Positron emission tomography and recovery following revascularization (PARR-1): the importance of scar and the development of a prediction rule for the degree of recovery of left ventricular function

Evaluation of functional capacity and quality of life

Canadian Cardiovascular Society angina classification and New York Heart Association dyspnea classification were recorded based on patient self-report. The Minnesota Living with Heart Failure questionnaire was used to assess health-related quality of life (26,27). Functional score using this approach ranges from 0 (good) to 105 (poor). Submaximal exercise capacity was determined using the 6-min walk test as a safe and objective means for evaluating the functional status of patients (28,29).

RNA

The RNAs were acquired using a standard electrocardiogram-gated equilibrium blood pool imaging protocol, using Tc-99m-labelled red blood cells (6,13–15,17,22). The EF was measured from the left anterior oblique 45° acquisition. The RNA analysis was performed in a central core laboratory by observers (R.M. Iwanochko and R. J. Burns, Toronto Hospital) who were blind to any clinical or imaging data (30,31).

FDG PET

A standardized protocol for static FDG PET imaging was used. Patients without diabetes were studied in the postprandial state after a 50-g oral glucose load. Those patients with diabetes or glucose intolerance also received insulin according to a standardized protocol (2,4–6,14,15,22,32,33).

Initial PET transmission scanning was performed to allow for attenuation correction. A 30-min emission scan was acquired starting 40 min after injection of 75 to 370 MBq of FDG in a whole-body PET camera (CTI/Siemens ECAT ART, Knoxville, Tennessee). The FDG uptake images were reconstructed using a Hann window cutoff at 0.2 cycles/pixel.

Perfusion imaging

Where PET perfusion imaging was available, this was performed using either N-13 ammonia or rubidium-82. Both of these have been used routinely in previous studies for perfusion imaging, often interchangeably (2,4,5,7,13–15,32,34,35). A dose of 370 to 740 MBq of rubidium-82 or 300 to 370 MBq of N-13 ammonia was administered intravenously. Standard perfusion imaging acquisition protocols were used (32,36–41). Sites without the capability of PET perfusion imaging performed Tc-99m sestamibi single photon emission computed tomography (SPECT) to assess perfusion (4,6,17,22,42). A dose of 740 MBq of Tc-99m sestamibi was administered intravenously, followed 1 h later by standard 30-min 180° SPECT imaging protocol. All patients were studied under resting conditions. The SPECT images were reconstructed with the same resolution as the PET images and transferred to the PET computer system for analysis.

Perfusion/FDG image data analysis

The heart was divided into 460 sectors by an automated analysis program (41,43). For each sector, tracer activity was expressed as a percentage of the maximum uptake for the patient’s myocardium. The sum of the percentages of all sectors, corrected for the maximum possible score (maximum score: 46,000 = 460 sectors × 100%), defined the total “raw perfusion”score and the total “raw FDG”score, respectively. To normalize the FDG to perfusion, sectors that were ≥80% of the maximum perfusion were defined as normal. FDG uptake was then normalized by scaling the percent FDG uptake to be equal to perfusion in the “maximum zone” sectors. After this normalization, sectors with a value >100% FDG were assigned a value of 100%. The sums of all sectors were calculated again to define the total “normalized perfusion”and the total “normalized FDG”uptake, respectively.

Tissue characterization

In the “abnormal” sectors with <80% perfusion, the extent and severity of hibernating myocardium (“mismatch”score) and “scar”score could be defined as a percentage of the total LV myocardium. Thus, for abnormal sectors:Scar score=∑(100−FDG)46,000in abnormal sectors×100[%ofLD]

In sectors with FDG < perfusion, the perfusion score was substituted for the FDG score in the above equation.

In sectors with FDG > perfusion:Mismatch score=∑(FDG-perfusion)46,000in abnormal sectors×100[%of LD]

The score parameters consider the combination of extent and severity. Using this method, sectors with reduced perfusion can have a mixture of both mismatch and scar scores. The sum of perfusion, mismatch, and scar scores for any abnormal sector equals 100%. To evaluate size alone, the number of sectors defined as scar or mismatch were considered. Figure 1shows examples of the polar maps to determine scar and mismatch scores in three patients.

• Download figure
• Open in new tab
• Download powerpoint

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Examples of reconstructed polar maps for three patients: A, B, C. In each set, the top panelis the raw perfusion (left)and raw F-18-fluorodeoxyglucose (FDG) uptake (right)polar maps; middle panelis the normalized perfusion and FDG uptake; and the lowest panelis the scar score (left)and mismatch score (right). (A)Predominantly scar in the inferolateral and anteroseptal walls and apex. Of the total left ventricle (LV) myocardium, 53% was normal, 42% was scar, and 5% was mismatch. The model predicted a change in ejection fraction (EF) of 0%; observed change was from 26% to 25% = −1%. (B)Partial mismatch (mixture of scar and hibernating myocardium) in the large defect involving the inferolateral wall and apex extending to the distal anteroseptal wall. Of the total LV myocardium, 62% was normal, 23% was scar, and 15% was mismatch. Model predicted change in EF of 4%; observed change was from 23% to 28% = 5%. (C)Small basal inferior scar is noted: 89% was normal, 9% was scar, and 2% was mismatch. Model predicted change in EF was 15%; observed change was from 23% to 37% = 14%.

Statistical analysis

Univariable comparisons evaluated possible significant associations between potential covariates and absolute changes in EF, the response variable. Separate analyses considered relative changes in EF. Size and score values were considered for the following PET-related variables: raw FDG, normalized FDG, raw perfusion, normalized perfusion, mismatch score, and scar score. Each of these variables was modeled as a normal distribution. The following demographic and clinical variables were considered: age, gender, baseline EF, previous revascularization, time to revascularization, symptom class, and diabetes. Candidate demographic and clinical variables were selected based on prior evidence of important prognostic factors in the setting of revascularization (44,45). Although associations with other variables were plausible, the study had sample size constraints. Therefore, in addition to obtaining a satisfactory fit of the data, a secondary goal of the analysis was to limit the number of variables in the model so that there would be at least 10 cases per variable in order to avoid overfitting (44,45).

Multivariable analyses were performed by using stepwise multiple regression methods. The criterion for possible inclusion in the multivariable model were univariable p value <0.2. The criterion for remaining in the multivariable model was an adjusted p value <0.05. Goodness of fit was based on assessment of overall model fit and graphic comparison of observed versus adjusted values residual plots to identify outlying or influential observations or systematic patterns (46). To illustrate the association between scar score and the absolute change in EF on RNA between baseline and post-revascularization follow-up, the scar score was compared across tertiles. Interaction terms were considered. Thus, an indicator variable was included in the model to account for the potential differences in perfusion tracer, as the use of PET or SPECT tracers could potentially affect the degree of scar or mismatch defined (8,47,48)(tracer variable = 1 for PET, 0 for SPECT perfusion).

Independent predictors of change in LV function

The following factors were independently and significantly associated with absolute change in EF after adjustment for other factors (Table 2) : scar score (p = 0.001), tracer (p = 0.043), time to operation (within 6 weeks) (p = 0.008), and diabetes (p = 0.029). The independent effect of the mismatch score was not significant. After adjustment for a statistically significant interaction between mismatch and perfusion tracer, the net effect of the mismatch score when using PET perfusion was significant (p = 0.085 for mismatch score, p = 0.043 for tracer, p = 0.021 for the interaction term). Because of this interaction the net effect of mismatch is better understood by considering the fitted values (discussed in the following text and in Table 3). Prior coronary artery bypass graft was included in the model because it confounded the estimate for age.

Multivariable correlates of change in LV function

Although multivariable models of absolute and relative changes in EF were similar, a better goodness of fit was achieved with absolute change in EF. Therefore, adjusted values were calculated for absolute change in EF (Table 3). The overall fit of the model was significant (F = 4.35, df = 8, p = 0.001). No observations were identified as being outlying or influential. Table 3describes the overall effect of an incremental change in selected variables on the fitted absolute change in EF while holding all other variables in the model constant. For this analysis mean values were used for continuous variables and modal values were used for categorical variables (tracer = 1, PET). The mean absolute change in EF was 4.3 ± 1.68%. A 10-point incremental increase in the extent of scar would reduce the change in EF by 3.38% to 0.94 ± 2.16%. A 10-point increase in mismatch using PET perfusion would increase the change in EF by 1.99% to 6.31 ± 2.37%. The complete model explained 36.4% of the variance in EF. The observed versus the fitted absolute changes in EF calculated from the multiple linear regression are shown in Figure 2(r = 0.60; p = 0.0003).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Prediction model: observed vs. fitted change in absolute ejection fraction. Correlation coefficient: r = 0.60; p = 0.0003.

Extent of scar and LV function recovery

To illustrate the relationship between scar score and change in EF, scar scores were divided into tertiles: (I, small: 0% to 16%; II, moderate: 16% to 27.5%; III, large: 27.5% to 47%). The respective changes in EFs were 9.0 ± 1.9%, 3.7 ± 1.6%, and 1.3 ± 1.5% (mean ± SD). There was a statistically significant difference in the absolute change of EF between patients with small scar score (0% to 16% of LV) compared with large scar score (27.5% to 47% of LV) (overall p = 0.006 by analysis of variance; p = 0.002 for tertile I [small] vs. III [large] by ttest) (Fig. 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Absolute change in ejection fraction (EF) versus scar scores: small (0% to 16% of left ventricle [LV]), moderate (16% to 27.5% of LV), and large (27.5% to 47% of LV); * p = 0.002, small versus large scar score.

Utility of the prediction model

In this study, several clinical parameters had impact on LV function recovery. The presence of higher risk parameters, including diabetes, age, and previous revascularization, increased the degree of recovery. At first glance, this may appear paradoxical. However, other studies (55)have shown that the patients at greatest risk are often those who gain the most benefit from treatment. It is possible that older patients and those with previous bypass have better developed collateral flow, which could facilitate recovery. Alternatively, a selection bias may have occurred whereby subjective criteria for revascularization decisions are stricter with the higher risk patients. In the case of diabetes, it may also indicate that the level of viable myocardium is greater than apparent on imaging because patients with diabetes tend to be more difficult to image (56–60). Examples of applying the model to three patients in the current study who had small, moderate, and large scar areas are shown in Figure 1.

Study limitations

Some limitations of this approach are worth considering. These primarily relate to the patient population in this study, which included patients who were predominantly men, predominately between 53 and 71 years of age (1 standard deviation from the mean), had multivessel disease, and had bypassable vessels. The generalizability of the prediction model to a wider patient population is being tested in a large randomized controlled trial (PET and Recovery following Revascularization–Phase 2 [PARR-2]).

The current study was intended as a pilot study to establish the prediction model for subsequent trials. The sample size was larger than many PET studies evaluating recovery of function. However, the sample size limited the parameters that could be included in the prediction model. Thus, some potentially important parameters, such as stress-induced ischemia and ventricular volumes, were not included. The number of parameters used simplifies the model for research and clinical use but also means that prediction rule for a given patient must be taken in the context of clinical judgment and image interpretation. The impact of some of these variables on the model will be tested in the larger ongoing randomized control trial.

The use of both PET and SPECT perfusion agents in the current study represents the clinical reality, as many institutions do not have easy access to PET perfusion agents. This may account for the lack of significance of the independent mismatch term. To account for potential differences in mismatch due to the type of perfusion tracer (PET or SPECT), an interaction variable (tracer variable) was used in the model. When PET perfusion was used, the overall effect of mismatch was for a predicted increase in EF. When SPECT perfusion was used, the mismatch effect was not significant and the confidence limits of the point estimate overlap 0. Thus, interpretation of the effect of mismatch with SPECT perfusion must be done cautiously. This limitation of the model likely reflects the small sample size. Ongoing clinical trials will allow evaluation of each PET variable with greater accuracy.

Although improvement in LV function has been noted at three months of follow-up in many previous studies, recent data suggest that more recovery may be observed with longer follow-up time (51,52,61). The impact of follow-up time on the model could not be addressed in the current study but will be considered in the larger ongoing randomized control trial that will follow up patients up to two years.

It could be argued the angiographic data should be included in the model. Clearly this is an important parameter in decision making. However, patients in this study did have suitable anatomy for revascularization. If they did not, they were not generally considered for revascularization and the model would not be applicable. Second, many patients referred for PET (about 50% in our institution) have not had a recent angiogram. Therefore, coronary anatomy data would not always be available to incorporate into the model. The prospective randomized controlled trial (PARR-2) includes patients who have as well as those who have not had recent angiography before the PET study: so the model can be validated in this patient population.