| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
ORIGINAL ARTICLES |
From the University of British Columbia, Department of Psychology, Vancouver, British Columbia, Canada.
Address correspondence and reprint requests to Wolfgang Linden, PhD, Psychology/UBC, 2136 West Mall, Vancouver BC V6T 1Z4 Canada. E-mail: wlinden{at}psych.ubc.ca
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Methods: At baseline, BP and HR measurements were recorded during three 5-minute laboratory challenges and three 5-minute recovery periods after each challenge. Measurements of systolic BP, diastolic BP, and HR were collected throughout this baseline protocol and also at 3-year and 10-year follow up by ambulatory monitoring.
Results: After adjustment for traditional biologic predictors, reactivity was found to explain significant variance in follow-up data across all 3-year indices and two of the 10-year indices. Recovery, entered in a following step after reactivity, was found to explain additional significant variance across all 3-year indices but none of the 10-year indices. Family hypertension history data were not found to be significantly associated with reactivity or recovery nor were they predictive of longitudinal ambulatory data after adjustment for initial resting cardiovascular levels.
Conclusion: Overall, from a hierarchical regression model perspective, the data support the use of both reactivity and recovery in clinical predictions of proximal BP and HR and generally support the use of reactivity (but not recovery) in long-term BP predictions.
Key Words: reactivity • recovery • hypertension • blood pressure • ambulatory monitoring • family history
Abbreviations: BP = blood pressure; HR = heart rate; SBP = systolic blood pressure; DBP = diastolic blood pressure; CVD = cardiovascular disease.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Animal research has achieved considerable success in verifying associations among chronic stress, noted risk factors, and peripheral alterations characteristic of hypertension. It has been much more difficult, however, to verify such associations in humans. This is in large part the result of methodological obstacles such as difficulty administering chronic stress, controlling stress confounds, and monitoring long-term changes. Researchers have therefore directed their efforts at showing relations between laboratory stress-induced cardiovascular changes such as cardiovascular reactivity and cardiovascular recovery and future hypertension status given that such changes (when well elevated) may foreshadow later physiological permutations (3,4).
Cardiovascular Reactivity (‘reactivity’)
Reactivity is defined as the magnitude of elevation in an individual’s BP or heart rate (HR) in response to an aversive, challenging, or engaging laboratory stressor (4). It is conceptualized as a relatively stable individual trait characteristic and has been shown to achieve acceptable levels of reliability (surpassing the 0.8 mark) when values are aggregated across laboratory tasks (5).
From their extensive review of the literature, Treiber et al. concluded that heightened reactivity may be considered an independent risk factor for hypertension (4). A meta-analytic review by Fredriksen and Matthews also found that heightened reactivity to cognitive, emotional, and active physical laboratory stressors was related to hypertension development in longitudinal studies with follow-up intervals of up to 45 years (6). In such studies, reactivity typically explains 4% to 12% of BP variance after adjustment for standard clinical risk factors (7,8). In the well-known CARDIA studies, active tasks (i.e., video games) were concluded to be superior to passive tasks (i.e., cold pressor challenge) in predicting 5-year increases in BP (9). In a recent study that used a large battery of behavioral tasks, a follow-up interval of around 10 years, and ambulatory BP measurement, Tuomisto et al. found that only systolic BP (SBP) reactivity significantly improved prediction models (10). Although no subset of tasks was found to be superior in predicting long-term BP, reactivity from active tasks was found to be better in predicting need for antihypertensive medication.
Nevertheless, there is still considerable doubt about the predictive ability of reactivity because inconsistent findings are prevalent. However, these have generally come from studies having low statistical power as a result of smaller sample sizes or shorter follow-up intervals (4). Additionally, when positive associations are found in the context of well-designed studies, the value of reactivity data is still questioned because many researchers feel these data provide minimal increases in predictive power beyond what is provided by traditional predictors (11). In the Whitehall studies, for example, the clinical use of stress testing was questioned because only SBP reactivity to a problem-solving task significantly predicted 4.9-year and 10.8-year SBP, and it merely explained approximately 1% of the follow-up variance (11). Moreover, a recent study by Stewart and France found that although HR reactivity to a math task predicted an additional 4% of the variance in 3-year follow-up resting SBP, reactivity from neither a math task nor an active or passive physical task added significantly to the prediction of 3-year diastolic BP (DBP) (12). In a more recent study by these same authors assessing reactivity to cognitive tasks in an older population, none of the reactivity measures were predictive of 3-year SBP or DBP (13).
In summary, although there seems to be a significant degree of association between laboratory-induced reactivity to active stress and later BP, uncertainty remains over the degree and significance of this information in light of the variance explained by standard risk factors.
Cardiovascular Recovery (‘recovery’)
A less well-known aspect of the cardiovascular stress response that is receiving increased attention is recovery, which refers to the process by which cardiovascular parameters reverse their activation after termination of stress (14). In contrast to reactivity, recovery uniquely taps into the duration of the stress response. Heightened recovery values indicate delayed return to normal cardiovascular levels after stress termination, and such slow recovery can be considered a visible marker of increased vascular resistance. Like reactivity, it has shown acceptable levels of reliability when values are aggregated across multiple tasks (15).
Delayed recovery has been associated with hypertension, borderline hypertension (possibly highlighting an early risk marker role for this parameter (16)), and familial hypertension (12,17). In some studies, recovery has been found to be superior to reactivity in predicting longitudinal BP and HR (18). However, there has also been substantial ambiguity with regard to both the significance and superiority (overreactivity) of these noted associations in longitudinal predictions (19). Thus, Schwartz et al. recently concluded that there is weak overall evidence for the predictive power of recovery (20).
One likely cause of the ambiguity has been the inability of laboratory stressors to effectively stress people to the extent that poor cardiovascular responses, particularly slow recovery curves, are revealed. Several researchers have criticized the weak ecologic validity of commonly used laboratory stressors and, in turn, the poor generalizability of reactivity and recovery data to real-world cardiovascular functioning (20). However, there is some thought that interpersonal tasks, particularly those evoking anger and hostility, may overcome this weakness. Such tasks tend to be more representative of chronic life stressors implicated in cardiovascular disease (CVD) development (i.e., job strain, marital stress) than the other cognitive and physical tasks commonly used. Interpersonal tasks have been shown to yield elevated reactivity patterns as well as (particularly) delayed recovery patterns (21–23), the latter possibly reflecting their ability to elicit rumination, a maladaptive process that recovery is likely sensitive to. Findings from our laboratory have also shown that relative to other tasks, emotion-evoking interpersonal tasks display high reproducibility over time (24) and yield superior predictions of baseline ambulatory BP (25).
In summary, although promising findings linking reactivity and recovery profiles with future BP changes have emerged, ambiguity still remains. Methodological variability is one likely cause for this. Subject samples tend to vary in age and risk status, and laboratory stress packages frequently vary in their incorporation of physical, cognitive, passive, or psychosocial types of tasks. Measures of follow-up data alternate between being resting and ambulatory, and many studies fail to examine meaningful interactions.
Present Study
The present study builds on previous research by Stewart and France (12) and Stewart et al. (13). By incorporating more stringent methodological features, however, it offers unique strengths. One unique advantage of our study is that it examines an equal number of normotensives and normotensives at risk for disease (resulting from familial hypertension). This is important because many studies have shown that individuals at risk for disease tend to exhibit not only higher resting BP and HR, but also relatively worse cardiovascular reactivity and recovery profiles.
A second strength of our study is that it examines both young and older individuals. This is important given that discrepant findings have been observed across studies assessing different age cohorts. Third, the present study evaluates both 3-year and 10-year follow-up data, and it uses ambulatory monitoring to determine BP and HR end points. This form of measurement provides a better understanding of cardiovascular responses to stress, because sampling in the natural environment allows for aggregation of findings across multiple stimuli (5). Ambulatory measures have also been found to yield better long-term predictions of hypertension development than laboratory resting measures (26). Finally, in addition to using a standard physical and cognitive stress task, we are using a psychosocial task involving anger provocation to enhance ecologic validity. By using these distinct tasks, we hope to better capture real-life variability in stress responses.
Purpose, Objectives, and Hypotheses
The purpose of our study is to examine whether measures of reactivity to, and recovery from, laboratory-based challenges are clinically useful in predicting 3-year and 10-year ambulatory BP and HR among initially normotensive, community-dwelling adults and university students. The particular objectives of this study are to answer the following questions:
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
|
Procedure
Baseline testing involved a 2-hour block of laboratory assessment. Participants were encouraged to avoid ingesting alcohol, caffeine, or nicotine or exercising strenuously for the 2 hours before testing. Participants first completed questionnaire data on demographic and clinical risk factors. They then had their body fat levels assessed by a female experimenter using skinfold measurement. After this, participants sat alone during a 20-minute resting phase while BP and HR measurements were taken. After the final resting phase reading, task instructions were delivered over an intercom system and stress testing commenced. Specifically, participants engaged in a counterbalanced set of 5-minute behavioral stress tasks (i.e., mental arithmetic, handgrip testing, and an anger-recall discussion) during which time reactivity measurements were taken. Recovery was assessed during the 5-minute periods after each task during which time participants were instructed to read magazines. At 3-year and 10-year follow up, ambulatory BP and HR was assessed over the course of 10 to 12 hours and 24 hours, respectively. For these sessions, participants were asked to choose a typical day free of any specific stressors. Monitors were fitted to participants, pretested on the spot, and returned after monitoring for data analysis. During monitoring, ambulatory readings were obtained every 20 minutes.
Physiological Measurement
Blood pressure and HR information was collected in the natural environment using SpaceLabs 90207 ambulatory monitors (Labs Medical Inc., Redmond, WA), which weigh approximately 0.7 kg and are worn in a protective pouch. Use and accuracy of these monitors is supported by previous validation work (27). Participants were explicitly instructed to minimize physical activity during measurement cycles. Only approximately 10% of attempted measures were unusable, and these were identified by error codes. In the laboratory, BP and HR information was obtained by Dinamap 845 Vital Signs Monitors using oscillometry (Critkon Corporation, Tampa, FL). Previous validation work has shown Dinamap BP values to be highly correlated with intraarterial measurements (28).
Behavioral Laboratory Tasks
For the isometric handgrip task, participants were instructed to maintain handgrip tension on a standard dynomanometer (29) at 20% maximum for 3 minutes followed by 2 minutes at 30% maximum. For the mental arithmetic task, participants read a set of problems aloud from a television and then verbalized their answers. These problems were presented at 5-second intervals, and participants answered approximately 65% of these correctly. For the anger-recall discussion task, participants were given 2 minutes to recall an anger-provoking situation from their work or personal life after which time they discussed the event with a same-sex research assistant for 3 minutes (30). Cardiovascular readings were collected at 1.5 and 3.5 minutes during each exercise and recovery period. Additional details regarding the measurement, reliability, and predictive value of these tasks are reported in previous articles (24).
Measurement of Coronary Risk Factor Variables
A questionnaire was administered before stress testing to gather information on age, gender, cardiovascular medication use, smoking status, alcohol consumption, exercise frequency, and family hypertension history. A binary yes/no format was used to record information on cardiovascular medication use and smoking status. Information on alcohol consumption related to the approximate number of alcoholic drinks consumed per week, in which a drink was considered any glass of wine or hard alcohol or any bottle of beer or cider. Information on exercise frequency related to the approximate number of hours spent exercising per week, in which exercise was defined as any activity involving voluntary bodily movement in the context of a sport or workout environment that results in substantially increased energy expenditure. The family history item inquired about whether participants had a parent with hypertension; we did not discriminate whether they had one or two hypertensive parents as a result of reasons of statistical power. For measurements of daily stress, we used the Daily Stress Inventory (31), a 60-item self-report measure that describes a range of frustrating events and asks: a) whether or not the event has occurred within the past 24 hours, and if so, b) what was the stress severity on a 1 to 7 Likert scale. Lastly, body fat levels were determined using skin calipers. The amount of fat just under the skin was measured at six different sites on the body, and this test was executed twice and values were averaged for reliability before being summed (32). Although slightly less accurate than hydrostatic body fat measures, skin calipers are vastly more practical and provide more specific body fat data compared with other crude indices such as body mass index.
For descriptive purposes, similar information to that discussed here was collected again at 3-year and 10-year follow up with two exceptions. First, the baseline measure of daily stress was replaced by a measure of chronic disease at 3-year and 10-year follow up, in which participants were asked whether they had experienced chronic disease during the elapsed time since the previous testing session. This change was made so that a more representative indication of stress over the elapsed time period between measurement phases could be obtained. Second, there was no six-site skin caliper test performed at 10-year follow up.
Reactivity and Recovery Calculations
Both reactivity and recovery scores were computed for each cardiovascular index (HR, DBP, SBP) on each of the three stress tasks yielding a total of 18 stress response scores. We computed individual task scores by a) averaging the two readings obtained on each task and b) subtracting this average score from the respective mean resting level. The mean resting level was itself calculated by averaging the last two readings taken during the initial adaptation phase of the laboratory protocol.
Given that research by Kamarck has shown that the reliability of stress reactivity responses is improved with the aggregation of measures across multiple tasks (33), we also computed three aggregate reactivity values and three aggregate recovery values. Each aggregate value represented a summation of the (reactivity or recovery) scores obtained on a given cardiovascular index across the three individual stress tasks.
Although reactivity and recovery scores can be measured in a variety of ways, we considered simple change scores to be appropriate. Our lack of moment-to-moment recovery readings reduced the conceptual value of a slope measure, time to recovery measure, or measure using a curve-fitting estimation. Although residualized change scores have the advantage of explicitly adjusting for preexisting BP and HR differences in comparison groups, correlations between residualized scores and simple change scores are known to be extremely high (>0.9). Previous studies in our own laboratory as well as those of others have shown that these two types of scores yield similar findings and conclusions in regression analyses (34). Thus, we felt that we would not be sacrificing significant statistical power by using the more basic simple change scores.
Statistical Analyses
Justification of Aggregate Data
Pearson r correlations were computed between the aggregate data and the respective individual task data to determine the extent to which the aggregate data were representative of data from the individual tasks. Across both reactivity and recovery measures, data from the individual tasks correlated fairly highly with respective aggregate data with reactivity correlations ranging from 0.63 to 0.86 and recovery correlations ranging from 0.84 to 0.89. Although the lower end correlation of 0.63 is relatively weaker than desired, we nevertheless felt justified in using the aggregate data given that a) aggregate data generally demonstrate enhanced reliability (33) and b) our aggregate data appeared sufficiently representative of the individual task data. (Nevertheless, for reasons discussed later, supplemental post hoc regression analyses were conducted using individual task data, and the relatively weaker results of these, as discussed in the "Results" section, further justified our use of the aggregate data in place of the individual task data.)
Predicting Longitudinal Cardiovascular Data
We conducted a series of hierarchical regression analyses to determine the additional contribution of cardiovascular reactivity and recovery data to predictions of 3-year and 10-year ambulatory BP and HR after adjustment for standard clinical predictors. A total of six regression models were created, one for every cardiovascular outcome index (SBP, DBP, HR) at each of the two follow-up phases. For each:
Although our method of calculating recovery values (i.e., subtracting raw scores from the resting level average) increased their independence from reactivity values, our recovery values nevertheless remain somewhat dependent on our reactivity values. This is because reactivity scores effectively determine the starting point from which the recovery process commences. To determine the degree of interdependence between our (aggregate) reactivity and recovery scores, Pearson r correlations between these constructs were calculated (with correlations ranging from 0.68–0.71 for SBP, 0.67–0.74 for DBP, and 0.58–0.66 for HR). These correlations suggest that although our reactivity and recovery constructs are correlated at the 0.01 level, they nevertheless exhibit some independence and therefore offer some unique information. Because this is what is of crucial interest to us in our regression analyses, we felt justified in using these constructs in separate steps in our prediction models, regardless of some loss of power resulting from collinearity. This analytical strategy follows previously established traditions for evaluating the independent contributions of reactivity and recovery in predicting longitudinal BP (13).
It should also be noted that in all of our analyses, reactivity was always entered into the regression models before recovery. We felt this was most practical given that reactivity data will likely always be used because a) it will always be available when recovery data are collected, and b) it has already been concluded to be a valuable predictor of hypertension development. Thus, our key question is whether recovery data can provide useful information above and beyond reactivity data, and this is why recovery data were forced into the regression models in a separate step after the reactivity data.
Statistical Power
Because of the large sample at baseline, power levels to detect medium effect sizes met or exceeded 0.9 for tests of interrelationships with risk factor and psychological variables. However, as a result of smaller follow-up sample sizes, and the relative infrequency of high BP, power levels approximated 0.7 for detecting medium effect sizes in regression analyses.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
2 and analysis of variance techniques were used to assess significant differences across the samples on all other categorical control variables. Of note, the 3-year sample had a moderately (2 = 3.084, p = .07) higher ratio of FH+ individuals (i.e., individuals having a family history of hypertension) to FH- individuals.
Cardiovascular Variables
Table 1 also displays average resting and ambulatory BP and HR values across our three sample sets. Analyses of variance in this regard revealed that all baseline resting cardiovascular data were significantly lower than 3-year and 10-year follow-up resting data. Interestingly, the 10-year resting HR data were also found to be significantly lower than the 3-year resting HR data. Similarly, 10-year ambulatory DBP and HR were significantly lower than the respective 3-year measures (F = 7.4, p < .05 and F = 13.2, p < .001, respectively).
Participant Attrition
Given that there was a high participant dropout rate at 3-year (62%) and 10-year (65%) follow up, analyses were performed to determine whether the "dropouts" significantly differed in any way from those remaining in the study. Independent samples t-tests and 2 tests were used to assess differences in demographic and lifestyle baseline data at each follow-up period. Significant findings revealed that study completers at both follow-up dates tended to be older (t = 4.2, p < .001 and t = 3.5, p < .001, respectively) and more likely to have a family history of hypertension (2 = 6.95, p < .05 and 2 = 4.05, p < .05, respectively). Additionally, 10-year sample completers tended to have a significantly higher resting SBP (t = 1.98, p < .05). When independent samples t-tests were used to compare aggregate reactivity and recovery data across dropouts and completers, only one of the 12 comparisons revealed a significant difference (t = 2.0, p < .05), which showed the 3-year dropouts to have significantly lower aggregate SBP reactivity scores.
Correlations Between Control Variables and Aggregate Reactivity and Recovery Data
Pearson r correlations were computed to assess relationships between baseline reactivity and recovery data and control variables measured at each of the three testing sessions. As a result of the numerous correlations computed, significance was set at .01 to adjust for type 1 error inflation. Analysis of the correlation matrix revealed that heightened reactivity was significantly associated with male gender, lower age, and (surprisingly) nonsmoking status, whereas delayed recovery was only found to be significantly associated with male gender.
Family Hypertension History
As previously mentioned, one of our study goals was to assess whether at-risk status significantly associates with reactivity and recovery data, and thus whether it may be useful in improving predictor models of longitudinal BP and HR. From the previous paragraph, we know that family history status did not significantly associate with reactivity or recovery. From further analyses that examined stress response curves as well as total area under the curve data across FH+ and FH- individuals, it became clear that: a) FH+ individuals consistently display higher resting BP and HR, and b) FH+ and FH- individuals show negligible differences in reactivity and recovery information when data are independent of resting cardiovascular levels.
Determination of Control Variables Used in Hierarchical Regression Models
Two-step regression models were constructed to determine which control variables explained unique variance in longitudinal BP and HR beyond what was already explained by resting BP and HR. Variables found to explain significant unique variance (at the 0.05 level) were retained for subsequent hierarchical regression analyses, whereas the rest were excluded to avoid weakening analytical power, a justification method borrowed from previous research (10,13). The variables found to be significant are listed in step 1 of the regression models in Tables 2 and 3. Although some of these variables correlated significantly with each other (specifically, age and body fat, r = 0.24, p < .01; body fat and exercise frequency, r = –0.14, p < .05; and exercise frequency and smoking status, r = 0.14, p < .01), none of these correlations were above the 0.25 level and so they were not excluded from the main regression analyses. In contrast, (student versus community) location, a variable found to be predictive in the 10-year DBP model, did correlate highly with age (r = 0.74, p < .01), and so it was excluded from future regression analyses to avoid redundancy.
|
|
Predictions of 3-Year Heart Rate and Blood Pressure
Table 2 presents the results of the 3-year hierarchical linear regression analyses. Across each step of all three models, predictions are highly significant (p < .001). In addition, step 1 is consistently found to explain the highest proportion of ambulatory (BP or HR) variance.
Significant improvement to all three predictor models is also consistently observed after the addition of reactivity in step 2 as well as recovery in step 3. Moreover, across all three models, reactivity and recovery are noted to predict ambulatory variance in the expected direction (i.e., larger stress responses predict higher ambulatory levels).Overall, the DBP model has the highest predictive power at 44% explained variance.
Predictions of 10-Year Heart Rate and Blood Pressure
Table 3 presents the results of the 10-year hierarchical linear regression analyses. Like with the 3-year data, all three models are highly significant (p < .001) in their predictions at each of the three steps of the model. Moreover, reactivity and recovery are once more observed to consistently predict ambulatory variance in the expected direction, and step 1 again appears to explain a highly significant proportion of ambulatory variance. Furthermore, the DBP model shows the highest overall predictive power at 44%.
In contrast to the 3-year data, only the DBP and SBP models show reactivity significantly improving ambulatory predictions. Moreover, recovery fails to explain significantly unique variance in any of the three 10-year models.
To determine the extent to which the 12 participants on antihypertensive medication affected our 10-year results, a further set of analyses was conducted with these individuals removed. Similar trends were observed, with the DBP model showing the highest overall predictions, the reactivity data showing significant contributions to the BP models, and the recovery data failing to contribute significantly to any of the models. For the DBP and SBP models, the variance explained by reactivity was higher (by 2.2% and 2.8%, respectively), and the total variance explained by the models increased (by 5.6% and 4.9%, respectively). Interestingly, the HR model showed a 6.7% reduction in total explained variance. Regardless of these minor differences, these analyses confirm that the inclusion of those 12 individuals on antihypertensive medication did not significantly affect our conclusions.
To determine whether our alternating use of 12 hours versus 24 hours ambulatory data (as dependent variables for the 3-year and 10-year follow-up periods, respectively) greatly affected our analyses, the 10-year models were recreated exclusively using daytime 12-hour ambulatory data. In these revised models, lower overall predictions were observed for the DBP (7.4% reduction), SBP (0.7% reduction), and HR (5.2% reduction) data. Moreover, reactivity was no longer significantly predictive in the SBP model. It therefore appears that using the 24-hour ambulatory data at 10-year follow up improved, rather than reduced, the efficiency of our predictor models.
Using Equivalent Control Variables
When analyses were conducted using equivalent control variables across both models for each index (first using the 3-year control variables, then using the 10-year control variables), no major differences were found. Across four of the six models, the total explained variance declined (with reductions ranging from 6% for the 3-year SBP model to 1.9% for the 10-year SBP model). Although both the 3-year HR model and 10-year DBP model showed an increase in total explained variance, these values were minor (1.6% and 0.6% improvement, respectively). It therefore appears that our choice to use different control variables across our six models did not significantly weaken our statistical power.
We also performed analyses in which gender and family history data were exclusively used as control variables across all six models. Across five of these models, gender, family history status, and the interaction of these variables with reactivity or recovery did not significantly contribute to predicting explained variance. Not surprisingly, the 10-year HR model was the exception because it showed gender to contribute significantly. In terms of total explained variance, similar trends to those observed in our original six models were seen. Although half of these models displayed slightly reduced predictions and half displayed slightly increased predictions, the sizes of these changes were minor. The one notable difference is that recovery no longer explained significant unique variance in the 3-year HR model (p = .054). Overall, these analyses suggest that using gender and family history data in predictor models does not significantly improve their usefulness.
Role of Interactions Between Stress Response Variables and Control Variables
Across all six of our original models, first-order interactions (between stress response data and control variables) were entered in step 4 but were found to be insignificant in enhancing predictions. This suggests that the use of reactivity and recovery in predicting longitudinal ambulatory data does not vary on account of such demographic and risk factor information.
Ability of Individual Stress Data to Predict Longitudinal Blood Pressure and Heart Rate
Given that our anger-recall discussion task was consistently found to yield the largest BP stress response data, and as a result of the fact that there is speculation in the literature that interpersonal tasks may be superior in eliciting vulnerable cardiovascular response profiles, a further set of regression models were constructed (see Table 4).
|
In these, individual stress task data were used in place of aggregate composites in our six original models. Results indicated that individual stress task data did not generally improve the longitudinal BP and HR predictions. Of the 18 models created, only three yielded larger overall predictions ranging from 0.2% to 1.5% improvement. Interestingly, not one of the six models pertaining to the anger-recall discussion task was responsible for any of these improvements.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Strengths of Our Study
There were several noteworthy features to our study, namely prospective examination across both 3-year and 10-year follow up, incorporation of ambulatory monitoring, and use of three distinct stress tasks. Two other unique features were that we started with equal amounts of men and women as well as equal numbers of FH+ and FH– individuals, enabling sufficient power for assessing moderator effects. This is the first study of its kind to begin with equal-sized cells on these critical participant characteristics. Additionally, we enhanced data reliability by aggregating scores across individual laboratory tasks. This was particularly beneficial given that reactivity and recovery data are expected to change somewhat over time. Moreover, predictions yielded by our aggregate data were found to be superior to individual task data predictions, further justifying our use of the aggregate data.
Key Findings
Our study revealed several interesting findings. First, across all six models, the predictions yielded by the (step 1) traditional predictors fell in the 29.4% to 38.8% range for explained variance. This overlaps with previous studies in which findings were also in the 30% range.
Second, both reactivity and recovery significantly improved 3-year prediction models, thus supporting the use of these constructs in predicting proximal BP and HR changes. The degree of model improvement produced by reactivity agrees with the current literature, which typically reports improvements in the 4% to 12% range (8). Our 3-year recovery findings, in contrast, are particularly impressive given the conservative nature of our regression models (stemming from the fact that we examined the predictive use of recovery above and beyond reactivity).
Third, our results generally support the use of reactivity data in predicting long-term cardiovascular changes, but do not support the use of recovery data in this regard. Although reactivity was significantly predictive when added in step 2 of the BP predictor models, this was not the case for the HR predictor model nor was it the case for any of the predictor models when recovery was additionally incorporated. The reactivity BP findings are in agreement with previous conclusions in the literature (4,6) that have noted such relationships to hold over long periods of time. (We are not overly surprised or concerned by the relatively weaker findings of the HR data given that the significance of this cardiovascular index for hypertension development is unclear (36)). With respect to the recovery findings, this is the first study to assess long-term hierarchical predictions of recovery above and beyond predictions by reactivity. Our findings suggest that recovery is not a powerful additional predictive tool in long-term predictions. We speculate that failing to find a significant additional effect of recovery on 10-year BP is the result of a combination of reasons: a) limited statistical power, especially in light of our conservative analytic strategy; b) the influence of reactivity on subsequent recovery; and c) the much greater opportunity for other changes affecting BP to accrue over 10 rather than 3 years, which may include weight or exercise changes, changes in allostatic load, or maturation of the body’s self-regulatory capacity. Because this is speculative, further research needs to explore the processes by which recovery works to affect BP over time.
Fourth, although we found significant differences in resting cardiovascular data between FH+ and FH– individuals, we did not find significant differences in their reactivity or recovery data. Thus, stress response data do not appear to be useful in determining which FH+ individuals are more likely to develop hypertension.
Lastly, the interpersonal (anger-recall) task appeared to be a particularly effective stressor because it was found to yield the highest reactivity and recovery scores across both BP indices and it produced the highest individual task-aggregate composite correlations for SBP and DBP reactivity and HR recovery. Clearly, using this task was beneficial in maximizing the predictive power of our aggregate data. When used independently in regression analyses, however, neither this task nor the handgrip or arithmetic task proved superior to the aggregate composite in predicting longitudinal BP and HR. This is not surprising given that aggregate data are likely more clinically meaningful than elevated responses to any one task.
Study Limitations
A few limitations to our study can be noted, which may have played a role in weakening power in our analyses. First, our failure to control for caffeine intake or body posture during ambulatory monitoring may have added error to our data, because these factors are known to influence BP. It is possible that the declines in average ambulatory BP and HR observed across the 3-year to 10-year follow-up periods were attributable to participants becoming increasingly more sedentary in their daily activities, perhaps as a result of entering more time-consuming senior positions. Because we did not control for body posturing, we have no way of proving this. Alternatively, these changes may be explained by declining stimulant hormone levels (e.g., stress hormones, sex hormones, growth hormone, thyroid hormone), which typically occurs with increasing age, particularly around the age of 40. These changes tend to produce generalized physiological slowing, including decreased metabolic rate and ultimately reduced BP and HR. This hypothesis also offers a reason for why resting HR, specifically, showed a significant average reduction from 3-year to 10-year follow up; intuitively, it seems reasonable that the pacemaker muscles of the heart are more sensitive to age-related physiological changes than the blood vessels. Perhaps if more time had surpassed between follow-up periods, significant declines would have also been observed on the resting BP indices, thus paralleling the (more variable) ambulatory data changes.
Another limitation to our study is that our use of the handgrip task as a physical stressor may have prevented us from observing more significant relationships, because other similar studies in the field have tended to use more demanding physical tasks such as bicycle exercise. In addition, our decision to aggregate across tasks may have weakened our recovery protocol, because only one of our three stress tasks (namely, anger-recall) yielded slow recovery. It is possible that if we had used more challenging laboratory stressors (i.e., bicycle exercise, a timed mathematics examination, a public speaking presentation), we would have produced slower (and more vulnerability-revealing) recovery profiles and, in turn, found recovery to explain more variance in long-term cardiovascular predictions. Nevertheless, these limitations are not very threatening to our conclusions because they actually render our significant findings more conservative. Moreover, we felt it was important to aggregate because we wanted to provide data offering unique information relative to similar studies (12,13) that have only analyzed individual task predictions. Moreover, although psychosocial stressors have been reported to be superior in cardiovascular stress testing as a result of increased ecologic validity, this was not found to be the case in our study because none of the anger-recall task data yielded superior predictions to the aggregate data. In our study, aggregating was also useful because it improved the reliability of our data, made it more generalizable (as a result of the greater variability sampled), and was found to generally yield superior BP and HR predictions to the individual task data.
A further limitation of our study relates to the association between reactivity and recovery. This dependency creates some redundancy in our regression models and thus prevents us from making conclusions about the total predictive ability of recovery. Our recovery step, therefore, speaks to the unique additional explanatory power of recovery beyond reactivity. As earlier mentioned, we feel these findings are informative because we believe reactivity data will always be collected and used in the realm of recovery data assessment. Moreover, the approach we used to calculate recovery was specifically decided on the basis of wanting to parallel other methodological designs such as that by Stewart et al. (13) to enable more standardized comparisons between study findings. Had this not been an objective, we likely would have used other techniques (such as piecewise latent growth curve modeling) to reduce the redundancy in our models and thereby improve our analytical power.
Finally, our statistical analyses lost power for several reasons. First, relatively few participants (i.e., 20%) developed hypertensive disease over the course of the study. In addition, participant attrition resulted in much smaller follow-up samples than existed at baseline. Analyses did reveal that study completers participating in both the 3-year and 10-year testing dates tended to be older and more likely to have a family history of hypertension, completers in the 10-year sample tended to have significantly higher resting SBP, and 3-year dropouts tended to have significantly lower aggregate SBP reactivity. However, closer analysis reveals that these significant differences did not seriously threaten study generalizability. The significant differences found with respect to age, resting SBP, and 3-year aggregate SBP reactivity are all likely clinically insignificant given the relatively small size of the raw average differences. Completers and dropouts only differed on average by 3.74 years at 3-year follow up and by 4.82 years at 10-year follow up. Moreover, age was found to explain relatively little variance in all the final predictor models. With respect to resting SBP, the 10-year completers and dropouts differed on average by only 2.8 U. As for 3-year aggregate SBP reactivity, dropouts only differed from completers by 4.3 U. Although statistically significant, these difference are likely not clinically significant given the natural variability in BP measurements, particularly SBP measurements, that tends to occur throughout the day. The family history differences are also nonthreatening on account of two reasons; 1) "at-risk" status was not found to be significantly associated with longitudinal cardiovascular data in our preliminary analyses, and 2) the family history differences between study completers and dropouts did not translate into significant differences across the three samples. Thus, it appears that our high attrition rate did not significantly change the general composition of our study sample and thus did not seriously threaten the generalizability of our study findings.
Value of Study and Future Directions
Despite the limitations noted previously, our findings are informative. If cardiovascular reactivity and recovery can significantly improve prediction models of longitudinal BP and HR (as increasing evidence suggests) as well as disease status and mortality, then their utilization in standard clinical practice may be warranted (10). Through such improved screening methods, vulnerable individuals may be more easily identified and directed toward appropriate treatments. Future research should aim at replicating our findings as well as performing similar analyses using disease status (i.e., hypertension, CVD) as an outcome variable. An evaluation of the degree to which predictor models are improved through using ambulatory (rather than resting) cardiovascular data would also provide valuable insight; this is our next research focus.
Lastly, reactivity and recovery information is valuable because it extends beyond the cardiovascular system to other bodily systems (i.e., endocrine, immune) and chronic diseases. It is possible, for example, that combining reactivity and recovery information across multiple bodily systems may provide superior information for use in CVD predictor models. Alternatively, stress response data from other bodily systems (i.e., blood glucose, muscle tension) may be used to improve predictor models for other types of disease in which stress is known to play a role (37).
|
NOTES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Received for publication June 28, 2005; revision received June 15, 2006.
DOI:10.1097/01.psy.0000238453.11324.d5
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  O. V. Shcheslavskaya, M. M. Burg, P. S. McKinley, J. E. Schwartz, W. Gerin, C. D. Ryff, M. Weinstein, T. E. Seeman, and R. P. Sloan Heart Rate Recovery After Cognitive Challenge Is Preserved With Age Psychosom Med, February 1, 2010; 72(2): 128 - 133. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. A. Scheuer Regulation of the stress response in rats by central actions of glucocorticoids Exp Physiol, January 1, 2010; 95(1): 26 - 31. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. A. Sbarra, R. W. Law, L. A. Lee, and A. E. Mason Marital Dissolution and Blood Pressure Reactivity: Evidence for the Specificity of Emotional Intrusion-Hyperarousal and Task-Rated Emotional Difficulty Psychosom Med, June 1, 2009; 71(5): 532 - 540. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
