Comparison Between the Prognostic Value of Left Ventricular Function and Myocardial Perfusion Reserve in Patients with Ischemic Heart Disease

Patients and Study Design

This study prospectively included, with retrospective analysis, 480 subjects with advanced IHD who underwent rest ¹³N-ammonia, dipyridamole stress ¹³N-ammonia, and gated ¹⁸F-FDG PET, between 1995 and 2003 at the PET center of the University Medical Center Groningen, for evaluation of stress-induced ischemia and myocardial viability. Patient data were collected from the hospital information system (Table 1). PET-driven intervention was defined as any cardiac (surgical or percutaneous) procedure performed within the first 6 mo after the PET study date. If no intervention was performed, the patients were considered to be only medically treated.

Endpoints

All causes of mortality were assessed. Cardiac death was defined as sudden death, death after the onset of symptoms suggestive of cardiac ischemia, or death due to heart failure. Cardiac events included cardiac death, myocardial infarction, and non–PET-driven revascularization. Myocardial infarction was defined as an increase in cardiac enzymes (>2× the upper limit of normal), new pathologic Q waves on the electrocardiogram, or both. A major adverse cardiac event was defined as cardiac death, myocardial infarction, percutaneous coronary intervention (PCI), coronary artery bypass grafting (CABG), or hospitalization for an acute coronary syndrome or heart failure.

PET

The patients underwent dynamic rest ¹³N-ammonia, dipyridamole stress ¹³N-ammonia, and gated ¹⁸F-FDG PET using a 1-d protocol, as described previously (12). Briefly, PET studies were performed after the patients had discontinued vasoactive medication for 5 plasma half-lives and had refrained from caffeinated beverages for a minimum of 12 h. Imaging was performed with the patient supine and used an ECAT 951 positron camera (Siemens CTI). Thirty-one planes were measured simultaneously over a length of 10.8 cm. The measured resolution of the system was 6 mm in full width at half maximum. Data were automatically corrected for accidental coincidence and dead time. Patients were positioned with the help of a rectilinear scan. Photon attenuation was measured using a retractable external ring source filled with ⁶⁸Ge/⁶⁸Ga. Perfusion imaging was performed after dipyridamole had been infused (0.56 mg/kg in 4 min). Imaging was started by injecting 400 MBq of ¹³N-ammonia 6 min after the start of dipyridamole infusion and continued for 15 min.

To stimulate ¹⁸F-FDG uptake, patients were given 75 g of glucose orally just before scanning or were given 500 mg of acipimox (Nedios; Byk Pharmaceuticals) orally 90 min before scanning to lower circulating free fatty acids (13). To prevent side effects of acipimox (e.g., skin rash), 250 mg of aspirin were administered orally 5 min before acipimox. In diabetic patients, ¹⁸F-FDG imaging was done with hyperinsulinemic euglycemic glucose clamping (14). After the ¹³N-ammonia data had been acquired, 200 MBq of ¹⁸F-FDG were injected intravenously, followed by a PET dynamic acquisition. The total ¹⁸F-FDG PET acquisition time was 40 min, with the last 20 min acquired in gated mode with 16 frames per cardiac cycle. The length of each gate was based on the current R-R interval. The R-R interval was allowed to vary by ±10%. Data were corrected for attenuation using the transmission scan and were reconstructed using filtered backprojection (Hann filter, 0.5 pixels/cycle).

Kinetic Models and Data Analysis

From the PET data, dynamic parametric polar maps were constructed (12). PET perfusion data at rest were corrected for rate–pressure product. Myocardial blood flow data were corrected for partial-volume effect and spillover and quantified by the model of Hutchins et al. (6). Briefly, myocardial and blood time–activity curves derived from regions of interest over the heart and ventricular chamber are fitted using a 3-compartment model for ¹³N-ammonia, yielding rate constants for tracer uptake and retention. Perfusion flow reserve (dipyridamole-to-rest ratio) was calculated by dividing the dipyridamole ¹³N-ammonia stress study by the ¹³N-ammonia rest study.

Data analysis of ¹⁸F-FDG was performed with PATLAK analysis (15).

Mismatch was quantified by first normalizing the ¹⁸F-FDG uptake polar map and the dipyridamole blood flow polar map to their means. Then, a difference polar map was created by subtracting the normalized dipyridamole blood flow polar map from the normalized ¹⁸F-FDG uptake polar map. Mismatch was calculated as the percentage myocardium above the 95% confidence interval of the normal database, and results were expressed as percentage of the total myocardium. Similarly, matching areas were quantified by constructing a product polar map; the normalized dipyridamole blood flow polar map was multiplied by the normalized ¹⁸F-FDG uptake polar map. Match was defined as the percentage myocardium below the 95% confidence interval. The extent of mismatching areas (viable myocardium) and matching areas (nonviable myocardium) was calculated from these data as previously described (12).

The last frames (20-min acquisition time) of the dynamic gated ¹⁸F-FDG PET studies were summed and transformed into static studies and used for further data analysis with the help of the quantitative gated SPECT program (15). Based on the gated ¹⁸F-FDG images, left ventricular end-systolic and end-diastolic volumes, as well as LVEF, were computed.

Statistical Analysis

Descriptive results are expressed as mean ± SD. Categoric measures are presented as frequencies with percentages. Crude data were compared across tertiles of MPR, defined as perfusion during dipyridamole divided by resting perfusion, with the χ² test for trend (dichotomous variables) and generalized linear models (continuous variables). The significance of MPR, controlled for important risk modifiers as presented in Table 1 (P < 0.20), was examined with multivariable Cox proportional hazards regression analyses by using fractional polynomials (16). Results are summarized by hazard (risk) ratios with confidence intervals based on robust SE estimates. To assess the prognostic value of LVEF and MPR adjusted for age and sex, Harrell's C-statistic was computed (comparable to the area under the receiver-operating-characteristic curve). Model fit was assessed with Bayesian information criterion statistics, which are goodness-of-fit measures adjusted by degrees of freedom and sample size. Smaller Bayesian information criterion values indicate that the model fits better. A difference of 10 points or more between a given model and the other model is strong evidence for a significantly better goodness of fit. The significance level was set at 0.05. Observations with missing values for contributing variables in the multivariate model were excluded. The statistical analysis was performed with SPSS (SPSS Inc.), version 9.1, and STATA statistical software, release 10.0 (StataCorp LP).

Patient Characteristics

Between 1995 and 2003, 480 patients underwent a ¹³N-ammonia rest, a dipyridamole stress, and a gated ¹⁸F-FDG PET scan. In 17 patients, gating was not possible because of heart rate irregularities occurring during the scan. In 463 patients, valid MPR could be measured. One hundred nineteen patients (368 men; mean age, 66 ± 11 y; LVEF, 35% ± 15%) underwent a PET-driven revascularization (67 through PCI and 52 through CABG). Patients with a PET-driven intervention were comparable to the study group with respect to age (66 ± 11 y vs. 68 ± 10 y), sex (22% female vs. 18% female), risk factors, previous myocardial infarction (71% vs. 78%), and previous PCIs (45% vs. 36%) but had significantly more previous CABGs (28% vs. 15%, P = 0.028) and a higher LVEF (36 ± 16 vs. 32 ± 14, P = 0.007). The remaining 344 patients were the subject of this study. The baseline characteristics of these 344 patients are shown in Table 1: 14% of the patients were in NYHA class I, 49% in class II, 31% in class III, and 5% in class IV. Overall, the MPR was 1.71 ± 0.50 (intertertile boundaries, 1.49 and 1.84). Areas of mismatch were found in 91 patients (27%), areas of matching defects in 267 (78%), both mismatching and matching defects in 80 (23%), and no defects at all in 66 (19%). Mean percentage match in the study group was 27% ± 16%, mean percentage mismatch was 9% ± 10%, and mean LVEF was 36% ± 15%. Coronary artery disease was present in the left anterior descending artery in 66% of patients, in the right coronary artery in 43%, and in the circumflex coronary artery in 30%.

Outcome Event

The median follow-up among survivors was 85 mo (range, 1–138 mo). Among the 344 patients in this study, there were a total of 85 deaths (25%), of which 60 (17%) were cardiac deaths. Twenty-five patients (7%) experienced a nonfatal myocardial infarction. A total of 71 patients (21%) underwent a PCI and 27 a CABG (8%) during follow-up.

Hazard Ratio of MPR

Table 2 summarizes the results of the Cox regression analysis for cardiac death. In the univariate analysis, the parameters significantly associated with cardiac death were MPR; family history; previous myocardial infarction; LVEF; left ventricular end-diastolic volume; the use of aspirin, diuretics, or digoxin; and matching. After controlling for age and sex, the following parameters were associated with cardiac death: MPR, family history, previous myocardial infarction, LVEF, left ventricular end-diastolic volume, aspirin, diuretics, and digoxin. MPR was associated with a hazard ratio for cardiac death of 4.11 (95% confidence interval, 2.98–5.67) per SD decrease, whereas the risk for LVEF was 2.76 (2.00–3.82) per SD decrease. Interestingly, the prognostic value of MPR was independent of the extent of matching and mismatching defects. Survival data for each MPR tertile are shown in Figure 1. The hazard function of MPR, when compared with LVEF, was steeper in a prognostic model adjusted for age and sex, resulting in improved C-statistics and Bayesian information criterion statistics (0.83, 605 vs. 0.77, 620) (Fig. 2). Finally, in a secondary mutually adjusted multivariate analysis of MPR; family history; previous myocardial infarction; LVEF; left ventricular end-diastolic volume; the use of aspirin, diuretics, or digoxin; and percentage matching defects, the parameters that remained statistically significant in the model were MPR, LVEF, and the use of diuretics (hazard ratios of 4.08 [2.50–6.65, P < 0.001], 1.91 [1.10–3.31, P = 0.021], and 2.19 [1.07–4.97, P = 0.033], respectively).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

Cardiac death-free survival curves for median values of MPR tertiles are shown. Median MPR values of each tertile (intertertile boundaries, 1.49 and 1.84) were taken to illustrate differences between these groups. Numbers indicate patients at risk and number of events per tertile (T).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2. 

Hazard ratio curves and percentage distribution for MPR (A) and for LVEF (B).

Hazard Ratio for Major Adverse Cardiac Event

After univariate analysis of baseline demographics (including PET parameters) MPR had a hazard ratio of 1.60 (1.31–1.94, P < 0.001) for major adverse cardiac events. Finally, in a secondary mutually adjusted multivariate analysis, MPR remained statistically significant in the model (hazard ratio, 1.44 [1.14–1.84, P = 0.003]).

Interestingly, the prognostic value of MPR for cardiac death and major adverse cardiac events was independent of whether patients received a PET-driven medical strategy or a revascularization strategy and of the extent of matching or mismatching defects.