Causality and Association: Two Terms to Use and Interpret Judiciously

Dear Editor, In medical research, the distinction between causality and association is crucial yet often misunderstood or misrepresented. While association refers to a statistical relationship between variables, causality implies that one variable directly influences another.[1] Misinterpreting these concepts can lead to flawed conclusions, affecting clinical decision-making and policy formulation. Table 1 presents a brief overview of the differences between causality and association. Observational studies, such as cross-sectional and cohort designs, often identify associations but cannot establish causality due to potential confounders (variables that influence both the exposure and outcome, making it difficult to identify the true effect) and biases. In contrast, experimental designs like randomized controlled trials (RCTs) provide stronger evidence of causality by minimizing these confounding effects.[2]Table 1: Difference between causality and association inference in biomedical studiesFor example, a study on causality may investigate whether regular physical exercise causes a reduction in blood pressure. This can be examined using an RCT, where participants are randomly assigned to either an exercise intervention group or a control group. After six months, a significant reduction in blood pressure in the intervention group suggests a causal relationship between exercise and lower blood pressure. In contrast, a study on association would explore whether exercise and blood pressure are merely related without implying causation. A cross-sectional study may collect data on individuals’ exercise habits and blood pressure, finding that those who exercise more tend to have lower blood pressure. However, since confounding factors like age, diet, and genetics are not controlled, the study only establishes an association, not causation. Sir Austin Bradford Hill proposed nine criteria to help assess whether an observed association is likely to be causal. These criteria can be grouped into two broad categories: those based on empirical evidence, such as data and observations, and those grounded in theoretical reasoning and scientific understanding, as outlined in Table 2. While not absolute rules, these criteria offer a valuable framework for researchers interpreting causality in studies.[3]Table 2: Bradford Hill’s criteria, meaning, and potential actions that help to assess the strength of evidence for a causal relationship[ 3 ]Among the nine, the “analogy” criterion is often excluded in practice because it is considered less scientifically robust. Despite their value, Hill’s criteria alone are not sufficient to establish causality. Real-world data often includes hidden or unmeasured factors, known as confounding variables, that can distort associations. Additional challenges, such as measurement errors or selection bias (in the patient recruitment phase), further complicate causal inference. Many complex diseases involve multiple interacting factors, making it difficult to identify a single clear cause. These complexities highlight the importance of rigorous study design and cautious interpretation. Although a perfectly conclusive study is rarely possible due to practical and ethical constraints, researchers should strive to approximate this ideal as closely as possible. While ongoing discussions continue around which criteria are most essential,[4,5] in my view, clinicians and readers should at least consider the core criteria shown in Figure 1 when designing or interpreting studies that aim to explore causal relationships.Figure 1: Three major characteristics to categorize a study reporting causalityOverstating causality from associative data can mislead public health recommendations, while underestimating strong associations may delay important interventions.[6] Thoughtful study design and judicious interpretation of statistical analyses are essential to uphold scientific integrity and ensure evidence-based medical practice. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest. Use of artificial intelligence (AI) The author declares the use of a generative artificial intelligence – DeepSeek-V3 for correction of grammar and language and NapkinAI for visualization.