Article Text

Download PDFPDF

OP115 Improving the assessment of causality in population health: should bradford hill be revisited to incorporate developments in causal inference?
  1. M Shimonovich,
  2. H Thomson,
  3. A Pearce,
  4. V Katikireddi
  1. MRC/CSO Social and Public Health Sciences Unit, University of Glasgow, Glasgow, UK


Background Bradford Hill’s (BH) guidelines are the traditional approach to causal assessment in population health and epidemiology. However, assessments can be inconclusive; there is no consensus on the thresholds required for components. Some have proposed incorporating more recent developments in causal thinking to BH guidelines to improve assessment of causality. This study aims to understand how traditional approaches to causal assessment can be refined by incorporating alternative causal methods. We will do this by understanding the similarities and differences of these approaches to BH.

Methods We mapped each BH component against three subsequent, prominent causal inference approaches: directed acyclic graphs (DAGs), grading of recommendations, assessment, development and evaluation methodology (GRADE), and sufficient-component cause models (SCC, also referred to as ‘causal pies’), drawing upon existing studies that had assessed the overlap between one or more of these approaches. Existing studies were found through targeted searching and snowballing, with no a priori list of inclusion/exclusion criteria.

Results The approaches can be grouped into two categories: models (DAGs and SCC) and assessment guidelines (BH and GRADE). The literature does not necessarily explicitly make this distinction, but the identified literature largely restricted comparisons within each of these categories.

We found that some components overlap between the guidelines and models, while some are specific to certain approaches. For example, BH causal assessment considers if an increased exposure corresponds with increased incidence of the disease (dose-response). Similarly, GRADE will upgrade evidence from an observational study with evidence of dose-response. However, testing dose-response for DAGs may not be helpful. A dose-response may be demonstrated for different exposure levels due to a confounder that has the same impact on the exposure and the outcome. Thus, it would be the confounder causing the dose-response, not the causal relationship. The SCC model is often drawn with binary exposures and outcomes where dose-response is not considered. However, it can be incorporated by including dose as providing different contributions to the causal pie. Similar comparisons were made for the remaining BH components.

Conclusion Assessing causal relationships is challenging, yet of fundamental importance. There have been limited efforts to incorporate insights from DAGs and SCC into BH guidelines. However, our review did not investigate all potential approaches to assessing causality (e.g. International Agency for Research on Cancer) and the comparisons require further analysis. Nevertheless, this detailed exploration improves the potential for refining our approach to making judgements about causal relationships in public health.

  • causal assessment approaches

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.