Search

A broad review is given of the general principles underlying study design with emphasis on applications in medical and epidemiological contexts. The main theme of the paper is that, while the distinction between interventionist studies, that is experiments, and purely observational ones is important, there are many common threads. A wide range of specific applications are used in outline to illustrate the discussion. [ABSTRACT FROM AUTHOR]

A review of epidemiological papers conducted in 2009 concluded that several studies employed variable selection methods susceptible to introduce bias and yield inadequate inferences. Many new confounder selection methods have been developed since then. The goal of the study was to provide an updated descriptive portrait of which variable selection methods are used by epidemiologists for analyzing observational data. Studies published in four major epidemiological journals in 2015 were reviewed. Only articles concerned with a predictive or explicative objective and reporting on the analysis of individual data were included. Method(s) employed for selecting variables were extracted from retained articles. A total of 975 articles were retrieved and 299 met eligibility criteria, 292 of which pursued an explicative objective. Among those, 146 studies (50%) reported using prior knowledge or causal graphs for selecting variables, 34 (12%) used change in effect estimate methods, 26 (9%) used stepwise approaches, 16 (5%) employed univariate analyses, 5 (2%) used various other methods and 107 (37%) did not provide sufficient details to allow classification (more than one method could be employed in a single article). Despite being less frequent than in the previous review, stepwise and univariable analyses, which are susceptible to introduce bias and produce inadequate inferences, were still prevalent. Moreover, 37% studies did not provide sufficient details to assess how variables were selected. We thus believe there is still room for improvement in variable selection methods used by epidemiologists and in their reporting. [ABSTRACT FROM AUTHOR]

Matching by a confounder in a case–control study nearly always produces a control-selection bias that mixes with the confounding to produce a net bias. Previous theoretical work has assumed that control for a single confounder, the matching factor, is sufficient to remove all the confounding and that the confounder-exposure, confounder-outcome and exposure-outcome associations are monotonic. Under these conditions: (a) The net bias is toward the null if the exposure affects the outcome and nil if it does not. (b) If the confounding is away from the null, the selection bias is toward the null. (c) If the confounding is toward the null, the selection bias can be in any direction or even nil. If more than one confounder needs to be controlled to remove all the confounding, the net bias from matching by one of them can be away from the null, whether the exposure affects the outcome or not. An influential heuristic, that matching controls to cases by a variable associated with exposure always brings the marginal exposure distributions of the case and control groups closer together, turns out to be faulty. The implications of matching by confounders in case–control studies are less straightforward than previously thought. Suggestions are offered for advancing the methodologic literature on this topic. [ABSTRACT FROM AUTHOR]

We thank Dr. Karp for his interest [1] in our paper [2]. We agree on some points, but our theoretical description differs from his in ways leading to important divergences for teaching and practice. We also see a danger of overextending abstract theory (with its inevitable and extensive simplifications) into practice [3], especially when the practical questions are causal but the theory applied lacks an explicit, sound longitudinal causal model to address these questions. As we will explain, a defect in the “study base” theory Dr. Karp adopts as a foundational belief system is that it takes as a foundation a parameter affected by baseline risk factors—including exposure when that has effects on follow-up or disease. It consequently leads to biases and misconceptions of the sort documented elsewhere [4, 5] and below, which require a coherent theory of longitudinal causality to address. Our divergence from Dr. Karp thus raises the issue of the role of theory and methods in research, although matching serves to illustrate our points in a familiar epidemiologic context.

Misconceptions about the impact of case–control matching remain common. We discuss several subtle problems associated with matched case–control studies that do not arise or are minor in matched cohort studies: (1) matching, even for non-confounders, can create selection bias; (2) matching distorts dose–response relations between matching variables and the outcome; (3) unbiased estimation requires accounting for the actual matching protocol as well as for any residual confounding effects; (4) for efficiency, identically matched groups should be collapsed; (5) matching may harm precision and power; (6) matched analyses may suffer from sparse-data bias, even when using basic sparse-data methods. These problems support advice to limit case–control matching to a few strong well-measured confounders, which would devolve to no matching if no such confounders are measured. On the positive side, odds ratio modification by matched variables can be assessed in matched case–control studies without further data, and when one knows either the distribution of the matching factors or their relation to the outcome in the source population, one can estimate and study patterns in absolute rates. Throughout, we emphasize distinctions from the more intuitive impacts of cohort matching. [ABSTRACT FROM AUTHOR]

Women who drink light-to-moderately during pregnancy have been observed to have lower risk of unfavourable pregnancy outcomes than abstainers. This has been suggested to be a result of bias. In a pooled sample, including 193 747 live-born singletons from nine European cohorts, we examined the associations between light-to-moderate drinking and preterm birth, birth weight, and small-for-gestational age in term born children (term SGA). To address potential sources of bias, we compared the associations from the total sample with a sub-sample restricted to first-time pregnant women who conceived within six months of trying, and examined whether the associations varied across calendar time. In the total sample, drinking up to around six drinks per week as compared to abstaining was associated with lower risk of preterm birth, whereas no significant associations were found for birth weight or term SGA. Drinking six or more drinks per week was associated with lower birth weight and higher risk of term SGA, but no increased risk of preterm birth. The analyses restricted to women without reproductive experience revealed similar results. Before 2000 approximately half of pregnant women drank alcohol. This decreased to 39% in 2000-2004, and 14% in 2005-2011. Before 2000, every additional drink was associated with reduced mean birth weight, whereas in 2005-2011, the mean birth weight increased with increasing intake. The period-specific associations between low-to-moderate drinking and birth weight, which also were observed for term SGA, are indicative of bias. It is impossible to distinguish if the bias is attributable to unmeasured confounding, which change over time or cohort heterogeneity. [ABSTRACT FROM AUTHOR]

Definitions and conceptualizations of confounding and selection bias have evolved over the past several decades. An important advance occurred with development of the concept of exchangeability. For example, if exchangeability holds, risks of disease in an unexposed group can be compared with risks in an exposed group to estimate causal effects. Another advance occurred with the use of causal graphs to summarize causal relationships and facilitate identification of causal patterns that likely indicate bias, including confounding and selection bias. While closely related, exchangeability is defined in the counterfactual-model framework and confounding paths in the causal-graph framework. Moreover, the precise relationships between these concepts have not been fully described. Here, we summarize definitions and current views of these concepts. We show how bias, exchangeability and biasing paths interrelate and provide justification for key results. For example, we show that absence of a biasing path implies exchangeability but that the reverse implication need not hold without an additional assumption, such as faithfulness. The close links shown are expected. However confounding, selection bias and exchangeability are basic concepts, so comprehensive summarization and definitive demonstration of links between them is important. Thus, this work facilitates and adds to our understanding of these important biases. [ABSTRACT FROM AUTHOR]

In observational studies, selection of confounding variables for adjustment is often based on observed baseline incomparability. The aim of this study was to evaluate this selection strategy. We used clinical data on the effects of inhaled long-acting beta-agonist (LABA) use on the risk of mortality among patients with obstructive pulmonary disease to illustrate the impact of selection of confounding variables for adjustment based on baseline comparisons. Among 2,394 asthma and COPD patients included in the analyses, the LABA ever-users were considerably older than never-users, but cardiovascular co-morbidity was equally prevalent (19.9% vs. 19.9%). Adjustment for cardiovascular co-morbidity status did not affect the crude risk ratio (RR) for mortality: crude RR 1.19 (95% CI 0.93-1.51) versus RR 1.19 (95% CI 0.94-1.50) after adjustment for cardiovascular co-morbidity. However, after adjustment for age (RR 0.95, 95% CI 0.76-1.19), additional adjustment for cardiovascular co-morbidity status did affect the association between LABA use and mortality (RR 1.01, 95% CI 0.80-1.26). Confounding variables should not be discarded based on balanced distributions among exposure groups, because residual confounding due to the omission of confounding variables from the adjustment model can be relevant. [ABSTRACT FROM AUTHOR]