Does Aggressive Care Following Acute Myocardial Infarction Reduce Mortality? Analysis with Instrumental Variables to Compare Effectiveness in Canadian and United States Patient Populations

Most regions of the United States tend to adopt an aggressive approach to care for acute myocardial infarction (AMI)—using invasive procedures such as cardiac catheterization in all patients and revascularization in most patients—while most Canadian regions favor a conservative approach—using invasive procedures more selectively (Pilote, Racine, and Hlatky 1994; Pilote, Granger et al. 1995; Pilote et al. 1998; Pilote, Bourassa et al. 1995). Whether or not the aggressive approach reduces mortality in comparison to more conservative approaches remains a topic of intensive investigation (Tu et al. 1997). There is therefore increasing interest in using data from administrative sources to evaluate the effectiveness of AMI treatment approaches in “real-world” patient populations. However, one important limitation of administrative database research is a strong potential for confounding bias, due to differences between comparison groups in terms of patient characteristics that have not been captured in the database (Ray 1997; Byar 1991). One approach that has been proposed to deal with this bias is the use of instrumental variables (Newhouse and McClellan 1998; Zohoori and Savitz 1997). In the instrumental variable estimation strategy, an instrumental variable is used in analyses to form groups of subjects that are unrelated to confounding variables, but that have different probabilities of receiving a particular treatment (Ho, Hamilton, and Roos 2000; Ettner, Hermann, and Tang 1999; Gowrisankaran and Town 1999). In this sense, instrumental variable estimation allows a pseudorandomization of study subjects. For this pseudorandomization to occur, the instrumental variable must be associated with the main independent variable but not be directly associated with the outcome variable of interest (Greenland 2000). The differences in care received across instrumental variable groups allows for unbiased estimation of the effects of a treatment for the “marginal” subpopulation of subjects whose type of care received was dependent on the instrumental variable (Harris and Remler 1998). McClellan et al. used instrumental variable estimation to investigate whether the aggressive approach reduced mortality in marginal, elderly U.S. Medicare beneficiaries who were admitted for AMI in 1987 (McClellan, McNeil, and Newhouse 1994). This study confirmed the presence of appreciable bias in standard outcome measures due to unobserved differences between groups of subjects receiving aggressive or conservative care, and the likelihood of less bias in outcome measures following the application of the instrumental variable approach. Standard analytical methods indicated that there were large benefits from aggressive care, while outcome measures obtained using instrumental variable methodology showed only minimal benefits. Two later studies showed small, but statistically significant, mortality benefits using this methodology (McClellan 1996; Brooks, McClellan, and Wong 2000). To more fully evaluate the usefulness of instrumental variable methodology, it is important that it be reapplied in different patient populations. In particular, it should be reapplied in populations from regions that adopt different approaches to post-AMI care, as the regions will be operating on different margins. In each region, there will be a group of patients that would clearly benefit from aggressive care, a group that would clearly not benefit, and a group for whom a benefit is possible but not clear. It is likely that the marginal patients are drawn from this latter group. As the approach to care becomes more aggressive overall within a region, the expected benefits for the marginal patients become smaller. Thus, we would expect the marginal mortality benefits resulting from aggressive care in regions with more conservative care (lower margin) to be greater than the marginal mortality benefits in regions with more aggressive care (higher margin). For instance, the marginal benefits of aggressive care should be greater in Canada than in the United States. Although there are striking regional differences in the rate of use of aggressive post-AMI care within the United States (Pilote, Racine, and Hlatky 1994; Pilote, Granger et al. 1995; Pilote et al. 1998; Pilote, Bourassa et al. 1995; Tu et al. 1997), the effect of these differences on the marginal benefit of such care has not been previously explored. The main objective of this study was to evaluate the marginal effects of an aggressive approach to post-AMI care on mortality in a Canadian patient population. We used an administrative database of all patients sustaining a first AMI in Quebec in 1988. By obtaining data from this time period, and by using the analytic approach used by McClellan et al. (McClellan, McNeil, and Newhouse 1994), we were able to compare our results to those previously obtained for the U.S. Medicare population. Unlike the U.S. Medicare database, the Quebec database includes AMI patients of all ages and not only patients aged 65 years and older. Therefore, a secondary objective of this study was to compare the effects of more aggressive care in patients younger than 65 years old to the effects in older patients.