Utilização de diagramas causais e a comunicação de fontes de incerteza em estudos observacionais.

Abstract

Causal diagrams (Directed Acyclic Graphs- DAG) have been suggested as a potential tool to improve methodological and reporting quality of observational studies. However, little is known about their empirical applications. A literature review was performed to describe when, where and how causal diagrams have been applied in analytical observational studies in the past 18 years. Potential studies were retrieved through a citation search and a keyword search within PubMed and Web of Science. For a sample of 100 articles that provided the causal structure, the graphical model's features, and reporting of confounding were assessed. In addition, reporting limitations and use of uncertainty markers (hedges) were compared between articles with and without causal DAG. From 2000 to 2017, 1034 publications were identified, including 5021 authors and 85 country affiliations. Only 430 articles (42%) provided the graphic structure. The majority of studies contain only one DAG, and few graphical models included unobserved confounders (23%), or measurement errors (6%) and selection mechanisms (3%). Modifications in the adjustment set were reported in 19% of the articles. Besides, 20% were classified as a possible occurrence of Table 2 fallacy. The number of study limitations acknowledged by the authors and the frequency of uncertainty markers was similar in articles with and without causal diagrams. However, the reporting of quantitative assessments of study limitations were more frequent among studies with DAG (52% vs. 21%). There is a need for further discussion and research concerning the construction and analysis of causal models and the development of reporting guidelines of causal