Testing the cross-stressor hypothesis under real-world conditions: Exercise as a moderator of the association between momentary anxiety and cardiovascular responses

Abstract

The cross-stressor adaptation hypothesis of exercise training has not been investigated under real-life conditions. Using ecological momentary assessment (EMA), we tested whether usual exercise level moderates the relationship of self-reported anxiety to concurrent ambulatory heart rate (HR) and systolic/diastolic blood pressure (SBP/DBP). Participants (N=832) completed 24-hour ambulatory monitoring of HR/BP, using a brachial BP cuff that took readings at 28-minute intervals. Anxiety levels were concurrently reported on a visual analog scale (VAS) using a Palm Pilot. Usual exercise behavior was assessed by a self-report questionnaire. Random coefficients linear regression models predicting momentary HR/BP readings from time-matched anxiety scores were estimated, yielding the average within-person effect (slope) of anxiety. The interaction of exercise level (i.e., no weekly exercise, 1–149, and ≥150 minutes/week; a between-person factor) with anxiety was added to the model in order to estimate the average anxiety slope for participants in each exercise category. The relationship of HR/BP to anxiety did not differ significantly among exercise categories, hence not providing evidence for the cross-stressor hypothesis. In an exploratory analysis of the difference in HR/BP between occasions when anxiety was in the top versus bottom person-specific quintiles of responses, the difference in HR (but not SBP or DBP) varied significantly by exercise level (F(2,625)=4.92, p=0.008). Though our pre-specified analysis did not support the hypothesis, we provide some post-hoc evidence supporting the cross-stressor hypothesis of exercise training for the HR response to anxiety.

Introduction

Evidence suggests that chronic exposure to psychological stress is linked to increased risk of cardiovascular (CV) and psychological disorders (Dimsdale, 2008; Gerber, Börjesson, Ljung, Lindwall, & Jonsdottir, 2016; B. G. Schwartz et al., 2012). Accordingly, there is a need to identify modifiable behaviors amenable to intervention to mitigate the risk incurred from psychological stressors, such as anxiety-inducing (i.e., anxiogenic) stimuli (Heidenreich et al., 2011; Kessler, 2007). Exercise, defined as a subset of physical activity (PA) that is planned, structured, and repetitive and has as a final or an intermediate objective to improve or maintain physical fitness, has been demonstrated to be beneficial for psychological and cardiovascular health (Haskell et al., 2007; Penedo & Dahn, 2005) and thus might be one such modifiable health behavior (Ensari, Greenlee, Motl, & Petruzzello, 2015; Petruzzello, Landers, Hatfield, Kubitz, & Salazar, 1991). Individuals engaging in regular exercise have been reported to have a lower incidence of anxiety-related disorders (Strohle et al., 2007) and experimental studies further indicate that regularly-exercising individuals show an attenuated physiological response (e.g., smaller increase in heart rate and blood pressure) to anxiogenic stimuli (Forcier et al., 2006; Klaperski, von Dawans, Heinrichs, & Fuchs, 2013; Salmon, 2001). These findings have led to the “cross-stressor adaptation” hypothesis of exercise training (Sothmann, 2006), which postulates that voluntary regular exercise (a physical stressor in a basic, physiological sense) induces adaptations in the physiological stress response systems of the body (e.g. the CV system) that then helps buffer the adverse responses to exposures to other taxing stressors, such as anxiogenic stimuli (Salmon, 2001; Sothmann et al., 1996).

Existing literature on the cross-stressor adaptation hypothesis of exercise training has thus far been inconclusive (Forcier et al., 2006; Jackson & Dishman, 2006) and is largely limited to laboratory studies in which some have demonstrated that regularly exercising and/or aerobically fit individuals experience diminished CV reactivity when exposed to anxiogenic stressors (e.g., verbal and mental arithmetic tests, the Trier Social Stress Test [TSST]) (Rimmele et al., 2009; Zschucke, Renneberg, Dimeo, Wüstenberg, & Ströhle, 2015), as well as to occupational anxiogenic stressors (Throne, Bartholomew, Craig, & Farrar, 2000). However, laboratory-based studies have various shortcomings that may limit the generalizability of the observed findings supporting the cross-stressor adaptation hypothesis. First, it is unknown how well the laboratory stressors used in such studies mimic real-life stressors (Kamarck & Lovallo, 2003). Second, laboratory protocols typically hold a variety of factors constant (e.g., posture, activity, food/caffeine/alcohol consumption) whereas these and other factors vary over the ambulatory monitoring session (Dunton, 2017). Similarly, the single time point design of laboratory studies only provides cross-sectional information concerning the physiological and psychological stress response profiles to one (or a couple of) particular type(s) of stressor. It does not capture the variability in the responses to a wider range of stressors or fluctuations that operate across time and location, such as over the course of a day of an ambulatory individual (Dunton, 2017; Kamarck, Schwartz, Janicki, Shiffman, & Raynor, 2003). Therefore, laboratory-based anxiogenic stressors suffer from a potential lack of generalizability to the real world.

One method to circumvent these limitations is to employ ecological momentary assessments (EMA) of affect, which can capture both between- and within-person fluctuations through collection of repeated measures over time under real world conditions (Dunton, 2017; Schwarz, 2007) and address the issue of real-world generalizability. A previous study (von Haaren et al., 2016) investigated the cross-stressor adaptation hypothesis of exercise training under real life settings among 61 male college students. The authors reported that in a 20-week randomized controlled trial testing an aerobic exercise-training intervention, those in the intervention arm experienced significantly smaller increases in HR and smaller decreases in HRV (LF/HF ratio and RMSSD) during a 2-day final exam period, compared to those in the control group (von Haaren et al., 2016). As noted, the sample size was small (N=61), and the study had limited generalizability (same anxiogenic stimulus for all participants, male college students).

Outside of studies examining the cross-stressor adaptation hypothesis, previous EMA-based studies have provided evidence for a within-person association of momentary anxiogenic stress levels and concurrent CV parameters (e.g., HR and BP) in various samples (Johnston et al., 2016; Johnston, Tuomisto, & Patching, 2008; Kamarck et al., 1998; Vella, Kamarck, & Shiffman, 2008; Zanstra & Johnston, 2011), and collectively suggest a link between exposure to anxiogenic stressors and increase in concurrently assessed CV function. Some of these studies have further reported considerable individual differences in the direction and magnitude of these associations (Kamarck et al., 1998). However, these studies have not examined the possible moderating effect of exercise on the observed between-person variability in the CV response (i.e., “reactivity”) to changing anxiety levels. Investigations are thus still needed to elucidate whether habitual exercise levels blunt the CV response to anxiogenic stimuli in daily life.

For the purposes of this study, we chose anxiety as a predictor because it is 1) a type of negative psychological stress (i.e., “distress”) that previous literature indicates is influenced by acute(Ensari et al., 2015; Herring, Hallgren, & Campbell, 2017) and chronic (e.g., habitual)(Asmundson et al., 2013) exercise, and 2) because it is most consistent with the practical implications of the cross-stressor hypothesis, which focuses on distress. For example, a recent EMA study (Bernstein, Curtiss, Wu, Barreira, & McNally, 2019) exploring the relationship between voluntary exercise and the temporal dynamics of daily emotions reported that anxiety and other emotions were not significnatly associated with average voluntary leisure time exercise; however, more physically active individuals exhibited less anxiety-specific emotional inertia. Anxiety is further linked to a variety of stress-related measures (e.g., salivary cortisol, hormonal response, self-reported perceived stress measures) (González-Cabrera et al., 2018), and some have suggested that anxiety is a marker of psychological stress (Armario, Marti, Molina, De Pablo, & Valdes, 1996; Rashkova, Ribagin, & Toneva, 2012). From a theoretical construct perspective, stress-tension and anxiety have been reported to be significant and moderately strong predictors of each other based on a causal inference approach (Davey et al., 2016). As such, measuring anxiety may be the most suitable alternative to measuring stress directly. Here we aim to assess whether CV responses to fluctuations in anxiety in a large sample under ambulatory conditions are moderated by habitual levels of exercise, as predicted by the cross-stressor hypothesis.

Accordingly, the purpose of this study is to investigate whether self-reported exercise levels moderate the relationship of changes in momentary HR and BP to changes in self-reported anxiety (i.e., the cross-stressor adaptation hypothesis of exercise training) under real world (i.e., ambulatory) conditions. We investigate this aim in 2 steps/hypotheses. First, we test the previously supported hypothesis that overall, increases in self-reported momentary anxiety are significantly associated with increases in concurrent HR and BP, and that there are considerable individual differences (i.e., heterogeneity) in the magnitude of this response. Next, we hypothesize that those who report engaging in regular exercise have an attenuated increase in their HR and BP responses to changes in momentary anxiety compared to the responses of non-exercisers. Finally, as the available evidence for the cross-stressor adaptation hypothesis is predominately from laboratory-based stressors known to yield sizeable anxiogenic responses, we conduct an exploratory analysis based on an idiographic approach (i.e., study of within-person variability over time and space) (Conner, Tennen, Fleeson, & Barrett, 2009; Hsueh et al., 2017) investigating whether the moderating effect of exercise on the CV response to momentary anxiety is observed when we use a person-specific approach to identifying “high anxiety” and “low anxiety” moments.

Methods

The following analyses use data from the Masked Hypertension Study (Schwartz et al. 2016), a multi-site observational study comprised of employed adults who were recruited between 2005 and 2011 from four workplaces to 1 of 2 university sites: Stony Brook University, University Hospital at Stony Brook, Columbia University Irving Medical Center, and a private hedge fund management organization in New York (all participants were seen at one of the 2 university sites). The purpose of the Masked Hypertension Study was to investigate the predictors and prognosis of elevated out-of-office BP. Participants completed 5 separate study visits and a 24-hour ambulatory BP recording with simultaneous self-ratings of various emotions (e.g., anxiety, anger, frustration, depressed, happy) via EMA. The study was conducted in adherence to the guidelines set forth by the Declaration of Helsinki and was approved by the institutional review boards of Stony Brook University and Columbia University Irving Medical Center.

Sample

Recruitment and enrollment procedures have been described in detail elsewhere (Schwartz et al. 2016). Briefly, 1988 potentially eligible individuals were identified from BP screenings that were conducted at the university campuses, and from the private hedge fund management organization following a public presentation about the study. Study inclusion criteria assessed during the BP screenings were: 1) able to speak English, 2) age 21 or older, 3) employed at least 17.5 hours/week, and 4) had a screening BP of less than 160/105 mmHg. Exclusion criteria were: 1) use of an antihypertensive medication, 2) evidence of secondary hypertension, 3) history of overt cardiovascular disease, 4) chronic renal, liver, thyroid, or adrenal disease, and 5) cancer not in remission for at least 6 months. Interested participants were followed up via telephone (N=1699) to confirm the above criteria and further screen for the last three exclusion criteria: 6) active substance abuse or a serious mental health illness, 7) being on any CV medication (other than a statin), and 8) pregnancy. Of those, 1254 met all eligibility criteria and 1011 enrolled in the study. A total of 893 wore the ambulatory blood pressure monitor and simultaneously completed the EMA protocol – visits 3 and 4 – and 842 completed Visit 5 during which exercise behavior and body mass index (BMI) were assessed.(Schwartz et al., 2016) After exclusion due to missing data for BP (defined as less than 10 valid awake ambulatory readings), usual exercise behavior and body mass index (BMI), a sample of 832 individuals were included in the analyses reported here.

Procedures

A timeline of study procedures is reported in detail elsewhere (Schwartz et al. 2016). The baseline phase of the Masked Hypertension Study was cross-sectional in design; it included 5 visits to the lab/clinic (See Figure 1), with the first 3 visits, separated by approximately 1 week, focused on enrollment and assessments of in-clinic BP. Participants were fitted with the 24-hour ambulatory BP monitor and instructed on the use of the EMA diary at the end of visit 3, returning the equipment the next day (Visit 4). At the 5th visit, in conjunction with a cardiovascular evaluation that included anthropometric measurements, an electrocardiogram and an echocardiogram, we conducted a 12-page medical history interview, which was followed by questions asking about health behaviors. Physical activity was assessed as a part of these health behaviors (e.g., smoking, nutrition).

Figure 1.

Flowchart of parent study (“Masked Hypertension”) protocol.

For the ambulatory assessment between visits 3 and 4, participants were trained on the use of a pre-programmed Palm Pilot electronic diary on which they were asked to answer questions after each BP reading regarding their affect immediately prior to the reading. They were further asked to complete a device log to record the times that they went to sleep, woke up, took a nap, and/or removed any of the equipment (e.g., to exercise or shower). The record from each electronic diary entry included a date/time stamp which was used to match each EMA report to the nearest ABP reading; only those ABP readings with an EMA report made within 8 minutes were included in the analysis. Participants wore the equipment for 24 hours on a week day (including either 1 entire work shift or parts of 2 consecutive work shifts) and returned it the next day (Visit 4). At a subsequent visit (~1 week later; Visit 5 in Figure 1), height and weight were measured for calculation of body mass index (BMI) (weight/height-squared, kg/m²). Participants further provided self-reported information on habitual exercise levels via a health questionnaire during this visit. The analyses for the present study include data from the 24-hour ambulatory/EMA assessment, the physical activity questionnaire and height and weight assessments collected during the final study visit (i.e., Visit 5), and the BP readings taken during Visits 1–3 (3 readings taken on each of 3 visits) used to calculate the average clinic BP for each participant. We chose a 24-hour ambulatory data collection period because this is what is recommended by both U.S. and European guidelines for ambulatory BP monitoring. The EMA component – completion of a 30-item diary entry every 28–30 minutes added substantially to participant burden (Collins & Muraven, 2007). Of note, despite the 24-hour time window, we have on average of 21 completed EMA reports per person, which is more than many EMA-based studies that collect 3–6 reports/day for 2–4 weeks (e.g., Jones, Taylor, Liao, Intille, & Dunton, 2017). All participants completed the ABPM and EMA monitoring on a work day, or parts of two consecutive work days, to further standardize the protocol.

Measures

Ambulatory Blood Pressure (BP) and Heart Rate (HR)

Ambulatory SBP, DBP and HR were measured automatically using a 24-hour ambulatory BP monitor (Spacelabs 90207, Snoqualmie, WA) worn on the non-dominant arm. Measurements were taken every 28–30 minutes to vary the time of data collection each hour and thereby reduce the likelihood of anticipatory effects in the participants. These CV measures were analyzed based on extensive evidence of their responsiveness to psychological stress (Forcier et al., 2006; Hinz, Seibt, & Scheuch, 2001), and substantial variability (Forcier et al., 2006; Wright, O’Brien, Hazi, & Kent, 2014) over a 24-hour period.

Ecological momentary assessment of self-reported anxiety levels

Anxiety level was measured after each BP reading using a Visual Analog Scale (VAS) presented on a pre-programmed electronic diary (Palm Pilot Tungsten 3). The VAS is a single item measure (“Just before [the] BP [reading], how anxious/tense were you feeling?”); participants placed a mark on a line with anchors labeled “Not at all” and “Very much”) at the endpoints; the placement of the mark was converted to a numerical score ranging from 0 to 100. The VAS has been demonstrated to have good psychometric properties (Davey, Barratt, Butow, & Deeks, 2007); its scores correlate strongly with scores from other similar scales, including the Spielberger State Anxiety Inventory (SSAI) (Bond, Shine, & Bruce, 1995; Chlan, 2004; Davey et al., 2007; Facco et al., 2013) and provides comparable scores with the SSAI on anxiety in both experimental/laboratory (e.g., in response to anxiogenic stimuli) (Ensari, Petruzzello, & Motl, 2019) and ambulatory (i.e., outside of the laboratory) settings in both healthy and clinical populations (e.g., Chlan, 2004; Ensari et al., 2019; Facco et al., 2013). The emotions angry/hostile and frustrated were also assessed by EMA in this study. Of the 3 self-reported negative affect measures, we a priori chose to examine only anxiety (i.e., anxious/tense) in the current analyses given the strong evidence for the cross-stressor adaptation hypothesis observed for anxiety-related constructs (Bernstein et al., 2019; Salmon, 2001; Sothmann et al., 1996). Importantly, the computerized version of the VAS for anxiety has been reported to be sensitive to measuring responses to psychosocial stress (Abend, Dan, Maoz, Raz, & Bar-Haim, 2014), thus making it appropriate for the purposes of this study.

Exercise level

Exercise level was based on the participants’ responses to 3 questions: 1) “Do you regularly exercise?”, 2) [If yes to item #1] “How many times on average do you engage in at least 15 minutes of exercise per week?”, and 3) [If yes to item #1] “How long (in minutes) do you exercise for during each session?”. Responses to items 2 and 3 were multiplied to estimate the total minutes of exercise/week. This study used a modified version of the Physical Activity Vital Sign (PAVS), which was created to measure habitual leisure time physical activity (i.e., exercise) levels (Haskell et al., 2007), and is also similar to the Speedy Nutrition and Physical Activity Assessment (SNAP) (Ball et al., 2015; Golightly et al., 2017) based on its focus on only weekly frequency and minutes of exercise per week, without explicit mention of intensity (Ball et al., 2015). The PAVS was created by a primary care provider and has evidence of adequate construct validity (Greenwood, Joy, & Stanford, 2010). It further has criterion validity with accelerometry (r=0.51) and adequate agreement with accelerometry-based classification of whether someone meets current ACSM/AHA aerobic PA recommendations (Ball et al., 2015; Ball, Joy, Gren, Cunningham, & Shaw, 2016). Similarly, SNAP has statistically significant, moderate correlations with accelerometer counts of physical activity (r=0.31) (Ball et al., 2015) and was created to be understood at a 5^th grade literacy level (Golightly et al., 2017). Consistent with the American Heart Association’s Life’s Simple 7 metric (Lloyd-Jones et al., 2010), participants were divided into the following 3 categories; those who reported not exercising regularly, those who reported between 1 and 149 minutes of exercise per week, and those who reported at least 150 minutes of exercise per week. Similar brief instruments of exercise behavior have been used in previous studies (Craig et al., 2003; Kurtze, Rangul, Hustvedt, & Flanders, 2008; K Milton, Bull, & Bauman, 2010), and have been shown to possess adequate validity and strong reliability in adult samples (Craig et al., 2003; Kurtze et al., 2008; Milton et al., 2010; Karen Milton, Clemes, & Bull, 2013).

Data Analysis

We obtained percentages or means and standard deviations (SD) for demographic characteristics and all outcome measures. Self-reported anxiety scores were considerably right-skewed. A cube root power transformation approximately normalized the distribution of this measure. Accordingly, subsequent analyses were conducted using the cube root transformed anxiety scores (i.e., anxiety^1/3). On average, there were about 24 anxiety EMA reports time-matched (within 8 minutes) to HR, SBP and DBP observations per person (Mean=24.5, SD=7.4).

A random coefficients linear regression model was used to test our hypotheses. First, we estimated regression models predicting HR, SBP and DBP (separately) from concurrent anxiety ratings and exercise level (3 categories: no weekly exercise, 1–149 mins/week, ≥150 mins/week), which yielded estimates of the average within-person slope (change in the outcome per unit increase in anxiety), the variability of these slopes, and the main effect of exercise category. To test the moderating effect of exercise level on the relationship between momentary anxiety and CV outcomes, we added the multiplicative interaction of exercise level with anxiety which yielded estimates of the average slope for the anxiety→HR, anxiety→SBP, and anxiety→DBP relationships for those in each exercise category. The general Satterthwaite approximation for the denominator degrees of freedom was used in the F-test for the interaction (numerator DF=2), and the t-tests and confidence intervals for the average slopes within each exercise category. All models were adjusted for age, sex, race, ethnicity, years of education, and BMI. To estimate the proportion of variance in the within-person effects of anxiety on CV outcomes that could be explained or accounted for by exercise group, we computed a pseudo-R² statistic as:

$$pseudo{R}^2=1-\frac{Var\left(B_{anxiety|exercisegroup\times anxietyinteraction}\right)}{Var\left(B_{anxiety}\right)},$$

where “|” stands for “controlling for”. The square-root of this value is a measure of the association/correlation between exercise group and the BP/HR response to change in anxiety.

Exploratory analyses

As an exploratory analysis, we conducted a within-person “extreme-group” analysis (Preacher, Rucker, MacCallum, & Nicewander, 2005) by a) identifying those EMA reports for which the anxiety rating was in the participant-specific top and bottom quintiles of anxiety, b) modeling the effect of quintile (top vs bottom) on HR/BP as a random effect, and c) testing for effect modification by exercise group. This “idiographic” approach (Conner et al., 2009) to the analysis, wherein we compared the difference in HR (or BP) between participant-specific high versus low anxiety reports, allows for possible individual differences in the way participants used the VAS to report anxiety (e.g., 2 different individuals placing a mark at a score of 20 might be reporting that they are “moderately anxious” vs only “mildly anxious”). As such, this person-specific approach to analyzing the VAS scores may be an advantageous alternative to the traditional “nomothetic” (i.e., group-level) approach (Conner et al., 2009; Hsueh et al., 2017).

In a second exploratory analysis, we conducted a sensitivity analysis in which we re-examined the moderating effect of exercise by running the same effect modification model while treating exercise as a continuous measure, instead of as a categorical measure. SAS 9.4 was used for all data analyses.

Results

Sample characteristics are provided in Table 1. Characteristics of study participants stratified by study site are provided in Supplemental Table 1. The mean (± standard deviation) age was 45.3 ± 10.3 years, 7.2% were black, 11.7 % were Hispanic, and the mean BMI was 27.6 ± 5.4 kg/m². The majority (59.8 %) had normal clinic BP (<120/80 mmHg), but 34.8% of participants were pre-hypertensive (≥120/80 and ≤139/89 mmHg) and 5.2% were hypertensive (≥140/90 mmHg) based on the mean of 9 readings taken over 3 visits to the clinic. Among the 832 participants included in the current analysis, 43.9%, 24.6%, and 31.5% reported no weekly exercise, 1–149 mins/week of exercise, and ≥150 mins/week of exercise, respectively. The mean of the 20,349 momentary anxiety VAS ratings across all participants was 16.4 on the 0–100 scale; the median value was 7 and the quartiles were 0 and 24. Of note, the mean anxiety and anxiety^1/3 scores (i.e., averaged across all the time points per person) did not differ significantly among the three exercise subgroups (p=0.378 and p=0.633, respectively, by 1-way ANOVA). This indicates that at the day-level (i.e., as opposed to moment-level), habitual exercise levels were not differentially associated with the 24-hour average of self-reported momentary anxiety ratings.

Table 1.

Participant characteristics (N=832).

Measure	Mean(SD) or N (%)
Age (years)	45.3 (10.4)
Female	493 (59.25%)
Black	60 (7.2%)
Hispanic	97 (11.7%)
BMI (kg/m2)	27.6 (5.3)
Years of Education	16.5 (2.9)
Mean awake SBP (mmHg)	123.1 (10.4)
Mean awake DBP (mmHg)	77.5 (7.4)
Mean awake HR (bpm)	76.5 (9.2)
Clinic BP Category*
Normotensive	498 (59.9%)
Pre-hypertensive	291 (34.9%)
Hypertensive	43 (5.2%)
Exercise Category
None	365 (43.9%)
1–149 mins/week	205 (24.6%)
≥150 mins/week	262 (31.5%)

BMI: Body Mass Index. SBP: Systolic Blood Pressure, DBP: Diastolic Blood Pressure, HR: Heart Rate, bpm: beats per minute.

^*Normotensive, prehypertensive and hypertensive categories were defined as clinical BP readings of <120/80, 120/80–139/89, and ≥140/90 mmHg, respectively.

There was a statistically significant (p<0.001) association, on average, between self-reported momentary anxiety scores and concurrent HR, SBP and DBP (Table 2). Specifically, the fixed effects (average within-person slope) for these associations indicate that on average, HR, SBP and DBP increased as self-reported momentary anxiety level increased. The SD of the slopes (a measure of heterogeneity among persons) was also statistically significant for HR and SBP, but not DBP, indicating that there was considerable variability (individual differences) with respect to the HR-anxiety slope and the SBP-anxiety slope, but not for the DBP-anxiety slope. Histograms of the distributions of empirical best linear unbiased predictions of participants’ HR-anxiety, SBP-anxiety, and DBP-anxiety slopes, along with superimposed normal distribution curves, are presented in Figure 2. Finally, there was a statistically significant main effect of exercise category on time-matched HR and DBP, but not SBP (Supplemental Table 2), indicating that engaging in any exercise is associated with a lower average awake HR and DBP, but not SBP in comparison to those who reported no weekly exercise (i.e., reference group) (F(2,833)=19.59, p<0.001; F(2,832)=5.87, p=0.003, vs F(2,834)=2.86, p=0.058; respectively).

Table 2.

Average within-person relationships of heart rate and blood pressure to momentary anxiety^* and between-person variability in these relationships.

	Average Effect				Variance
Outcome	Average B	SE	F-value**	p-value	SD (B)	p-value
HR (bpm)	0.76	0.09	77.19	<0.001	0.95	<0.001
SBP (mmHg)	0.83	0.09	90.82	<0.001	0.91	0.0013
DBP (mmHg)	0.72	0.07	115.88	<0.001	0.37	0.199

^*Cube-root transformed anxiety scores were used in the analyses.

^**F-statistic for partial fixed effects. Models are adjusted for: age, sex, race/ethnicity, years of education, body mass index, and exercise category. SE=Standard Error, HR=Heart Rate, SBP=Systolic Blood Pressure, DBP=Diastolic Blood Pressure.

Figure 2.

Distribution of person-specific slopes* for cube-root transformed anxiety when predicting, from left to right in the panel, (a) heart rate, (b) systolic blood pressure, and (c) diastolic blood pressure (N=832), each with a superimposed normal curve.

* empirical best linear unbiased predictor (EBLUP)

In the moderation analyses performed as the primary test of the cross-stressor adaptation hypothesis, the association of anxiety to HR, SBP, and DBP did not vary by exercise level (F(2,563)=1.51, p=0.221; F(2,484)=0.72, p=0.488; F(2,443)=0.73, p=0.484, respectively). This indicates that in comparison to those who reported no weekly exercise (i.e., reference group), those who engaged in any weekly exercise did not experience a significant attenuation in their HR, SBP or DBP responses to changes in anxiety (See Table 3). The pseudo-R² statistics associated with HR, SBP, and DBP were 0.04, 0.04 and −0.01, respectively, indicating that at most ~4% of the between-person variation in the HR and SBP responses/reactivity to changes in anxiety could be accounted for by differences in exercise level.

Table 3.

Moderating effect^* of exercise category on relationships of heart rate and blood pressure to momentary anxiety; differences among exercise categories in the average within-person relationships of heart rate and blood pressure to momentary anxiety (B_anxiety), as well as the F-test of the null hypothesis that the average is constant across the 3 categories.

Exercise category	HR		SBP		DBP
	Banxiety (SE)	Diff (SE)	Banxiety (SE)	Diff (SE)	Banxiety (SE)	Diff (SE)
No weekly Exercise	0.90 (0.13)	Ref.	0.94 (0.13)	Ref.	0.65 (0.10)	Ref.
1–149 mins/week	0.75 (0.17)	−0.15 (0.22)	0.70 (0.18)	−0.23 (0.22)	0.85 (0.14)	−0.20 (0.17)
≥150 mins/week	0.55 (0.16)	−0.35 (0.20)	0.75 (0.16)	−0.18 (0.20)	0.72 (0.12)	−0.07 (0.16)
F-test	F(2,563)=1.51, p=0.221		F(2,484)=0.72, p=0.488		F(2,443)=0.73, p=0.485

^*interaction term of exercise level with cube-root transformed anxiety. Models adjusted for: age, sex, race/ethnicity, years of education, body mass index, and exercise category. Ref.= Referent group (i.e., no weekly exercise). SE=Standard Error, HR=Heart Rate, SBP=Systolic Blood Pressure, DBP=Diastolic Blood Pressure.

Exploratory analyses

In reflecting on why our planned analysis did not support the cross-stressor adaptation hypothesis, we speculated that this might be due to differences in the way participants used the VAS scale to report anxiety levels. To examine this possibility, we utilized an ideographic approach in which we identified the person-specific top and bottom quintiles of each participant’s distribution of anxiety ratings (coded 1 and 0, respectively). Re-estimating the prior models after replacing the continuous anxiety^1/3 score with this binary score, we found that 1) the mean HR/BP was higher for readings obtained concurrently with anxiety ratings in the top quintile compared to readings obtained concurrently with anxiety ratings in the bottom quintile (see Supplemental Table 3) and 2) there was a statistically significant interaction of exercise level with top-vs-bottom anxiety quintile for HR, but not for SBP or DBP (F(2,659)=4.92, p=0.008; F(2,636)=1.21,p=0.298; F(2,612)=0.22, p=0.806, respectively; see Table 4). The interaction was such that an individual engaging in at least 150 mins/week of exercise had a significant attenuation in the difference in their HR between when they were feeling relatively more anxious (i.e., their top quintile) vs less anxious (i.e., their bottom quintile), compared to an individual engaging in no weekly exercise. The pseudo-R² statistics associated with HR, SBP, and DBP were 0.13, 0.03 and 0.02 respectively, indicating that 13% of the between-person variation in the HR difference between top and bottom quintiles and 2–3% of the between-person variation in SBP/DBP responses/reactivity to changes in anxiety could be accounted for by exercise level when the anxiety scores were limited to the top and bottom quintiles for each participant. In the sensitivity analyses, all results were similar when exercise was treated as a continuous predictor (results not shown).

Table 4.

Moderating effect^* of exercise category on relationships of heart rate and blood pressure to momentary anxiety when the anxiety scores are limited to those in the top and bottom quintiles for each participant (i.e., “within-person extreme group analysis”). Differences among exercise categories in the average within-person relationships of heart rate and blood pressure to momentary anxiety (B_anxiety), as well as the F-test of the null hypothesis that the average is constant across the 3 categories.

Exercise category	HR		SBP		DBP
	Banxiety (SE)	Diff (SE)	Banxiety (SE)	Diff (SE)	Banxiety (SE)	Diff (SE)
No weekly Exercise	1.99 (0.28)	Ref.	2.02 (0.29)	Ref.	1.46 (0.23)	Ref.
1–149 mins/week	1.56 (0.37)	−0.44 (0.46)	1.42 (0.39)	−0.60 (0.49)	1.67 (0.31)	0.22 (0.38)
≥150 mins/week	0.64 (0.33)	−1.35 (0.43)	1.39 (0.36)	−0.63 (0.46)	1.42 (0.28)	−0.04 (0.36)
F-test	F(2,659)=4.92, p=0.008		F(2,636)=1.21, p=0.289		F(2,612)=0.22, p=0.806

^*interaction term of exercise levels with anxiety quintile (highest vs lowest). Models adjusted for: age, sex, race/ethnicity, years of education, body mass index, exercise categories. Ref.= referent group (i.e., no weekly exercise). SE=Standard Error, HR= Heart Rate, SBP=Systolic Blood Pressure, DBP=Diastolic Blood Pressure, Ref.= Referent group.

Discussion

This study investigated the cross-stressor adaptation hypothesis of exercise training under real-world conditions using 24-hour EMA of self-reported anxiety coupled with ambulatory measures of HR, SBP, and DBP in a sample of 832 working adults. Our findings indicate that increases in self-reported momentary anxiety levels were associated with increases in concurrent HR, SBP and DBP. We further report significant heterogeneity in the HR- and SBP-, but not the DBP-anxiety slopes, providing evidence that the CV responses to fluctuations in self-reported anxiety vary considerably across individuals. However, contrary to the prediction of the cross-stressor hypothesis, the differences in the HR and SBP responses to fluctuations in anxiety were not significantly related to usual exercise level.

Existing research investigating the interplay between fluctuations in anxiety, CV responses and exercise has been largely limited to laboratory-based, single time point assessments and therefore does not account for within-person variability across time and/or setting. This observational study is among the first to overcome these weaknesses through an EMA-based approach by including frequent (i.e., every 28 minutes) assessments to investigate the cross-stressor adaptation hypothesis of exercise training under real-life conditions in a relatively large sample of employed adults.

In the present study, when the entire distribution of VAS scores was considered, we did not find statistically significant moderation by exercise category of the association of self-reported momentary anxiety with time-matched HR, SBP and DBP. Given the limited variability in the momentary anxiety scores in the present study (median=7, upper quartile=24 on a 100-point scale), it is possible that the range of anxiety experienced on a regular, working day is limited. Furthermore, the average effects in our study were relatively small in magnitude (i.e., 0.76 and 0.83 increase in HR and SBP, respectively, for each unit increase in anxiety^1/3 [total range of 0 to 4.64, when anxiety=100]). In comparison, the magnitude of effect observed in response to experimentally administered anxiogenic tasks is often considerably larger (e.g., effect sizes of 1.2, 0.39 and 0.62, for anxiety, HR and SBP, respectively, in response to a simulated public speaking task based on a meta-analysis of 11 experimental studies and 143 participants) (Zuardi, Crippa, Hallak, & Gorayeb, 2013). Importantly, these studies assess the effect of the anxiogenic task on self-reported anxiety (as a manipulation check), but not the direct relationship of the self-reported anxiety to the physiological response (i.e., BP, HR). Our study, distinct from this common practice in the existing literature, directly investigates how the fluctuations in self-reported anxiety are linked to concurrently measured BP and HR. In sum, laboratory-based studies and observational ambulatory studies in people’s normal everyday settings use different methods to investigate the cross-stressor hypothesis of exercise, and the results of the former may not generalize to the latter (Zuardi et al., 2013).

That increases in self-reported anxiety levels were associated with increases in CV responses is in accordance with results from other studies that have been conducted in both laboratory and field settings (Kamarck et al., 2003; Klaperski et al., 2013). Our analyses further reveal substantial individual differences (i.e., heterogeneity) in these CV responses to changes in self-reported momentary anxiety. Similar findings on inter-individual heterogeneity in the CV response to other types of daily psychosocial stressor experiences (e.g., arousal, negative affect, social conflict) have previously been reported by others (Kamarck et al., 2003). Collectively, these findings underscore the benefits of using an EMA-based approach that enhances ecological validity, presumably reducing recall errors and biases (by collecting self-reported information in real time) (Schwartz & Stone, 1998; Schwarz, 2007) and permitting multiple assessments per person to permit evaluation of within-person effects.

In an effort to avoid a Type II error, we conducted an exploratory, idiographic extreme groups analysis in which we examined whether within-person mean differences in HR/BP between those occasions when the person was most anxious (top quintile) and those occasions when he/she was least anxious (bottom quintile) were associated with exercise category. We found that the difference in heart rate (between high anxiety and low anxiety moments) was smallest for those who exercised ≥150 minutes/week, consistent with the cross-stressor hypothesis (p=0.002), though the differences in SBP and DBP were again not statistically significant. Our findings are consistent with those of Bernstein et al. (Bernstein et al., 2019), in that we did not see an association between habitual exercise and average level of anxiety. Thus, while the present study provides limited support for the cross-stressor hypothesis, our results support this idiographic approach – premised on the possibility that participants may each use the VAS scale somewhat differently – and merit further investigation. This was a secondary analysis, and the sample was not selected to have high levels of anxiety (e.g., people with high anxiety sensitivity or those with clinically meaningful levels of anxiety). While it appears that in a general healthy population, reports of anxiety tend to be low, a strength of our study is that it was based on an employed, community sample, thus potentially representative of a larger segment of the population. Future studies could include those with higher anxiety in order to further investigate the cross-stressor hypothesis in high-anxious populations, given preliminary evidence that among samples pre-selected to have high anxiety-sensitivity, participants tend to have exaggerated physiological responses, independent of changes in the self-reported measures of anxiety based on the VAS (Ensari et al., 2019).

It has been argued that real-life stressor reactivity might provide a better marker of disease-causing stress-related processes than is possible via artificial laboratory stressors (Zanstra & Johnston, 2011), though meta-analytic evidence based on prospective, longitudinal cohort studies (N=31) demonstrate that greater stress reactivity during laboratory stressors is significantly associated with poorer cardiovascular risk status (Chida & Steptoe, 2010). In contrast, studies investigating the stress-CV reactivity relationship under ambulatory/real-world conditions have not established a similar, significant association between CV reactivity and subsequent CVD risk (Kamarck et al., 2005), which has led some to suggest that this might be influenced by whether or not the anxiogenic stimulus involved a social situation or not (Smith, Limon, Gallo, & Ngu, 1996), Collectively, given that evidence on the ambulatory CV reactivity-CVD association is still modest, elucidation of the considerable inter-individual heterogeneity documented in the present and previous studies requires further investigation.

A unique and important aspect of our study is that anxiety is not the outcome of our study; rather, it is the fluctuations in HR and BP and their association with fluctuations in anxiety. Using change in a self-reported emotion (i.e., anxiety) as a predictor/mediator of change in a physiological outcome caused by an experimental stressor, rather than just as a manipulation check for the stressor, is not common in the literature. For example, whether people who report greater changes in negative emotions also exhibit higher CV responses has not been comprehensively assessed, nor have others asked whether exercise moderates this association. Stated differently, our outcomes of interest (i.e., dependent variables) are BP and HR, not anxiety. Through our hypotheses and mixed-level modeling approach we tackle this question and that is a strength of our work. Our post-hoc analyses provide preliminary evidence that there might be merit in conducting within-person analyses using the extreme scores for delineating this relationship, and future studies can further measure self-reported stress (e.g., Jones et al., 2017; Þórarinsdóttir et al., 2019) to investigate whether this association changes when stress, instead of anxiety, is used as the self-report measure of negative affect.

This study has several limitations. First, habitual exercise levels were assessed by self-report, a methodology subject to reporting bias and recall error, and our questionnaire did not explicitly ask about intensity. Though not the gold standard, self-report questionnaires are a non-costly, fast, and non-invasive method of measuring exercise. Future studies with objective measures of exercise may be warranted to confirm the observed findings. Similarly, the exercise categories were relatively crude even though we used physical activity guidelines to categorize the participants. However, it should be noted that when the exercise level was entered as a continuous variable, the results were substantively unchanged. Second, our sample consisted of individuals who were at least half-time employed (nearly all were employed full-time), of good overall health and from a single geographic region and relatively homogeneous demographic. Therefore, our findings might not be generalizable to other populations such as those with ethnic/racial minority status, chronic diseases, or older, retired individuals. However, to increase the diversity of our sample, we recruited participants at 2 research sites, one of which is located in an ethnically and racially diverse location. This enabled us to increase the proportion of Hispanic and black participants. Participants at one site were also more likely to meet criteria for pre-hypertension. While these site differences might have influenced the outcomes in the study, all models were adjusted for all these factors, as well as age, sex, BMI and years of education. Third, these analyses were based on a single, 24-hour assessment of anxiety and HR and BP completed on a week (i.e., working) day. Accordingly, it is possible that different results would have been observed if the analyses were conducted over a longer period of time (e.g., 1 week) or on a weekend day.

Previous studies investigating anxiety responses have distinguished between a “threat” response and a “challenge” response to potential stressors (Blascovich & Mendes, 2000; Mendes, Blascovich, Major, & Seery, 2001). Both are characterized by elevated HR, but threat responses are marked by peripheral vasoconstriction, whereas challenge responses are marked by peripheral vasodilation (Jamieson, Mendes, & Nock, 2013). It is possible that, compared to the other two groups, the high-volume exercisers might more often be exhibiting a challenge versus a threat response to an anxiety provoking situation. Alternatively, those reporting higher levels of self-reported exercise may be fostering a faster rebound from a distress episode relative to those who are less active, based on levels of post-stressor rumination (Bernstein & McNally, 2017). While these are plausible explanations for our findings, we did not measure ambulatory peripheral vasodilation or the type of stressor (i.e., threat vs challenge) in our study, and thus are unable to empirically investigate these speculations. These are potential limitations that warrant future research under ambulatory conditions.

In conclusion, we report an association between fluctuations in self-reported momentary anxiety and fluctuations in concurrently assessed HR and BP in a sample of 832 ambulatory, working adults, and significant inter-individual variability in these associations. We furthermore report that engaging in regular exercise largely did not buffer the cardiovascular response to momentary anxiety (and hence was not a contributor to the observed inter-individual variability). However, we provide some post-hoc evidence in support of the cross-stressor hypothesis of exercise training for the HR response to relatively high stressful stimuli, which supports further investigation into a person-specific analytic approach. Possible directions for future studies include assessment of exercise levels by objective measures (e.g., accelerometers), comparison of these associations between week days and weekend days, and inclusion of cardiorespiratory fitness (CRF) levels measured via CRF testing as possible moderators of the changes in the CV markers in response to increases in anxiety. Rapid growth in mobile health technology, and especially wearables and sensors (e.g., smartwatches) capable of passively collecting participant data over time at frequent intervals, has great potential to facilitate and refine our understanding of the ecological fluctuations in psychological symptoms and their physiological correlates, and ultimately improve behavioral (e.g., physical activity) intervention approaches for cardiovascular health.

Abbreviations

SBP

Systolic Blood Pressure

DBP

Diastolic Blood Pressure

HR

Heart Rate

CV

Cardiovascular

EMA

Ecological Momentary Assessment

PA

Physical Activity

TSST

Trier Social Stressor Test

VAS

Visual Analog Scale

BMI

Body Mass Index