Long-term MI Outcomes at Hospitals With or Without On-site Revascularization FREE

Mark Twain reputedly said: "To a man with a hammer, every nail looks like it needs driving." This aphorism is reflected in the oft-replicated finding that, when patients with acute myocardial infarction (AMI) are admitted to hospitals with on-site revascularization facilities, they undergo percutaneous coronary intervention (PCI) and coronary artery bypass graft (CABG) surgery much more often than those admitted to hospitals without such facilities.¹ Many studies have now compared patient outcomes in these 2 practice settings, and most have demonstrated similar survival but better quality of life or lower rates of recurrent cardiac admissions for patients experiencing higher rates of revascularization.^2- 13

These comparisons and conclusions rest on the assumption that a single ecological variable (being admitted to a hospital with or without PCI and CABG surgical capacity) and the associated variation in one aspect of process of care (higher or lower rates of revascularization) account for the outcome differences.^14- 15 By design, other differences in characteristics of the patients, the admitting hospital, or the attending physicians, along with myriad other potential differences in follow-up care, are given much less attention as determinants of outcome. Our study examines the validity of such assumptions using a cohort of AMI patients from Ontario. We hypothesized that ecological variables may be highly correlated with other factors and that procedural differences alone are unlikely to explain the outcome differences between hospitals (and, by extension, regions) with high vs low rates of cardiac procedures after AMI.^16- 18

To date, no study has compared the outcomes of hospitals with high and low rates of invasive cardiac procedures within Canada, where the intervention rates are markedly lower than in the United States. Therefore, we first set out to confirm whether patients with AMI at hospitals with on-site revascularization facilities have outcome advantages vs hospitals without on-site revascularization facilities in Ontario. If such findings could be confirmed, we would then test our primary hypothesis that the reasons the differences might be attributed wrongly to a single distinguishing characteristic of the admitting hospital—a high rate of performance of revascularization procedures—when many other patient, physician, and process-related differences might play as much a role.

The Ontario Myocardial Infarction Database (OMID) links a variety of population-based administrative data sources. Hospital discharge abstracts compiled by the Canadian Institutes of Health Information yielded data pertaining to the index admission: patient demographics, illness severity, comorbidity, in-hospital procedure use, and mortality. Hospital readmissions were identified using longitudinal linkages. Physician claims data for the entire province was obtained from the Ontario Health Insurance Plan and were used to determine emergency and postdischarge physician visits. Both physician claims and hospital discharge abstracts were used to determine cardiac procedure use. Patients aged 65 years or older were linked to the Ontario Drug Benefit Program to identify discharge rates of evidence-based secondary-prevention therapies (eg, aspirin and β-blockers). The Ontario Registered Persons Database allowed us to determine out-of-hospital deaths. To maintain patient confidentiality, all linkages were conducted using scrambled health card numbers.

We constructed a cohort of all patients admitted with a most responsible diagnosis of AMI (International Classification of Diseases, 9th Revision, Clinical Modification [ICD-9-CM] code 410) in Ontario between April 1, 1992, and December 31, 1993. The accuracy of AMI coding in the OMID database has been previously validated through large province-wide chart audits.^19- 20 We excluded non-Ontario residents, those with invalid Ontario Health Card numbers, those younger than 20 or older than 105 years, those discharged alive whose total length of hospital stay was fewer than 4 days, those for whom AMI was coded as a hospital complication, those transferred from another acute care facility, and those with no link postal code information for income data. To reduce the chances that subgroups within the cohort varied in severity of cardiovascular disease, we also excluded any patient who had been hospitalized with AMI in the year preceding the index admission. Details about the OMID database and the rationale for these criteria have been published elsewhere.^19- 21

Patients were categorized by on-site procedural characteristics at their admitting hospitals (eg, on-site invasive cardiac procedure facilities vs no on-site invasive cardiac procedure facilities, herein also respectively termed invasive- vs noninvasive-procedure hospitals) regardless of potential downstream hospital transfers. During the study period, there were 197 acute care hospitals in Ontario. Three rural isolated hospitals were excluded from the analysis because census data information needed to obtain socioeconomic status in isolated rural communities is suppressed. Four institutions had on-site angiography-only facilities comprising 3.5% of the sample population and were also excluded from the analysis because of small sample size. Accordingly, comparisons were made between 2 hospital groups: the 9 hospitals with on-site angiography and revascularization capabilities and the 181 hospitals with no invasive-procedure facilities.

Since neighborhood income level has been shown to be an important predictor of outcomes following AMI in Ontario,²¹ we used the 1991 official Canadian census data to calculate the average household income for each patient's postal code. To control for variations in patient severity of illness on admission, we used the Ontario AMI mortality prediction rule for 30-day and 1-year mortality rates.²⁰ The variables in this model include age, sex, cardiac severity (eg, congestive heart failure, cardiogenic shock, arrhythmias), and comorbid status (eg, diabetes mellitus, stroke, acute and chronic renal disease, and malignancy), as derived from the ICD-9 codes present in the 15 secondary diagnostic fields of the hospitalization database. This prediction rule was derived in a different subset of the OMID database (ie, all AMI patients admitted between April 1, 1994, and March 31, 1997) with areas under the receiver operating characteristic curve (AUROC) of 0.775 for 30-day mortality and 0.793 for 1-year mortality rates. The predictive accuracy of the model was confirmed in the cohort of patients with AMI used in this study (ie, AUROC were 0.76 for 30-day mortality and 0.78 for 1-year mortality) and independently validated on 4836 patients with AMI from Manitoba and 112 234 patients with AMI from California AUROC of 0.77 in both samples).^20,22

Coronary angiography and revascularization procedure rates were determined using procedure codes in the hospitalization and physician claims databases. The exact date of referrals for procedures could not be determined from the administrative data. However, as reasonable surrogates for waiting times, we calculated the number of days from the index AMI admission to the date of coronary angiography and the number of days from angiography to revascularization. Generally, these surrogate measures will slightly overestimate true procedural waiting time. We divided these process factors into 2 periods: those occurring during the index admission and those occurring within the first 6 months of follow-up.

Administrative data lack detailed information about processes of care during the index admission. Nonetheless, post-MI outcomes may be influenced by other physician and hospital characteristics, which include attending physician specialty, hospital volumes (based on the annual number of patients admitted to the MI facility²⁰), hospital teaching status, and geographical proximity to the nearest tertiary facility (from latitudinal and longitudinal coordinates or "as the crow flies"²³). Such ecological indicators may serve as proxy measures of intersite differences in processes of care. Attending physician specialty and physician specialty visits were identified using hospital discharge abstracts and Ontario Health Insurance Plan claims data, respectively.

Although the majority of invasive-procedure centers are also large-volume institutions and teaching hospitals, the majority of large-volume and teaching hospitals are not invasive-procedure centers. Moreover, formal diagnostic testing for collinearity across hospital and physician-level variables did not reveal any variance inflation factor greater than 5.0. Therefore collinearity was not a significant issue for this analysis (ie, the maximum variance inflation factor across explanatory variables for the 190 hospitals was 2.6).²⁴

All patients were tracked for a minimum of 5 years. Long-term outcomes included mortality, cardiac readmissions, and emergency department visits. Cardiac readmissions were identified when patients were readmitted to hospital with a most responsible diagnosis of recurrent MI (ICD-9 code 410), congestive heart failure (ICD-9 code 428), or angina (ICD-9 code 411 or 413). In total, 96.9% of patients readmitted to hospital also had concomitant emergency assessments, suggesting that the vast majority of readmissions reflected urgent clinical need rather than an elective indication (eg, elective angiography). Emergency department visits were identified using claims data.

Clinical characteristics of patients were compared between the 2 groups. Differences in procedural rates, outpatient visits, mortality, and cardiac readmission were examined using χ² analyses. Intergroup hospital differences in procedural waiting times, outpatient visits, and hospital bed-days were tested using the t test.

Crude mortality rates were examined at 5 years to ensure equivalent follow-up duration for all patients. Long-term survival and time-to-first-cardiac readmission were analyzed by hospital group and compared using Kaplan-Meier plots and the log-rank test. Mortality comparisons censored patients at 5 years although time-to-first-cardiac readmission censored patients at 5 years or death. Nonfatal outcomes were examined both individually and combined (ie, the first cardiac readmission or emergency department visit).

To determine which factors best accounted for the overall invasive-procedure effect on fatal and nonfatal outcomes, we constructed multivariable models that sequentially added patient-level, physician-level, and hospital-level characteristics. Such analysis models the hierarchical nature of the data, which consist of patients nested in physicians, who in turn are nested in hospitals, with characteristics measured at different levels of this hierarchy.²⁵ A traditional regression model will tend to underestimate SEs for characteristics measured at higher levels of the hierarchy, thereby producing confidence intervals (CIs) that are artificially narrow. Multilevel analysis allows one to incorporate correctly variables measured at different levels of the hierarchy and produces more accurate estimates of significance. Multilevel analyses adjusted for age, sex, socioeconomic status (average household income), illness severity (ie, individual risk factors comprising the estimated 30-day mortality at the time of index admission), index revascularization procedures, as well as the 4 proxy measures of processes listed above. The order with which variables were entered into the models did not significantly alter the results. Multilevel analyses were implemented using HLM version 5.²⁶

Statistical significance was defined as P<.05 for all analyses. We used SAS statistical software version 6.12 (SAS Institute Inc, Cary, NC) for the remaining statistical analyses.

The cohort included 25 697 patients. Table 1 illustrates the distribution of baseline demographic, socioeconomic, clinical, and in-hospital process characteristics between patients at the 2 types of hospitals. Although patients at invasive-procedure hospitals had significantly higher rates of coronary angiography and myocardial revascularization procedures during the index hospitalization, there were many other factors that also differed between the 2 types of institutions. For example, although patients who were admitted to invasive-procedure centers were more affluent, they were also at higher risk of early mortality, with a predicted 30-day mortality rate of 16.2% vs 14.7% for noninvasive-procedure centers (P<.001). Patients admitted to invasive-procedure hospitals were also more likely to be admitted to a teaching facility and a large-volume hospital and assigned to an attending cardiologist.

Table 2 illustrates that follow-up and treatment factors also differed between the 2 types of institutions. For example, patients at invasive-procedure hospitals were more likely to be followed up by a cardiologist than a general practitioner while elderly patients admitted to noninvasive-procedure hospitals were less likely to receive a prescription for aspirin, β-blockers, angiotensin-converting enzyme inhibitors, or statins within 90 days following discharge.

Unadjusted 5-year survival was similar between groups. However, patients initially admitted to invasive-procedure hospitals had lower nonfatal adverse outcomes at 5 years following AMI compared with patients admitted to noninvasive-procedure hospitals (Figure 1). The nonfatal outcome advantages at invasive-procedure hospitals were driven by lower angina readmissions (log-rank χ², 17.86; P<.001) and emergency department visits (log-rank χ², 311.09; P<.001), with a weak trend to lower reinfarction rates (log-rank χ², 1.14; P = .29). Among patients experiencing nonfatal outcomes, the time to first event was significantly longer for those patients initially admitted to invasive-procedure hospitals (470 vs 375 days, P<.001). Patients admitted to invasive-procedure centers had fewer total cardiac readmissions (11 fewer cardiac readmissions per 100 AMI patients, P<.001), fewer angina readmissions (12 fewer angina readmissions per 100 AMI patients, P<.001), and fewer emergency department visits (57 fewer emergency visits per 100 AMI patients, P<.001). There were no significant differences in the number AMI readmissions (P = .43), congestive heart failure readmissions (P = .93), or cumulative length of hospital stay after discharge (P = .92).

Figure 2 examines residual invasive-procedure effects on 5-year outcomes when sequentially adjusted for patient-level, physician-level, and hospital-level characteristics. All nonfatal outcomes were aggregated, given that the determinants for each nonfatal outcome were similar to one another. Figure 2A examines the invasive-procedure hospital effects on mortality, and Figure 2B examines the invasive-procedure hospital effects on nonfatal outcomes.

Figure 2A illustrates that adjustments for age, sex, and socioeconomic status do not significantly alter the odds ratio (OR) of death for patients admitted to invasive-procedure vs noninvasive-procedure hospitals (adjusted OR, 1.07; P = .15). After adding clinical factors into the model (ie, those clinical variables comprising the predicted 30-day risk of mortality), admission to invasive-procedure hospitals (adjusted OR, 0.86; P = .003) becomes protective against mortality. Some of the protective effects of invasive-procedure hospitals are explained by index revascularization procedures themselves, as illustrated by the attenuated effect when index revascularization procedures are added to the model (adjusted OR, 0.92 for invasive-procedure hospitals; P = .07). However, there are no incremental protective effects of invasive-procedure hospitals on mortality after adjusting for the remaining physician-level and hospital-level characteristics (adjusted OR, 1.01 for invasive-procedure hospitals; P = .82).

Figure 2B illustrates that the unadjusted invasive-procedure effects on nonfatal outcomes (unadjusted OR, 0.65; 95% CI, 0.52-0.82; P<.001) is not significantly altered when sequentially adjusting for patient-level (adjusted OR, 0.68; 95% CI, 0.53-0.89; P<.001), and physician-level (adjusted OR, 0.72; 95% CI, 0.56-0.92; P<.001) characteristics. Index revascularization procedure use does not explain any of the morbidity advantages at invasive-procedure hospitals. However, the protective effects of invasive-procedure hospitals on nonfatal outcomes are partially negated when adding geographical proximity (adjusted OR for invasive-procedure hospitals, 0.84; 95% CI, 0.63-1.12; P = .24), and entirely negated when adding teaching status into the multilevel model (adjusted OR for invasive-procedure hospitals, 0.98; 95% CI, 0.73-1.30; P = .87). In contrast, the protective effects of teaching status on nonfatal outcomes were not negated when adding either invasive-procedure hospitals or geographical proximity into the models (Figure 3). Moreover, nonfatal outcome benefits of teaching status persisted when excluding patients referred for index revascularization procedures (adjusted OR for teaching status, 0.71; 95% CI, 0.56-0.90; P = .005), and when excluding those patients transferred from noninvasive- to invasive-procedure hospital facilities (adjusted OR for teaching status, 0.68; 95% CI, 0.53-0.88; P = .004). Most of the nonfatal outcome advantage initially observed for invasive-procedure hospitals was explained by the hospital's teaching status and not because of higher revascularization rates themselves.

To explore the relationship between teaching status, invasive-procedure centers, and nonfatal outcomes further, we first compared baseline and follow-up differences between hospital categories. Differences in follow-up specialty care most distinguished teaching from invasive-procedure hospitals (Table 3). We then performed subgroup analyses subdividing hospitals into 4 groups: teaching and invasive-procedure, teaching and noninvasive-procedure, nonteaching and invasive-procedure, nonteaching and noninvasive-procedure. The postdischarge rates of aspirin (67% vs 61.5%, P<.001), β-blockers (48.4% vs 39.6%, P<.001), cardiology follow-up rates at 90 days (33.4% vs 19.8%, P<.001), and the total number of follow-up cardiology visits (2.8 vs 1.9, P<.001) were all higher in teaching and noninvasive-procedure hospitals than in nonteaching and noninvasive-procedure hospitals. Both teaching and invasive-procedure (P = .01) and teaching and noninvasive-procedure (P = .04) hospitals were associated with lower nonfatal event rates, whereas the 1 nonteaching and invasive-procedure hospital had higher nonfatal event rates (P = .003) after AMI. After eliminating the 1 nonteaching and invasive-procedure hospital from the analysis, the results did not change. That is, teaching status still accounted for the invasive-procedure effect in nonfatal outcomes.

To further explore the relationship among teaching hospitals, postdischarge cardiac care, and nonfatal outcomes after AMI, a supplementary time-dependent covariate analysis was performed since outpatient physician visits and therapies occur at different time intervals after discharge. This analysis was confined to the subgroup of patients aged 65 years and older given that outpatient medication data were only available for these patients. Due to the limitations of the HLM software package,²⁶ a hierarchical analysis could not be used for this analysis. Accordingly, conventional statistical techniques using Cox regression models were incorporated with postdischarge medication use, physician visits, and downstream revascularization procedures modeled as time-varying covariates. Independent predictors of lower risk of nonfatal outcomes included postdischarge use of aspirin (adjusted hazard ratio [HR], 0.94; 95% CI, 0.90-0.98; P = .002), β-blockers (adjusted HR, 0.89; 95% CI, 0.85-0.93; P<.001), and statins (adjusted HR, 0.84; 95% CI, 0.78-0.92; P<.001). All variables were associated with a lower risk for nonfatal outcomes. In contrast, postdischarge use of calcium channel blockers (adjusted HR, 1.19; 95% CI, 1.14-1.24; P<.001), and follow-up general practitioner visits (adjusted HR, 1.24; 95% CI, 1.17-1.31; P<.001) were associated with a higher risk of nonfatal outcomes among elderly patients. Undergoing a revascularization procedure after discharge was not a significant predictor of nonfatal outcomes after AMI (adjusted HR, 0.98; 95% CI, 0.85-1.14; P = .83). In total, downstream processes of care had a marginal impact on explaining the protective effects of teaching hospitals among elderly patients (adjusted HRs for teaching hospitals, 0.75; 95% CI, 0.69-0.81 before and 0.78; 95% CI, 0.72-0.85 after adjustments for downstream processes of care, respectively; P<.001 in both models).

Our study demonstrated that Ontario patients admitted to hospitals with on-site revascularization facilities had lower recurrent cardiac hospitalization rates, but similar rates of survival compared with patients admitted to hospitals without on-site revascularization facilities. The determinants of fatal outcomes were different from those for nonfatal outcomes. In contrast to fatal events, the effect of invasive-procedure hospitals on nonfatal outcomes was explained by other ecological variables—hospital teaching status and to a lesser extent geographical proximity (to tertiary centers). Despite the variation in cardiac intervention rates that existed between invasive-procedure and noninvasive-procedure hospitals, differences in the utilization rates of invasive cardiac procedures themselves did not account for outcomes at the hospital level.

Our results are consistent with the vast majority of studies demonstrating no significant relationship between interhospital (or regional) variations in procedure rates and survival following AMI.^{2- 9,12- 13} Our findings are also consistent with studies that have demonstrated lower rates of recurrent cardiac admissions for groups incorporating higher rates of cardiac procedures after MI.^4- 6,10,13

Many investigators have used such ecological designs to draw conclusions about the relative merits of higher vs lower cardiac intervention rates after MI. For example, a recent substudy of the Global Use of Strategies to Open Occluded Arteries in Acute Coronary Syndromes (GUSTO-II) clinical trial examining non–Q-wave MI and unstable angina demonstrated that at 6 months, US patients had an absolute rate reduction of 3% for recurrent MI compared with their Canadian counterparts.¹⁰ The authors concluded that the different outcomes may have been attributed to differences in procedure rates between the 2 countries.

Our study serves to illustrate the complex interrelationships between procedure use, secondary process characteristics, case-mix factors, and outcomes. Many other factors apart from differences in the use of procedures on the index admission distinguish hospitals with from those without on-site revascularization facilities. For example, AMI patients admitted to invasive-procedure hospitals in this study had a more adverse risk profile, were wealthier, had longer initial lengths of hospital stay, and were more likely admitted to an attending cardiologist than were patients admitted to noninvasive-procedure hospitals. Moreover, different process measures may have also existed during follow-up. Although outpatient prescription rates of evidence-based therapies were higher for those patients initially admitted to invasive-procedure hospitals, specialty follow-up care varied less markedly than during the index admission. Postdischarge myocardial revascularization rates were similar between the 2 groups. Accordingly, 2 groups of hospitals (or regions) cannot be distinguished solely based on a single process characteristic (eg, rates of revascularization procedures).

We demonstrated that the benefits of invasive-procedure hospitals on nonfatal outcomes were largely explained by teaching status. In our statistical models, teaching status likely served as a proxy for an uncertain number of unmeasured process characteristics. One distinguishing process difference between teaching hospitals and invasive-procedure institutions was in their respective downstream processes of care after discharge from the index AMI admission. Patients admitted to teaching hospitals were more likely to be followed up by cardiac specialists after discharge from hospital, independent of whether hospitals had on-site revascularization facilities. Indeed, our subgroup analysis determined that the use of evidence-based therapies predicted lower rates of recurrent cardiac admissions and emergency department visits and accounted, albeit marginally, for some of the protective effects of teaching status on nonfatal outcomes. It is plausible that follow-up care is an important determinant of nonfatal outcomes after AMI. Although a relationship between outpatient care and morbidity outcomes after AMI is speculative, such a relationship has been demonstrated to exist in other disease-specific cohorts (eg, congestive heart failure).^27- 28 Moreover, at least one other study has noted an interaction among coronary intervention rates, on-site procedural capacity, and teaching status after AMI.²⁹ Most studies comparing hospitals with on-site revascularization facilities with those without have not adjusted for differences in teaching status or downstream processes of care.^2- 4,14

One noteworthy study limitation involves the lack of clinical detail specifically related to the initial infarct characteristics, postinfarct left ventricle function, and in-hospital pharmacological therapies. Although we adjusted for a number of clinical, institutional, and physician-related characteristics, we acknowledge that additional clinical and process measures, both during the index admission and following discharge, may have altered our results or may have helped explain why teaching status accounted for the morbidity advantages of invasive-procedure hospitals. The time-dependent covariate analysis incorporated conventional rather than hierarchical techniques. Accordingly, we caution against overinterpreting the results of our subgroup analysis. Nonetheless, the inability of procedures themselves to have accounted for hospital-related outcome differences was a finding independent of the type and number of variables entered into our statistical models. Furthermore, since hospital-associated outcomes varied according to a limited set of covariates, it seems unlikely that additional covariates would have changed our overall conclusions. Finally, our data were comprehensive, consecutive, and comprised patients with AMI that are highly representative of the Canadian population.

In conclusion, our findings demonstrate that patients with AMI admitted to hospitals with on-site procedural capacity in Ontario had similar rates of mortality but lower rates of recurrent cardiac hospitalization and emergency care compared with hospitals without such capacity. Despite vastly different cardiac intervention rates, similarities and differences in outcomes could not be explained by variations in procedure rates themselves. We believe that these results may be applicable to many other interjurisdictional or interinstitutional comparisons. We suggest that misleading conclusions drawn from such studies be taken as examples of the Mark Twain fallacy, which occurs when outcomes are compared across practice sites or regions, with a narrow focus on a single ecological variable that apparently distinguishes the 2 settings without fully accounting for myriad patient and process-of-care variables that may actually explain any observed outcome differences.