Evaluating the Prognostic Value of Positron-Emission Tomography Myocardial Perfusion Imaging Using Automated Software to Calculate Perfusion Defect Size

Authors


Abstract

Background:

Myocardial perfusion imaging by positron-emission tomography (PET MPI) is regarded as a valid technique for the diagnosis of coronary artery disease (CAD), but the incremental prognostic value of PET MPI among individuals with known or suspected CAD is not firmly established.

Hypothesis:

Myocardial perfusion defect sizes as measured by PET MPI using automated software will provide incremental prognostic value for cardiac and all-cause mortality.

Methods:

This study included 3739 individuals who underwent rest-stress rubidium-82 PET MPI for the evaluation of known or suspected CAD. Rest, stress, and stress-induced myocardial perfusion defect sizes were determined objectively by automated computer software. Study participants were followed for a mean of 5.2 years for cardiac and all-cause mortality. Cox proportional hazards models were developed to evaluate the incremental prognostic value of PET MPI.

Results:

A strong correlation was observed between perfusion defect sizes assessed visually and by automated software (r = 0.76). After adjusting for cardiac risk factors, known CAD, noncoronary vascular disease, and use of cardioprotective medications, stress perfusion defect size was strongly associated with cardiac death (P < 0.001). Rest perfusion defects demonstrated a stronger association with cardiac death (P < 0.001) than stress-induced perfusion defects (P = 0.01), yet both were highly significant. Similar patterns held for all-cause death.

Conclusions:

The current study is the largest to date demonstrating PET MPI provides incremental prognostic value among individuals with known or suspected CAD. Automated calculation of perfusion defect sizes may provide valuable supplementary information to visual assessment.

This work was partially funded by a predoctoral fellowship grant awarded to the first author by the American Heart Association's Founders' Affiliate. Additional funding was provided by Niagara Falls Memorial Medical Center, Positron Corporation, the University at Buffalo, and Niagara University. The authors have no other funding, financial relationships, or conflicts of interest to disclose.

Introduction

Noninvasive myocardial perfusion imaging with positron-emission tomography (PET MPI) is recognized as an accurate technique for the diagnosis of coronary artery disease (CAD), but the prognostic value of PET MPI among individuals with known or suspected CAD has not been firmly established.1–4 The acknowledged technical superiority of PET MPI has yet to translate into widespread clinical implementation due largely to PET's technical complexity, greater initial testing costs, and lack of a firm research base guiding proper incorporation of PET MPI in clinical practice.2,3,5 To the latter point, only a small number of modestly sized studies have evaluated the prognostic capacity of PET MPI.6–9 Furthermore, these studies incorporated visual subjective semiquantitative assessment of myocardial perfusion defect sizes that may be subject to inter- and intrareader measurement variability. Large prognostic studies of PET MPI are important (1) to determine whether normal PET MPI findings portend low cardiac event risk and (2) to evaluate whether perfusion defects as determined by PET MPI provide incremental prognostic value over established risk factors with respect to important clinical outcomes. The current study attempts to address these deficiencies in the literature, and furthermore uses automated computer software in addition to visual assessment to calculate perfusion defect sizes.

Methods

The current study comprised 3739 individuals with known or suspected CAD who underwent clinically indicated rest-stress rubidium-82 PET MPI at a single clinical center, the Heart Center of Niagara (HCON). Individuals undergoing PET MPI for evaluation of possible acute coronary syndromes were excluded. Descriptions of the PET camera and the PET MPI protocol in place at the HCON have been described.10 Measurement of myocardial perfusion defect sizes was accomplished objectively through automated computer software (Positron Corp, Westmont, IL), which estimates the percentage of left ventricular (LV) myocardial mass having relative tracer activity below a severity threshold of 60% (Figure 1).11–14 The automated software computes the size of rest and stress perfusion defects; stress-induced perfusion defect size is the difference between stress and rest defect sizes (stress-induced = stress − rest). Notably, stress perfusion defects include any combination of rest and/or stress-induced perfusion defects (stress = rest + stress-induced). In addition, visual estimation of perfusion defect sizes in percent of LV was performed by a single experienced reader blinded to software results (M.E.M.). Thus, all PET images were evaluated by both visual estimation and automated software, although the primary analyses reported here are restricted to perfusion defects measured by automated software.

Figure 1.

Typical positron-emission tomography myocardial perfusion image with output from automated software. Perfusion defect sizes are defined by the percentage of the left ventricular (LV) myocardium with relative tracer activity <60% of maximum. In this example, the perfusion defect size during pharmacologic stress is 12.00% of the LV; the perfusion defect size at rest is 1.07% of the LV. Abbreviations: SD, standard deviation.

Clinical variables corresponding to the index PET MPI were abstracted retrospectively from medical records maintained at the HCON. Variables gathered included demographics, physical examination findings, traditional cardiovascular disease risk factors, medical history, medications, indications for PET MPI, pretest likelihood of significant CAD, and PET MPI procedural characteristics. Pretest likelihood of significant CAD was estimated using the method of Diamond et al considering age, gender, risk factors, symptoms, and results from previous noninvasive testing when available.15 Mortality information was obtained through the National Death Index (NDI), with a mean follow-up time among presumed survivors of 5.2 (±1.7) years (range, 2.5–8.7 years). Cardiac death was defined by NDI-provided International Classification of Diseases, 10th Revision codes for underlying cause of death with prefixes I20-I25 (ischemic heart disease) and I30-I52 (other forms of heart disease). According to NDI records, 510 (14%) cohort members were identified as deceased, with 187 identified as cardiac deaths (37% of all deaths). Information on early revascularization following PET MPI was not universally available; therefore, early revascularizations were not removed from the analysis.

All analyses incorporate myocardial perfusion defect sizes as measured by automated computer software. Perfusion defect sizes were measured on a continuous scale by the automated software and grouped into 0%, >0% to <5%, 5% to <10%, 10% to <20%, and ≥20% of the LV myocardium to facilitate statistical analysis. Continuous baseline characteristics are reported as means and standard deviations with trends across perfusion defect groups tested for significance by linear regression. Categorical baseline characteristics are reported as percentages with trends across groups tested for significance by Mantel-Haenzel χ2 trend tests. The primary outcome variable (cardiac death) was defined as time to event, measured as the number of days from the index PET MPI to death from a cardiac cause as defined above. Individuals dying from a noncardiac cause were censored at the date of the noncardiac death; individuals presumed alive were censored at the last date vital status was assumed to be known through the NDI. All-cause death was also defined as a time to event variable with individuals presumed alive censored at the last date vital status was assumed to be known. Annual death rates were calculated on a group-specific basis by dividing the number of group-specific deaths by the sum of group-specific follow-up times and multiplying by 100%. Differences in annual death rates across groups were tested for significance by log-rank tests. Cox proportional hazards models were developed to estimate the association between perfusion defect sizes and mortality end points, with the absence of perfusion defects (0% of the LV) serving as the referent.

The incremental prognostic value of PET MPI-determined perfusion defect sizes was assessed by evaluating the gain in prognostic information provided by PET MPI perfusion parameters beyond that provided by a thorough collection of a priori-selected variables having the strongest established associations with cardiac death.16,17 This pre-PET baseline model included age, gender, body mass index, resting electrocardiogram abnormalities, hypertension, dyslipidemia, diabetes, smoking history, parental history of CAD, prior myocardial infarction (MI), prior coronary artery bypass graft surgery (CABG), prior percutaneous coronary intervention (PCI), cerebrovascular disease, and peripheral vascular disease. Perfusion defect sizes as measured by PET MPI were then forced into the baseline model, and tests for trend were performed. Adjusted hazard ratios and 95% confidence intervals were estimated for the perfusion defect size groups. Separate models are reported for (1) stress perfusion defects (includes fixed rest and/or stress-induced perfusion defects together) and (2) rest and stress-induced perfusion defects as separate terms in the model. Sensitivity analyses were performed, whereby the pre-PET model was modified to evaluate the robustness of findings to variations in the pre-PET model. The final conclusions were deemed not appreciably altered by slight variations to the pre-PET model; therefore, only results related to the aforementioned model are reported.

This study was approved by the institutional review board at the University at Buffalo who granted a waiver of patient consent. This investigation was conducted in accordance with the Declaration of Helsinki.

Results

The PET MPI study cohort had a mean age of 62 years (±13 years), and 53% were male. The prevalence of known CAD (prior MI, PCI, or CABG) was 29%. The most common primary indications for PET MPI were chest pain (52%) and shortness of breath (29%). Among individuals without known CAD, the pretest likelihood of significant CAD was 31% (±15%). Using automated software, the prevalence of stress perfusion defects (>0% of the LV hypoperfused) was 60%, and the prevalence of stress-induced perfusion defects was 35%. Visual estimation of stress perfusion defect sizes by an experienced reader showed a strong correlation with automated computer software (r = 0.76).

Larger stress perfusion defects were associated with a worse cardiovascular disease risk profile (Table 1). Larger stress perfusion defects were associated with increased age, male gender, and an increased prevalence of several cardiovascular disease risk factors including hypertension, dyslipidemia, diabetes, and smoking history. Individuals with larger stress perfusion defects were also more likely to have a history of CAD, including prior MI, PCI, and CABG. Accordingly, individuals with larger stress perfusion defects were also more likely to be taking cardioprotective medications such as statins, aspirin, and angiotensin-converting enzyme inhibitors. Similar patterns were noted when examining baseline characteristics across rest and stress-induced perfusion defect groups respectively (data not shown).

Table 1. Baseline Variables by Size of Stress Perfusion Defects as Measured by PET MPI
 Percent of Left Ventricle With Stress Perfusion Defect 
 0% (n = 1501)>0%–<5% (n = 1421)5%–<10% (n = 297)10%–<20% (n = 237)≥ 20% (n = 283)P Value
  1. Abbreviations: ACE, angiotensin-converting enzyme; CAD, coronary artery disease; DBP, diastolic blood pressure; ECG, electrocardiogram; PCI, percutaneous coronary intervention; PET MPI, positron-emission tomography myocardial perfusion imaging; SBP, systolic blood pressure.

Demographics
Age, y61 ± 1361 ± 1363 ± 1367 ± 1267 ± 12<0.001
Male, %4256616772<0.001
Physical examination
Body mass index, kg/m229 ± 731 ± 731 ± 730 ± 629 ± 60.31
Heart rate, bpm69 ± 1267 ± 1267 ± 1267 ± 1269 ± 130.10
Systolic blood pressure, mm Hg130 ± 22132 ± 23133 ± 22133 ± 23129 ± 230.39
Diastolic blood pressure, mm Hg73 ± 1073 ± 1172 ± 1171 ± 1170 ± 11<0.001
Abnormal rest ECG, %1718222627<0.01
Cardiovascular risk factors, %
Hypertension5963666864<0.01
Dyslipidemia6262677373<0.001
Diabetes2024303735<0.001
Smoking status     0.02
Current1617171615 
Former4543495156 
Never3940343329 
Parental history of CAD53524849490.08
Medical history, %
Myocardial infarction713284652<0.001
Coronary bypass surgery711242937<0.001
PCI712181623<0.001
Cerebrovascular disease88111213<0.01
Peripheral vascular disease101081315<0.01
Kidney disease7771480.02
Lung disease18171814160.16
Medications, %
Statin4346566072<0.001
Aspirin4448476262<0.001
Nitrate1517243436<0.001
Digoxin4681116<0.001
Diuretic2227333238<0.001
Beta blocker3440566163<0.001
Calcium channel blocker20222829210.02
ACE inhibitor2433394148<0.001
Angiotensin-receptor blocker991011100.51
Anticoagulant78101112<0.001
Antiplatelet810181818<0.001
PET procedural characteristics
Heart rate at peak stress, bpm92 ± 1789 ± 1687 ± 1687 ± 1787 ± 17<0.001
SBP at peak stress, mm Hg126 ± 21128 ± 21128 ± 22128 ± 22123 ± 220.26
DBP at peak stress, mm Hg70 ± 1170 ± 1169 ± 1268 ± 1267 ± 12<0.001

Significant trends in crude annual death rates were observed across stress, rest, and stress-induced perfusion defect groups respectively, for both cardiac and all-cause death (Tables 2 and 3). After considering a thorough collection of preselected variables having the strongest established associations with cardiac death, stress, rest, and stress-induced perfusion defects all remained independently associated with both mortality outcomes (Tables 2 and 3). Of note, small resting perfusion defects involving <5% of the LV conferred a significant 70% increased risk of cardiac death compared to no resting perfusion defects (0% of LV). Furthermore, despite observing a significant trend between the size of stress-induced perfusion defects and cardiac death, the hazard ratio for the largest stress-induced perfusion defect group (≥20%) was below the hazard ratio for the 5% to 10% perfusion defect size group.

Table 2. Annual Death Rates and Hazard Ratios by Size of Stress Myocardial Perfusion Defects as Measured by Positron-Emission Tomography Myocardial Perfusion Imaging
  Cardiac DeathAll Cause Death
Stress Perfusion Defect SizeNo.ADRHR (95% CI)aHR (95% CI)bADRHR (95% CI)aHR (95% CI)b
  1. Abbreviations: ADR, annual death rate; CI, confidence interval; HR, hazard ratio.

  2. a

    aUnadjusted. bAdjusted for age, gender, body mass index, resting electrocardiogram findings, hypertension, dyslipidemia, diabetes, smoking status, parental history of coronary artery disease, myocardial infarction, coronary bypass surgery, percutaneous coronary intervention, cerebrovascular disease, and peripheral vascular disease.

0%15010.5%1.0 (−)1.0 (−)1.9%1.0 (−)1.0 (−)
>0%–<5%14210.8%1.5 (1.0–2.2)1.4 (0.9–2.1)2.3%1.2 (1.0–1.5)1.2 (1.0–1.5)
5%–<10%2971.2%2.2 (1.3–3.9)1.8 (1.0–3.2)3.4%1.8 (1.3–2.5)1.5 (1.1–2.1)
10%–<20%2372.7%5.0 (3.1–8.0)3.3 (2.0–5.5)4.9%2.7 (2.0–3.7)1.9 (1.4–2.7)
≥20%2832.8%5.2 (3.4–8.1)3.4 (2.1–5.6)6.4%3.7 (2.8–4.8)2.5 (1.9–3.4)
P value <0.001<0.001<0.001<0.001<0.001<0.001
Table 3. Annual Death Rates and Hazard Ratios by Size of Rest and Stress-Induced Myocardial Perfusion Defects as Measured by Positron-Emission Tomography Myocardial Perfusion Imaging
  Cardiac DeathAll Cause Death
 No.ADRHR (95% CI)aHR (95% CI)bADRHR (95% CI)aHR (95% CI)b
  1. Abbreviations: ADR, annual death rate; CI, confidence interval; HR, hazard ratio.

  2. a

    Only rest and stress-induced perfusion defect sizes included in model.

  3. b

    Adjusted for age, gender, body mass index, resting electrocardiogram findings, hypertension, dyslipidemia, diabetes, smoking status, parental history of coronary artery disease, myocardial infarction, coronary bypass surgery, percutaneous coronary intervention, cerebrovascular disease, and peripheral vascular disease.

Rest perfusion defect size       
 0%12140.6%1.0 (−)1.0 (−)2.3%1.0 (−)1.0 (−)
 >0%–<5%17371.0%1.4 (0.9–2.0)1.7 (1.2–2.6)2.5%1.0 (0.8–1.2)1.3 (1.0–1.6)
 5%–<10%3700.9%1.2 (0.7–2.2)1.4 (0.8–2.6)2.6%1.1 (0.8–1.5)1.3 (0.9–1.8)
 10%–<20%2571.9%2.5 (1.5–4.2)2.6 (1.5–4.5)3.7%1.5 (1.0–2.0)1.6 (1.1–2.3)
 ≥20%1612.8%3.5 (2.1–6.0)3.4 (1.9–5.9)6.6%2.7 (2.0–3.8)2.8 (2.0–4.0)
 P value <0.001<0.001<0.001<0.001<0.001<0.001
Stress-induced perfusion defect size       
 0%24050.8%1.0 (−)1.0 (−)2.2%1.0 (−)1.0 (−)
 >0%–<5%8641.0%1.4 (1.0–2.0)1.1 (0.8–1.6)2.9%1.4 (1.2–1.7)1.2 (0.9–1.4)
 5%–<10%1921.8%2.1 (1.2–3.5)1.6 (0.9–2.7)4.4%2.0 (1.4–2.7)1.6 (1.1–2.2)
 10%–<20%1552.8%3.0 (1.9–4.9)1.8 (1.1–3.0)5.1%2.1 (1.5–3.0)1.2 (0.9–1.8)
 ≥20%1232.0%2.1 (1.1–3.8)1.5 (0.8–2.8)4.7%1.9 (1.3–2.8)1.4 (0.9–2.0)
 P value <0.001<0.0010.010<0.001<0.0010.022

Discussion

The current study is the largest to date reporting the incremental prognostic value of PET MPI for hard clinical outcomes. In particular, the size of rest, stress-induced, and stress myocardial perfusion defects as measured objectively by automated computer software in percent of LV myocardium were significantly associated with both cardiac and all-cause death after considering a thorough collection of important pretest clinical risk factors. We noted a high level of correlation between perfusion defect sizes as estimated by automated computer software and a blinded experienced reader. Interestingly, among individuals with the largest stress-induced perfusion defects, we detected a level of risk comparable to individuals with much smaller perfusion defects, suggesting a possible benefit of aggressively treating these large ischemic zones. These findings may have implications for guiding clinical management of individuals with known or suspected CAD using PET MPI.

The current findings confirm and extend the few previous studies evaluating the prognostic value of PET MPI in cohorts with known or suspected CAD. In accord with previous studies, this study demonstrated low cardiac death risk with normal PET MPI findings and progressively increasing risk as perfusion defects worsened.6–9 The current study is over 2 times larger than any previous PET MPI prognostic study, permitting more comprehensive assessment of the association between perfusion defects and outcomes. The incorporation of automated software allows more reproducible measurement of perfusion defect sizes compared with visual subjective measurement.17–19 These unique study attributes coupled with the advanced PET technology allowed evaluation of the prognostic impact of smaller perfusion defects.

The current findings may prove valuable in differentiating those individuals who can be safely managed medically from those who may need further diagnostic testing and/or procedural intervention to lower mortality risk.3,20 In the current study, having <5% of the LV hypoperfused during pharmacologic stress was associated with a <1% annual risk of cardiac death (low risk). Numerous previous studies have concurred that normal perfusion imaging confers low risk of subsequent cardiac events among individuals with known or suspected CAD.8,21 Evidence suggests these low-risk individuals can be safely managed medically irrespective of the extent and severity of anatomic coronary disease, thereby foregoing further invasive diagnostic testing and/or treatments that have poor cost-benefit and risk-benefit ratios among lower-risk patients.21,22 Importantly, PET MPI has a recognized advantage for identifying these low-risk patients, given PET's lower false positive rate and higher specificity.23

In contrast, abnormal PET MPI findings were associated with a significantly increased risk of cardiac death with annual cardiac death rates approaching 3% (high risk) among individuals with stress perfusion defects involving 10% or more of the LV. As such, these high-risk findings as revealed by PET MPI may identify a group of individuals who may benefit from further diagnostic evaluation and possible revascularization. As suggested by previous studies, the current study revealed resting perfusion defects as a stronger contributor to the high short-term cardiac death risk state.8,24 Individuals with sizeable fixed rest-stress perfusion defects may benefit from myocardial viability testing to distinguish necrotic from hibernating myocardium, the latter being an amenable target for revascularization with beneficial outcomes compared to medical therapy alone.25–28 A recent study by D'Egidio et al using PET with the F-18-fluorodeoxyglucose radiotracer for evaluating myocardial viability among patients with ischemic LV dysfunction showed the benefit from revascularization on cardiac outcomes increased proportionally with the amount of viable myocardium.27

Stress-induced perfusion defects were also significantly associated with cardiac death in the current study. This increased risk was evident for stress-induced defects involving 5% or more of the LV. Interestingly, the hazard ratio for stress-induced defects involving 20% or more of the LV was slightly lower than the hazard ratios for defects involving 5% to 10% and 10% to 20% of the LV. Reasonable speculation suggests this observation could be attributed to more aggressive post-PET therapy in patients with larger stress-induced perfusion defects. Although we were unable to collect follow-up revascularization data on the entire cohort, among a subset of 552 subjects with follow-up data, we noted post-PET revascularization rates of 3%, 10%, 25%, 27%, and 59% across the 5 stress-induced perfusion defect groups. The importance of stress-induced perfusion defects in risk stratification and guiding revascularization decisions is well established.19,29,30 An ischemia-based paradigm for guiding revascularization decisions has been widely endorsed over a pre dominantly anatomy-based one, with Hachamovitch et al suggesting an ischemic threshold of 10% or more of the LV for which revascularization may confer a clinical benefit.26,30–32 Our data provide some support for this assertion, and furthermore raise the question of whether increased utilization of revascularization among individuals with even smaller ischemic defects would lower risk.

The current study is timely in light of recent reports by nuclear cardiology experts projecting an impending shift away from SPECT MPI toward PET MPI for the evaluation of known or suspected CAD.2,3,33,34 An on-going debate persists regarding the relative merits of each modality for this purpose.5,20 The dearth of sizeable PET MPI prognostic studies has precluded a comprehensive comparison between PET and SPECT—a void in the research literature partially filled by the current study. Though PET is generally perceived as providing better images, improved diagnostic capacity, and more sensitive characterization of perfusion defects, widespread assimilation of PET into clinical practice has yet to be realized, attributed to PET's higher initial testing costs, greater technical complexity, and a shortage of significantly sized research cohorts for guiding its clinical integration.1,2,5,23,34 Inevitably, cost-effective supremacy in a randomized setting may be needed for PET to assume a larger role in the clinical evaluation of individuals with known or suspected CAD. Multiple nonrandomized studies have suggested PET MPI may be a more cost-effective strategy for guiding CAD management, especially among individuals with an intermediate pretest likelihood of CAD.10,35 The greater 1-time initial testing costs with PET were found to be offset by a subsequent avoidance of downstream invasive testing and procedures without compromising outcomes.10 Whether these findings persist when evaluated within a rigorous randomized trial is an important area of research with significant clinical and cost implications.

The primary limitation of the current study is the lack of complete information on LV ejection fraction (LVEF), nonfatal outcomes including myocardial infarction and late revascularization, and PET-driven therapies including medication changes and early revascularization. LVEF is regarded as one of the strongest predictors of cardiac death, and studies capable of evaluating both LVEF and perfusion defects have typically found both provide strong independent prognostic information for clinical outcomes.7,36 In a subset of the current cohort, LVEF demonstrated modest correlation with both stress perfusion defect size (r = −0.39) and rest perfusion defect size (r = −0.31). The prognostic strength of perfusion defects would likely have been modestly attenuated had LVEF been universally available and included in multivariable models. Furthermore, this study was not designed to account for PET-driven therapies that may have influenced the association between perfusion defect measures and mortality. We noted earlier in a subset analysis that revascularization rates increased as stress-induced perfusion defects worsened. We cannot determine convincingly whether salutary benefits of revascularization made any contribution to the observed effects.

Conclusion

The current study is the largest to date demonstrating the incremental prognostic value of PET MPI for hard clinical outcomes. Using automated computer software for objectively measuring the size of myocardial perfusion defects, PET MPI was able to provide incremental prognostic value for cardiac death, allowing differentiation between lower- and higher-risk patients. Future studies should evaluate the value of PET MPI in facilitating cost-efficient CAD care relative to competing techniques.

Ancillary