Applying the Bradford Hill criteria in the 21st century: how data integration has changed causal inference in molecular epidemiology.

Fedak KM, Bernal A, Capshaw ZA, Gross S. · 2015

Read Applying the Bradford Hill criteria in the 21st century: how data integration has changed causal inference in molecular epidemiology. by Fedak KM, Bernal A, Capshaw ZA, Gross S. (2015) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4589117.txt.

Introduction

Kristen M. Fedak, Autumn Bernal, Zachary A. Capshaw, and Sherilyn Gross revisit a classic checklist for proving cause, in Emerging Themes in Epidemiology in 2015. In 1965 Austin Bradford Hill published nine viewpoints for judging whether an observed association is causal. They became epidemiology's most cited framework, yet they were written before modern genetics, molecular toxicology, and computing existed.

Today's studies fuse classic population data with gene, exposure, and mechanism data, and each Hill viewpoint reads differently against those new sources. The paper works through the criteria one by one using modern examples, showing where data integration strengthens a causal call and where it can mislead. The aim is a 21st-century reading of Hill, not a replacement.

Catalog record

CREATORFedak KM, Bernal A, Capshaw ZA, Gross S.
YEAR2015
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC4589117

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Applying the Bradford Hill criteria in the 21st century: how data integration has changed causal inference in molecular epidemiology By Kristen M Fedak, Autumn Bernal, Zachary A Capshaw, Sherilyn Gross Published in: Emerging Themes in Epidemiology DOI: 10.1186/s12982-015-0037-4 ABSTRACT Abstract In 1965, Sir Austin Bradford Hill published nine “viewpoints” to help determine if observed epidemiologic associations are causal. Since then, the “Bradford Hill Criteria” have become the most frequently cited framework for causal inference in epidemiologic studies. However, when Hill published his causal guidelines—just 12 years after the double-helix model for DNA was first suggested and 25 years before the Human Genome Project began—disease causation was understood on a more elementary level than it is today. Advancements in genetics, molecular biology, toxicology, exposure science, and statistics have increased our analytical capabilities for exploring potential cause-and-effect relationships, and have resulted in a greater understanding of the complexity behind human disease onset and progression. These additional tools for causal inference necessitate a re-evaluation of how each Bradford Hill criterion should be interpreted when considering a variety of data types beyond classic epidemiology studies. Herein, we explore the implications of data integration on the interpretation and application of the criteria. Using examples of recently discovered exposure–response associations in human disease, we discuss novel ways by which researchers can apply and interpret the Bradford Hill criteria when considering data gathered using modern molecular techniques, such as epigenetics, biomarkers, mechanistic toxicology, and genotoxicology. Keywords: Causation, Causal inference, Data integration, Bradford Hill, Molecular epidemiology BACKGROUND In 1965, Sir Austin Bradford Hill gave the first President’s Address to the newly formed Section on Occupational Medicine, which was published within the Proceedings of the Royal Society of Medicine [1]. Hill began his address by pointing out a fundamental problem facing the Section members: how could they effectively practice preventative occupational medicine without a basis for determining which occupational hazards ultimately cause sickness and injury? Namely, Hill asked, “In what circumstances can [one] pass from [an] observed association to a verdict of causation?” [1]. He proceeded to propose nine “aspects of association” for evaluating traditional epidemiologic data. These aspects, which have since become fundamental tenets of causal inference in epidemiology, are often referred to as the Bradford Hill Criteria. The nine “aspects of association” that Hill discussed in his address (strength of association, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy) have been used to evaluate countless hypothesized relationships between occupational and environmental exposures and disease outcomes. Yet, when Hill conceived these nine aspects (hereafter referred to as criteria), the mechanistic connections between exposure and disease were not well understood. Consider that Hill published his criteria just 12 years after Watson and Crick first suggested the double-helix model for DNA. Traditional epidemiologic study designs that were developed and used around the time of Hill’s speech treated the connection between exposure and disease as a ‘black box’—meaning that the biological mechanisms that occur between exposure and disease onset were unknown and therefore omitted in study design [2]. Over the past 50 years, advances in scientific fields (e.g., molecular genetics, genomics, molecular toxicology) and technology (e.g., computers, software, statistics, analytical methods) have provided researchers with a much deeper and more complex understanding of how diseases initiate and progress, effectively allowing researchers to glimpse into the ‘black box’ of the exposure-to-disease paradigm. As a result, researchers considering causal inference have new and more diverse types of information to consider when establishing causality beyond the traditional epidemiologic study designs that were available when Hill wrote his causal criteria. Data integration refers to the incorporation of data, knowledge, or reasoning from across multiple disciplines or approaches, with the goal of generating a level of understanding or knowledge that no discipline achieved alone [3, 4]. Data integration, while not always referred to by that term, has been discussed in light of causal inference of disease for over a decade, and the epidemiologic community has generally welcomed these interdisciplinary collaborations [5–7]. For example, the preface of the 5th edition of the Dictionary of Epidemiology directly acknowledges the “positive blurring of the boundaries of epidemiological research methods” into other scientific disciplines. The preface welcomes non-epidemiologists to contribute to and use the Dictionary and inversely invites trained epidemiologists to utilize the concepts within the Dictionary in non-epidemiological initiatives [4]. Furthermore, numerous agencies, organizations, and academics have recently attempted to establish frameworks or guidelines for data integration in the field of human health and ecological risk assessment. These frameworks consider how researchers should address, compare, and contrast the value and contributions of data that come from different evidence streams or scientific disciplines [8–11]. Hill aptly stated at the end of his speech that “[a]ll scientific work is incomplete… [and] liable to be upset or modified by advancing knowledge” [1]. Today, researchers considering causal inference must integrate data from a variety of scientific disciplines. Herein, we discuss how data integration in the field of causal inference of diseases affects the application and interpretation of each of Hill’s criteria. CRITERIA 1: STRENGTH OF ASSOCIATION Hill’s first criterion for causation is strength of the association. As he explained, the larger an association between exposure and disease, the more likely it is to be causal. To illustrate this point, Hill provided the classic example of Percival Pott’s examination of scrotal cancer incidence in chimney sweeps. The tremendous strength of association between that occupation and disease—nearly 200 times greater than seen in other occupations—led to a determination that the chimney soot was likely a causal factor. Contrarily, Hill suggested that small associations could more conceivably be attributed to other underlying contributors (i.e. bias or confounding) and, therefore, are less indicative of causation. Defining what constitutes a “strong” association is critical to the assessment of potentially causal relationships. Advances in statistical theory and the computational processing power have allowed scientists to delineate strong versus weak associations using more defensible mathematical criteria than Hill had in mind. Strength is no longer interpreted as simply the magnitude of an association. Furthermore, researchers have gained a greater appreciation for multi-factoral diseases and the existence of determinant risk factors that are small in magnitude yet statistically strong. Today, statistical significance—not the magnitude of association—is the accepted benchmark for judging the strength of an observed association, and thus its potential causality. Yet, these same statistical and computational advances necessitate an added degree of scrutiny when interpreting study results. Modern tools have enabled researchers to collect much larger datasets, access wide ranges of metadata, employ complex algorithms, and choose from a multitude of statistical approaches. As such, statistically significant results presented within a study are not always biologically meaningful or methodologically appropriate for contributing to causal inference. Conversely, failure to mathematically demonstrate statistical significance in a single study does not preclude the possibility of a meaningful exposure–response relationship in reality. Thus, assessing strength of association in causal inference requires examination of underlying methods, comparison to the weight of evidence in the literature, and consideration of other contextual factors including the other criteria discussed herein. An example can be seen in the analysis and subsequent re-analysis of pulmonary function in a cohort of 106 workers at a flavorings manufacturing facility that used a variety of chemicals, including acetaldehyde, acetoin, benzaldehyde,

Read the complete text