chaves et al.33 showed that, among 389 women aged 65 years and older, a lower LF and SDNN were associated cross-sectionally with physical frailty with poor functional capacity. This association was replicated in another cross-sectional study of 276 women aged 65 years and older34. Interestingly, Ogliari et al.45 showed that, among 5042 older adults aged 70 years and older, a lower SDNN at baseline examination tended to be associated with greater odds of impairment in perceived (subjective) functional independence. Unlike LF, we did not observe a significant association of baseline LF/HF ratio with change in chair rise time. Although this finding still needs to be confirmed by future studies of a larger scale, it is generally consistent with previous studies showing that a lower LF at baseline was associated with a higher risk of adverse health outcomes, whereas a lower LF/HF ratio was not46,47.

On the other hand, for the acute decline in functional capacity, we noted that an increase in HF and a decrease in LF/HF ratio, reflecting an intercurrent increase in parasympathetic nervous system activity and a sympathovagal balance skewed toward the parasympathetic arm, was associated with an acute decline in the chair-rising tests performance. In agreement with the observed association of a decrease in LF/HF ratio with an acute decline in functional capacity, a lower LF/HF ratio was cross-sectionally associated with physical frailty with poor functional capacity33,34. Also, in agreement with the observed association of an increase in HF with an acute decline in functional capacity, Varadhan et al. noted that a lower level of a principal component of HRV measures consisting of a negatively weighted HF was associated with a greater odd of physical frailty with poor functional capacity34. Moreover, the association of a sympathovagal balance that is skewed toward the sympathetic arm with a preserved functional capacity is consistent with the important roles of autonomic nervous system activity plays in maintaining muscle function19. It is interesting to note that aging is associated with an increase in autonomic support of blood pressure that is correlated with an increase in muscle sympathetic nerve activity48.

There are several possible explanations for the observed association between autonomic nervous system function and functional capacity changes. First, the lower HRV at baseline could indicate poor health37,46,47, which could then contribute to a decline in functional capacity through multiple mechanisms. For instance, the lower HRV may be accompanied by impaired cardiac autonomic control, which could contribute to a decline in muscle strength and power by failing to maintain adequate organ perfusion16. The lower HRV at baseline may also be associated with subclinical chronic metabolic disorders, such as insulin resistance, adiposity, hyperglycemia, and inflammation21,22,23,28,29,49, which in turn could lead to a decline over time in muscle mass, strength, and power chronically. Second, the observed lower LF range variance in heart rate at baseline and the temporal increase in HF range variance and a decrease in LF/HF ratio during follow-up may be associated with biological aging processes that underly the functional capacity decline and physical frailty development during aging. Third, for the associations of the temporal increase in HF range variance and a decrease in LF/HF ratio with acute decline in the chair-rising tests performance, the possibility of impaired skeletal muscle function contributed by an altered autonomic nervous system activity could not be excluded19.

The results of this study could have clinical implications. First, changes in autonomic function with an increase in HF and a decrease in LF/HF ratio may serve as a risk marker of functional capacity decline with usefulness in risk assessment. Second, the association between autonomic function and the chair-rising test performance suggests the possibility that the chair-rising test might be used as a stress test during which the body was exposed to a specific stress. The body’s responses to this stress, including those involving autonomic function, might be a major determinant of the test performance. This possibility, once proved, may open up a direction of clinical application. For instance, the chair-rising test, currently used as an objective physical performance test, might be used as a test for autonomic function that could be easily applied in many clinical settings for screening or monitoring purposes, whereas the formal autonomic function test might be reserved for frail older adults who could not perform the task. Future research that examines this possibility would be valuable.

To our knowledge, this study represents the first attempt to test the temporally ordered association between the autonomic nervous system function and functional capacity. Other strengths include objective, standardized, and serially conducted measurements of the autonomic nervous system function and functional capacity. Despite these strengths, there are several limitations. First, this was an observational study. Causality inference still should be made with caution. Second, despite adjustment for multiple confounders, the possibility of biased estimates caused by residual confounding could not be eliminated. It is possible that other factors, such as severity of chronic diseases (e.g., diabetes mellitus, visceral adiposity), target organ damages, medications and time-varying covariates, may have confounded the observed associations. Although we attempted to adjust for medications that might have effects on the autonomic nervous system, we assessed medication use mainly based on self-report. The possibility of misclassification could not be excluded. Nevertheless, literature suggested that, unlike HRV at a single time point, HRV change is less likely to be affected by a wide range of confounding factors, including medication use50. In a sensitivity analysis, we further adjusted for the adjudicated β-blocker usage. The results were not materially different after this additional adjustment. Third, our study population consisted partly of older adults who regularly used mobile phones or tablets with easy access to internet connectivity. The generalizability of the study results to other populations remains to be confirmed. Fourth, to facilitate clinical translation, more research on the normative HRV change is necessary. This data would be valuable in defining physiological and pathological changes during aging. Fifth, HRV was assessed using a 5-minute resting single-lead electrocardiogram, and its change was assessed by repeated measurements performed three times during the six months follow-up period. Although it was assessed according to the guideline37, further investigations of HRV changes in which HRV is assessed by using longer-term, or even continuously51, measured heart rate signals are worthwhile. Finally, HRV measures were repeatedly assessed at the same time as the functional capacity. As such, the possibility of effect cause could not be totally excluded.

In conclusion, we showed that autonomic nervous system function, measured by HRV, was longitudinally associated with changes in functional capacity in older adults. A lower autonomic nervous system activity with a lower LF range variance in heart rate was associated with an accelerated rate of gradual decline in functional capacity. A change in autonomic nervous system activity with an increase in HF range variance in heart rate and a decrease in LF/HF ratio was associated with an acute decline in functional capacity. Changes in autonomic nervous system function may precede the decline in functional capacity in older adults and may serve as a risk marker of functional capacity decline in this population. The autonomic nervous system function may be associated with the mechanisms of the functional capacity change in older adults.