Introduction

In patients undergoing both exercise treadmill testing (ETT) and exercise stress single photon emission computed tomography (SPECT), myocardial perfusion imaging (MPI), blood pressure (BP) response (the difference between peak stress and resting BP) have prognostic value, independent of other clinical and MPI variables.1,2 In contemporary practice, a large proportion of patients are unable to exercise, and in these patients, pharmacologic vasodilator SPECT or positron emission tomography (PET) MPI provides a useful alternative for risk stratification and for diagnosis of coronary artery disease (CAD).3 However, there is a scarcity of literature describing whether BP response has the same independent prognostic value in patients undergoing vasodilator PET MPI as it has in patients undergoing ETT or exercise SPECT MPI.4–6 A prior study has previously demonstrated that heart rate response is a significant independent predictor of cardiac death in patients undergoing vasodilator PET MPI,7 but blood pressure response was not studied.

We utilized data from the Positron Emission Tomography (PET) Prognosis registry in order to address this gap in knowledge. The PET Prognosis Registry is a prospective, multicenter registry that has reported prognostic variables in a large set of patients that underwent vasodilator rubidium (Rb)-82 stress PET MPI for the evaluation of suspected myocardial ischemia. We sought to achieve the following objectives: (1) Examine the relationship between mortality and BP response, i.e., peak stress minus resting systolic blood pressure (∆SBP), and peak stress minus resting diastolic blood pressure (∆DBP); (2) Examine the relationship between mortality and both resting systolic blood pressure (rSBP) and resting diastolic blood pressure (rDBP); and (3) Determine the prognostic value of BP response (∆SBP and ∆DBP) and resting BP (rSBP and rDBP) after adjusting for both clinical and MPI factors. We hypothesized that BP response would have added prognostic value independent of resting BP, clinical factors, and MPI findings in patients undergoing vasodilator stress Rb-82 PET MPI.

Methods

PET Prognosis Registry

The main results from the PET Prognosis Registry have been previously published.8 The PET Prognosis Registry enrolled 7061 consecutive patients undergoing pharmacologic stress Rb-82 PET MPI with either dobutamine or vasodilator stress. This analysis included the subset of 3413 patients who were documented to have undergone vasodilator stress with adenosine or dipyridamole (those stressed by dobutamine or by an undocumented stressor were excluded), as previously described.4 All participating centers obtained institutional review board approval for the stress imaging procedures and follow-up methods.

Collection of Past Medical History Data

Data collection for relevant clinical history variables was conducted in a standardized fashion among all centers.8 Study staff collected data on demographic and clinical factors including gender, age, height, weight, diabetes, hypertension, dyslipidemia, smoking, prior coronary revascularization procedures, and history of myocardial infarction. Ongoing uses of medications including beta blockers, calcium channel blockers, nitrates, ACE inhibitors, and diuretics were also recorded in the database.9–11

MPI Protocol and Interpretation

Study participants underwent standardized PET MPI protocols as per the guidelines of the American Society of Nuclear Cardiology.12 Patients were instructed to fast for 6 hours and withhold caffeine-containing beverages (24 hours before the test) and antianginal medications (beta blockers, calcium blockers, and nitrates) on the morning of the test.9 Patients underwent Rb-82 MPI PET using dedicated PET (ECAT ART; Siemens-CTI, Knoxville, TN; Posicam HZL/R, Positron Corporation, Houston, TX) or hybrid PET/CT cameras (Discovery Rx or STE Light Speed (16, 64 slice CT), GE Healthcare, Milwaukee, WI; Biograph 64, Siemens, Knoxville, TN). Infusion of 20-60 mCi Rb-82 occurred during both rest and stress. Vasodilator stress was induced by either intravenous dipyridamole (142 mcg·kg·minutes for 4-5 minutes) or adenosine (140 mcg·kg·minutes). Peak stress blood pressure measurements were obtained at or around the time of Rb-82 injection depending on site-specific protocols. This was approximately around the 7-minute mark for patients receiving dipyridamole and at the mid-point of the infusion for patients receiving adenosine.7,8 Myocardial perfusion images were interpreted using a standardized approach, including the 17 segmental scoring system. Based on the 17-segment scores, the summed rest and stress scores were calculated and categorized into <5%, 5-9.9%, and ≥10% abnormal myocardium.12,13

Blood Pressure Protocol and Blood Pressure Response

Resting and peak stress systolic and diastolic BP were measured in millimeters of mercury (mmHg) per study protocol using an automatic BP cuff with digital display or manual cuff. BP measurements were obtained at baseline and every 2 minutes during pharmacologic stress and recovery. Manual or automatic monitoring continued until the end of stress image acquisition, at which point the patient was removed from the PET gantry. A final BP with a 12-lead ECG was obtained before the patient left the imaging suite. For each patient, we defined ∆SBP as peak stress minus resting systolic BP, and ∆DBP as peak stress minus resting diastolic BP. Resting and maximum heart rate measurements were also recorded before and following intravenous vasodilator stress.

Follow-Up Methods

Detailed descriptions of the follow-up methodologies have been previously reported.8 The primary endpoint for this analysis was all-cause mortality for an average follow-up of 1.9 years with a maximum of 5.1 years. The timing of all-cause mortality was documented for all patients. We chose all-cause mortality (instead of cause-specific death) as a primary endpoint because cause of death may not have been classified correctly in all patients, and using all-cause death would potentially avoid misclassification bias.14 Prior studies have also demonstrated that the majority of deaths in patients with suspected coronary artery disease (such as those that might be referred for vasodilator stress Rb-82 PET MPI) are actually cardiovascular in nature, hence we considered all-cause death a reasonable endpoint.15–17

Each institution arranged a scripted telephone interview with the patient or family member. This was supplemented with review of patient’s electronic medical record or by confirmation by the patient’s primary care physician. For US centers, in order to confirm patient’s survival status during the study time period, the national death index was queried. The follow-up time period was, on average, 1.9 ± 0.9 years. Survival analysis was performed by documenting the number of patients at risk during each year of follow-up.7,8

Statistical Methods

Descriptive statistics were used to generate histograms of our four blood pressure variables: ∆SBP, ∆DBP, rSBP, and rDBP. Even though the focus of this analysis was on BP response (∆SBP and ∆DBP), we also evaluated the association of resting BP (rSBP and rDBP) with mortality to provide context to the BP response variables. We then created categories for ∆SBP (≤−20, >−20 to ≤−10, >−10 to ≤0, and >0 mmHg); ∆DBP (≤−10, >−10 to ≤0, and >0 mmHg); rSBP (<90, ≥90 to <120, ≥120 to <140, ≥140 to <160, and ≥160 mmHg); and for rDBP (<60, ≥60 to <90, and ≥90 mmHg). These categories were chosen because they were clinically relevant and easy to remember, while also providing for a sufficient number of patients in each category for statistical analysis. We then described baseline characteristics among the four groups of ΔSBP, and separately among the three groups ΔDBP, using ANOVA to compare continuous variables and Chi-square tests to compare categorical variables across groups.

We plotted all-cause mortality rates by categories of each BP variable: ∆SBP, ∆DBP, rSBP, and rDBP. For each plot, we showed both crude mortality by BP cut points, as well as the predicted mean mortality calculated from univariate Cox models. We then generated univariate and multivariate models using Cox regression analysis. Covariates that had univariable P values <.10 were entered into the multivariable model. Covariates included in the final multivariable analysis were as follows: age, sex, history of diabetes, history of hypertension, home use of beta blocker, home use of diuretics, home use of lipid-lowering agents, history of prior MI, history of prior revascularization, resting heart rate, heart rate response (HRR), percent abnormal myocardium at rest, percent ischemic myocardium, systolic blood pressure response (∆SBP), resting systolic blood pressure (rSBP), and resting diastolic blood pressure (rDBP). Diastolic blood pressure response (∆DBP) was not included in the final multivariable model because it did not predict mortality in univariate analysis (P = .96). Unadjusted and adjusted hazard ratios and 95% confidence intervals were presented for each blood pressure group. The proportional hazards assumption was checked both visually and through the use of residuals. We then repeated the univariate and multivariate Cox models using continuous (instead of categorical) variables for each of the four BP variables, and found no meaningful difference in the results. We used SPSS 21 statistical software for most of the statistical analyses. SAS® statistical software was used to generate Martingale residuals to confirm the linear fit of the BP variables. We also examined all our variables for co-linearity utilizing variance inflation factors, which indicated no significant co-linearity of any of the variables entered into our Cox models. An alpha of 0.05 was used to designate statistical significance for covariates entered in the multivariable model.

Results

Baseline Characteristics

We examined baseline characteristics of this cohort both by categories of ∆SBP (Table 1) and ∆DBP (Table 2). Patients with the greatest drop in systolic BP with stress (∆SBP ≤ −20 mmHg) tended to be older than patients with a less marked drop in systolic BP (P < .001, Table 1). This relationship was also seen by categories of ∆DBP, i.e., patients with the greatest drop in diastolic BP with stress (∆DBP ≤ −10 mmHg) also tended to be older (P < .001, Table 2). Patients who had the greatest drop in systolic BP and diastolic BP also had higher prevalence of hypertension (P < .001 across both ∆SBP and ∆DBP categories). Those with the greatest drop in systolic BP and diastolic BP also had a lower prevalence of smoking history (P < .001 across both ∆SBP and ∆DBP categories). As expected, patients with the greatest drop in systolic BP (∆SBP ≤ −20 mmHg) had higher resting systolic and diastolic BP values, and lower stress systolic and diastolic BP values, compared with patients in the ∆SBP > 0 mmHg group (P < .001, Table 1). A similar relationship for resting and stress systolic and diastolic BP values was seen across the groups of ∆DBP (P < .001, Table 2). There was no significant association between either ∆SBP or ∆DBP and percent abnormal resting myocardium. Patients with a greater drop in systolic BP did have a higher percent of ischemic myocardium compared with patients with a lesser drop in systolic BP (P < .001, Table 1). However, there was no association between ∆DBP and percent ischemic myocardium (P = .296, Table 2).

Table 1 Clinical descriptors of the 3413 patients referred for stress myocardial perfusion Rb-82 PET by peak stress minus resting systolic blood pressure categories
Full size table
Table 2 Clinical descriptors of the 3413 patients referred for stress myocardial perfusion Rb-82 PET by peak stress minus resting diastolic blood pressure categories
Full size table

Baseline characteristics across categories of rSBP and rDBP are also presented (Supplementary Table 1 and Supplementary Table 2). With increasing rSBP, age and the proportion of female patients tended to increase (P < .001 for both variables, Supplementary Table 1). In contrast, with increasing rDBP, age and the proportion of female patients tended to decrease (Supplementary Table 2). Increasing categories of rSBP and rDBP were both associated with lesser degrees of abnormal myocardium at rest (P < .001 for both rSBP and rDBP). However, only increasing categories of rDBP were associated with percent ischemic myocardium (P = .049), whereas rSBP was not (P = .983).

Mortality and Blood Pressure Response

We plotted crude mortality and mean predicted mortality from the Cox survival functions by groups of both change in systolic BP (∆SBP) and change in diastolic BP (∆DBP) (Figure 1). Both the crude mortality and predicted mortality curves were similar. For ∆SBP, mortality was highest in the first group (≤−20 mmHg) but lower in the subsequent three groups. For ∆DBP, there was no appreciable difference in mortality across the groups.

Figure 1
figure 1

Peak stress minus resting blood pressure mortality curves. Crude mortality (blue bars) and mean predicted mortality from the Cox survival function (green line) for systolic blood pressure response (∆SBP) (A) and diastolic blood pressure response (∆DBP) (B). Blood pressure response is calculated as the peak stress minus the resting (pre-vasodilator infusion) blood pressure

Full size image

Mortality and Resting Blood Pressure

Figure 2 shows plots of both crude mortality and mean predicted mortality from the Cox survival functions by categories of rSBP and rDBP. For rSBP, crude rates of mortality were 9.1% in <90 mmHg group, 9.1% in the ≥90 to <120 mmHg group, 8.3% in the ≥120 to <140 mmHg group, 5.3% in the ≥140 to <160 mmHg group, and 9.3% in the ≥160 mmHg group. For rDBP categories, crude mortality rates were 11.9% in the <60 mmHg group, 7.0% for the ≥60 to <90 mmHg group, and 7.2% for the ≥90 mmHg group. Similar to BP response, the crude and mean predicted mortality curves were similar in shape for the rSBP and rDBP curves. Only the lowest category of diastolic blood pressure was associated with an increase in mortality.

Figure 2
figure 2

Resting systolic blood pressure mortality curves. Crude (blue bars) and predicted (green line) mortality curves for resting systolic blood pressure (rSBP) (A) and resting diastolic blood pressure (rDBP) (B), i.e., prior to vasodilator infusion

Full size image

Univariable and Multivariable Cox Models for Mortality by Blood Pressure Response and Resting Blood Pressure Categories

After median 1.9 years follow-up, 270 deaths from any cause were observed. In univariable analyses, ∆SBP (P = .082), rSBP (P = .008), and rDBP (P < .001) were of potential prognostic value (using P < .10 as the cutoff for entry into multivariable analysis), but ∆DBP was not (P = .96) (Table 3). After adjustment for other clinical and MPI variables, ∆SBP was no longer an independent predictor (P = .287), while rSBP (P = .026) and rDBP (P = .045) remained independent predictors of mortality (Table 3). In the adjusted models, the hazard ratios for categories of rDBP were consistent with the shape of the crude and predicted mean mortality curves. The rDBP group of <60 mmHg observed a hazard ratio of 1.77 (95% CI 1.35-2.32, p = <.001) for all-cause mortality compared with the reference group of ≥60 to <90 mmHg. There was no difference in mortality for the ≥90 mmHg group of rDBP (adjusted HR 1.00, 95% CI 0.63-1.59, P = .99) compared with the reference group (Table 3). For rSBP, compared with the reference group of ≥120 to <140 mmHg, most BP categories were associated with adjusted hazard ratios >1.0, although these were not significant (P > .005). However, those with rSBP ≥ 140 to < 160 demonstrated an adjusted HR of 0.60 (95% CI 0.42-0.86, P = .005). We also evaluated rDBP, rSBP, and ∆SBP in the multivariable model as continuous variables and our results were similar, in which ∆SBP was not a significant multivariate predictor of mortality; only rSBP and rDBP remained significant predictors (Table 3).

Table 3 Predictors of mortality in patients undergoing vasodilator stress Rb-82 PET MPI
Full size table

Discussion

In this cohort study of 3413 consecutive patients undergoing vasodilator Rb-82 PET MPI, we hypothesized that BP response (∆SBP and ∆DBP) during pharmacologic stress would be an independent predictor of all-cause mortality. We found that rDBP, rSBP, and ∆SBP predicted mortality in univariable analysis, and ∆DBP did not. However, when entered into a multivariable mortality model along with other clinical, hemodynamic, and MPI variables, only lower rDBP and rSBP remained independently prognostic, whereas neither BP response variables (∆SBP nor ∆DBP) were prognostic. These findings are in contrast to results from studies of exercise SPECT MPI and exercise treadmill stress testing, in which BP response has been shown to be a predictor of mortality independent of resting BP.1–3

Multiple prior studies have found that both resting BP and BP response are associated with mortality in patients undergoing EET or exercise SPECT MPI. However, little is published describing whether these associations hold in patients undergoing vasodilator PET MPI. In these patients, our study demonstrated that only resting BP but not BP response was associated with mortality. There are potential explanations for this discrepancy in the prognostic value of BP response in patients undergoing exercise versus vasodilator stress.

First, the expected BP response is inherently different between these two stress modalities: SBP is expected to rise during exercise testing, whereas it may not rise (and often decreases) during vasodilator stress. Second, patients referred for vasodilator MPI tend to have a greater burden of comorbidities that either preclude or limit exercise, or are more deconditioned and unlikely to achieve an adequate heart rate response during exercise. In these “sicker” patients referred for vasodilator stress MPI, there may be diminishing prognostic value of hemodynamic variables that are known to be prognostic in exercise testing, including BP response. Third, as previously shown by our group, heart rate response has been shown to be prognostic in vasodilator stress PET MPI.7 Compared with the reference group of ≤4 beats per minute (bpm) heart rate increase with vasodilator stress, patients with a 5-14 bpm increase had an adjusted HR for CAD mortality of 0.58 (95% CI 0.44-0.77; P < .001), and patients with a ≥ 15 bpm increase had an adjusted HR of 0.30 (95% CI 0.21-0.43; P < .001). This association between heart rate response and mortality perhaps leaves little room for other hemodynamic response variables (such as BP response) to confer any additional prognostic information, given that changes in heart rate and blood pressure tend to be strongly inversely linked during vasodilator stress and therefore statistically collinear.

This relationship between heart rate response and vasodilators has been studied extensively in the diabetic population, a population that is prone to blunting of their hemodynamic responses.18–20 These prior studies suggest that diabetic patients have altered sympathetic response to vasodilators, and this can significantly alter these patients’ hemodynamic response with regards to heart rate response.21 Given that nearly one third of the patients in this cohort had a known diagnosis of diabetes, we performed additional sub-analyses comparing diabetic patients to non-diabetic patients using our blood pressure covariates. This sub-analysis did not reveal any significant differences between the diabetic and non-diabetic subgroups.

Many of the known limitations to the main PET Prognosis Registry have been previously described.8 Prior studies have demonstrated some differences in observed hemodynamic responses among the different vasodilator agents.22,23 All patients included in this analysis underwent vasodilator stress, but the specific vasodilator agent used (dipyridamole or adenosine) was not recorded in the registry. Given this limitation, it was not possible to control for the specific vasodilator agent administered, or describe any differences among agents that might exist when determining the relationship between BP response and mortality. Furthermore, we cannot comment on whether our findings would apply to patients undergoing regadenoson stress, as patients in the PET Prognosis Registry did not undergo regadenoson stress. Regadenoson has a similar expected effect on hemodynamics (increase in heart rate and drop in blood pressure) compared with dipyridamole and adenosine, and therefore we might speculate that our findings would likely not change if patients in this study also underwent regadenoson stress. However, future studies are required to address this issue conclusively.

The registry collected data on BP-lowering medications. However, we did not collect data on medication dosing, or whether patients held certain medications (e.g., beta blockers, nitrates, centrally acting calcium channel blockers, etc.) within 24-48 hours of their PET MPI study, as recommended by the American Society of Nuclear Cardiology guidelines. A concern common to all prospective cohort studies is selection bias; however, this bias was reduced by the consecutive enrollment of patients undergoing vasodilator PET MPI at participating centers. Finally, this is an observational study that can only suggest associations, and causal inferences cannot be drawn between any variables in the model and the outcome of mortality.

Finally, it is worth noting that while the results of this analysis indicate that BP response during vasodilator stress imaging does not provide added prognostic value at the population level, this does not mean that monitoring and recording BP changes during the course of the exam should not be performed. Indeed, in individual patients, measuring BP response following vasodilator stress is essential to ensure patient safety and to provide context for clinical interpretation of PET studies.

In summary, our study shows that lower resting systolic and diastolic BP but not BP response conferred independent prognostic value for mortality in patients undergoing vasodilator PET MPI, when considered in addition to resting BP and other clinical, hemodynamic stress, and MPI variables. Our findings differ from studies of exercise SPECT MPI in which BP response has been shown to be prognostic. Additional studies are needed to determine whether the prognostic value of BP response varies according to which vasodilator stress agent is used, and based on the profile of BP-lowering medications (dose, and timing of last dose in relation to the actual stress MPI study).

New Knowledge Gained

Vasodilator stress MPI is increasingly used as a method for evaluating cardiovascular risk. Blood pressure response is a known independent predictor of mortality in exercise MPI studies and is routinely reported in a similar manner for patients undergoing vasodilator MPI stress as well. However, our analysis using the PET Multicenter Registry indicates that lower resting BP but not BP response had independent prognostic value beyond resting BP in this patient population undergoing vasodilator stress MPI.