Directed acyclic graphs: a tool for causal studies in paediatrics

Abstract

Many paediatric clinical research studies, whether observational or interventional, have as an eventual aim the identification or quantification of causal relationships. One might ask: does screen time influence childhood obesity? Could overuse of paracetamol in infancy cause wheeze? How does breastfeeding affect later cognitive outcomes? In this review, we present causal Directed Acyclic Graphs (DAGs) to a paediatric audience. DAGs are a graphical tool which provide a way to visually represent and better understand the key concepts of exposure, outcome, causation, confounding, and bias. We use clinical examples, including those outlined above, framed in the language of DAGs, to demonstrate their potential applications. We show how DAGs can be most useful in identifying confounding and sources of bias, demonstrating inappropriate statistical adjustments for presumed biases, and understanding threats to validity in randomised controlled trials. We believe that a familiarity with DAGs, and the concepts underlying them, will be of benefit both to researchers planning studies, and practicing clinicians interpreting them.

Relationships of maternal body mass index and plasma biomarkers with childhood body mass index and adiposity at 6 years: The Children of SCOPE study

“…To aid the selection of confounders in the analysis of the relationship between maternal BMI and childhood BMI and adiposity, linear regression was used to identify maternal clinical factors and biomarkers associated with childhood adiposity outcomes, independent from maternal BMI. Confounders selected for model B (Table S1) were chosen a priori based on known associations with obesity or implication in the causal pathway between maternal BMI of childhood obesity. These were decided upon with the use of directed acyclic graphs (Figure S1).…”

“…These were decided upon with the use of directed acyclic graphs (Figure S1). Results were corrected for multiple testing using Benjamini and Hochberg's false discovery rate . The variables used in the adjusted model were infant mode of feeding, infant sex, and maternal: ethnicity, socioeconomic status, green leafy vegetable intake pre‐pregnancy, multivitamin use pre‐pregnancy, alcohol intake pre‐pregnancy, cigarette smoking in first trimester, time spent watching television in first trimester, depression in first trimester, sleep assessed at 20 weeks' gestation, maternal hypertension, GDM diagnosis, and placental growth factor.…”

Alcohol Intake in Early Pregnancy and Spontaneous Preterm Birth: A Cohort Study

“…A directed acyclic graph (Howards, ; Williams et al, ) was used to select the following self‐reported confounders a priori: maternal age (years), cohabitation status ( cohabiting/living alone ), highest attained educational level ( none, skilled training, <3 years of higher education, 3 to 4 years of higher education, and ≥ 5 years of higher education ), pregestational body mass index (BMI) based on self‐reported height and weight (kg/m 2 ), recreational drug use prior to pregnancy ( yes/no ), physical activity in pregnancy ( <1, 1 to 5, and >5 h/wk ), smoking in pregnancy ( yes/no ), parity ( nulliparous/multiparous ), and assisted reproductive technology in the current pregnancy ( yes/no ). Based on the most prevalent chronic disease categories among Danish pregnant women (hypertension, heart disease, lung disease including asthma, diabetes mellitus, thyroid disorders, rheumatoid arthritis, and psychiatric disorder) (Jolving et al, ), self‐reported chronic disease ( yes/no ) was also included as a confounder.…”

Reduced survival of total knee arthroplasty after previous unicompartmental knee arthroplasty compared with previous high tibial osteotomy: a propensity-score weighted mid-term cohort study based on 2,133 observations from the Danish Knee Arthroplasty Registry

“…PS were estimated with logistic regression and applied by IPTW with stabilized weights aiming to estimate the average effect of treatment (Austin 2014). Based on the considerations depicted in the directed acyclic graph (Williams et al 2018) ( Figure 2, see Supplementary data), the following covariates were included in the model: sex, age (quantiles), weight (quantiles), KSSs, Charnley class, level of constraint, patella resurfacing, fixation, hospitals annual arthroplasty volume, and period of…”