70 - A Model Evidence Framework for Causal Modeling of the Effect of Midline Elevated Head Positioning on Risk of Intraventricular Hemorrhage in Preterm infants.

Poster Number: 70
Publication Number: 70.437

Benjamin al-Haddad, University of Washington School of Medicine, Seattle, WA, United States; Kendell German, University of Washington, Seattle, WA, United States; Sarah E. Kolnik, University of Washington School of Medicine, Seattle, WA, United States; Krystle M. Perez, University of Washington School of Medicine, Snohomish, WA, United States; Sandra E. Juul, University of Washington, Bellevue, WA, United States

Background: Directed Acyclic Graphs (DAG) are analytical tools for causal models that make assumptions explicit and directly inform statistical analyses. Although best practices suggest that DAG representations be informed by evidence, there is a critical lack of formal methods to represent the relationship evidence for clinically important questions. For example, midline elevated positioning in the first 72 hours of life may reduce the risk of intraventricular hemorrhage (IVH) in preterm and or small neonates. Although this practice has been widely adapted, there is limited evidence whether it is effective in decreasing risk of IVH.

Objective: Our objective was to develop a Model Evidence Framework to inform the causal DAG model to determine whether midline elevated positioning decreased risk of IVH in infants born before 30 weeks gestation or whose birth weight was less than 1250g.

Design/Methods: We defined 27 variables (e.g. maternal and infant demographics, delivery complications, and neonatal morbidities, etc.) hypothesized to be on the causal pathway between midline elevated head positioning and IVH based on literature review and author consensus. PubMed searches were performed for each combination of variables (325 searches) to determine: (1) subjective level of evidence for each relationship, (2) evidence of causality and (3) direction of causality. We used DAGitty graph theory software to determine the DAG implications for statistical analysis.

Results: See Figure 1 for Model Evidence Framework summary table and Figure 2 for the DAG. After graph theory relationship analysis of the 27 variables, 10 were ancestors in the causal chain of the exposure (midline elevated positioning) and the outcome of interest (IVH). Twelve factors were causal ancestors of the outcome (IVH). The model contained 477 testable assumptions. Based on the graph theory analysis, the minimal sufficient adjustment set for estimating the relationship between midline elevated head positioning and risk of IVH was birth weight, gestational age and year.Conclusion(s): The Model Evidence Framework is a novel rubric for the formal representation of evidence and assumptions to inform causal models of clinically significant questions. This framework may be applied in the analysis of a broad range of clinical questions in pediatrics. We find that given the evidence based causal assumptions, a statistical analysis of the relationship between midline elevated positioning and risk of IVH must at least control for birth weight, gestational age and year of birth.
cv_academic_12032021_umn.pdf
Figure 2. Directed Acyclic Graph based on the Model Evidence Framework for the relationship between Midline Elevated Positioning and Intraventricular Hemorrhage produced using DAGitty graph theory analysis software.Each arrow or “edge” represents a causal relationship assumption from the Model Evidence Framework.