Tools
Principles of Causation
Summary / Explanation
Causation refers to a process wherein an initial or inciting event (exposure) affects the probability of a subsequent or resulting event (outcome) occurring.[1][2] Epidemiologists' definitions of causation and methods for establishing causal relationships (causality) have evolved. Contemporary studies involving causality require strong assumptions, causal-structural subject-matter knowledge, careful statistical analysis, and considerations for alternative explanations.[3] The following models demonstrate the core principles of causation.
Bradford Hill Criteria for Causation
First outlined in 1965 by Sir Austin Bradford Hill to demonstrate the link between tobacco smoking and lung cancer, these 9 aspects of the association have historically been used as a quasi-checklist to assess causal relationships.[4][5]
- Temporality: The effect follows the cause. Of all the criteria, this is the most widely accepted requirement of a causal relationship.
- Strength of association: A stronger relationship between the exposure and outcome implies a higher likelihood of causality. A relationship's strength is typically measured by statistical analysis, and modern computing has significantly contributed to this effort.
- Consistency: Similar relationships between the exposure and outcome are observed in different studies and settings.
- Specificity: An exposure is linked to a specific outcome in a particular population. Previously believed to be a weak criterion, modern data integration now allows for causation to be assessed based on the presence or absence of relationships between variables.
- Biological relationship: An increased level of exposure results in an increased frequency of the effect, also known as the dose-response relationship. This monotonic description is now regarded as overly simplistic, and Hill acknowledged the possibility that more complex dose-response relationships exist in his original outline.
- Plausibility: The proposed exposure-outcome relationship is consistent with biological knowledge. Hill remarked that he was restricted by the biological knowledge of his time, and modern technology allows for continuous reassessment of causation based on discoveries, particularly in molecular biology.
- Coherence: New evidence aligns with existing findings.
- Experiment: Studies support the presumed causal relationship. Hill asserted that reducing or completely removing an exposure produces the same effect on outcome. However, modern experimenters acknowledge that disease processes can occur due to various exposures and complex mechanisms, accounting for conditions that do not resolve or worsen by manipulating a single exposure variable.
- Analogy: A known causal relationship between an exposure and outcome suggests a similar relationship for another exposure and outcome. This argument remains valid if the exposure and outcome are similar to the original, although the evidence for this new relationship may be weaker.
These criteria have since been scrutinized, with some researchers asserting that temporality based on causal-structural subject-matter knowledge is the only essential criterion for establishing causality.[4][6]
Sufficient and Component Cause Model
Proposed by Rothman, this model defines cause as an event, condition, or characteristic necessary for disease occurrence, emphasizing that a disease results from multiple components acting together.[7][8] This model aids in understanding the multifactorial nature of disease causation in epidemiology.A cause (or set of causes) may contribute to a causal relationship if it is:
- Sufficient: A set of minimal conditions or events inevitably leads to an outcome. Each condition is necessary. Completion of a sufficient cause is equivalent to disease onset.
- Necessary: A component present in every sufficient cause for a particular disease.
- Component: An individual event, condition, or characteristic a sufficient cause requires. A single component cause can be part of multiple sufficient causes.
Understanding how these descriptions function together in disease causation involves several key elements as follows:
- Strength of a cause: Determined by the prevalence of its components. A rare factor can be strong if its complementary causes are common.
- Interactions among causes: Mutual biological interactions exist within a sufficient cause.
- Induction period: The time from causal action to disease initiation.
- Latent period: The interval between disease occurrence and its detection.
- Synergism: The combined effect of causes exceeds the sum of their individual effects.
Counterfactual Model
The counterfactual (or potential outcomes) model has become a standard approach for inferring causality in healthcare-related sciences.[9][10] This model defines true causal effects as the differences between observed outcomes in exposed individuals and their counterfactual outcomes if unexposed, all else being equal.[8][9][11] As it is impossible to designate the same population to be exposed and unexposed simultaneously, epidemiologists compare the disease risk in an exposed group to the risk in an unexposed group that is as similar as possible (exchangeable).[10][12] Non-exchangeability occurs when the substitute population does not accurately represent the target population's counterfactual experience, leading to invalid or biased effect estimates.
Bias in Causal Inference
Bias is critical for establishing causation and must be considered during a study's design and analysis phases. There are 2 primary forms of bias—random and structural.[13] Random bias often occurs due to chance in studies with a small number of subjects but can be mitigated by increasing this number.[14] Structural bias persists regardless of study size and includes confounding, selection bias, and information bias.[13][15][16][17]
Confounding occurs when the effect of an exposure on the outcome intermingles with the effect of another factor (confounder) not under study, leading to an inaccurate assessment of the true causal effect of interest. Confounding can be addressed through study design, such as randomization, and analysis techniques, such as stratification, regression, and propensity-score methods.[2][18][19][20]
A confounder meets the following 3 criteria:
- Associated with the exposure of interest
- Associated with the outcome of interest independent of the exposure
- Not a downstream consequence of the exposure or outcome of interest
Selection bias is a potential problem in all study designs. This problem arises from non-representative sample selection or non-random participant dropout and can be prevented by careful participant selection.[2][13]Information bias results from erroneous data collection, commonly due to measurement error or misclassification, and is best mitigated during the study design stage.[16][17] Addressing this bias during analysis is more challenging and often requires advanced methods.[21]
Conclusion Establishing causation in epidemiology requires strong assumptions, causal-structural subject-matter knowledge, and careful study design and statistical analysis considerations.[22][23][24][25] Addressing challenges, such as confounding and other forms of bias, is essential for ensuring the validity of effect estimates, ultimately guiding effective public health interventions and improving patient care.[26][27]
Register For Free And Read The Full Article
Search engine and full access to all medical articles- 10 free questions in your specialty
- Free CME/CE Activities
- Free daily question in your email
- Save favorite articles to your dashboard
- Emails offering discounts
Learn more about a Subscription to StatPearls Point-of-Care
References
Hernán MA, Robins JM. Estimating causal effects from epidemiological data. Journal of epidemiology and community health. 2006 Jul:60(7):578-86 [PubMed PMID: 16790829]
Level 2 (mid-level) evidenceVanderWeele TJ, Shpitser I. On the definition of a confounder. Annals of statistics. 2013 Feb:41(1):196-220 [PubMed PMID: 25544784]
Joffe M, Gambhir M, Chadeau-Hyam M, Vineis P. Causal diagrams in systems epidemiology. Emerging themes in epidemiology. 2012 Mar 19:9(1):1. doi: 10.1186/1742-7622-9-1. Epub 2012 Mar 19 [PubMed PMID: 22429606]
Fedak KM, Bernal A, Capshaw ZA, Gross S. Applying the Bradford Hill criteria in the 21st century: how data integration has changed causal inference in molecular epidemiology. Emerging themes in epidemiology. 2015:12():14. doi: 10.1186/s12982-015-0037-4. Epub 2015 Sep 30 [PubMed PMID: 26425136]
Shimonovich M, Pearce A, Thomson H, Keyes K, Katikireddi SV. Assessing causality in epidemiology: revisiting Bradford Hill to incorporate developments in causal thinking. European journal of epidemiology. 2021 Sep:36(9):873-887. doi: 10.1007/s10654-020-00703-7. Epub 2020 Dec 16 [PubMed PMID: 33324996]
Höfler M. The Bradford Hill considerations on causality: a counterfactual perspective. Emerging themes in epidemiology. 2005 Nov 3:2():11 [PubMed PMID: 16269083]
Level 3 (low-level) evidenceRothman KJ. Causes. American journal of epidemiology. 1976 Dec:104(6):587-92 [PubMed PMID: 998606]
Pearl J. An introduction to causal inference. The international journal of biostatistics. 2010 Feb 26:6(2):Article 7. doi: 10.2202/1557-4679.1203. Epub 2010 Feb 26 [PubMed PMID: 20305706]
Höfler M. Causal inference based on counterfactuals. BMC medical research methodology. 2005 Sep 13:5():28 [PubMed PMID: 16159397]
Flanders WD. On the relationship of sufficient component cause models with potential outcome (counterfactual) models. European journal of epidemiology. 2006:21(12):847-53 [PubMed PMID: 17048084]
Greenland S, Robins JM. Identifiability, exchangeability, and epidemiological confounding. International journal of epidemiology. 1986 Sep:15(3):413-9 [PubMed PMID: 3771081]
Level 2 (mid-level) evidenceMittleman MA, Mostofsky E. Exchangeability in the case-crossover design. International journal of epidemiology. 2014 Oct:43(5):1645-55. doi: 10.1093/ije/dyu081. Epub 2014 Apr 22 [PubMed PMID: 24756878]
Level 3 (low-level) evidenceHernán MA, Hernández-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology (Cambridge, Mass.). 2004 Sep:15(5):615-25 [PubMed PMID: 15308962]
Rothman KJ, Greenland S. Planning Study Size Based on Precision Rather Than Power. Epidemiology (Cambridge, Mass.). 2018 Sep:29(5):599-603. doi: 10.1097/EDE.0000000000000876. Epub [PubMed PMID: 29912015]
VanderWeele TJ, Hernán MA, Robins JM. Causal directed acyclic graphs and the direction of unmeasured confounding bias. Epidemiology (Cambridge, Mass.). 2008 Sep:19(5):720-8. doi: 10.1097/EDE.0b013e3181810e29. Epub [PubMed PMID: 18633331]
Althubaiti A. Information bias in health research: definition, pitfalls, and adjustment methods. Journal of multidisciplinary healthcare. 2016:9():211-7. doi: 10.2147/JMDH.S104807. Epub 2016 May 4 [PubMed PMID: 27217764]
Maclure M, Schneeweiss S. Causation of bias: the episcope. Epidemiology (Cambridge, Mass.). 2001 Jan:12(1):114-22 [PubMed PMID: 11138805]
Kahlert J, Gribsholt SB, Gammelager H, Dekkers OM, Luta G. Control of confounding in the analysis phase - an overview for clinicians. Clinical epidemiology. 2017:9():195-204. doi: 10.2147/CLEP.S129886. Epub 2017 Mar 31 [PubMed PMID: 28408854]
Level 3 (low-level) evidenceAustin PC. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate behavioral research. 2011 May:46(3):399-424 [PubMed PMID: 21818162]
VanderWeele TJ. Principles of confounder selection. European journal of epidemiology. 2019 Mar:34(3):211-219. doi: 10.1007/s10654-019-00494-6. Epub 2019 Mar 6 [PubMed PMID: 30840181]
Greenland S, Morgenstern H. Ecological bias, confounding, and effect modification. International journal of epidemiology. 1989 Mar:18(1):269-74 [PubMed PMID: 2656561]
Level 2 (mid-level) evidenceMathur MB, VanderWeele TJ. Methods to Address Confounding and Other Biases in Meta-Analyses: Review and Recommendations. Annual review of public health. 2022 Apr 5:43():19-35. doi: 10.1146/annurev-publhealth-051920-114020. Epub 2021 Sep 17 [PubMed PMID: 34535060]
Nowinski CJ, Bureau SC, Buckland ME, Curtis MA, Daneshvar DH, Faull RLM, Grinberg LT, Hill-Yardin EL, Murray HC, Pearce AJ, Suter CM, White AJ, Finkel AM, Cantu RC. Applying the Bradford Hill Criteria for Causation to Repetitive Head Impacts and Chronic Traumatic Encephalopathy. Frontiers in neurology. 2022:13():938163. doi: 10.3389/fneur.2022.938163. Epub 2022 Jul 22 [PubMed PMID: 35937061]
Koopmans E, Schiller DC. Understanding Causation in Healthcare: An Introduction to Critical Realism. Qualitative health research. 2022 Jul:32(8-9):1207-1214. doi: 10.1177/10497323221105737. Epub 2022 Jun 1 [PubMed PMID: 35649292]
Level 2 (mid-level) evidenceEtminan M, Collins GS, Mansournia MA. Using Causal Diagrams to Improve the Design and Interpretation of Medical Research. Chest. 2020 Jul:158(1S):S21-S28. doi: 10.1016/j.chest.2020.03.011. Epub [PubMed PMID: 32658648]
Raita Y, Camargo CA Jr, Liang L, Hasegawa K. Big Data, Data Science, and Causal Inference: A Primer for Clinicians. Frontiers in medicine. 2021:8():678047. doi: 10.3389/fmed.2021.678047. Epub 2021 Jul 6 [PubMed PMID: 34295910]
Roy J, Mitra N. Measured and accounted-for confounding in pharmacoepidemiologic studies: Some thoughts for practitioners. Pharmacoepidemiology and drug safety. 2021 Mar:30(3):277-282. doi: 10.1002/pds.5189. Epub 2021 Jan 6 [PubMed PMID: 33372303]