Prediction of 2-year Major Adverse Cardiac Events from Myocardial Perfusion Scintigraphy and Clinical Risk Factors

Abstract

Background:

We developed CRAX2MACE, a new tool derived from clinical and SPECT myocardial perfusion imaging (MPI) variables, to predict 2-year probability of major adverse cardiac event (MACE) comprising death, hospitalized acute myocardial infarction or coronary revascularization.

Methods:

Consecutive individuals with SPECT MPI 2001–2008 had two-year MACE determined from population-based health services data. CRAX2MACE included age, sex, diabetes, recent cardiac hospitalization, pharmacologic stress, stress total perfusion deficit (TPD), ischemic (stress-rest) TPD, left ventricular ejection fraction and transient ischemic dilation ratio. Two-year event rates were classified as low (<5%), moderate (5.0–9.9%), high (10–19.9%) and very high (20% or greater).

Results:

The study population comprised 3,896 individuals for the development and 1,946 for the validation subgroups with subsequent MACE in 589 (15.1%) and 272 (14.0%), respectively. CRAX2MACE, derived from the development subgroups, accurately stratified MACE risk in the validation subgroup (area under the receiver operating characteristics curve 0.79) with stepwise increase in the observed event rate with increasing predicted risk category (low, 2.3%; moderate, 5.5%; high, 18.8%; very high 33.2%; p-trend <0.001).

Conclusions:

A simple tool based upon clinical risk factors and MPI variables predicts 2-year cardiac events. Risk stratification between the low and very high groups was greater than 10-fold.

Introduction

Coronary artery disease (CAD) remains the principle cause of death in developed countries despite major advances in both the medical and interventional approaches to management, with an emerging epidemic among low- and middle-income countries which are now responsible for overwhelming majority of cardiovascular disease worldwide.^1,2 Myocardial perfusion imaging (MPI) is a powerful technique to assess for the presence and severity of CAD, and is predictive of both short term and long term cardiac events.^3,4 Clinical risk factors and derived composite scores are also widely used to stratify risk for CAD and related outcomes.⁵ To date, there have been limited attempts to quantitatively integrate MPI and clinical risk factors into a unified risk measure that could potentially be used to generate a personalized risk profile and guide treatment, although several prognostic risk models have been developed and adopted for non-MPI laboratory and imaging measures.⁶

There is increasing consensus that therapeutic intervention should be personalized based upon an individual’s level of risk and anticipated benefit.^7–9 An accurate assessment of baseline risk is therefore an essential component of this process. A previous study provided proof-of-concept that a prediction tool, CRAX, for five year death and 5-year acute myocardial infarction (AMI) based upon fully-automated MPI analysis and clinical risk factors was feasible.¹⁰ Limitations of CRAX included the relatively long time horizon of 5 years. Moreover, diabetes was not included in this model though studies document the importance of diabetes as an important risk factor for adverse cardiac outcomes even when adjusted for MPI variables, and as a CAD equivalent.^11–13

The objective of the current analysis was to develop a new clinical prediction tool, CRAX2MACE, to predict 2-year probability of major adverse cardiac event (MACE) that comprised death, hospitalized AMI or coronary revascularization. Clinical risk factors and MPI variables were included in the prediction algorithm which was developed and internally validated in a large clinical cohort.

Materials and Methods

Demographics

The study cohort and methods have been previously described in detail.^10,14 Briefly, we identified consecutive individuals who underwent stress-rest SPECT myocardial imaging at St. Boniface Hospital, Winnipeg, the regional cardiac center for the province of Manitoba (1.3 million), for suspected coronary artery disease between February 2001 and July 2008. We excluded subjects due to incomplete or corrupted scan data. In addition, studies being performed as follow-up to a previous scan were excluded from consideration. This retrospective study was approved by the institutional review board and the need for written informed consent was waived.

Stress Testing and SPECT Imaging

Standard 99mTc-sestamibi or tetrofosmin rest/stress protocols were employed as previously described with treadmill testing or dipyridamole infusion with low-level exercise.^10,14 Adjunctive low-level exercise was performed when feasible and not contraindicated (e.g., left bundle branch block). Dual-detector scintillation cameras with low-energy, high-resolution collimators (Vertex; Philips Medical Systems, Milpitas, CA) were used to acquire MPI. Tomographic reconstruction was performed by AutoSPECT and Vantage Pro programs (Philips Medical Systems). Emission images were automatically corrected for non-uniformity, radioactive decay, center-of-rotation, and motion during acquisition. Filtered back-projection and Butterworth filters were applied to obtain the non-attenuation corrected MPI with an order of 10 and cut-off of 0.50 for rest MPI, and an order of 5 and cut-off of 0.66 for stress MPI. Non-attenuation correction images were used for the current analysis as we have previously reported that the use of attenuation did not significantly improve the prognostic performance of MPI.¹⁴

Automated Image Analysis

Left ventricular segmentation, left ventricular contour quality control, quantitation of left ventricular ejection fraction (LVEF) and global total perfusion deficit (TPD) as a percentage of the myocardium were performed with automated software without user intervention (QGS-QPS software, Cedars-Sinai Medical Center, Los Angeles, CA, USA).¹⁵ All image analyses were performed by researchers at Cedars-Sinai Medical Center as previously described using de-identified image files to ensure blinding to all clinical information and outcomes.¹⁴ Only 10% of rest/stress studies required visual correction of contours. Stress and rest TPD were measured independently from the stress and rest scans, respectively, and ischemic TPD was defined as their difference: (stress TPD minus rest TPD, negative results set to zero i.e., no ischemia).

Clinical Risk Factors

In the Province of Manitoba, Canada, health services are provided to virtually all residents through a single public health care system. Registry data are regularly updated to capture loss of coverage due to migration or death (Vital Statistics). Manitoba Health maintains computerized databases of the physician services and hospitalizations provided for all persons registered with the system. Each physician service record includes information on the identity of the patient, the date of service, services provided, and a diagnosis, which is subsequently coded to a 3-digit International Classification of Disease-9-Clinical Modification (ICD-9-CM) code. After each hospitalization, a detailed abstract is created; prior to 2004, this included up to 16 diagnoses as 5- digit ICD-9-CM codes, and from 2004 onwards included up to 25 diagnoses as International Classification of Diseases-10-Canadian Enhancements (ICD-10-CA) codes. This data repository allows for the creation of a longitudinal record of health service utilization for an individual through a unique personal health identification number, anonymized to preserve patient confidentiality. The accuracy of administrative data has been established for a wide range of clinical disorders, including cardiovascular outcomes.^16,17

Outcomes

The primary endpoint of this study was the composite endpoint of MACE within 2 years, which included death from any cause, hospitalized AMI, or late coronary revascularization (including both percutaneous coronary intervention and coronary artery bypass grafting >90 days following MPI). The individual endpoints were examined as secondary outcomes. Finally, we also examined MACE within 5 years as a secondary outcome.

Model Development

The Cardiovascular Risk Assessment for 2-year MACE (CRAX2MACE) tool was modeled with the goal of combining relevant clinical and imaging data in order to optimize the prediction of cardiovascular outcomes. Subjects were randomly assigned to the development (66.7%) or validation groups (33.3%). Based upon our previous work, variables were limited to those that had previously shown the strongest independent associations with adverse outcomes with the addition of diagnosed diabetes mellitus.¹⁰ Diabetes was ascertained from physician services and hospitalizations in the prior 3 years using a validated definition.^18,19 The only other change in the predictor variables was the use of recent cardiac hospitalization (within the last 3 years) rather than all-cause hospitalizations in the model. Due to non-normality and zero-inflation in the TPD measures, stress TPD was categorized as <5% (referent), 5 – 9%, 10 – 19%, 20 – 29%, and 30% or greater. Ischemic TPD was categorized as < 5% (referent), 5 – 9%, 10 – 19% and 20% or greater. Left ventricular ejection fracture (LVEF) from the 8-bin gated rest images was categorized as ≥ 45% (referent), 40 – 44%, 30 – 39% and < 30%.^20,21 The ratio of left ventricular volume post-stress versus rest was used as an index of transient ischemic dilation (TID) from the summed images. The laboratory normal ranges for normal TID are <1.20 for exercise and <1.25 for pharmacologic stress derived from almost 300 individuals with normal MPI and LVEF.

Statistical Analysis

Statistical analyses were performed using Statistica (Version 12.0, StatSoft, Inc., Tulsa, OK, USA). Categorical variables are presented as frequencies, and continuous variables as mean ± SD. All statistical analyses were two-tailed, and a p value of < 0.05 was considered significant. Logistic regression models were constructed to estimate odds ratios (OR) for each of the prediction variables. Model stratification was assessed using area under the curve (AUC) and calibration was assessed with the Hosmer-Lemeshow test. Model outputs for 2-year MACE was categorized as <5% (low), 5.0–9.9% (moderate), 10–19.9% (high) and 20% or greater (very high) which were found to give roughly equal sized groups (approximating quartiles). Linear trend for risk category versus predicted outcome percentages was tested using the Cochran-Armitage test.

Results

The study population consisted of 3,896 individuals for the development subgroup (mean age 64.7 ± 12.0 years, 52.9% male) and 1,946 for the validation subgroup (mean age 65.1 ± 11.5 years, 51.3% male). The development and validation subgroups were well balanced in terms of the model variables (Table 1).

TABLE 1:

Characteristics of the development and validation subgroups.

All cases	Development	Validation	p-value
N=	3896	1946
Age (years)	64.7 ± 12.0	65.1 ± 11.5	0.193
Male sex (vs female)	2061 (52.9)	998 (51.3)	0.244
Diabetes	1225 (31.4)	613 (31.5)	0.964
Pharmacologic stress (vs exercise)	1357 (34.8)	677 (34.8)	0.975
Recent cardiac hospitalization	2328 (59.8)	1189 (61.1)	0.322
Stress TPD	10.3 ± 11.9	9.7 ± 11.5	0.080
Ischemic TPD [stress-rest]	3.1 ± 4.9	3.0 ± 4.7	0.230
LVEF	55.6 ± 14.8	56.1 ± 14.7	0.319
TID ratio	1.03 ± 0.13	1.03 ± 0.13	0.941
Primary end point, MACE 2 years	6.1 ± 2.7	6.1 ± 2.6	0.872

Data are mean standard deviation or N (percent). p-value from t-test. TPD= total perfusion deficit, LVEF=left ventricular ejection fraction, TID=transient ischemic dilatation, AMI=acute myocardial infarction, MACE=major adverse cardiac event (AMI, death, revascularization).

During two years of follow-up, the primary MACE outcome occurred in 589 (15.1%) and 272 (14.0%) of the development and validation subgroups, respectively. The secondary endpoint of AMI or death occurred in 401 (10.3%) and 179 (9.2%), respectively. The percent of individuals experiencing these endpoints was similar in the development and validation subgroups. Older age, male sex, diabetes, recent cardiac hospitalization, pharmacologic stress, stress TPD, ischemic TPD, LVEF and TID ratio were significantly associated MACE (Table 2).

TABLE 2:

Baseline characteristics of the development subgroup according to outcome.

All cases	No MACE	MACE	p-value
N=	3307	589
Age (years)	64.0 ± 12.1	68.3 ± 11	<0.001
Male sex (vs female)	1683 (50.9)	378 (64.2)	<0.001
Diabetes	957 (28.9)	268 (45.5)	<0.001
Recent cardiac hospitalization	948 (28.7)	409 (69.4)	<0.001
Pharmacologic stress (vs exercise)	1869 (56.5)	459 (77.9)	<0.001
Stress TPD	9.0 ± 11.0	17.7 ± 13.9	<0.001
Ischemic TPD [stress-rest]	2.6 ± 4.3	6.0 ± 6.7	<0.001
LVEF	57.0 ± 14.1	48.0 ± 16.7	<0.001
TID ratio	1.02 ± 0.13	1.06 ± 0.13	<0.001

Data are mean standard deviation or N (percent). p-value from t-test. TPD= total perfusion deficit, LVEF=left ventricular ejection fraction, TID=transient ischemic dilatation, MACE=major adverse cardiac event (acute myocardial infarction, death, revascularization).

Results from the logistic regression in the development subgroup are summarized in Table 3. Age, diabetes, recent cardiac hospitalization and pharmacologic stress (inability to perform treadmill exercise) were independently associated with the primary outcome; sex was not independently predictive of MACE but was forced into the final model. Stress TPD, ischemic TPD, LVEF and TID ratio were all independently predictive of the primary outcome. These variables resulted in good model fit according to the Hosmer-Lemeshow test (P-value = 0.28). The model parameters and construction of the 2-year prediction probability is provided in the Appendix. AUC for 2-year MACE prediction was 0.79 in the development subgroup was 0.79 for the validation subgroup. MACE risk stratification for the validation subgroup persisted up to 5 years (AUC 0.80). MACE risk stratification for the validation subgroup was similar in sex-stratified analyses (AUC males 0.77, females 0.78, p-interaction=0.72).

TABLE 3:

Odds ratios for MACE from multivariable logistic regression (developmental group).

Characteristic	Level	Odds Ratio (95% CI)	p-value
Age	Per 10 years	1.18 (1.08–1.29)	<0.001
Sex	Male vs female	1.08 (0.86–1.34)	0.507
Stress procedure	Pharmacologic vs exercise	1.29 (1.01–1.65)	0.038
Diabetes	Present vs absent	1.44 (1.18–1.75)	<0.001
Recent cardiac hospitalization	Present vs absent	3.45 (2.80–4.25)	<0.001
Stress TPD	<5%	REFERENT	<0.001
	5–9%	1.20 (0.89–1.64)
	10–19%	1.43 (1.02–2.01)
	20–29%	1.50 (1.01–2.24)
	30%+	1.82 (1.19–2.80)
Ischemic TPD	<5%	REFERENT	<0.001
	5–9%	1.40 (1.06–1.85)
	10–19%	2.47 (1.75–3.48)
	20%+	2.69 (1.43–5.06)
LVEF	>45%	REFERENT	<0.001
	40–44%	1.20 (0.80–1.81)
	30–39%	1.47 (1.03–2.09)
	<30%	1.96 (1.35–2.85)
TID	Elevated vs normal	1.47 (1.02–2.14)	0.041

TPD= total perfusion deficit, LVEF=left ventricular ejection fraction, TID=transient ischemic dilatation, MACE=major adverse cardiac event (acute myocardial infarction, death, revascularization).

Two-year event rates according to CRAX2MACE were classified as low (< 5%), moderate (5.0 – 9.9%), high (10 – 19.9%) and very high (20% or greater). Approximately equal numbers feel within each risk range, with a similar breakdown for the development and validation subgroups (Figure 1).

Figure 1.

Percentages of individuals classified by CRAX2MACE as low (<5%), moderate (5.0–9.9%), high (10–19.9%) and very high risk (20% or greater).

There was a stepwise increase in the observed event rate with increasing risk category (Cochran-Armitage p-trend <0.001). The observed fraction of individuals experiencing MACE fell within the predicted range (Figure 2). For the validation subgroup risk increased as follows: low, 2.3%; moderate, 5.5%; high, 18.8%; very high 33.2%. Using the same risk thresholds, a similar stepwise increase in observed risk was seen for the individual 2-year secondary outcomes of AMI (low, 0.5%; moderate, 1.3%; high, 4.3%; very high 6.4%; p-trend <0.001), coronary revascularization (low,1.1%; moderate, 1.2%; high, 9.9%; very high 13.4%; p-trend <0.001), and death (low, 0.9%; moderate, 4.0%; high, 6.9%; very high 17.4%; p-trend <0.001), and also for 5-year MACE (low, 6.5%; moderate, 14.7%; high, 34.9%; very high 59.6%; p-trend <0.001).

Figure 2.

Observed 2-year risk for the primary outcome of MACE (upper left panel) and individual secondary outcomes (AMI, death, coronary revascularization) according to CRAX2MACE predicted risk category. Error bars are 95% confidence intervals.

Discussion

A simple tool for prediction of 2-year major adverse cardiac events (MACE) was constructed and internally validated based upon relevant clinical risk factors and MPI variables. Risk stratification between the low risk and very high risk groups was greater than 10-fold for the primary combined endpoint. Good risk stratification was also seen for the individual outcomes, and when MACE outcomes were evaluated over 5 years.

Many studies have demonstrated the ability of both visual (semi-quantitative) and quantitative MPI to detected CAD and predict important clinical outcomes when adjusted for baseline characteristics including those with diabetes.^{14,15,22–28} The current study extends previous work that demonstrated proof-of-concept that integrating clinical and MPI variables lead to better prediction of outcomes than clinical risk factors or MPI variables alone.¹⁰ Although CRAX was developed to predict long-term (5 year) hard outcomes (AMI or death), CRAX2MACE is potentially more useful for identifying individuals at imminent risk of adverse events or requiring coronary revascularization. These two approaches, assessment of long-term and short-term risk, may be complementary.

A machine learning (ML) approach has even greater potential to exploit the value of combining clinical and MPI data. Betancur et al.²⁹ trained an ensemble (“boosted”) algorithm to predict 3-year risk of MACE using 22 imaging data, 8 stress test, and 17 clinical variables in 2,619 consecutive patients. Prediction was significantly higher for the combined model than models built from imaging data alone, physician diagnosis, automated stress TPD, or automated ischemic TPD. CRAX2MACE was engineered to easily integrate into the routine clinical workflow within a nuclear medicine laboratory, with limited ancillary clinical data collection. The value of integrating quantitative SPECT variables with simple clinical risk factors to generate a personalized risk profile and guide treatment could be of particular value in developing countries, where nuclear medicine utilization remains low due to lack of needed equipment and experienced readers.³⁰

Limitations to this analysis are acknowledged. We were parsimonious in the selection of candidate variables, acknowledging that increasing the complexity of the model with additional variables would undoubtedly improve performance, but at the expense of becoming more cumbersome for clinical application.³¹ Our dataset did not include measures for height, weight, adiposity, blood pressure, lipid status, diabetes duration or control, smoking or other lifestyle variables. We also acknowledge that therapies and nuclear imaging technology have advanced since 2001–2008, and it is uncertain whether this would affect model performance in the current era. The prediction model was developed and internally validated within the same dataset. External validation in other cohorts and in dedicated camera imaging systems would clearly be an important step before widespread clinical application. Whether CRAX2MACE reporting will actually help to guide clinical decision making would need to be prospectively evaluated.

New Knowledge Gained.

A simple prediction tool based upon a small number of clinical and MPI variables was found to stratify 2-year risk for MACE. Further studies are warranted to see whether this can be incorporated into routine clinical reporting of MPI and ultimately contribute to individualized patient decision-making and improved clinical outcomes.

Abbreviations

SPECT

Single photon emission computed tomography

MPI

Myocardial perfusion imaging

AMI

Acute myocardial infarction

MACE

Major adverse cardiac event

LVEF

Left ventricular ejection fraction

TID

Transient ischemic dilatation

TPD

Total perfusion defect

CRAX

Cardiovascular risk assessment

CRAX2MACE

Cardiovascular risk assessment for MACE at 2 years

APPENDIX:

Calculation of CRAX2MACE for estimating 2-year probability of MACE.

• Calculate logodds =

  • −4.35770
  • +Age*0.0164
  • +Sex=male*0.0745
  • +Stress=pharmacologic*0.2568
  • +Diabetes=present*0.3622
  • +Recent cardiac hospitalization=present*1.2391
  • +(stressTPD=‘5–9%’)*0.1852
  • +(stressTPD=‘10–19%’)*0.3592
  • +(stressTPD=‘20–29%’)*0.4068
  • +(stressTPD=‘30%+’)*0.6004
  • +(ischemicTPD=‘5–9%’)*0.3378
  • +(ischemicTPD=‘10–19%’)*0.904
  • +(ischemicTPD=‘20%+’)*0.9884
  • +(LVEF=‘40–44%’)*0.1816
  • +(LVEF=‘30–39%’)*0.3858
  • +(LVEF=‘<30%’)*0.6745
  • +(TID=‘high’)*0.3884
• Convert to Probability =

  • exp(logodds) /[1 + exp(logodds)]