Methodological developments in constructing casual diagrams with counterfactual analysis of adolescent alcohol harm

Ferguson, Karl D. (2020) Methodological developments in constructing casual diagrams with counterfactual analysis of adolescent alcohol harm. PhD thesis, University of Glasgow.

Abstract

Background and aims: Causal diagrams, or Directed Acyclic Graphs (DAGs), are mathematically formulated networks of nodes (variables) and arrows which rigorously identify adjustment sets for statistical models. They are thus promising tools for improving statistical analysis in health and social sciences. However, a lack of pragmatic yet robust guidance for building DAGs has been identified as problematic for their use in applied research. This thesis aims to contribute an example of such guidance in the form of a novel research method, and to demonstrate this method’s utility by applying it to observational data. Design: This thesis introduces ‘Evidence Synthesis for Constructing Directed Acyclic Graphs’ (ESC-DAGs) as a protocol for building DAGs from research evidence. It is demonstrated here in the context of parental influences on adolescent alcohol harm and the resulting DAGs are used to inform analysis of data from the Avon Longitudinal Study of Parents and Children (ALSPAC). Methods: ESC-DAGs integrates evidence synthesis principles with classic and modern perspectives on causal inference to produce complex DAGs in a systematic and transparent way. It was applied here to a subset of literature identified from a novel review of systematic reviews, which identified 12 parental influences on adolescent alcohol harm. ESC-DAGs was then further applied to the ALSPAC data to produce a ‘data integrated DAG’. The outcome measure was the Alcohol Use Disorders Identification Test (AUDIT) administered to adolescent participants at age 16.5 years. Nine parental influences were measured, alongside 22 intermediates (variables lying on the causal pathway between parental influences and AUDIT score). The DAGs were then used to direct two stages of analysis: 1) weighting and regression techniques were used to estimate Average Causal Effects (ACEs) for each parental influence and intermediate; and 2) causal mediation analysis was used to decompose the effect of maternal drinking on adolescent AUDIT score to estimate Natural Indirect Effects (NIEs) for the intermediates and the other parental influences. Findings: Evidence for an ACE was found for each parental influence. Parental drinking, low parental monitoring, and parental permissiveness towards adolescent alcohol use had larger effects that were more robust to sensitivity analysis. Several peer and intrapersonal intermediates had higher effects. There was little evidence of an NIE of maternal drinking through other parental influences. There were substantial NIEs for substance-related behaviours of the adolescent and their peers. Conclusions: ESC-DAGs is a promising tool for using DAGs to improve statistical practices. The DAGs produced were transparent and able to direct various forms of data analysis in an immediate sense while differentiating between a comparatively large volume of confounders and other covariates. Future development is possible and should focus on efficiency, replicability, and integration with other methods, such as risk of bias tools. ESC-DAGs may thus prove a valuable platform for discussion in the DAG and wider quantitative research communities. The statistical analyses were performed with methods that were novel to the literature and findings triangulated with the wider evidence base. Mediation analysis provided novel evidence on how parental drinking influences adolescent alcohol harm.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Keywords: Causal diagrams, Directed Acyclic Graphs, Evidence Synthesis for Constructing Directed Acyclic Graphs (ESC-DAGs), causal mediation analysis, potential outcomes framework, adolescent alcohol harm.
• Subjects: H Social Sciences > H Social Sciences (General); R Medicine > RA Public aspects of medicine > RA0421 Public health. Hygiene. Preventive Medicine
• Colleges/Schools: College of Medical Veterinary and Life Sciences > School of Health & Wellbeing > MRC/CSO Unit
• Supervisor's Name: Lewsey, Professor Jim, Mark, Dr. McCann and Danny, Professor Smith
• Date of Award: 2020
• Depositing User: Karl Ferguson
• Unique ID: glathesis:2020-81271
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 06 Apr 2020 15:24
• Last Modified: 29 Apr 2020 11:26
• Thesis DOI: 10.5525/gla.thesis.81271
• URI: https://theses.gla.ac.uk/id/eprint/81271

Related URLs

• Article DOI
• Enlighten Publications Record