Assessing causality in epidemiology: revisiting Bradford Hill to incorporate developments in causal thinking

Abstract

The nine Bradford Hill (BH) viewpoints (sometimes referred to as criteria) are commonly used to assess causality within epidemiology. However, causal thinking has since developed, with three of the most prominent approaches implicitly or explicitly building on the potential outcomes framework: directed acyclic graphs (DAGs), sufficient-component cause models (SCC models, also referred to as ‘causal pies’) and the grading of recommendations, assessment, development and evaluation (GRADE) methodology. This paper explores how these approaches relate to BH’s viewpoints and considers implications for improving causal assessment. We mapped the three approaches above against each BH viewpoint. We found overlap across the approaches and BH viewpoints, underscoring BH viewpoints’ enduring importance. Mapping the approaches helped elucidate the theoretical underpinning of each viewpoint and articulate the conditions when the viewpoint would be relevant. Our comparisons identified commonality on four viewpoints: strength of association (including analysis of plausible confounding); temporality; plausibility (encoded by DAGs or SCC models to articulate mediation and interaction, respectively); and experiments (including implications of study design on exchangeability). Consistency may be more usefully operationalised by considering an effect size’s transportability to a different population or unexplained inconsistency in effect sizes (statistical heterogeneity). Because specificity rarely occurs, falsification exposures or outcomes (i.e., negative controls) may be more useful. The presence of a dose-response relationship may be less than widely perceived as it can easily arise from confounding. We found limited utility for coherence and analogy. This study highlights a need for greater clarity on BH viewpoints to improve causal assessment.

SARS-CoV-2 infection during pregnancy and risk of preeclampsia: a systematic review and meta-analysis

“…When assessing causality, evidence of a dose-response relationship suggests that the association is less likely to be due to confounding. 116,117 We found evidence for a dose-response gradient effect on the basis that patients with symptomatic illness were more likely to develop preeclampsia (OR¼2.11) than those with asymptomatic illness (OR¼1.59). Recently, a multicenter cohort study by Metz et al, 118 which did not include women without SARS-CoV-2 infection, reported that the frequency of hypertensive disorders of pregnancy among women with asymptomatic SARS-CoV-2 infection, mild or moderate disease, and severe or critical disease was 18.8%, 23.8%, and 40.4%, respectively.…”

“…Temporality is considered fundamental when assessing the likelihood of a causal relationship between an exposure and an outcome. 116,117 Evidence that the participants were exposed before the occurrence of the outcome strengthens a causal relationship argument. The study by Rosenbloom et al 108 provided clear evidence to support a meaningful temporal relationship between SARS-CoV-2 infection and preeclampsia.…”

Impact of educational level and employment status on short-term and long-term pain relief from supervised exercise therapy and education: an observational study of 22 588 patients with knee and hip osteoarthritis

“…According to the temporal assumption, some of the variables measured at baseline were therefore not considered to be confounders in the relationship between educational level and pain intensity, in contrast to the relationship between current employment status and pain intensity. 16 In the model with education, we adjusted for age (continuous) and gender (male/female). In the model with employment, we additionally adjusted for BMI (continuous), educational level (categorical), SF-12 MCS (continuous), number of comorbidities (discrete) and prior surgery (no/yes).…”

Causal attribution fractions, and the attribution of smoking and BMI to the landscape of disease incidence in UK Biobank

“…Disagreement between studies may indicate incomplete understanding of the underlying causal model, with inappropriate or insufficient adjustment for confounding factors. In the common situation where uncertainty of the causal processes linking exposure X to disease risk remain, then the standard methods and cautious reporting of conventional epidemiology must remain [2, 8], and data from these observational studies cannot reliably be used in meta-analyses.…”

“…Although epidemiologists are careful to describe their results in terms of “associations”, the purpose of epidemiology is to detect and quantify causal associations, e.g. between lifestyles and health [2, 8]. Recently the science of causal inference [6, 7, 9], has developed to identify circumstances where causal estimates are possible using observational data.…”