Prediction is very difficult, especially if it is about the future (Niels Bohr)

Assessment of risk of future events in patients with suspected or known coronary artery disease (CAD) is challenging. Many studies are published every month on prognostic factors, and novel prognostic models and flow-chart are proposed for clinical use. A quantity of data accumulated since the 70’ demonstrates the strength of stress single-photon emission-computed tomography myocardial perfusion imaging (MPI) in predicting outcome.1,2 MPI is also commonly used as a gatekeeper to select patients for coronary revascularization.3,4 The degree of left ventricular (LV) dysfunction and the extent and magnitude of inducible myocardial ischemia are important prognostic variables that can be assessed by MPI. Transient ischemic LV cavity dilation is another high-risk variable associated with the presence of severe angiographic CAD5 and with an adverse outcome.6,7 Serial MPI imaging has been used to compare the effectiveness of intercurrent treatments, within the framework of randomized controlled trials.8-10 Moreover, an analysis of the Detection of Ischemia in Asymptomatic Diabetics (DIAD) study not including patients with intervening cardiac events or revascularization found unexpected resolution of ischemia in most of these patients, potentially because of more aggressive medical treatment of cardiovascular risk factors.11

Prognostic Value of Regadenoson MPI

Regadenoson is the first selective adenosine A2A receptor agonist approved by the Food and Drug Administration and is the most widely used stress agent for MPI in the United States. It has many of the characteristics of an ideal stress perfusion agent, being a potent and a selective coronary vasodilator with a rapid onset of action, a short duration of action, and being administered as a fixed-dose bolus (not weight-based). Further, it has a good safety and tolerability profile including in patients with reactive airway disease, and its side effects can be readily reversed by an antagonist if needed.12 The diagnostic accuracy of regadenoson is comparable to adenosine MPI.13 Regadenoson MPI also provide powerful prognostic information that has important implications in patient management and can guide clinical practice.14 Hage et al15 categorized 1400 patients (42% male, 37% diabetes, 21% heart failure, 26% end-stage renal disease) based on the perfusion defect size (PDS) using automated quantitative analysis. The primary outcome was a composite of cardiac death, myocardial infarction, and late coronary revascularization (>90 days after MPI). The primary outcome occurred in 23% of the patients during 46 ± 18 months of follow-up and 8% had early coronary revascularization (within 90 days of MPI). In an adjusted Cox proportional model, the hazard ratio for the primary outcome progressively increased with the extent of PDS.

Serial MPI Studies and Prognosis

The prognostic significance of serial change of imaging parameters has been previously reported.16-18 Iskandrian et al19 in a landmark study first comprehensively reviewed where serial MPI might be useful clinically and in research studies. They also highlighted strategies for addressing the various issues that are unique to serial testing in order to derive more valid and robust data from the serial scans. In fact, criteria for defining a real MPI change in an individual patient beyond variability are paramount to interpret and use information from serial testing. In this issue of the Journal, El-Hajj et al20 evaluated the effect of changes in PDS and LV ejection fraction (EF) during serial myocardial perfusion imaging on cardiovascular outcomes in high-risk patients. The authors retrospectively evaluated 698 patients (61 ± 11 years, 53% male) who underwent two regadenoson MPI studies within 16 ± 9 months for clinical indications. Patients were stratified into 3 groups: group 1 included patients with normal perfusion on MPI-1 and MPI-2; group 2 included patients with abnormal perfusion on MPI-1 and with improvement (≥5% decrease in PDS) on MPI-2; group 3 included patients with abnormal perfusion on MPI-1 and with either no improvement (<5% change) or worsening (≥5% increase in PDS) on MPI-2 and patients with normal perfusion on MPI-1 and abnormal perfusion on MPI-2. Changes in PDS and EF on serial MPI studies provided incremental prognostic information to initial and follow-up MPI findings. A new perfusion abnormality or an increase in PDS and a drop in LVEF identified high-risk patients. In the period between the 2 MPI studies, 80 (12%) patients underwent coronary revascularization: most of these patients were in group 2. The primary endpoint was a composite of all-cause death, non-fatal myocardial infarction or coronary revascularization after the second MPI. During 24 ± 16 months of follow-up after the second MPI, the primary outcome occurred in 167 (24%) patients (8% death, 9% myocardial infarction, 15% coronary revascularization). Interestingly, the study cohort consisted of high-risk patients defined with high prevalence of end-stage renal disease (21%), diabetes (48%), prior myocardial infarction (25%), and prior coronary revascularization (49%). However, it appears that 437 (63%) patients had a normal first study. In particular, in 399 patients both MPI studies were normal. Our group recently performed a meta-analysis of published studies including diabetic patients with known or suspected CAD to assess the predictive value of normal stress MPI.21 During a weighted mean follow-up of 36.2 months, the negative predictive value for cardiac death and non-fatal myocardial infarction of normal MPI was 94.92% [95% confidence interval (CI) 93.67-96.05], resulting in an estimated annualized event rate after a negative test of 1.60% (95% CI 1.21-2.04). Thus, in the study of El-Hajj et al,20 a significant number of deaths were probably due to non-cardiac cause, questioning the link between MPI results and outcome. Ideally, competing risk analysis could have given further insights into the prognostic value of serial assessment myocardial perfusion.22

Warranty Period of Normal MPI Study

El-Hajj et al20 found that only 38 out of 437 patients with normal perfusion on MPI-1 had abnormal perfusion on MPI-2. The authors clearly stated that patients in groups 2 and 3 were significantly older, had more co-morbidities, higher prevalence of prior myocardial infarction and coronary revascularization, and were on more cardiovascular medications as compared to group 1, but they did not report if some covariates might predict the switch from normal to abnormal MPI. Instead, any effort to clarify the warranty period of a normal MPI is welcome. It must be outlined that the warranty period of a normal stress MPI varies according to clinical variables and post-stress LVEF. In the study of El-Hajj et al,20 193 (28%) patients, proportionately distributed among the 3 groups, had a drop in LVEF ≥5%. When the drop in LVEF was considered in the multivariate Cox model, it was independently associated with the primary outcome [hazard ratio (HR) 1.5, 95% CI 1.1-2.1, P = .01]. These results agree with those reported by Acampa et al23 who evaluated the relationship between diabetes and temporal characteristics of cardiac risk at long-term follow-up in a propensity score-matched cohort of diabetic and non-diabetic patients with normal MPI. After matching, clinical characteristics were comparable in 260 diabetic and 260 non-diabetic patients. All patients were followed for at least 1 year (median 53 months). End-point events were cardiac death or nonfatal myocardial infarction. At Cox multivariable analysis, diabetes and post-stress LVEF ≤45% were independent predictors of events. At parametric analysis, non-diabetic patients with post-stress LVEF >45% remained at low risk for the entire length of follow-up, while the highest probability of events and the major risk acceleration was observed in patients with diabetes and post-stress LVEF ≤45%. Thus, the warranty period of a normal stress MPI varies according to diabetic status and post-stress LVEF.

Challenges of Longitudinal Studies

Longitudinal study refers to investigations where participant outcomes and possibly treatments or exposures are collected at multiple follow-up times, yielding repeated measurements on each subject. In longitudinal clinical study, it is also common to observe relevant events, generating time-to-event data. If the interest is to analyze the longitudinal outcome response variable with dropout at the time of event, longitudinal data analysis is performed. Unlike cross-sectional designs, where observations from study subjects are available only at a single time point, individuals in longitudinal or cohort studies are assessed repeatedly over time. By taking advantages of multiple snapshots of a group over time, data from longitudinal studies capture both between-individual differences and within-individual dynamics, affording the opportunity to study more complicated biological, psychological, and behavioral hypotheses than their cross-sectional counterparts. However, longitudinal data present multiple methodological challenges in study designs and data analyses, such as the correlation among the repeated responses of the same subject, heterogeneous variability (the variance of the response changes over the study), and the presence of missing data.24 Approaches to modeling continuous longitudinal data are the analysis of response profiles, linear mixed-effects models, and generalized estimating equations. Another problem is that only patients who survived to further observations may be included in the analysis (an example of informative censoring). This is more evident in retrospective study and is known as “immortal time,” a period of follow-up during which, by design, death or the study outcome cannot occur.25,26 This bias may have serious consequences in particular in the evaluation of life-extension benefits of therapy (Figure 1). Several statistical approaches have been suggested to avoid this source of bias and these include Cox’s proportional hazards regression with time-varying covariates, a modified form of the Kaplan-Meier analysis, and the Poisson regression.27 More recently, joint modeling of longitudinal and survival data has been proposed to investigate the relationship between a repeatedly measured marker, subject to measurement error, and the time to an event of interest.28

Figure 1
figure 1

Types of immortal time bias. The period between entry in the study and date of first exposure (to a drug, a procedure to a diagnostic test), during which death has not occurred, is either misclassified (A) or simply excluded and not accounted for in the analysis (B). Both “misclassified immortal time” (A) and “excluded immortal time” lead to a bias in favor of the intervention group

Full size image

Future Directions

It should be noted that in trials using MPI to assess the effects of therapy, serial imaging would be stronger if at least moderate ischemia is an inclusion criterion, exceeding the variability of MPI and thus assuring that the enrolled patients actually have ischemia.19 Farzaneh-Far et al29 identified 1425 consecutive patients with angiographically documented CAD who underwent 2 serial MPI scans within a 36-month time frame. They found that ischemia worsening is an independent predictor of death or myocardial infarction, resulting in significantly improved risk reclassification when added to previously known predictors. Despite this finding, the authors concluded that their results could not be used as justification for performing serial MPI scans and that randomized prospective trials are required before any such recommendations can be proposed. The ongoing International Study of Comparative Health Effectiveness With Medical and Invasive Approaches (ISCHEMIA) Trial attempts to demonstrate the extent to which an angiographic-driven strategy for higher risk stable ischemic heart disease patients with moderate-severe ischemia will or will not improve clinical outcomes.4 This trial will be completed in ~2019. In the meantime, the data of El-Hajj et al20 further suggest that if a patient has had 2 consecutive MPI studies performed for appropriate clinical reasons, the information regarding perfusion change may be used to improve prognostication and can guide clinical practice.