For Percutaneous Coronary Interventions in the Clinical Outcomes Assessment Program

ABSTRACT: Published mortality models for percutaneous coronary intervention (PCI), including the Clinical Outcomes Assessment Program (COAP) model, have not considered the effect of out-of-hospital cardiac arrest. The primary objective of this study was to determine if the inclusion of out-of-hospital cardiac arrest altered the COAP mortality model for PCI. The COAP PCI database contains extensive demographic, clinical, procedural and outcome information, including out-of-hospital cardiac arrest, which was added to the data collection form in 2006. This study included 15,586 consecutive PCIs performed in 31 Washington State hospitals in 2006. Using development and test sets, the existing COAP PCI logistic regression mortality model was examined to assess the effect of out-of-hospital arrest on in-hospital mortality. Overall, 2% of individuals undergoing PCI had cardiac arrest prior to hospital arrival. Among 8 hospitals with PCI volumes < 120 cases per year, 4 had cardiac arrest volumes that exceeded 10% of total volume, whereas none of the centers with > 120 cases per year did. In-hospital mortality was 19% in the arrest group and was 1.0% in remaining procedures (p < 0.0001). In the new multivariate model, out-of-hospital cardiac arrest was highly associated with mortality (odds ratio = 5.50; 95% confidence interval [CI] = 3.28–9.25). When evaluated in the test set, the new model had excellent discrimination (c-statistic = 0.89; 95% CI = 0.85–0.93). Out-of-hospital cardiac arrest is an important determinant of risk-adjusted in-hospital mortality for PCI, particularly for hospitals with low volumes and relatively high volumes of cardiac arrest cases.

J INVASIVE CARDIOL 2009;21:1–5

Over the past 15 years, state and national organizations have devoted considerable effort and resources to measuring hospital and operator performance for percutaneous coronary interventions (PCI).1–10 One of these organizations, the Clinical Outcomes Assessment Program (COAP), measures performance as part of a collaborative quality improvement program for hospitals performing cardiac revascularization procedures in Washington State. Published mortality models for PCI, including the COAP model, have not considered the effect of out-of-hospital cardiac arrest. There is minimal information about the use of PCI in patients with out-of-hospital cardiac arrest, and most of it is from descriptive studies conducted outside the United States.11–14

Since several of the 31 participating COAP hospitals have relatively high proportions of patients with out-of-hospital cardiac arrest, application of available risk stratification models could potentially penalize these sites by underestimating their expected mortality rates. With this in mind, the purpose of this study was to: 1) determine in COAP the proportion of cases with out-of-hospital cardiac arrest; 2) compare baseline and procedural characteristics as well as hospital mortality in cases with and without the condition; and 3) assess whether out-of-hospital cardiac arrest added to the predictive power of the COAP mortality model for PCI.10

Methods

Patient population. This study included 15,586 PCIs performed in 31 Washington State hospitals between January 1, 2006 and December 31, 2006. With the exception of 1 military hospital, all consecutive procedures performed in the state during 2006 were captured.

Study variables. The COAP PCI database contains extensive information which is collected by skilled medical records abstractors. Data are entered at the local hospital on case report forms and are electronically transmitted to a central contractor that in turn carefully checks the completeness and accuracy of data before constructing analytic files. This process of chart abstraction, data transmission and management has been in place since 1999.9,10

Variables in the published COAP PCI mortality model are cardiogenic shock, age, nonelective procedure (including urgent and emergent categories), baseline creatinine > 2.0 mg/dL, ejection fraction, acute myocardial infarction < 24 hours from admission, history of chronic obstructive pulmonary disease, male gender, history of peripheral vascular disease, history of PCI and history of congestive heart failure.10

In 2006, a variable describing cardiac arrest was added to the COAP data collection form. The definition of cardiac arrest specifies that chest compressions and/or intubation for resuscitation were required and includes events that occurred from the time the patient started seeking care. Examples are out-of-hospital arrest at home, in public, during transport to the hospital and both witnessed and unwitnessed events. Whether the arrest occurred prior to or after hospital arrival was recorded on the case report form. Other details including type of arrhythmia and pre-hospital treatment by bystanders, emergency medical technicians or medics were not recorded on the case report form. For purposes of this analysis, the concern is only with out-of-hospital cardiac arrest.

Statistical methods. We compared baseline and procedural characteristics for those with and without cardiac arrest; the chi-square test was used for categorical variables and the two-sample t-test for continuous variables. Our objective was to determine if the cardiac arrest variable augmented the existing COAP PCI mortality model, whose development has been described in detail.10 Over the past several years, the existing model has yielded high levels of discrimination with c-statistics in the range of 0.87–0.89. The published PCI mortality model was refined by adding out-of-hospital cardiac arrest to the list of predictor variables. These 12 variables were entered into a multivariate model and backwards stepwise logistic regression was used to select the final model. Variables with p > 0.05 were removed from the model. The model was developed in a set of 10,149 cases that had complete information received as of January 2007 and was tested in 5366 cases that had complete information received as of April 2007. The first data set essentially included the first three quarters of 2006, and the second included the last quarter of 2006. Model discrimination and calibration were assessed with the c-statistic and the Hosmer-Lemeshow test, respectively. Standard errors of the regression coefficients were corrected for clustering of patients within hospitals.

There were relatively few cases with missing data, as most predictors were absent in fewer than 10 cases; exceptions to this were elevated creatinine missing in 4% of cases and ejection fraction, which was unknown in 23% of cases. Hospital mortality in those with and without creatinine values was similar (1.4% versus 1.6%; p = 0.66), although mortality was higher in the missing ejection fraction group as opposed to ejection fraction > 30% (2.6% versus 0.7%; p < 0.0001). With the exception of age and out-of-hospital cardiac arrest, dichotomous variables were imputed to the category with the greater numbers; for example, missing creatinine was defined as creatinine < 2.0 mg/dL. Ejection fraction was categorized as: 1) < 30%; 2) missing; or 3) > 30%.

Results

Overall, 2% of individuals undergoing PCI had cardiac arrest prior to hospital arrival. Patients with cardiac arrest were younger and were more likely to be men than their counterparts without cardiac arrest (Table 1). More than 85% of the group with cardiac arrest had a previous myocardial infarction, including those that occurred within 24 hours of admission. In addition, the prevalence of chronic obstructive pulmonary disease and congestive heart failure was higher, but the group without cardiac arrest had more hypertension and previous cardiac revascularization.

As Table 2 shows, procedural characteristics were significantly different in the two groups. The vast majority of cardiac arrest patients had myocardial infarction within 24 hours of hospital arrival, and > 30% had cardiogenic shock as well. Consistent with an increased prevalence of congestive heart failure, a higher proportion of patients with cardiac arrest had ejection fractions < 30%. Although not all procedures performed in this group were defined as primary PCI, most had an urgent or emergency priority level. Not surprisingly, in-hospital mortality for out-of-hospital cardiac arrest was 19%, as opposed to only 1% in the group without out-of-hospital cardiac arrest (p < 0.0001).

In addition to these patient level differences, there was a distinctive trend concerning the distribution of cardiac arrest cases with respect to medical center, as seen in Figure 1. Among 8 medical centers with annual procedure volumes of < 120 cases, 4 had cardiac arrest volumes that exceeded 10% of caseload, whereas none of the sites with > 120 procedures per year had such cardiac arrest volumes (p < 0.0001).

Given these findings, we refined the COAP PCI mortality model by adding out-of-hospital cardiac arrest to determine its association with hospital mortality. As seen in Table 3, out-of-hospital cardiac arrest was a highly significant predictor of hospital mortality. The model had both good discrimination and calibration with c-statistic or area under the receiver operating characteristic curve = 0.90 (95% confidence interval [CI] = 0.87–0.92) and Hosmer-Lemeshow statistic p = 0.38. When evaluated in the test set, the c-statistic was 0.89 (95% CI = 0.85–0.93). Moreover, when backwards stepwise logistic regression was performed, the revised model contained only 6 predictors in comparison to 11 in the original model.

Discussion

While the incidence of out-of-hospital cardiac arrest in patients undergoing PCI in Washington State was 2%, centers with low PCI volumes had higher proportions with out-of-hospital cardiac arrest. It is possible that risk-adjusted performance for these hospitals could be falsely reported by not considering out-of-hospital cardiac arrest as a covariate in the multivariate logistic regression model. Because of this possibility, out-of-hospital cardiac arrest was added to the existing COAP PCI mortality model, which has been described and tested in an independent data set.10,15 Compared to the old model with a c-statistic of 0.87, the new PCI mortality model had fewer variables and similar discrimination, as indicated by the c-statistic of 0.89.

There are several risk-adjustment models for in-hospital PCI mortality including those from the American College of Cardiology National Cardiovascular Data Registry (ACC-NCDR)3,4 and the Mayo Clinic.7,8 Although these models have many predictor variables in common, the COAP model is the only one of these to include out-of-hospital cardiac arrest as a predictor. Both the ACC-NCDR and the Mayo Clinic models had high levels of discrimination (c-statistic = 0.89 for both models) similar to that observed in both the old and new COAP models. It is important to recognize that the change in the COAP model was made not to improve overall model performance, but to more accurately assess hospital performance for low-volume centers who had a higher proportion of patients with cardiac arrest.

Relatively little is known about out-of-hospital cardiac arrest in patients undergoing PCI.11–14 Most studies are either descriptive natural history reports or compare modes of reperfusion therapies for these patients. There is a consistent finding of excess mortality for patients undergoing PCI for ST-elevation myocardial infarction after out-of-hospital cardiac arrest followed by successful resuscitation. These studies have identified mortality rates ranging from 21–71%. In one study, the hospital mortality rate for patients with successful out-of-hospital resuscitation was 27.5% versus 4.9% for no cardiac arrest.12 In another sample of patients who were unconscious at the time of the PCI, there was a 49–71% mortality rate depending on the level of cerebral performance compared to no deaths in those conscious at the time of the procedure.14

This study is bolstered by the fact that COAP is in its ninth year of collecting high-quality data regarding all cardiac revascularization procedures performed in 31 Washington State hospitals.

Study limitations. One shortcoming of the study was the limited information about the circumstances and field treatment of out-of-hospital cardiac arrest. It is possible that the occurrence of cardiac arrest could have been overestimated. However, COAP has recently implemented an audit process in which selected charts from all hospitals are reviewed annually. So far, there has been no evidence that the incidence of cardiac arrest has been inflated. Another limitation was the relatively brief time between hospital admission and discharge. A 30-day mortality window would be optimal, but currently in COAP, it is not possible to easily follow patients once they leave the hospital. In addition, patients with cardiac arrest have a longer length of stay, thus expanding the window for the occurrence of adverse in-hospital events including death. Also, nearly 25% were missing ejection fraction data, which could result in inaccurate estimates of risk. Ejection fraction was removed from the model and regression coefficients for the remaining variables were recalculated. When this model was applied to the test set, the c-statistic was 0.87, indicating that model performance did not change when ejection fraction was removed from the model. This is not to suggest, however, that ejection fraction should be removed from the model, as it is a key predictor of mortality. Finally there could be unmeasured confounders for which we could not adjust.

Conclusion

As the practice of PCI is continually evolving, it is important that organizations charged with performance reporting have the capacity to add new variables to data collection forms, as well as to modify statistical models used to assess performance. The current study demonstrates the need for this adaptability, as it has shown that out-of-hospital cardiac arrest is a key factor in assessing risk-adjusted hospital mortality, particularly for low-volume centers with a relatively high number of cardiac arrest cases. Cardiac arrest as a predictor of mortality may become more important as the use of both field and hospital cooling could lead to increased numbers of cardiac arrest patients undergoing PCI.16