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
When planning a study measuring the effects of a neurodevelopmental treatment (NDT), we were confronted with the methodological problem that while measuring the effects of NDT, a rival hypothesis is that the decision to implement the NDT might be related to the quality of nursing care. Therefore, we measured the quality of nursing care as a possible confounding variable in relation to this outcome study. The quality of nursing care was measured on 12 wards participating in the experimental and control groups of the outcome study. Data were collected from 125 patients and 71 nurses and patients' records. The findings showed no significant differences in the quality of nursing care between the 2 groups of wards (P = .49). This method may be useful to other researchers conducting outcome research and who are confronted with a similar methodological problem.
BackgroundE-mental health holds promise for people with severe mental illness, but has a limited evidence base. This study explored the effect of e-health added to face-to-face delivery of the Illness Management and Recovery Programme (e-IMR).MethodIn this multi-centre exploratory cluster randomized controlled trial, seven clusters (n = 60; 41 in intervention group and 19 in control group) were randomly assigned to e-IMR + IMR or IMR only. Outcomes of illness management, self-management, recovery, symptoms, quality of life, and general health were measured at baseline (T0), halfway (T1), and at twelve months (T2). The data were analysed using mixed model for repeated measurements in four models: in 1) we included fixed main effects for time trend and group, in 2) we controlled for confounding effects, in 3) we controlled for interaction effects, and in 4) we performed sub-group analyses within the intervention group.ResultsNotwithstanding low activity on e-IMR, significant effects were present in model 1 analyses for self-management (p = .01) and recovery (p = .02) at T1, and for general health perception (p = .02) at T2, all in favour of the intervention group. In model 2, the confounding covariate gender explained the effects at T1 and T2, except for self-management. In model 3, the interacting covariate non-completer explained the effects for self-management (p = .03) at T1. In model 4, the sub-group analyses of e-IMR-users versus non-users showed no differences in effect.ConclusionBecause of confounding and interaction modifications, effectiveness of e-IMR cannot be concluded. Low use of e-health precludes definite conclusions on its potential efficacy. Low use of e-IMR calls for a thorough process evaluation of the intervention.
MULTIFILE
