Use of directed acyclic graphs (DAGs) in applied health research: review and recommendations

Abstract

BACKGROUND: Directed acyclic graphs (DAGs) are an increasingly popular approach for identifying confounding variables that require adjustment when estimating causal effects. This review examined the use of DAGs in applied health research to inform recommendations for improving their transparency and utility in future research. METHODS: Original health research articles published during 1999-2017 mentioning "directed acyclic graphs" or similar or citing DAGitty were identified from Scopus, Web of Science, Medline, and Embase. Data were extracted on the reporting of: estimands, DAGs, and adjustment sets, alongside the characteristics of each article's largest DAG. RESULTS: A total of 234 articles were identified that reported using DAGs. A fifth (n=48, 21%) reported their target estimand(s) and half (n=115, 48%) reported the adjustment set(s) implied by their DAG(s). Two-thirds of the articles (n=144, 62%) made at least one DAG available. Diagrams varied in size but averaged 12 nodes (IQR: 9-16, range: 3-28) and 29 arcs (IQR: 19-42, range: 3-99). The median saturation (i.e. percentage of total possible arcs) was 46% (IQR: 31-67, range: 12-100). 37% (n=53) of the DAGs included unobserved variables, 17% (n=25) included super-nodes (i.e. nodes containing more than one variable, and a 34% (n=49) were arranged so the constituent arcs flowed in a consistent direction. CONCLUSIONS: There is substantial variation in the use and reporting of DAGs in applied health research. Although this partly reflects their flexibility, it also highlight some potential areas for improvement. This review hence offers several recommendations to improve the reporting and use of DAGs in future research.

Assessing Knowledge, Attitudes, and Practices towards Causal Directed Acyclic Graphs: a qualitative research project

“…Our findings build upon the findings of a recent, large, systematic review of published causal graphs in the epidemiologic literature (Tennant et al 2020). There, the authors found that 38% of papers which claimed to have used DAGs did not include the model in the main text or supplementary material.…”

“…Causal graphs, such as causal directed acyclic graphs, are a useful tool for improving the clarity of the assumptions required for valid causal inference, interrogating those assumptions, and designing appropriate data collection and analytic plans in light of the required assumptions (Greenland, Pearl, and Robins 1999;Tennant et al 2020). Although the mathematical foundations of causal graphs have been well-specified, they have not been widely adopted in epidemiology and medical research (Greenland, Pearl, and Robins 1999;Hernan and Robins 2020;Pearl 1995;Tennant et al 2020). We believe the reason for this implementation lag is the lack of available tools for the application of causal graphs in the study design and analysis phases of applied research studies.…”

Introducing causal inference to the medical curriculum using temporal logic to draw directed acyclic graphs

“…Of course it is possible that the apparent improvements in covariate classification observed following DAG specification in the present study simply reflected the tougher conceptual challenge involved in the application of temporality during the preceding task (temporality-driven covariate classification), particularly since the undergraduates involved had limited expertise in the conceptualisation and operationalisation of quantitative variables and the impact thereof on the opacity of temporal relationships amongst and between these. Indeed, these are also exacting challenges for competent analysts with advanced training and substantial experience (Tennant et al 2020); and there are well-established (if contentious and contested) concerns that drawing DAGs might actually obfuscate rather than elucidate the critical insights and associated thinking required to design analytical models capable of supporting causal inference with observational data (e.g. Krieger and Davey Smith 2016).…”

“…Elsewhere, 16 (24.2%) DAGs used composite 'super-nodes' (i.e. a single node with which two or more covariates were associated; Tennant et al 2020), and in these 16 DAGs the median number of arcs drawn was just 6 (range: 3-9), while for the remaining 50 (75.8%) DAGsall of which had separate nodes for each of the selected covariatesthe median number of arcs was 12 (4-22). Neither approach to DAG specification (using super-nodes or separate nodes) generated DAGs that were assessed as being 'forward saturated' (i.e.…”

The role of welfare regimes in the relationship between childhood economic stress and adult health: a multilevel study of 20 European countries

“…The diagram showed that the minimal set of control variables necessary to include to investigate the association between CES and SRH/cancer was welfare regime, family socioeconomic conditions, and sociodemographics (age, gender, immigrant status) ( Fig. 1 ), where the latter is included in the diagram as a “super-node” ( Tennant et al, 2019 ). The diagram shows that SES is an intermediate variable, and we included SES in separate models.…”