[Analysis with the propensity score of the association between likelihood of treatment and event of interest in observational studies. An example with myocardial reperfusion].

INTRODUCTION AND OBJECTIVES Analysis of the effect of treatment in observational studies is complex due to differences between treated and nontreated patients. Calculating the probability of receiving treatment conditioned on relevant covariates (propensity score [PS]) has been proposed as a method to control for these differences. We report an application of PS to assess the association between reperfusion treatment and 28-day case fatality in patients with acute myocardial infarction (AMI). METHOD We describe the procedure used to calculate PS for receiving reperfusion treatment, and different strategies to analyze the association between PS and case fatality with regression modeling and matching. Data were from a population-based registry of 6307 patients with AMI in Spain during 1997-98. RESULTS The PS for reperfusion was calculated in 5622 patients. In the multivariate analysis, reperfusion was associated with lower case fatality (OR = 0.59; 95% confidence interval [95% CI]: 0.46-0.77). When PS was included as a covariate, this association became non- significant (OR = 0.76; 95% CI: 0.57-1.01). In the subgroup of matched patients with a similar PS (n = 3138), treatment was not associated with case fatality (OR = 0.95; 95% CI: 0.72-1.26). When the influence of cases with missing data on PS was controlled for, reperfusion treatment was associated with lower fatality (OR = 0.66; 95% CI: 0.55-0.80). CONCLUSIONS Calculating propensity score is a method that controls for differences between treated and nontreated patients. This score has limitations when matching is incomplete and when data are missing. Results of the present example suggest that reperfusion treatment reduces AMI case fatality.