What Is behind Changes in Resting Heart Rate and Heart Rate Variability? A Large-Scale Analysis of Longitudinal Measurements Acquired in Free-Living

he aim of this study was to investigate the relationship between heart rate and heart rate variability (HRV) with respect to individual characteristics and acute stressors. In particular, the relationship between heart rate, HRV, age, sex, body mass index (BMI), and physical activity level was analyzed cross-sectionally in a large sample of 28,175 individuals. Additionally, the change in heart rate and HRV in response to common acute stressors such as training of different intensities, alcohol intake, the menstrual cycle, and sickness was analyzed longitudinally. Acute stressors were analyzed over a period of 5 years for a total of 9 million measurements (320 374measurements per person). HRV at the population level reduced with age (p< 0.05, r = 0.35, effect size = moderate) and was weakly associated with physical activity level (p< 0.05, r = 0.21, effect size = small) and not associated with sex (p= 0.35,d = 0.02, effect size = negligible). Heart rate was moderately associated with physical activity level (p< 0.05, r = 0.30, effectsize = moderate) and sex (p< 0.05, d = 0.63, effectsize = moderate) but not with age (p= 0.35, r

with physical activity level (p< 0.05, r = 0.21, effect size = small) and not associated with sex (p= 0.35,d = 0.02, effect size = negligible). Heart rate was moderately associated with physical activity level (p< 0.05, r = 0.30, effectsize = moderate) and sex (p< 0.05, d = 0.63, effectsize = moderate) but not with age (p= 0.35, r = 0.01). Similar relationships between BMI, resting heart rate (p< 0.05, r = 0.19, effect size = small), and HRV (p< 0.05, r = 0.10, effect size = small) are shown. In response to acute stressors, we report a 4.6% change in HRV (p< 0.05, d = 0.36, effect size = small) and a 1.3% change in heart rate (p< 0.05, d = 0.38, effect size = small) in response to training, a 6% increase in heart rate (p< 0.05, d = 0.97, effect size = large) and a 12% reduction in HRV (p< 0.05, d = 0.55, effect size = moderate) after high alcohol intake, a 1.6% change in heart rate (p< 0.05, d = 1.41, effectsize = large) and a 3.2% change in HRV (p< 0.05, d = 0.80, effect size = large) between the follicular and luteal phases of the menstrual cycle, and a 6% increase in heart rate (p< 0.05,d = 0.97, effect size = large) and 10% reduction in HRV (p< 0.05, d = 0.47,effect size = moderate) during sickness. Acute stressors analysis revealed how HRV is a more sensitive but not speci c marker of stress. In conclusion, a short resting heart rate and HRV measurement upon waking using a smartphone app can effectively be used in free-living to quantify individual stress responses across a large range of individuals and stressors. Keywords:heart rate variability; heart rate; training; stress; sickness; menstrual cycle 1. Introduction Autonomic control impacts heart rhythm in response to stress [1–4]. In particular, the heart has its own pacemaker, beating at approximately 100 beats per minute. However, heart rate at rest is typically lower than the intrinsic ring rate of the sinoatrial node (the pacemaker) due to the in uence of the

heart rate; training; stress; sickness; menstrual cycle 1. Introduction Autonomic control impacts heart rhythm in response to stress [1–4]. In particular, the heart has its own pacemaker, beating at approximately 100 beats per minute. However, heart rate at rest is typically lower than the intrinsic ring rate of the sinoatrial node (the pacemaker) due to the in uence of the autonomic nervous system (ANS) via its two main branches: the parasympathetic system and the sympathetic system. Normally, the parasympathetic branch of the ANS slows down heart rate and increases heart rate variability (HRV), while the sympathetic branch of the ANS increases heart rate and reduces HRV [1,5]. When measuring physiology in a resting state, there are differences between Sensors2021,21, 7932.

Sensors2021,21, 7932 2 of 18 resting heart rate and HRV due to increased parasympathetic in uence during different phases of the cardiac cycle [6]. For example, parasympathetic activity is higher during the exhale phase of the breathing cycle [7,8]. Additionally, the in uence of the parasympathetic system on heart rhythm is almost instantaneous [3,9], resulting in large differences in beat- to-beat heart rate. However, by de nition, resting heart rate averages out any beat-to-beat differences in consecutive heartbeats. As a result, parasympathetic modulation in response to stress tends to be better captured by HRV with respect to heart rate [10]. For these reasons, resting heart rhythm and, in particular, HRV, have been widely investigated in relation to acute stressors. When considering training as an acute stressor, the rationale behind monitoring recov- ery using resting heart rate or HRV is therefore coupled with how the ANS responds to such stressors. In the context of physical exercise, intense training shifts the ANS towards a sympathetic dominance [11–13], which is re ected in the higher heart rate and in the lower HRV 24–48 h after training [11,14]. Reductions in HRV and increases in heart rate as mea- sured at rest rst thing in the morning on the day after high-intensity aerobic exercise have been reported across a wide range of individuals [10,14]. However, heart rate increases after training are often very small and of limited practical applicability [10,15]. Similarly, other stressors have been investigated acutely. For example, alcohol intake was reported to suppress HRV while impacting heart rate to a lesser extent [16]. On the other hand, resting heart rate has been proposed as a clear marker able to detect infections as well as recovery from an infection [17], changes that typically are also re ected in HRV [18]. When analyzing acute stressors, it is of interest to establish whether the relationship between resting heart rate, HRV, and the stressor is reproducible across a wide range of individuals. Unfortunately, most studies to date focused on a homogeneous sample, typically of male and relatively young individuals [19], therefore limiting our understanding of the relation-

also re ected in HRV [18]. When analyzing acute stressors, it is of interest to establish whether the relationship between resting heart rate, HRV, and the stressor is reproducible across a wide range of individuals. Unfortunately, most studies to date focused on a homogeneous sample, typically of male and relatively young individuals [19], therefore limiting our understanding of the relation- ship between, e.g., training, sickness, alcohol intake, and resting heart rate and HRV in other groups of the population. In the context of a longitudinal analysis in response to stressors, the menstrual cycle should also be considered, given that several studies have shown how HRV is slightly suppressed during the luteal phase of the menstrual cycle [20]. Apart from the mechanisms in uencing heart rhythm in response to stress, stratifying population-level data across different subgroups of individuals (e.g., based on sex, activity level, age, or body mass index (BMI)) can provide useful insights into the differences between resting heart rate and HRV. For example, previous research has highlighted how HRV reduces with age [21]. Additionally, the link between cardiorespiratory tness and resting heart rate seems stronger than for HRV, despite a few studies showing increased HRV in response to an exercise program [22]. Typically, women have higher resting heart rates than men [23]. It follows from the inverse relationship between resting heart rate and HRV that HRV should be slightly lower in women. However, according to published literature [24], this is not necessarily the case. In recent times, monitoring resting heart rate and HRV unobtrusively in real-life settings, outside of the lab, has nally become a practical possibility. Data can be acquired using validated smartphone apps [25,26] longitudinally over periods of weeks or months, providing novel insights on an individual's response to training and lifestyle stressors. As a result, monitoring physiological stress and recovery status by means of an HRV measure- ment is becoming more common among elite athletes as well as sports enthusiasts [15]. However, while technological advancements have made it very easy to acquire high-quality heart rate and HRV data in resting conditions, many questions remain unanswered when

response to training and lifestyle stressors. As a result, monitoring physiological stress and recovery status by means of an HRV measure- ment is becoming more common among elite athletes as well as sports enthusiasts [15]. However, while technological advancements have made it very easy to acquire high-quality heart rate and HRV data in resting conditions, many questions remain unanswered when it comes to the use of resting heart rate and HRV as well as their differences both at the population level and within individuals (i.e., in response to stressors). Thus, the aim of this cross-sectional and longitudinal analysis is to provide a more com- prehensive view of the relationship between resting heart rate, HRV, acute stressors, and population-level characteristics, analyzing data acquired on a large sample of individuals in real-life settings.

Sensors2021,21, 7932 3 of 18 2. Materials and Methods 2.1. Data Acquisition Physiological data (resting heart rate and HRV) and annotations (individual character- istics and stressors) were collected using the HRV4Training app, as detailed below. 2.1.1. Heart Rate and HRV Resting heart rate and HRV were collected daily using the validated HRV4Training app [25–27]. The HRV4Training app is a commercially available mobile phone app that allows for non-invasive measurements of resting heart rate and HRV using either the phone camera or an external sensor (see Figure). Users included in this study downloaded the HRV4Training app from the Apple Store or Google Play out of their own interest and explicitly agreed to provide collected measurements and annotations for research purposes via a consent form embedded in the app. The app instructed users to perform the measurement right after waking up while still lying down to limit the effect of other stressors (e.g., caffeine intake or physical activity). Instructions were provided to reproduce conditions similar to measurements at rest in supervised settings. HRV features that are representative of parasympathetic activity and that are typically reported in the scienti c literature are the high-frequency power (HF) and the square root of the mean squared difference between beat-to-beat intervals or rMSSD [28]. However, rMSSD might be preferable as it is less dependent on breathing rate [29] and better standardized. Thus, only rMSSD is reported in this work. Measurement duration was con gurable between 1 and 5 min since 1 min measurements were previously validated and considered of suf cient duration for accurate HRV analysis of time-domain features such as rMSSD [30,31]. Figure 1. Screenshots of the HRV4Training app. The image on the left side shows the measurement screen, displaying the photoplethysmographic signal acquired via the mobile phone camera. The middle image shows an example of the questionnaire that is used after the measurement to annotate stressors such as training intensity, alcohol intake, sickness, or the menstrual cycle. The third image, on the right, shows a historical view of the data. 2.1.2. Users Users with at least 60 resting heart rate and HRV measurements were

the mobile phone camera. The middle image shows an example of the questionnaire that is used after the measurement to annotate stressors such as training intensity, alcohol intake, sickness, or the menstrual cycle. The third image, on the right, shows a historical view of the data. 2.1.2. Users Users with at least 60 resting heart rate and HRV measurements were included in the analysis so that individual responses to various acute stressors (alcohol intake, sickness, the menstrual cycle, and training) could be investigated in relation to multiple instances

Sensors2021,21, 7932 4 of 18 of the same stressors, as opposed to individual instances typically reported in laboratory studies (see Figure). Additionally, all measurements that resulted in more than 10% RR intervals being discarded after applying the artefact correction method described in [25] were excluded. After applying the inclusion criteria, 28,175 users (22,750 male and 5425 female) were included, for a total of 9,032,749 measurements (320 364 measurements per user). Mean age across the entire population was 37 12 years, and mean BMI was 23.8 3.2 kg/m 2 . The user-reported training habits were as follows: 667 users reported training occasionally, 1551 users reported training 1–2 times per week, 11,973 users reported training3–4 timesper week, and 13,614 users reported training daily. The BMI categories were underweight (BMI below 18.5 kg/m 2 ,n= 314), normal (BMI between 18.5 kg/m 2 and 25 kg/m 2 ,n= 16,838), overweight (BMI between 25 kg/m 2 and30 kg/m 2 , n= 6808), and obese (BMI above 30 kg/m 2 ,n= 1145). 2.1.3. Individual Characteristics Users lled in a questionnaire upon registering an account in the HRV4Training app, including individual characteristics analyzed in this paper. In particular, users reported their age, physical activity level (one of the following: not training, training occasionally, training 2–3 times per week, training 4–5 times per week, and training daily), sex, weight, and height, from which BMI was derived. 2.1.4. Acute Stressors Training days were manually annotated in the app while answering a short ques- tionnaire, which is shown to the user right after the measurement. Training intensities were selected among four categories: rest, easy, average, and intense. Training intensities were then clustered in two groups: low intensity, comprising rest days and training days annotated as easy, and high intensity, comprising training days annotated as average or intense by users (see Figure). While training can be quanti ed in many different ways, the main goal of the proposed clustering was to quantify the effect of low intensity against high-intensity exercise on resting physiology, as typically adopted in a polarized training model [32]. Additionally, users reported alcohol intake, menstruation days, and sickness.1.

training days annotated as average or intense by users (see Figure). While training can be quanti ed in many different ways, the main goal of the proposed clustering was to quantify the effect of low intensity against high-intensity exercise on resting physiology, as typically adopted in a polarized training model [32]. Additionally, users reported alcohol intake, menstruation days, and sickness.1. Collectdaily HRV measurements and training intensities for > 60 days 2. Compute HRV difference from the previous day Lowload High load… …Difference after low load Difference after high load HRV measurements 3. Compute average difference for each category (e.g. low or high load) Average difference after low load Average difference after high load 4. Repeat for all users… Figure 2.Procedure used for the analysis of acute stressors. First, HRV (and resting heart rate) data and annotated training intensities were collected. Then, day-to-day differences in HRV (and resting heart rate) were computed. Differences were then averaged across categories, e.g., to compute the average day-to-day change in HRV (or heart rate) in response to either easy or high training intensity. The procedure is repeated for each individual so that we can determine the stress response for each stressor at the population level.

Sensors2021,21, 7932 5 of 18 2.2. Data Analysis 2.2.1. Population Level Analysis Individuals were clustered into different subgroups, two for sex (male/female), four for age (20 to 30 years old, 30 to 40 years old, 40 to 50 years old, and 50 to 60 years old), and four for tness level (training occasionally, 1–2 times per week, 3–4 times per week, or daily). Linear models using either heart rate or HRV as dependent variables and individual characteristics (sex, physical activity level, age, and BMI) as independent variables were built to determine the variance explained by such models. 2.2.2. Analysis of Acute Stressors The relation between physiological data and training was analyzed by rst computing day-to-day differences in heart rate and HRV for each individual. Subsequently, the change in resting heart rate and HRV on days following training of different intensities was analyzed for each user (see Figure). Additionally, the relationship between resting heart rate, HRV, and training was analyzed by age group and sex. The same procedure was used to analyze the relationship between heart rate, HRV, and alcohol intake. For sickness data, the percentage change was computed with respect to the non-sick condition. For the menstrual cycle, user-reported menstruation days were used to de ne the duration of a cycle and to determine the change in resting heart rate and HRV during the follicular and luteal phases, with respect to the users' average (see Figure). 2.2.3. Statistics Summary statistics are reported as mean standard deviations for each measure and subgroup. Mean resting heart rate and HRV were computed for each individual. The results for all acute stressor analyses are reported in percentage with respect to the user's average heart rate and HRV. Reporting the results as percentages can ease interpretation and comparison of the sensitivity of each marker with respect to a speci c stressor. Comparisons between two groups for the population level analysis as well as for acute stressors were carried out using t-tests, with a signi cance level of 0.05. One-way ANOVA was used to compare three or more groups. Effect sizes are reported using Cohen's d

interpretation and comparison of the sensitivity of each marker with respect to a speci c stressor. Comparisons between two groups for the population level analysis as well as for acute stressors were carried out using t-tests, with a signi cance level of 0.05. One-way ANOVA was used to compare three or more groups. Effect sizes are reported using Cohen's d (d) or Pearson's correlation coef cient (r) [33,34].1. Collect daily HRV measurements and annotated menstruation days for > 180 days per person 2. Estimate follicular and luteal phases (first and second half between tagged cycles) Menstruation tag: YES Menstruation tag: NO… Follicular phase Luteal phase HRV measurements for a typical cycle 3. Compute average HRV for each phase (follicular and luteal) Average HRV during follicular phase Average HRV during luteal phase 4. Repeat for all users… … 1st menstruation day Figure 3. The procedure used for the analysis of the menstrual cycle. First, HRV (and resting heart rate) data, and annotated menstruation days were collected. Then, the beginning of each cycle was de ned as the rst menstruation day, and the following days, up to the next cycle, were split into two to estimate the follicular and luteal phases. Average heart rate and HRV were computed for each phase (follicular and luteal) and for each user.

This study analyzes heart rate and HRV in response to various stressors over five years.