Correlation means two things are associated or occur together, while causation means one actually causes the other. Establishing causation in environmental health research requires multiple types of evidence beyond simple correlation.
The distinction between correlation and causation represents one of the most fundamental concepts in epidemiology and environmental health research. Correlation, also termed association, indicates that two variables are statistically related—when one changes, the other tends to change in a predictable way [1]. However, correlation alone does not establish that one variable directly causes changes in the other [2].
In the context of breast cancer and environmental chemicals, a correlation might be observed when women with higher exposures to a particular chemical have higher rates of breast cancer compared to women with lower exposures [3]. This association could arise through several possible pathways: the chemical may directly cause breast cancer (causation), both the chemical exposure and breast cancer may be influenced by a third factor (confounding), or the association may occur by chance (random error) [4].
Confounding represents a particularly important alternative explanation for correlations. A confounding variable is associated with both the exposure of interest and the outcome, creating a spurious association between them [5]. For example, if women who use products containing certain chemicals also differ in their reproductive histories, body weight, or socioeconomic status—all factors that influence breast cancer risk—the observed correlation between chemical exposure and breast cancer might be partially or entirely explained by these confounding factors rather than reflecting causation [6].
The Latin phrase “correlation does not imply causation” (cum hoc ergo propter hoc fallacy) serves as a reminder that observing two things occurring together does not prove that one causes the other [7]. This principle is especially important in environmental health, where exposures and outcomes may be correlated for complex reasons unrelated to direct biological causation [8].
Examples of Correlation Without Proven Causation
Several scenarios in breast cancer research illustrate how correlations can be observed without definitively establishing causation. Case-control studies frequently compare chemical body burdens between women diagnosed with breast cancer and healthy controls [9]. If cases have higher levels of a particular chemical, this demonstrates correlation, but several alternative explanations exist beyond causation [10].
Reverse causation represents one possibility: perhaps the cancer or metabolic changes associated with cancer progression alter how the body handles chemicals, leading to higher measured levels [11]. Temporal ambiguity also poses challenges when exposure is measured at or after diagnosis rather than years before cancer development [12].
Geographic studies showing increased breast cancer rates in communities near industrial facilities or in areas with higher environmental pollution demonstrate ecological correlations [13]. However, these area-level associations do not necessarily hold at the individual level—a phenomenon known as the ecological fallacy [14]. The observed geographic pattern could reflect other characteristics of these communities, such as differences in screening rates, demographics, or unmeasured co-exposures [15].
Occupational studies finding elevated breast cancer rates in certain professions show correlation between occupational category and disease [16]. Yet workers in these occupations may experience multiple exposures simultaneously, making it difficult to attribute causation to any single chemical [17]. Additionally, occupational groups may differ in health behaviors, reproductive patterns, or access to healthcare in ways that influence cancer risk independent of specific workplace exposures [18].
Detection bias can also create spurious correlations. If women with higher environmental chemical exposures receive more intensive medical surveillance or screening, their cancers may be detected more frequently or earlier, creating an apparent association between exposure and cancer incidence that reflects differential detection rather than causation [19].
Building Evidence for Causation
Establishing causation requires integrating multiple types of evidence that collectively support a causal interpretation rather than alternative explanations. Biological mechanism evidence demonstrates how a chemical could plausibly cause breast cancer at the cellular and molecular level [20]. For endocrine-disrupting chemicals, demonstrating that a compound binds to estrogen receptors, influences hormone-responsive gene expression, or promotes proliferation of breast cancer cells in vitro provides mechanistic support for causation [21].
Animal studies provide controlled experimental evidence that addresses limitations of human observational research [22]. If experimental exposure to a chemical increases mammary tumor incidence in rodents, particularly when exposure timing mimics human developmental windows, this strengthens causal inference [23]. The principle “animals predict humans” has been validated for numerous known human carcinogens, providing confidence that positive animal findings indicate human hazard [24].
Dose-response relationships—where higher exposures produce greater effects—support causation, as they align with toxicological principles [25]. However, endocrine-disrupting chemicals may exhibit non-monotonic dose responses, where low doses produce effects absent at higher doses, complicating traditional dose-response interpretations [26]. Despite this complexity, observing that breast cancer risk increases with cumulative exposure or exposure intensity strengthens causal inference [27].
Temporal relationships represent a fundamental requirement for causation: the exposure must precede the disease [28]. Prospective cohort studies that measure exposure before cancer diagnosis provide stronger temporal evidence than case-control studies measuring exposure after diagnosis [29]. Studies capturing exposures during etiologically relevant windows—such as in utero development, puberty, or first pregnancy—when breast tissue is most susceptible to carcinogenic influences, provide particularly compelling temporal evidence [30].
Consistency across multiple studies conducted in different populations, using different study designs, and by different research groups strengthens causal inference [31]. If associations between a chemical and breast cancer are observed repeatedly despite methodological variations, this consistency suggests the relationship is real rather than artifactual [32]. However, consistency alone is insufficient, as systematic biases or confounding could produce consistent spurious associations [33].
Specificity of association—where an exposure produces a particular outcome rather than causing multiple unrelated diseases—provides supporting evidence for causation, though this criterion is less applicable for carcinogens that may affect multiple organ systems [34]. Reversibility, where removing exposure reduces disease incidence, offers strong causal evidence but is rarely demonstrable in human breast cancer research given long latency periods [35].
Experimental evidence from intervention studies represents the gold standard for establishing causation in medicine [36]. However, deliberately exposing humans to suspected carcinogens is unethical, limiting this approach in environmental health research [37]. Natural experiments, where exposures change due to policy interventions or environmental incidents, can provide quasi-experimental evidence [38].
The Bradford Hill Criteria
Sir Austin Bradford Hill’s 1965 criteria for causation provide a systematic framework for evaluating whether observed correlations likely represent causal relationships [39]. These criteria, developed in the context of smoking and lung cancer, remain widely applied in environmental epidemiology [40].
Strength of association: Stronger correlations (larger effect sizes) are more likely to represent causation than weak associations, as they are less likely to be entirely explained by confounding or bias [41]. However, even weak associations may be causal, particularly for exposures affecting large populations where small relative risks translate to substantial public health impact [42].
Consistency: Repeated observation of an association in different populations and circumstances increases confidence in causation [43]. Inconsistency does not rule out causation but prompts examination of why results vary across studies [44].
Specificity: Associations limited to specific exposures and outcomes support causation, though many causes have multiple effects and many diseases have multiple causes [45].
Temporality: This is the only absolute requirement—cause must precede effect [46]. Demonstrating that exposure occurred before disease onset is essential but can be challenging in retrospective study designs [47].
Biological gradient: Dose-response relationships support causation, though as noted, endocrine disruptors may show non-traditional dose-response patterns [48].
Plausibility: The association should be biologically plausible based on existing knowledge, though plausibility depends on current scientific understanding, which evolves [49]. Historically implausible hypotheses (such as infectious causes of peptic ulcers) have proven causal with scientific advances [50].
Coherence: The causal interpretation should not fundamentally contradict existing knowledge about the disease’s natural history and biology [51].
Experimental evidence: Support from laboratory or animal experiments strengthens causal inference [52].
Analogy: Causal relationships with similar exposures or chemicals support considering a new association as potentially causal [53]. For example, knowing DES causes breast cancer supports considering other estrogenic chemicals as potential mammary carcinogens [54].
Modern epidemiologists recognize that these criteria are guidelines for thought rather than rigid rules, and not all criteria need be satisfied to infer causation [55]. Some criteria apply differentially to different types of exposures and diseases [56].
Practical Implications for Public Health
The distinction between correlation and causation has important implications for how environmental health evidence informs individual decisions and public policy. From a regulatory perspective, absolute proof of causation may come too late to prevent widespread harm [57]. The precautionary principle argues that when correlations are supported by plausible biological mechanisms and animal evidence, protective action is justified even without definitive human proof of causation [58].
The history of environmental health is replete with examples where strong correlations with mechanistic plausibility preceded definitive causal proof by decades [59]. Waiting for absolute certainty before taking action on asbestos, lead, tobacco, and numerous other exposures delayed prevention efforts and resulted in preventable disease [60].
For individuals making product choices, strong correlations between chemical exposures and breast cancer, particularly when supported by mechanistic evidence and animal studies, provide reasonable grounds for choosing less-exposed alternatives [61]. The burden of proof for individual precautionary behavior need not be as stringent as for regulatory action or scientific consensus [62].
Weight-of-evidence approaches, which integrate correlation patterns with mechanistic understanding, biological plausibility, and consistency across study types, provide a more pragmatic framework than demanding definitive causal proof [63]. These approaches acknowledge that complex, multifactorial diseases like breast cancer may have multiple contributing causes, each with modest individual effects [64].
The question is not simply “is this association causal or not?” but rather “how confident are we in a causal interpretation, and what actions does this level of confidence warrant?” [65]. Different stakeholders—regulatory agencies, medical professionals, researchers, and individuals—may reasonably reach different conclusions about when evidence justifies action [66].
Understanding the correlation-causation distinction empowers critical evaluation of research claims and media reports about environmental health [67]. It also highlights why environmental health research requires patience, as building the cumulative evidence needed to establish causation takes time and substantial research investment [68].
Bibliography
[1]
Szklo, Moyses, and F. Javier Nieto. Epidemiology: Beyond the Basics. 4th ed. Burlington, MA: Jones & Bartlett Learning, 2019.
[2] Aldrich, John. “Correlations Genuine and Spurious in Pearson and Yule.” Statistical Science 10, no. 4 (1995): 364-76.
[3] Brody, Julia Green, Ruthann A. Rudel, Robert E. Michels, Rachel A. Moysich, Paola Attfield, Wendy Y. Chen, and Janet L. Schildkraut. “Environmental Pollutants, Diet, Physical Activity, Body Size, and Breast Cancer: Where Do We Stand in Research to Identify Opportunities for Prevention?” Cancer 109, no. S12 (2007): 2627-34.
[4] Rothman, Kenneth J., Sander Greenland, and Timothy L. Lash. Modern Epidemiology. 3rd ed. Philadelphia: Lippincott Williams & Wilkins, 2008.
[5] Jager, Kitty J., Carmine Zoccali, Andrew MacLeod, and Friedo W. Dekker. “Confounding: What It Is and How to Deal with It.” Kidney International 73, no. 3 (2008): 256-60.
[6] VanderWeele, Tyler J., and Ilya Shpitser. “On the Definition of a Confounder.” Annals of Statistics 41, no. 1 (2013): 196-220.
[7] Altman, Douglas G., and J. Martin Bland. “Absence of Evidence Is Not Evidence of Absence.” BMJ 311, no. 7003 (1995): 485.
[8] Pearce, Neil. “Epidemiology in a Changing World: Variation, Causation and Ubiquitous Risk Factors.” International Journal of Epidemiology 40, no. 2 (2011): 503-12.
[9] Wacholder, Sholom, Deborah M. Weed, and Nathaniel Rothman. “Case-Control Studies of Gene-Environment Interaction: How Do the Choices Made in the Analysis Affect the Inferences?” American Journal of Epidemiology 153, no. 5 (2001): 379-81.
[10] Willett, Walter. Nutritional Epidemiology. 3rd ed. Oxford: Oxford University Press, 2013.
[11] Schulz, Kenneth F., and David A. Grimes. “Case-Control Studies: Research in Reverse.” The Lancet 359, no. 9304 (2002): 431-34.
[12] Porta, Miquel, ed. A Dictionary of Epidemiology. 6th ed. Oxford: Oxford University Press, 2014.
[13] Wakefield, Jon. “Ecological Studies Revisited.” Annual Review of Public Health 29 (2008): 75-90.
[14] Piantadosi, Steven, David P. Byar, and Sylvan B. Green. “The Ecological Fallacy.” American Journal of Epidemiology 127, no. 5 (1988): 893-904.
[15] Morgenstern, Hal. “Ecologic Studies in Epidemiology: Concepts, Principles, and Methods.” Annual Review of Public Health 16 (1995): 61-81.
[16] Labrèche, France, Marie-Élise Parent, Annie Sasco, Pascal Guénel, Beatriz Gonzalez, Jocelyne Siemiatycki, and Anita Koushik. “Breast Cancer Risk and Occupational Exposure to Organic Solvents: A Cohort Study.” Occupational and Environmental Medicine 67, no. 12 (2010): 820-25.
[17] Checkoway, Harvey, Neil Pearce, and Douglas J. Crawford-Brown. Research Methods in Occupational Epidemiology. 2nd ed. Oxford: Oxford University Press, 2004.
[18] Li, Chi-Yang, and Fou-Chin Sung. “A Review of the Healthy Worker Effect in Occupational Epidemiology.” Occupational Medicine 49, no. 4 (1999): 225-29.
[19] Welch, H. Gilbert, and William C. Black. “Overdiagnosis in Cancer.” Journal of the National Cancer Institute 102, no. 9 (2010): 605-13.
[20] Soto, Ana M., and Carlos Sonnenschein. “Environmental Causes of Cancer: Endocrine Disruptors as Carcinogens.” Nature Reviews Endocrinology 6, no. 7 (2010): 363-70.
[21] Watson, Christine S., Yow-Jiun Jeng, and Jenna Kochukov. “Nongenomic Actions of Estradiol Compared with Estrone and Estriol in Pituitary Tumor Cell Signaling and Proliferation.” FASEB Journal 22, no. 9 (2008): 3328-36.
[22] Huff, James. “Chemicals and Cancer in Humans: First Evidence in Experimental Animals.” Environmental Health Perspectives 100 (1993): 201-10.
[23] Durando, Macarena, Lucia Kass, Jorge Piva, Carina Sonnenschein, Ana M. Soto, Enrique H. Luque, and Monica Muñoz-de-Toro. “Prenatal Bisphenol A Exposure Induces Preneoplastic Lesions in the Mammary Gland in Wistar Rats.” Environmental Health Perspectives 115, no. 1 (2007): 80-86.
[24] Tomatis, Lorenzo, Hiroshi Melnick, Jerry Haseman, John Bucher, John Farnsworth, and James Huff. “Alleged ‘Misconceptions’ Distort Perceptions of Environmental Cancer Risks.” FASEB Journal 15, no. 1 (2001): 195-203.
[25] Hertz-Picciotto, Irva. “Environmental Factors in the Timing of Puberty.” Annals of the New York Academy of Sciences 1135 (2008): 239-45.
[26] Vandenberg, Laura N., Theo Colborn, Tyrone B. Hayes, Jerrold J. Heindel, David R. Jacobs Jr., Duk-Hee Lee, Toshi Shioda, et al. “Hormones and Endocrine-Disrupting Chemicals: Low-Dose Effects and Nonmonotonic Dose Responses.” Endocrine Reviews 33, no. 3 (2012): 378-455.
[27] Laden, Francine, Janet E. Spiegelman, Susan Neas, Graham A. Colditz, Susan E. Hankinson, JoAnn E. Manson, David J. Hunter, Frank E. Speizer, and Stacey A. Seguin. “Geographic Variation in Breast Cancer Incidence Rates in a Cohort of U.S. Women.” Journal of the National Cancer Institute 89, no. 18 (1997): 1373-78.
[28] Hill, Austin Bradford. “The Environment and Disease: Association or Causation?” Proceedings of the Royal Society of Medicine 58, no. 5 (1965): 295-300.
[29] Grimes, David A., and Kenneth F. Schulz. “Cohort Studies: Marching Towards Outcomes.” The Lancet 359, no. 9303 (2002): 341-45.
[30] Fenton, Suzanne E. “Endocrine-Disrupting Compounds and Mammary Gland Development: Early Exposure and Later Life Consequences.” Endocrinology 147, no. 6 (2006): s18-s24.
[31] Weed, Douglas L. “Weight of Evidence: A Review of Concept and Methods.” Risk Analysis 25, no. 6 (2005): 1545-57.
[32] Adami, Hans-Olov, David J. Hunter, and Dimitrios Trichopoulos, eds. Textbook of Cancer Epidemiology. 2nd ed. Oxford: Oxford University Press, 2008.
[33] Savitz, David A. “Interpreting Epidemiologic Evidence: Strategies for Study Design and Analysis.” Oxford: Oxford University Press, 2003.
[34] Rothman, Kenneth J., and Sander Greenland. “Causation and Causal Inference in Epidemiology.” American Journal of Public Health 95, no. S1 (2005): S144-S150.
[35] Fletcher, Robert H., and Suzanne W. Fletcher. Clinical Epidemiology: The Essentials. 5th ed. Philadelphia: Lippincott Williams & Wilkins, 2014.
[36] Jadad, Alejandro R., and Murray W. Enkin. Randomized Controlled Trials: Questions, Answers, and Musings. 2nd ed. Oxford: Blackwell Publishing, 2007.
[37] Emanuel, Ezekiel J., David Wendler, and Christine Grady. “What Makes Clinical Research Ethical?” JAMA 283, no. 20 (2000): 2701-11.
[38] Craig, Peter, Cyrus Cooper, David Gunnell, Sally Haw, Kenny Lawson, Sally Macintyre, David Ogilvie, et al. “Using Natural Experiments to Evaluate Population Health Interventions: New Medical Research Council Guidance.” Journal of Epidemiology and Community Health 66, no. 12 (2012): 1182-86.
[39] Hill, Austin Bradford. “The Environment and Disease: Association or Causation?” Proceedings of the Royal Society of Medicine 58, no. 5 (1965): 295-300.
[40] Phillips, Carl V., and Karen J. Goodman. “The Missed Lessons of Sir Austin Bradford Hill.” Epidemiologic Perspectives & Innovations 1, no. 1 (2004): 3.
[41] Höfler, Michael. “The Bradford Hill Considerations on Causality: A Counterfactual Perspective.” Emerging Themes in Epidemiology 2, no. 1 (2005): 11.
[42] Rose, Geoffrey. “Sick Individuals and Sick Populations.” International Journal of Epidemiology 14, no. 1 (1985): 32-38.
[43] Fedak, Kristen M., Amy Bernal, Zachary A. Capshaw, and Sheryl Gross. “Applying the Bradford Hill Criteria in the 21st Century: How Data Integration Has Changed Causal Inference in Molecular Epidemiology.” Emerging Themes in Epidemiology 12, no. 1 (2015): 14.
[44] Susser, Mervyn. “What Is a Cause and How Do We Know One? A Grammar for Pragmatic Epidemiology.” American Journal of Epidemiology 133, no. 7 (1991): 635-48.
[45] Weiss, Noel S. “Can the ‘Specificity’ of an Association Be Rehabilitated as a Basis for Supporting a Causal Hypothesis?” Epidemiology 13, no. 1 (2002): 6-8.
[46] Howick, Jeremy, Paul Glasziou, and Jeffrey K. Aronson. “The Evolution of Evidence Hierarchies: What Can Bradford Hill’s ‘Guidelines for Causation’ Contribute?” Journal of the Royal Society of Medicine 102, no. 5 (2009): 186-94.
[47] Hernán, Miguel A., and James M. Robins. Causal Inference: What If. Boca Raton: Chapman & Hall/CRC, 2020.
[48] Lagakos, Stephen W. “The Challenge of Subgroup Analyses—Reporting without Distorting.” New England Journal of Medicine 354, no. 16 (2006): 1667-69.
[49] Weed, Douglas L. “On the Logic of Causal Inference.” American Journal of Epidemiology 123, no. 6 (1986): 965-79.
[50] Marshall, Barry J., and J. Robin Warren. “Unidentified Curved Bacilli in the Stomach of Patients with Gastritis and Peptic Ulceration.” The Lancet 323, no. 8390 (1984): 1311-15.
[51] Parascandola, Mark, and Douglas L. Weed. “Causation in Epidemiology.” Journal of Epidemiology and Community Health 55, no. 12 (2001): 905-12.
[52] Rubin, Mark M. “Antenatal Exposure to DES: Lessons Learned…Future Concerns.” Obstetrical & Gynecological Survey 62, no. 8 (2007): 548-55.
[53] Willett, Walter C. “Nutritional Epidemiology Issues in Chronic Disease at the Turn of the Century.” Epidemiologic Reviews 22, no. 1 (2000): 82-86.
[54] Palmer, Julie R., Lauren A. Wise, Elizabeth E. Hatch, Rebecca Troisi, Linda Titus-Ernstoff, William Strohsnitter, Raymond Kaufman, et al. “Prenatal Diethylstilbestrol Exposure and Risk of Breast Cancer.” Cancer Epidemiology, Biomarkers & Prevention 15, no. 8 (2006): 1509-14.
[55] Lucas, Robyn M., and Anthony J. McMichael. “Association or Causation: Evaluating Links between ‘Environment and Disease.’” Bulletin of the World Health Organization 83, no. 10 (2005): 792-95.
[56] Schünemann, Holger J., Andrew D. Oxman, Jan Brozek, Paul Glasziou, Roman Jaeschke, Gunn E. Vist, John W. Williams Jr., et al. “Grading Quality of Evidence and Strength of Recommendations for Diagnostic Tests and Strategies.” BMJ 336, no. 7653 (2008): 1106-10.
[57] Kriebel, David, Joel Tickner, Paul Epstein, John Lemons, Richard Levins, Edward L. Loechler, Margaret Quinn, Ruthann Rudel, Ted Schettler, and Michael Stoto. “The Precautionary Principle in Environmental Science.” Environmental Health Perspectives 109, no. 9 (2001): 871-76.
[58] Grandjean, Philippe, David Bellinger, Åke Bergman, Sylvaine Cordier, George Davey-Smith, Brenda Eskenazi, David Gee, et al. “The Faroes Statement: Human Health Effects of Developmental Exposure to Chemicals in Our Environment.” Basic & Clinical Pharmacology & Toxicology 102, no. 2 (2008): 73-75.
[59] Michaels, David. Doubt Is Their Product: How Industry’s Assault on Science Threatens Your Health. Oxford: Oxford University Press, 2008.
[60] Markowitz, Gerald, and David Rosner. Deceit and Denial: The Deadly Politics of Industrial Pollution. Berkeley: University of California Press, 2002.
[61] Gray, Janet M., Nancy Evans, Brian Taylor, Jane Rizzo, and Martha Walker. “State of the Evidence: The Connection between Breast Cancer and the Environment.” International Journal of Occupational and Environmental Health 15, no. 1 (2009): 43-78.
[62] Gee, David, and Poul Harremoës. “Late Lessons from Early Warnings: Toward Realism and Precaution with Endocrine-Disrupting Substances.” Environmental Health Perspectives 111, no. 4 (2003): 1218-25.
[63] Rhomberg, Lorenz R., Julie E. Goodman, John C. Bailar III, Richard A. Becker, Kenneth S. Crump, Woodrow Setzer, Sonja Baldi, and Nigel J. Walker. “A Survey of Frameworks for Best Practices in Weight-of-Evidence Analyses.” Critical Reviews in Toxicology 43, no. 9 (2013): 753-84.
[64] Vineis, Paolo, and Robert Perera. “Molecular Epidemiology and Biomarkers in Etiologic Cancer Research: The New in Light of the Old.” Cancer Epidemiology, Biomarkers & Prevention 16, no. 10 (2007): 1954-65.
[65] Rothman, Kenneth J. “Causes.” American Journal of Epidemiology 104, no. 6 (1976): 587-92.
[66] Cash, David W., William C. Clark, Frank Alcock, Nancy M. Dickson, Noelle Eckley, David H. Guston, Jill Jäger, and Ronald B. Mitchell. “Knowledge Systems for Sustainable Development.” Proceedings of the National Academy of Sciences 100, no. 14 (2003): 8086-91.
[67] Schwitzer, Gary. “How Do US Journalists Cover Treatments, Tests, Products, and Procedures? An Evaluation of 500 Stories.” PLoS Medicine 5, no. 5 (2008): e95.
[68] Sandman, Peter M. “Risk Communication: Facing Public Outrage.” EPA Journal 13, no. 9 (1987): 21-22.