Examining the relationship between alcohol and mortality through the use of propensity scores

Abstract

Introduction: The relationship between alcohol and health has long been of interest to researchers. Over the past few decades, the risk associated with alcohol consumption has often been characterized as a J-shaped curve. Studies within this area have largely concluded that moderate alcohol consumption is associated with lower mortality. Despite the evidence in support of the J-curve hypothesis, the debates continue, suggesting that analytical approaches beyond standard survival analysis can be explored. Objective: To re-examine the relationship between moderate alcohol consumption and all-cause mortality through the utilization of propensity score matching, and to examine the effect the passage of time and the algorithm used have on the outcome. Results: When a matching algorithm which allowed for replacement was utilized, the difference in the risk of mortality for lifelong abstainers and moderate drinkers was no longer significantly different following a series of adjustments. By contrast, when replacement was restricted, the mortality risk for lifelong abstainers remained significantly greater (p>0.05), though it was reduced by a large margin. Sensitivity analyses revealed that if unobserved variables caused the odds ratios to differ by a factor of 1.25 (2009) or 1.15 (2003) the results would no longer be significant, suggesting that unobserved or excluded variables could reshape the results. Conclusions: The results of this analysis generally support the findings of much of the published literature in this area. Six additional years of mortality data served to offer more support for the J-curve hypothesis; however, this additional support is only apparent when replacement is restricted. This analysis underscores the importance the matching algorithm used can play in shaping the results.