BackgroundConfounding bias is a common concern in epidemiological research. Its presence is often determined by comparing exposure effects between univariable- and multivariable regression models, using an arbitrary threshold of a 10% difference to indicate confounding bias. However, many clinical researchers are not aware that the use of this change-in-estimate criterion may lead to wrong conclusions when applied to logistic regression coefficients. This is due to a statistical phenomenon called noncollapsibility, which manifests itself in logistic regression models. This paper aims to clarify the role of noncollapsibility in logistic regression and to provide guidance in determining the presence of confounding bias.MethodsA Monte Carlo simulation study was designed to uncover patterns of confounding bias and noncollapsibility effects in logistic regression. An empirical data example was used to illustrate the inability of the change-in-estimate criterion to distinguish confounding bias from noncollapsibility effects.ResultsThe simulation study showed that, depending on the sign and magnitude of the confounding bias and the noncollapsibility effect, the difference between the effect estimates from univariable- and multivariable regression models may underestimate or overestimate the magnitude of the confounding bias. Because of the noncollapsibility effect, multivariable regression analysis and inverse probability weighting provided different but valid estimates of the confounder-adjusted exposure effect. In our data example, confounding bias was underestimated by the change in estimate due to the presence of a noncollapsibility effect.ConclusionIn logistic regression, the difference between the univariable- and multivariable effect estimate might not only reflect confounding bias but also a noncollapsibility effect. Ideally, the set of confounders is determined at the study design phase and based on subject matter knowledge. To quantify confounding bias, one could compare the unadjusted exposure effect estimate and the estimate from an inverse probability weighted model.
MULTIFILE
Intra-ocular straylight can cause decreased visual functioning, and it may cause diminished vision-related quality of life (VRQOL). This cross-sectional population-based study investigates the association between straylight and VRQOL in middle-aged and elderly individuals. Multivariable linear regression analyses were used to assess the association between straylight modeled continuously and cutoff at the recommended fitness-to-drive value, straylight ≥ 1.4 log(s), and VRQOL. The study showed that participants with normal straylight values, straylight ≤ 1.4 log(s), rated their VRQOL slightly better than those with high straylight values (straylight ≥ 1.4 log(s)). Furthermore, multivariable regression analysis revealed a borderline statistical significant association (p = .06) between intra-ocular straylight and self-reported VRQOL in middle-aged and elderly individuals. The association between straylight and self-reported VRQOL was not influenced by the status of the intra-ocular lens (natural vs. artificial intra-ocular lens after cataract extraction) or the number of (instrumental) activities of daily living that were reported as difficult for the elderly individuals.
ABSTRACT It is unknown whether heterogeneity in effects of self-management interventions in patients with chronic obstructive pulmonary disease (COPD) can be explained by differences in programme characteristics. This study aimed to identify which characteristics of COPD self-management interventions are most effective. Systematic search in electronic databases identified randomised trials on self-management interventions conducted between 1985 and 2013. Individual patient data were requested for meta-analysis by generalised mixed effects models. 14 randomised trials were included (67% of eligible), representing 3282 patients (75% of eligible). Univariable analyses showed favourable effects on some outcomes for more planned contacts and longer duration of interventions, interventions with peer contact, without log keeping, without problem solving, and without support allocation. After adjusting for other programme characteristics in multivariable analyses, only the effects of duration on all-cause hospitalisation remained. Each month increase in intervention duration reduced risk of all-cause hospitalisation (time to event hazard ratios 0.98, 95% CI 0.97–0.99; risk ratio (RR) after 6 months follow-up 0.96, 95% CI 0.92–0.99; RR after 12 months follow-up 0.98, 95% CI 0.96–1.00). Our results showed that longer duration of self-management interventions conferred a reduction in allcause hospitalisations in COPD patients. Other characteristics are not consistently associated with differential effects of self-management interventions across clinically relevant outcomes.
