Authors' reply

We agree that in observational studies residual confounding bias may subsist. However, to take into account this potential drawback with anxiety being a consequence of prior vascular changes, we carefully adjusted the models for a large number of confounding factors including vascular risk factors and cardiovascular diseases, and the association in women remained significant. The second argument of failing health and proximity of death does not hold as at baseline (time of anxiety evaluation) our sample consisted of high-functioning community-dwelling elderly persons, physically and psychologically able to travel to the medical centre. A careful examination of the Kaplan-Meier curves (Fig. 1) also indicates that very few deaths occurred during the first 2 years of follow-up. Last, Euba raises the question of statistical power to explain the absence of a significant association in men. In survival analysis, the statistical power depends on the number of events (i.e. deaths) and in our sample the frequency of deaths is higher in men (162/702 v. 136/1006 in women). A power calculation for anxiety disorder shows that we could have detected an unadjusted relative risk of 1.65 in women and 1.71 in men, with an alpha risk of 0.05 and a power of 0.80. The analysis in men is thus not underpowered and if an association with mortality exists in men, it is less strong than in women.

With ageing, people face multiple adverse events including physical multimorbidity and loss of capacities. Personal resources, such as self-efficacy, sense of mastery or control beliefs, and psychological resilience are important in the process of coping with a chronic disease. On the other hand, anxiety disorder, irrespective of the aetiology, could clearly contribute to a worse outcome. This underlines the importance of developing interventions for older persons aimed at maintaining or improving psychological coping resources when health declines. Up to now, very few well-designed studies have been performed with such a large population sample, capable of controlling for main confounders and using a validated anxiety diagnosis including anxiety subtypes. Although future research is needed to confirm our results and the gender-specific association, our study also stresses the importance of including anxiety diagnostic tools in population-based cohorts to improve the understanding of the consequences of anxiety in late life.