Data on the treatment and clinical outcomes of all patients who sustained a first AMI in Quebec in 1988 (n
=8,995) were obtained retrospectively from two government administrative databases: the Quebec hospital discharge summary database (Med-Echo), and the Quebec Medicare database (la Régie de l'assurance maladie du Québec
[RAMQ]). The Med-Echo database was used to identify study patients with a main discharge diagnosis of AMI (ICD-9 code 410). The absence of a code for AMI was ascertained for at least three years preceding the diagnosis. The positive predictive value for coding an AMI for elderly survivors in this database has been evaluated to be 96 percent (95 percent CI: 94 percent to 98 percent) (Levy et al. 1999
). Patient demographic and hospital characteristics were identified from these data. Secondary diagnoses were used to obtain data on subjects' comorbid diseases. Postal codes (first three digits) for the patients' residence at the time of their discharge were also identified for 99.4 percent of the cohort. Canada Post's definition of a rural address (a zero in the second position of the postal code) was used to characterize each patient's residence as rural or urban.
The RAMQ database was used to obtain data on each cardiac catheterization, percutaneous transluminal coronary angioplasty (PTCA), and coronary artery bypass graft surgery (CABG) performed during the follow-up period. Complete four-year survival data were obtained for 99.7 percent of the AMI cohort by merging data from both the Med-Echo and the RAMQ databases. The methods used to ascertain accurate survival data have been published elsewhere (19). All follow-up data spanned the years from January 1, 1988, to December 31, 1992.
This study received ethical approval from the McGill University Institutional Review Board.
As a preliminary step in the creation of the instrumental variables, we classified each acute care hospital in Quebec in four ways: according to whether or not they had (1) availability of cardiac catheterization, (2) availability of PTCA, (3) availability of CABG, and (4) treated a high or low volume of first AMI patients during 1988. In 1988, there were 129 acute care hospitals admitting AMI patients in Quebec, and 13 (10 percent) offered cardiac catheterization. Of these 13 hospitals, 12 offered PTCA and 9 offered CABG. Thus, the hospital categories were not mutually exclusive.
To classify a hospital according to volume, we calculated the number of first AMI patients admitted in 1988 for each hospital. We classified any hospital treating a number of first AMI patients greater than or equal to the 75th percentile value for the distribution across all hospitals as a high-volume hospital.
The type of hospital of admission was classified based on the patient's initial hospitalization for AMI. Thus, if an AMI patient was admitted to a hospital without catheterization facilities and then later transferred to a hospital with catheterization facilities, the patient was considered admitted to a hospital without catheterization.
Similar to the approach used previously for the U.S. Medicare population (McClellan, McNeil, and Newhouse 1994
), the four instrumental variables used in our study corresponded to the subjects' “differential distances” to the four classifications of hospitals. One instrumental variable corresponded to the subjects' differential distance to a catheterization hospital. We created this variable by calculating the difference between the distance from a subject's residence to the nearest catheterization hospital, and the distance from this subject's residence to the nearest acute care hospital of any type. The three other instrumental variables corresponded to the difference between the distance from a subject's residence to (1) the nearest CABG hospital, (2) the nearest PTCA hospital, and (3) the nearest high-volume hospital, and the subject's distance to the nearest acute care hospital of any type. The choice of these instrumental variables was based on two main assumptions: (1) that AMI patients who lived relatively closer to catheterization, PTCA, CABG, or high-volume hospitals were more likely to receive aggressive care, and (2) that differential distances to each hospital type were not associated with any characteristics such as health status, which could be associated with the receipt of aggressive care and mortality.
To construct the instrumental variables, we collected latitude and longitude data from Statistics Canada. We used spherical geographic coordinates derived from these data to construct straight-line distances from the center of each patient's residential postal code region to the center of the postal code regions for each acute care hospital in Quebec. Previous work suggests that these straight-line distances are highly correlated with travel time (Phibbs and Luft 1995
To permit direct comparisons, the analytic approach was almost identical to that used for analyses applied to the U.S. Medicare population (McClellan, McNeil, and Newhouse 1994
). The main independent variable used in this study was a binary variable corresponding to whether or not subjects received cardiac catheterization within 90 days after their admission for AMI. Receipt of this procedure was used to indicate the receipt of aggressive care. Our data show that most AMI patients in Quebec who receive catheterization will receive this procedure within 90 days (median time in 1988=34 days). In addition, only small numbers of patients will sustain a recurrent AMI within this time period (7 percent in 1988).
There were seven outcome variables used in this study: binary variables corresponding to mortality at 1 day, 7 days, 30 days, and 1, 2, 3 and 4 years following the date of admission for AMI.
As a first step in the analytic approach, we compared demographic characteristics, comorbid diseases, invasive procedures received, and mortality between subjects who received cardiac catheterization within 90 days and subjects who did not.
Second, we used a standard statistical method—analysis of variance (ANOVA)—to estimate the association between catheterization within 90 days, and mortality. For each mortality variable, we created a model that adjusted for age, sex, rural or urban residence, and comorbid diseases.
Third, we placed subjects into two groups based on their differential distance to each type of hospital—“high” and “low” differential distance. We then compared the demographic and clinical characteristics of each group, as well as the invasive procedures received and mortality, across each differential distance group.
Finally, we used two-stage least squares regression analysis to estimate the average marginal effects of the aggressive approach to post-AMI care on mortality. For these analyses, we created four new sets of instrumental variables. Each set of instrumental variables corresponded to groups of subjects based on their differential distance to one of the four hospital types. For example, we created eight binary variables to form eight approximately equal-sized groups of subjects based on their differential distances to catheterization hospitals. Each variable was coded as 1 if the subjects' differential distance to a catheterization hospital fell within a specified range (in miles [1 mile=1.61 km] and rounded off: 0, 0.02–1.7, 1.7–2.8, 2.9–5.2, 5.4–18.7, 18.8–35.1, 35.3–67.4, 68.2–473.0), and 0 otherwise. We also created eight binary variables based on subjects' differential distances to CABG and PTCA hospitals. Because the differential distance groups for PTCA hospitals had ranges identical to those for catheterization hospitals (in miles and rounded off: 0, 0.02–1.7, 1.7–2.8, 2.9–5.2, 5.4–18.7, 18.8–35.1, 35.3–67.4, 68.2–473.0), we did not include the groups for PTCA hospitals in any subsequent analyses. We created three binary variables based on subjects' differential distance to a high-volume hospital (in miles and rounded off: 0, 0.05–5.5, 5.6–531.6).
Before running our two-stage least squares regression models, we examined F-statistics for the association between the instrumental variables and receipt of catheterization (first-stage regression equations). All F-statistics for patients of all ages were 13.6 or greater (range: 13.6–34.7). The models including receipt of catheterization within one day as an outcome measure, and the models examining only study subjects either <65 years old or ≥65 years old, corresponded with lower F-statistics. However, all models corresponded with a p-value <.05, except for some of the models including receipt of catheterization within one day as an outcome measure (for study subjects ≥65 years old). Mortality at one day after AMI was therefore not used as an outcome measure in the two-stage least squares regression analyses for subjects ≥65 years old. These results provided evidence to support the hypothesis that differential distance to different types of hospitals is associated with aggressive care. The fact that the proportions of patients who received cardiac catheterization within 90 days decreased across greater differential distance groups provided additional evidence (data not shown).
The two-stage least squares regression models estimated the average effects of aggressive care on mortality for marginal subjects within the same age group, and with the same sex and comorbid diseases. We included different combinations of instrumental variables in the different regression models in order to account for differential access to aggressive care at catheterization, CABG, and high-volume hospitals both singularly and simultaneously. The main independent variables included in models estimating effects on mortality at 1 day, 7 days, and 30 days were receipt of catheterization within 1 day, 7 days, and 30 days, respectively. To evaluate the marginal effects of aspects of aggressive care other than invasive treatments, such as emergency response systems (McClellan, McNeil, and Newhouse 1994
), some models also included rural residence or admission to a high-volume hospital as independent variables.
We completed each set of analyses for all study subjects, for subjects <65 years and ≥65 years of age at the time of admission for AMI. We performed all analyses using STATA 4.0 (Stata Press, College Station, Texas).