| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
ORIGINAL ARTICLES |
From the Departments of Psychiatry (R.G.J., L.K.T., D.M.W., Y.D.) and Medicine (M.F.M.), University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania; Department of Psychology, University of Pittsburgh, Pittsburgh, Pennsylvania (S.B.M.); National Institute of Aging, Gerontology Research Center, Baltimore, Maryland (J.F.T.); and Center for Statistical Science, Brown University, Providence, Rhode Island (C.G.).
Address reprint requests to: Rolf G. Jacob, MD, Department of Psychiatry, University of Pittsburgh School of Medicine, 3811 O’Hara Street, Pittsburgh, PA 51213-2593.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
METHODS: The primary sample consisted of 69 normotensive or borderline hypertensive but otherwise healthy adult males. The validation sample consisted of 85 healthy male undergraduate college students. Both samples underwent half-hourly 24-hour ambulatory blood pressure measurements on four separate workdays, 1 week apart. At each ambulatory measurement, subjects recorded their behavior, environment, and mood. The circular mood scale, a circular visual analogue scale based on the circumplex model of mood, was used to reflect the totality of a participant’s affective state space. Longitudinal random effects regression models were applied in the data analysis.
RESULTS: The results for both samples were quite similar. Sleep and posture had the greatest influence on ambulatory blood pressure and heart rate. The effects of the environmental setting, social setting, and consumption were modest but statistically significant. Independent of these covariates, mood exerted a significant effect on blood pressure and heart rate. Relative to the "mellow" default category, blood pressure increased both for "anxious/annoyed" and "elated/happy" and decreased during "disengaged/sleepy" mood. The range of mood-related blood pressure estimates was 6.0/3.7 mm Hg.
CONCLUSIONS: The pattern of blood pressure responses suggests that they were related to the degree of engagement of a mood rather than the degree of unpleasantness. The hypothesis that posits that negative affect–related cardiovascular reactivity mediates the observed correlation between negative affect and disease risk should be reconsidered.
Key Words: ambulatory blood pressure • ambulatory heart rate • ambulatory mood • circumplex mood model
Abbreviations: ABP = ambulatory blood pressure; HR = heart rate; CMS = circular mood scale; DBP = diastolic blood pressure; SBP = systolic blood pressure; bpm = beats per minute; MAP = mean arterial pressure; REML = restricted maximumlikelihood.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Clinical relevance is a second major reason to be interested in ambulatory cardiovascular correlates of mood. The physiological states denoted by certain moods, if excessive, may strain an individual’s resources. Thus, a high disposition for certain moods, such as hostility, may increase risk for negative health outcomes, including cardiovascular disease (5). Ambulatory monitoring of mood-related cardiovascular responses may shed light on the importance of possible mediating mechanisms in the real-life setting.
The study of ABP presents a number of theoretical and methodological problems. From a statistical perspective, ABP data are characterized by both between- and within-subject variability. Table 1 (last column) lists the statistical approaches of studies that examined the effect of mood on automated ABP (6–18). With one exception (15), the quantitative techniques commonly used, analysis of variance or regression, do not unambiguously partition these different sources of variability. As reviewed by Jaccard and Wan (19), the commonly chosen methods are less than optimal, including the aggregation of raw, centered, or standardized data across individual subjects and the use of meta-analytic strategies of within-subject statistics. In the present study, we made use of longitudinal random effects regression modeling, in which subject and population effects are estimated separately (20–22).
|
A third question is whether the effect of mood should be considered only after controlling for the effect of other variables. Because mood is a function of concomitant situations and behaviors, the uniquely predictive value of mood should diminish as more mood-relevant behavioral information becomes available. For example, Schwarz et al. (15) reported that mood effects were larger before controlling for body position and location than after. The decision to include or exclude a particular covariate depends on whether the covariate can be considered more proximal than mood in its effect on blood pressure. Thus, it would make sense to control for physical activity because blood pressure changes represent the cardiovascular aspect of exercise (cf. 23). On the other hand, the effects of certain other situations might be mediated by the different emotions that they induce. In this case, controlling for the situation might result in the explaining away of important effects of mood. In the present study, we examined the effects of mood, controlling for other covariates of both types. We also present results for models that included fewer covariates.
Language includes a large number of words describing emotions, all with somewhat different connotations. The question becomes how to arrive at a manageable number of mood variables while assessing an individual’s entire emotion state space in an unbiased manner. The studies listed in Table 1 vary with respect to the type and quantity of moods included. Some studies focused on moods deemed clinically important and therefore were limited to the assessment of negative moods. Shapiro et al. (14), in perhaps the only recent comprehensive approach, selected variables based on a selection from 100 emotion/mood descriptors in the literature about emotion. Regardless of whether previous studies have pursued the question of mood assessment adequately, however, it is important to enrich the literature with findings based on alternative approaches.
In the current study, we capitalized on the well-known principles in emotion theory, (1) that most of the qualitative variability in mood states can be reduced to variations along two independent dimensions and (2) that the locations of discrete emotions plotted according to their values on these dimensions tend to be distributed along the perimeter of a circle (24, 25). Emotions located close together on the circle are similar to one another (eg, happy and mellow), whereas those on opposite poles of the circle are opposites of each another (eg, happy and unhappy). This arrangement is analogous to the color wheel, on which red is similar to orange but opposite green.
Applying this circumplex model for the assessment of mood in ambulatory studies has several advantages. By accounting for most of the qualitative variability in mood states, it provides a parsimonious way of assessing mood (25, 26). In a previous study, we found that subjects readily agreed on the location of emotion stimuli on a circular visual analogue scale of mood called the circular mood scale (CMS) (27). Another advantage of the CMS is that it does not require an a priori commitment to a limited set of discrete mood variables. In the present study, the mood categories used were based on the six main modes of the multimodal circular distribution identified in preliminary data. Another advantage of the CMS is that the basic datum that the CMS requests from the subject is a mark representing a spatial location (on the circle, see Figure 1 ) rather than a verbal report. Therefore, the CMS may be less affected by factors that specifically affect verbal stimuli (eg, culture). For example, the word "sleepy" has connotations of low positive affect (ie, an unpleasant state) in American culture, whereas in Japanese culture, its counterpart has no such connotation (28). A modified version of the CMS using only pictorial anchor points was recently developed and is currently being used in a study of civil servants in Nigeria.
|
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Subjects
The primary sample consisted of 69 adult men who participated in a study on the relationship between heart size and ABP. The study included both normotensive individuals (N = 31) and individuals with a history of clinical blood pressure measurements in the borderline hypertensive range (ie, clinical DBP exceeding 85 mm Hg) (N = 38). In the latter individuals, secondary hypertension was ruled out clinically by means of a physical examination and routine laboratory tests. Potential subjects were excluded if they did not complete the full research protocol, had a history of current or past drug treatment for hypertension, had a history of alcohol abuse or obvious psychiatric disorder, or were engaging in heavy endurance physical exercise on a regular basis.
Initially, 86 eligible subjects entered the study; of these, three dropped out before starting the monitoring and six more dropped out after completing only 1 day of monitoring. Eight subjects were excluded for unreadable (N = 5) or abnormal (N = 3) echocardiograms, leaving a total sample of 69 subjects who completed the protocol. The age range of the final sample was 19 to 62 years (average = 37.0 years). Fifty-seven (74%) were white, 15 (22%) were black, and three (4%) were Asian. Average body mass index was 26.2 (range = 18.1–37.2). Casual blood pressures (median of three readings before ABP monitoring on day 1) were symmetrically distributed with a mean of 126/83 mm Hg, a standard deviation of 13/11 mm Hg, and a total range of 98 to 154 over 56 to 110 mm Hg.
The validation sample consisted of 85 male undergraduates who were participating in a study on the effect of family history of hypertension on cardiovascular reactivity and heart size. The population and inclusion and exclusion criteria were described in detail by Manuck et al. (35). Participants were normotensive, nonobese (<25% overweight according to American Heart Association (AHA) tables), white college students. Their average age was 21.6 years (range = 18–25 years), and average body mass index was 24.3 (range = 17–32). None had a history of kidney disease, diabetes, cancer, hypertension, or other cardiovascular disease. Their mean casual blood pressure was 116/72 mm Hg, with a standard deviation of 8/7 mm Hg and a range of 103 to 138 over 56 to 86 mm Hg.
ABP Monitoring
Four monitoring days were scheduled on workdays on the same day of the week. The monitor was set to record ABP in fixed time intervals of 30 minutes throughout the 24-hour period, including sleep. The Colin ABP monitor was used for all ABP recordings. This monitor has previously been shown to reliably measure ABP (36, 37) and provides recordings that are inaudible to others even in noise-free environments.
The Colin recorder provides data for SBP, DBP, and HR separately for an auscultatory method and an oscillometric method. Data analyses were limited to auscultatory data. This choice was based on prior analyses, which indicated that at low values, diastolic oscillometric pressures were not highly correlated with auscultatory DBP and associated with greater error variance in statistical models. Furthermore, the decision was informed by findings that auscultatory DBPs provide a valid reflection of intraarterial blood pressure (38).
Behavioral Diary
Immediately after each ABP measurement, subjects completed a page in a behavioral diary (39). The diary information used in this study included posture (supine, sitting, or standing), activity (none, mild, moderate, or heavy), setting (home, work, driving, riding, or other), social environment (alone, with one person, with several persons), consumption (any occurrence of smoking or intake of food, caffeine, or alcohol), time of day, and mood (see Table 2 ).
|
The midpoints of the intervals representing the mood categories were used in secondary analyses examining the relationship between ABP responses and underlying mood dimensions. For example, with 
representing the midpoint of a mood category, the mood category’s location on the engagement dimension = sin , pleasantness = cos , negative affect dimension = cos ( + 45°), and positive affect dimension = cos ( - 45°).
The diary for the primary and validation samples differed in the following two ways. First, a space for recording "talking during cuff inflation" was included for the validation sample. This item was included to determine whether the effect of elated/happy mood observed in the primary sample might have been related to the behavior of talking. Second, the drawings of the faces on the CMS shown (Figure 1) were included only for the validation sample, because this form of the diary had not yet been developed at the beginning of data collection for the primary sample. Additionally, data for the validation sample were coded to include an activity of napping (ie, short periods of sleep during daytime hours separated from the main sleep period of the day). A typical nap period would occur during the day and last less than 2 hours. Sleep periods of 4 or more hours, regardless of the time of day, were categorized as sleeping.
Data Management and Preprocessing
ABP data were downloaded onto a computer disk. A computer program performed an initial screen of the blood pressure data based on the algorithm of Marler et al. (40). A second data screen excluded readings made when the diary indicated that physical activity had been heavy. This was done because analyses of a preliminary data set indicated that such data were infrequent and associated with large error variances and at times caused the models not to converge. Diary data were either double-entered by hand or entered by using a digitizing tablet. The computer program servicing the latter included automatic range checks for the data as they were entered. Each record was inspected and edited to ensure proper lining up of blood pressure with its corresponding diary entry.
Diary data were coded as described in Table 2. Time of day was modeled by two trigonometric functions, sin (t) and cos (t), where t = 360°/24 x military time. With two exceptions, the k levels within a covariate were dummy-coded (resulting in k - 1 indicator variables with the default level being represented by zero in all of these). Position was effect-coded rather than dummy-coded (ie, coded in such a way that the sum of the weights representing supine, sitting, and standing was zero). Activity was also effect-coded. For data obtained during sleep, the values of effect-coded variables were set at zero; dummy-coded variables were set as missing. (To examine whether the method of coding of sleep values had any effect on the estimates, we also ran models using daytime readings only; these differed only trivially from those in the full model).
Statistical Analysis
Longitudinal random effects regression models were used to examine the effects of behavioral covariates, using an adaptation of the program by Chi and Reinsel (21) developed for the efficient handling of large data sets. The programming was performed by a statistical programmer (Y.D.) in collaboration with the statistician coinvestigator (C.G.). Longitudinal random-effects regression models provide estimates simultaneously for population (fixed) and individual (random) effects and can account for correlations in the repeated observations of the same subject. Suppose ambulatory monitoring of SBP is conducted on n individual subjects. A total of Ti successive measurements are taken on individual i. Values of p covariates are also recorded on each subject. These can either be fixed over the duration of the experiment (eg, treatment group or trait measures, such as sex, age at entry, or resting blood pressure) or could vary with time, as when they are recorded along with the ambulatory monitored blood pressure (eg, activity, setting, or mood). The investigator’s goal is to assess the effects of all p covariates at the population level and, in addition, the effects of k of these covariates on each individual (so k is at most equal to p). The model can be conceptualized in two levels.
Level I describes the way in which the variability of the data on a specific individual can be explained in terms of the covariates for that individual. The equation for level I is Yi = Xia + Zibi + ei, where Yi is the vector of length Ti made up by the measurements on the ith individual, a is the vector of the population coefficients for the p covariates under consideration, Xi is the Ti x p population design matrix, bi is the vector of the k individual effects (ie, the difference between the individual and the population coefficients for each of the k covariates under consideration for the individual), Zi is the individual’s design matrix, and ei is the vector of the random errors (ie, random fluctuations around the mean value specified by the previous terms in the model). For each individual, the error vector ei is assumed to have a multivariate normal distribution mean vector 0 and covariance matrix Ri. The errors for different individuals are assumed to be statistically independent.
Level II describes the variability of the individual effects over the whole population. The vectors of the individual effect bi are assumed to be a random sample from a multivariate normal distribution with mean vector 0 and covariance matrix D.
The program used in this study used the REML method of estimation. The output of the program included estimates for the effects of each level of a covariate, the standard deviation, and the covariance matrix for the fixed effects. The latter were used to determine the error term for pairwise comparisons between different moods. An F test was used to determine the significance level for sets of variables. The program output also included a measure analogous to multiple correlation (R2), which can vary between -1 and +1; we used this measure to obtain a sense of the overall fit of the model.
In the primary sample, an informal strategy reminiscent of the "all possible regressions" approach was used to determine which variables had independent relations with ABP. As a result of these analyses, a "standard model" was developed and included the covariates as listed in Table 2. The standard model did not include time series modeling of the error terms. The residual plots of the standard model were inspected. They did not show clear evidence of serial dependencies, perhaps because the random effects of time of day were included in the standard model (see below).
The next step involved elimination of statistical outliers. Outliers were defined as a difference between the observed and predicted values exceeding three standard deviations (±32 mm Hg for SBP, ±25 mm Hg for DBP, and ±27 bpm for HR). To ensure that the analyses of SBP, DBP, and HR were based on the same data, exclusion of an outlying data point for one modality (eg, SBP) also led to the exclusion of the corresponding data point in the two other modalities (eg, DBP and HR). The final step fitted the standard model to the outlier-free data set. The robust standard model yielded estimates similar to the one based on the data with outliers included. Nine different statistical tests were performed to test for the significance of the effects of the covariates. A Bonferroni-adjusted target significance level of p = .005 was chosen to keep the experiment-wise error rate at or below p = .05. In the Results section, we present the robust standard model as well as some simpler models fitted to the outlier-free data set, illustrating variations in mood effects as a function of the other covariates included in the model. Data from the validation sample were analyzed by the a priori robust fitting of the standard model, augmented by estimates for napping and talking. The inclusion of validation samples is a standard procedure in studies relying on multiple regression.
Standard Model
The standard model included three sets of random effects: individual level, time of day (two variables), and recording day (four levels dummy-coded into three variables). The random day effects were included to control for differences in circumstances specific to a particular recording day, such as weather or bias in the readings of a particular monitor. The random time of day effects permitted the fitting of models responsive to individuals with unusual time schedules (eg, shift work).
Among the fixed effects (Table 2), the standard model included two sets of interaction terms. A sleep x time of day interaction was included to avoid forcing the acrophase (peak) and nadir (trough) of the sinusoidal "wave" modeled as the time effect to be exactly 12 hours apart. In addition, a position x activity interaction was included on the basis of findings by Marler et al. (40). We also report the results for a number of alternative models to examine what the effect of mood would be when not controlling for all other covariates. Furthermore, to illustrate how the longitudinal regression procedure can be used to model the effect of individual difference variables, we also present a model in which the subject’s mean arterial pressure taken by the technician was included as a covariate.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Frequencies of Covariates
Table 3 shows the prevalence of different behavioral covariates included in the standard model. Except for the sleep/wake behaviors, the results are percentages of daytime (awake) readings. The table lists (1) the average percentage per subject, (2) the proportion of subjects showing the behavior at least once, and (3) the proportion of subjects showing the behavior at least three times.
|
Table 3 also shows that subjects were awake during approximately 70% of the readings. During waking hours, they spent 45% of their time at home, 29% at work, 20% in other situations, and 7% driving or riding. They were alone 46% of the time, with one other person 30% of the time, and with two or more persons 24% of the time. Most of the time, 59%, was spent in the sitting position, followed by standing (31%) and lying supine (9.9%). Activity level was rated as low/none 60% of the time, followed by light (33%) and moderate (8%). The table also lists activity levels by position. Not surprisingly, the combination of moderate activity and supine position rarely occurred (0.1% and in only 6% of subjects).
Effects of Covariates Other Than Mood
Table 4 presents the estimates (expressed in mm Hg or bpm), of the effects on ABP of the various behavioral covariates in the standard model. For ease of presentation, standard errors of the estimates are not included. The first line in the table lists the intercepts (eg, 131.0/77.6 mm Hg for blood pressure). These represent estimates of sample ABP at the default states of the independent variables (ie, awake, at home, alone, not consuming, and mellow mood). From these data, ABP for a particular situation or behavior can be estimated (see footnote of Table 4). Intercept = ABP or HR at default states. The other numbers represent estimated deviations from the intercept. For example, for an average subject in the sample, estimated SBP while standing, with light activity while awake at 5 PM at work, and in a group of people, not eating, and in an elated/happy mood would be 131 + 3.6 + 0 + (-0.96 x sin 360 x 17/24) + (0.59 x cos 360/17/24) + 0.9 + 0.7 + 0 + 1.5 = 138.5 mm Hg. To estimate ABP for an individual subject, his random intercept, day effect, and time of day effect (not shown) would need to be added to the population estimate.
|
|
The position x activity interaction was significant for DBP and HR. To illustrate this interaction, Table 4 shows the estimates for all nine (3 x 3) combinations of levels of position and activity, although only four indicator variables (two for position and two for activity) were estimated directly by the model. The third level was calculated simply by subtracting the sum of the two indicator variables from zero, for example, Standing = 0 - (Sitting + Supine). For DBP, the interaction indicated a reverse activity effect in the standing position (ie, DBP was higher for no activity than for the other activity categories). For HR, the interaction was related to HR being relatively less affected by differences in activity when the subject was in the sitting position than in the other two positions.
The effects of time of day and sleep x time of day reflect blood pressure trends across time common for all individuals in the population. For DBP, these effects were not statistically significant. For SBP, the simple effect of time of day did not meet the Bonferroni-adjusted target significance level (0.005 < p < .01). However, the sleep x time interaction was significant. The effect of time on sleep blood pressure can be estimated by adding the main and interaction effects of time of day, that is, -(1.83 + 0.96)sin t + (0.59 - 0.3)cos t. This sleep time curve describes a decelerating declining trend throughout the night, with amplitude of 2.8 mm Hg and a nadir in the early morning hour (6:30 AM). HR showed significant time effects both during the waking and sleep states. Sleep HR showed a decelerating declining trend until 6:15 AM, with an amplitude of 5.4 bpm. Waking HR was associated with a peak (acrophase) at 5 PM and an amplitude of 2.6 bpm.
The effects of environmental settings were statistically significant for SBP, DBP, and HR. Surprisingly however, the difference between home and work blood pressure, when controlling for the other effects in the standard model, was only 0.9/0.6 mm Hg. The significant setting effects in the model were mainly related to blood pressure and HR increases occurring during the transportation settings. Riding increased blood pressure by 4.0/4.2 mm Hg and HR by only 1.9 bpm. Driving changed blood pressure by only 0.7/-0.3 mm Hg but increased heart rate by 4.1 bpm.
The effect of social setting was statistically significant for both SBP and DBP but modest in size. Compared with being alone, being with two or more persons increased blood pressure by 0.7/1.1 mm Hg. However, being with one person did not significantly affect blood pressure compared with being alone. Consumption was associated with significant increases in blood pressure and HR in the amount of 1.5/1.8 mm Hg and 2.6 bpm, respectively.
Effect of Mood
Standard Model.
The effects of mood were statistically significant for all three cardiovascular variables. Compared with the default mellow category, the anxious/annoyed category was associated with an increase in blood pressure of 2.8/2.2 mm Hg. Elated/happy was associated with an increase in pressure of 1.5/0.80 mm Hg. Moods in the attentive category were associated with an increase of less than 1 mm Hg. Bored/sad did not differ from that of the mellow mood category. Finally, moods in the disengaged/sleepy category were associated with significant decreases in blood pressure (3.2/1.5 mm Hg). The HR differences between mood categories show similar trends but were generally less pronounced (<1 bpm). The difference between highest and lowest mood effects was 6/3.7 mm Hg for blood pressure and 2.5 bpm for HR.
To further examine the relation between mood and cardiovascular responses, we performed pairwise comparisons among the various mood categories (Table 6 ). A p value of <.01 was the target significance level. The most obvious finding is that, with one exception, disengaged/sleepy was associated with significantly lower values compared with each of the other mood categories. In addition, it is noteworthy that the difference between anxious/annoyed and elated/happy was not significant for SBP and HR, although it showed a trend (p < .05) for DBP.
|
of the mood categories and the corresponding ABP estimates. The circular–linear correlation involves two numbers, (1) the acrophase (ie, the orientation of the linear dimension providing the best fit with the circular data) and (2) the Pearson product–moment correlation between this dimension and the blood pressure estimates. The circular–linear correlations (r) were 0.83 for SBP, 0.86 for DBP, and 0.81 for HR. The acrophases were 96° for SBP, 100° for DBP, and 74° for HR.
Mood Effects in Other Models.
To examine the effect of controlling for the nonmood covariates, we estimated the mood effects in alternative models with fewer such covariates. The results are shown in Table 7 . In most of the smaller models, the mood estimates were largely similar to the standard model. When position is excluded, however, the range (highest minus lowest) of the mood-related blood pressure estimates increased to 10.4/8.2 mm Hg, and the range of the HR estimates increased to 8.9 bpm. In all of the alternative models, the relative ranking of the different mood estimates remained essentially the same.
|
Validation Sample
The family history study included 12,004 blood pressure measurements from which 338 data points (2.8%) were excluded as statistical outliers. The estimates of the effects of the behavioral covariates are listed in Table 8 . As can be seen, ABP at the default states of the independent variables is 127.6/77.6 mm Hg, and 76.6 bpm ABP at other states can be computed in the manner shown in the footnote of Table 4. The statistical tests of the covariates are presented in Table 9 .
|
|
The estimates of the effects of mood were statistically significant and showed the same profile as those for the primary sample. That is, anxious/annoyed was associated with the largest increases (1.0/1.5 mm Hg and 2.1 bpm), followed by elated/happy and attentive, whereas disengaged/sleepy was associated with decreases (2.9/1.3 mm Hg and 1.4 bpm). Pairwise comparisons among moods (shown in Table 6) revealed that for all three dependent variables, disengaged/sleepy, with few exceptions, was significantly different from each of the other moods. Anxious/annoyed and elated/happy were not different from one another. The HR differences for comparisons involving the categories of mellow moods vs. the categories of elated/happy, attentive, or anxious/annoyed were statistically significant.
The circular–linear correlation (r) between the midpoints of mood categories and blood pressure responses was 0.84 for SBP, 0.91 for DBP, and 0.96 for HR. The corresponding acrophases were 68, 89, and 99°, respectively.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
A limitation of this study is that the CMS, like the underlying circumplex model of affect, does not permit recording of mixed mood states. The CMS also does not distinguish between the negative affects of anger vs. anxiety. Therefore, the results of this study could not address the possibly unique aspects of anger or anxiety (ie, those aspects of these emotions not covered by the general label of negative affect).
The effects of the other covariates besides mood deserve comment. Sleep was observed to have the largest cardiovascular effects among all the covariates. Sleep is generally associated with decreases in sympathetic nervous system activity and increases in parasympathetic activity (42). In addition, increases in parasympathetic activity at night might be associated with physical fitness and thus seem cardioprotective (43). The findings of a gradual decline in blood pressure during the night while sleeping that is followed by an abrupt increase in blood pressure on awakening are consistent with results obtained with intraarterial recordings (44). The effects of time of day during the day hours were nonsignificant or modest. This is consistent with results in the literature that led to the conclusion that diurnal blood pressure trends are environmentally determined rather than an expression of an internally driven circadian rhythm. For example, the diurnal blood pressure curve immediately entrains to new diurnal behavioral patterns with shift work (45, 46).
Posture was associated with large changes in expected directions. A reversal of the effect of activity in the standing position was found for DBP. Although not statistically significant at the target p level, the trend in the validation sample was the same, so the findings from the latter sample do not refute this reverse activity effect. The effect can be explained by the fact that higher activity in the standing position involves larger muscle groups, resulting in a relative decrease in peripheral resistance as activity increases (47).
In our study, we did not find significant home vs. work differences in cardiovascular responses. This seems to be at variance with results from studies in which the work and home settings are compared, which typically show increases in blood pressure in the work environment, at least for individuals who experience stress or job strain (48, 49). These studies did not include as many covariates as the standard model in the present study. The home and work settings are associated with differences in other covariates included in this study, particularly time of day but also possibly posture, activity, and mood, and our study assessed the setting effects that were independent of these other covariates. Larger differences between work and home ABP would be expected in studies in which these covariates were not included.
An interesting finding was the differential effect of driving vs. riding on HR vs. DBP. Whereas driving was associated with an increase in HR, riding was associated with an increase in blood pressure. This differential effect may reflect the difference between "active" coping during driving and "passive" coping during riding (50). Active coping responses have been shown to be associated with parasympathetic withdrawal and ß-adrenergically mediated cardiovascular responses (51). Passive coping responses have been shown to be associated with 
-adrenergically mediated increases, particularly in DBP (51).
The effects of social setting appeared to be smaller than those in two other studies (23, 52). Although both studies presented evidence that the social setting effect was not confounded with that of physical activity, they did not control for as many other covariates as we did in the present study. Furthermore, when designing the ambulatory diary for the present study, we constructed the social setting categories using labels (alone, with one, and with group) connoting a sense of affective neutrality. The larger effects found in these studies were for the setting categories of alone, family, friends, and strangers. These conditions may differ with respect to the moods that they induce, and the social setting effects in these studies may thus have been augmented by changes in mood.
The findings of the present study are relevant to the debate regarding the relative importance of the engagement vs. negative affect dimensions in determining ABP responses. If ABP maps onto the negative affect dimension, the highest blood pressures should have occurred within the CMS anxious/annoyed category, and the lowest, within the mellow category. However, the data do not support this logic. The lowest blood pressures occurred within the disengaged/sleepy CMS category, corresponding to the low end of the engagement dimension. Increases in blood pressure were about equally predicted by the CMS categories of anxious/annoyed, elated/happy, and attentive. The common denominator of these categories is a high level of engagement. This pattern suggests that the blood pressure responses obtained were related to the degree of engagement.
This pattern of mood-related ABP responses was expressed quantitatively with the acrophases of the circular–linear correlations between mood and ABP. The rotations of the engagement and negative affect dimensions relative to the CMS are 90° and 135°, respectively. We found that almost all of the acrophases were close to 90° (ie, the direction of the engagement dimension). The two acrophases that deviated from this pattern (75° for HR in the primary sample and 69° for SBP in the validation sample) did so in the direction of positive affect rather than negative affect. Our findings thus are congruent with other evidence supporting cardiovascular responses being determined most accurately by the engagement (or arousal) dimension (34).
If mood-related ABP responses reflect the degree of engagement, why did the blood pressure responses in the attentive category not exceed those within elated/happy and anxious/annoyed categories? This would have been the pattern most consistent with the location of the attentive category at the high engagement pole of the circumplex. The inconsistency suggests a problem in the upper region of the circumplex model of mood. Words such as "joyful," and "ecstatic," or "anxious," and "furious" appear to connote an even higher level of engagement or arousal than high-engagement but affectively neutral words such as "attentive." In other words, for the highest levels of engagement to be reached, an element of either pleasantness or unpleasantness seems to be necessary. Both the emotion words and the ABP responses would better be represented if the shape of the circle were changed. The change would have the upper lateral regions of the circle balloon upward while keeping the attentive midpoint fixed. As a result, the CMS anchor points alarmed/angry and euphoric/elated (see Figure 1) would move upward beyond the level of attentive. The final shape would thus resemble that of the upper border of a heart. Incidentally, a heart shaped transformation of the circumplex would also provide a better fit at the lower pole. In the mood circumplex, differences between two points on the engagement dimension should be proportional to the vertical distance between them. However, disengaged/sleepy was associated with ABP decreases that were out of proportion to the small vertical distance between this mood category and the mellow or bored/sad mood categories. A better fit with this pattern would occur if the midpoint of the lower pole of the circle were pulled downward into a shape similar to the tip of a heart.
The frequency distribution of the various mood categories showed that the largest percentages of recorded moods were rated as mellow or attentive. Approximately 5% of the reported moods were in the anxious/annoyed category. Given this infrequent occurrence, the notion that the experience of negative affect per se has a major role in the etiology of cardiovascular disease may have to be modified. Other explanations, such as inhibited negative affect or interactions with personality traits and/or relevant family history, may have to be entertained (53, 54).
If it is true that certain discrete moods place undue strain on an individual’s resources, what might those moods be? If it is the negative mood states that are specifically associated with cardiovascular risk, the fact that their pressor responses were not particularly high (relative to positive affect) would question the value of the reactivity hypothesis or at least the "recurrent activation" form of it (55). On the other hand, if the highly engaging or high arousal moods are the ones that constitute the health risk, why has the risk of excessive joy not been more generally emphasized?
Furthermore, behavioral factors such as posture and wakefulness were associated with responses that were larger than the mood effects. These and similar findings concerning the effects of exercise, for example, have challenged the notion that reactive responses constitute a risk factor simply because they are large. Metabolic appropriateness (56) may present as an explanation for the lack of risk associated with these behavioral factors. However, metabolic appropriateness would not necessarily explain the lack of risk associated with the cardiovascular responses to positive moods. The use of a structural model of mood in the present study serves to bring these discrepancies into focus.
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Received for publication September 3, 1997.
Revision received December 15, 1998.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  J. R. Jennings Autoregulation of Blood Pressure and Thought: Preliminary Results of an Application of Brain Imaging to Psychosomatic Medicine Psychosom Med, May 1, 2003; 65(3): 384 - 395. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. G. M. Vrijkotte, L. J. P. van Doornen, and E. J. C. de Geus Effects of Work Stress on Ambulatory Blood Pressure, Heart Rate, and Heart Rate Variability Hypertension, April 1, 2000; 35(4): 880 - 886. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
