Mediation Analysis in Ageing Research
Key Takeaways
- Mediation analysis asks whether part of an exposure-outcome association appears to operate through an intermediate variable, often called a mediator. [1] [2]
- In ageing research, possible mediators include inflammation, frailty, epigenetic age acceleration, brain structure, metabolism, and functional decline. [8] [9] [10]
- A mediation estimate is not automatically a mechanism; causal interpretation depends on timing, measurement quality, confounding control, and model assumptions. [2] [3] [4]
- Longitudinal ageing studies are especially challenging because exposures, mediators, confounders, and outcomes can all change over time. [5] [6]
Who This Is Useful For
This page is useful for readers interpreting papers that say a biomarker, behavior, disease state, or social exposure works "through" another ageing-related variable. Mediation analysis is common in studies of socioeconomic disadvantage, biological ageing markers, dementia, frailty, inflammation, and mortality, but the interpretation is often more conditional than the headline suggests. [2] [8] [9] [10]
Mediation analysis tries to separate an association into pathways. In a simple example, an exposure such as socioeconomic disadvantage may be associated with later disease, and part of that association may appear to pass through accelerated biological ageing or inflammation. The mediator is the intermediate variable placed between the exposure and the outcome in the proposed causal sequence. [1] [2] [8]
The word "mediate" can sound mechanistic, but the statistical result alone does not prove a biological pathway. A mediation model becomes more informative when the exposure occurs before the mediator, the mediator occurs before the outcome, important confounders are measured, and the analysis is aligned with a clear causal question. [2] [3] [4]
What Mediation Analysis Is Asking
The basic question is whether an exposure is associated with a mediator and whether that mediator is associated with the outcome after accounting for the exposure and relevant covariates. Traditional regression approaches framed this as a sequence of associations, while modern causal mediation methods define direct and indirect effects more explicitly using counterfactual reasoning. [1] [2] [3]
In ageing research, that framing is attractive because many hypotheses are pathway hypotheses: chronic inflammation may link multimorbidity to disability, frailty may sit between biological stress and mortality, or brain structure may partly explain the association between physical vulnerability and dementia. Each case requires a defensible timeline and a reason to treat the proposed mediator as intermediate rather than as a co-occurring marker. [8] [9] [10]
Common Terms
| Term | Meaning | Ageing Example | Interpretation Caution |
|---|---|---|---|
| Exposure | The variable whose association with an outcome is being studied | Social disadvantage, frailty status, diet quality, or baseline biomarker level | It must be ordered before the mediator for a pathway claim to be plausible |
| Mediator | An intermediate variable proposed to carry part of the association | Inflammatory markers, epigenetic age acceleration, metabolomic signature, or brain structure | It may be a marker of the process rather than the mechanism itself |
| Direct effect | The part of the exposure-outcome effect not operating through the specified mediator under the model | Residual association between disadvantage and disease after the biological-age pathway is separated | It is not necessarily a single biological pathway |
| Indirect effect | The part of the exposure-outcome effect attributed to the specified mediator under the model | Part of a frailty-mortality association attributed to a metabolic signature | It depends strongly on mediator measurement and unmeasured confounding assumptions |
Why It Appears Often in Ageing Research
Ageing research often studies long chains of events rather than single-step effects. Social conditions, health behaviors, chronic diseases, immune changes, molecular ageing markers, and functional outcomes can be linked across decades, so researchers use mediation analysis to ask which intermediate variables statistically account for part of an observed association. [6] [8] [9]
This can be useful for organizing evidence, but it also creates a risk of overinterpreting pathway diagrams. In older cohorts, many candidate mediators are intertwined with baseline health, disease burden, treatment history, survival, and measurement timing. Those features make simple mediation language less secure unless the study design handles temporal order and confounding carefully. [4] [5] [6]
Ageing-Relevant Examples
| Research Question | Possible Mediator | What the Analysis Can Suggest | What It Cannot Prove Alone |
|---|---|---|---|
| Why social disadvantage is linked to age-related disease | Age-related proteins or epigenetic measures | Part of the association may align with measurable biological ageing pathways | That social exposures act only through the measured biological markers |
| How frailty relates to mortality | Metabolomic signatures or epigenetic ageing markers | Some mortality association may be statistically attributed to systemic biological profiles | That changing the mediator would necessarily remove the mortality risk |
| How multimorbidity relates to functional limitation | Inflammation | Inflammatory burden may account for part of the multimorbidity-disability association | That inflammation is the only pathway or that timing is fully resolved |
| How physical vulnerability relates to dementia | Brain structure or immunometabolic markers | Neural or systemic markers may help explain part of the observed association | That the mediator is independent of earlier disease processes |
Why Timing Matters
A mediator should occur after the exposure and before the outcome in the proposed causal sequence. This is a major issue in cross-sectional biomarker studies because exposure, mediator, and outcome may be measured at the same visit, making it difficult to distinguish mediation from correlation or reverse causation. [2] [3] [6]
Longitudinal data help but do not remove the problem. In ageing cohorts, health status can affect later behavior, later behavior can affect biomarkers, biomarkers can affect disease risk, and disease can alter subsequent biomarkers. Methods for time-varying exposures, mediators, and confounders exist because a single baseline exposure and single follow-up mediator often oversimplify these feedback loops. [5] [6]
Direct and Indirect Effects Are Model-Based
Modern mediation analysis often estimates direct and indirect effects. A natural indirect effect describes the part of an effect attributed to changing the mediator along the path produced by the exposure, while a natural direct effect describes the remaining exposure effect under the modeled mediator setting. These quantities require assumptions about confounding and the absence of certain post-exposure confounding structures. [2] [3] [4]
This matters for ageing research because post-exposure confounding is common. For example, an exposure may affect disease burden, disease burden may affect both the mediator and later mortality, and the mediator may still be part of the pathway under study. Specialized approaches can sometimes address these structures, but ordinary regression adjustment may not be enough. [4] [5] [6]
Multiple Mediators
Ageing pathways rarely involve one mediator in isolation. Inflammation, metabolic state, vascular disease, epigenetic ageing, brain change, and frailty can occur in overlapping sequences, so a single-mediator model may assign too much explanatory weight to whichever marker was measured. Multiple-mediator methods try to separate pathways more explicitly, but they require even stronger design, measurement, and modeling choices. [7] [8] [9]
Interpreting a "percent mediated" estimate is therefore delicate. A high percentage does not mean the mediator is the whole mechanism, and a low percentage does not mean the pathway is biologically irrelevant. The estimate depends on which mediators were included, how they were measured, when they were measured, and which effect scale was used. [2] [7]
How to Read a Mediation Result
| Question | Why It Matters | Warning Sign |
|---|---|---|
| Was the mediator measured after the exposure? | Temporal order is needed for a pathway interpretation | The exposure, mediator, and outcome are all measured at the same time |
| Were mediator-outcome confounders measured? | Unmeasured confounding can create an apparent indirect effect | The paper adjusts for age and sex only despite complex health differences |
| Could the exposure affect later confounders? | Post-exposure confounding can make standard mediation models biased | Health status after exposure influences both the mediator and outcome |
| Is the mediator biologically specific? | Broad biomarkers may reflect many processes at once | A general ageing clock is described as one precise mechanism |
What This Does Not Mean
- It does not mean mediation analysis is invalid; it means the causal interpretation depends on assumptions that should be stated and examined. [2] [3]
- It does not mean a mediator must be molecular; social, behavioral, physiological, and clinical variables can all be candidate mediators if the timeline is coherent. [1] [8]
- It does not mean a significant indirect effect proves that intervening on the mediator would reproduce the same effect. [2] [4]
- It does not mean non-significant mediation rules out a pathway; measurement error, timing, power, and omitted mediators can all weaken the estimate. [2] [7]
Practical Interpretation Examples
- If epigenetic age mediates a diet-mortality association: read that as evidence that the measured clock may account for part of the association, not as proof that the clock is the only causal route. [2] [11]
- If inflammation mediates multimorbidity and disability: check whether inflammation was measured before functional limitation and whether baseline disease severity was handled. [10]
- If frailty is treated as a mediator: ask whether frailty is an intermediate state, a summary of existing disease, or partly an outcome already in progress. [9] [12]
- If a paper reports a percent mediated: look for the effect scale, confidence interval, model assumptions, and sensitivity analysis before treating the number as a biological fraction. [2] [3]
Related Reading
Summary
Mediation analysis is a useful way to examine possible pathways in ageing research, especially when investigators want to connect exposures with intermediate biological or functional changes and later outcomes. Its main value is conceptual clarity about proposed pathways, but its main risk is causal overstatement. The strongest readings come from checking temporal order, confounding assumptions, mediator specificity, longitudinal feedback, and whether the reported indirect effect matches the biological claim being made. [2] [3] [5] [7]
References
- Baron, R. M., & Kenny, D. A. (1986). Journal of Personality and Social Psychology. https://doi.org/10.1037/0022-3514.51.6.1173
- VanderWeele, T. J. (2016). Annual Review of Public Health. https://www.annualreviews.org/content/journals/10.1146/annurev-publhealth-032315-021402
- Imai, K., Keele, L., & Tingley, D. (2010). Psychological Methods. https://pubmed.ncbi.nlm.nih.gov/20954780/
- VanderWeele, T. J., Vansteelandt, S., & Robins, J. M. (2014). Epidemiology. https://pmc.ncbi.nlm.nih.gov/articles/PMC4214081/
- VanderWeele, T. J., & Tchetgen Tchetgen, E. J. (2017). Epidemiology. https://pmc.ncbi.nlm.nih.gov/articles/PMC5560424/
- Lin, S. H., Young, J., Logan, R., & VanderWeele, T. J. (2017). Statistics in Medicine. https://pmc.ncbi.nlm.nih.gov/articles/PMC6242332/
- VanderWeele, T. J., & Vansteelandt, S. (2014). Epidemiologic Methods. https://pmc.ncbi.nlm.nih.gov/articles/PMC4287269/
- Kivimaki, M., et al. (2025). Nature Medicine. https://pmc.ncbi.nlm.nih.gov/articles/PMC12092251/
- Zhang, X., et al. (2026). npj Aging. https://pmc.ncbi.nlm.nih.gov/articles/PMC12894700/
- Harries, L. W., et al. (2015). Age. https://pmc.ncbi.nlm.nih.gov/articles/PMC4600657/
- Jiang, R., et al. (2022). Clinical Epigenetics. https://clinicalepigeneticsjournal.biomedcentral.com/articles/10.1186/s13148-022-01380-x
- Ye, B., et al. (2023). Age and Ageing. https://www.ukbiobank.ac.uk/publications/the-role-of-lifestyle-in-the-association-between-frailty-and-all-cause-mortality-amongst-older-adults-a-mediation-analysis-in-the-uk-biobank/
This content is provided for educational purposes only and does not constitute medical advice.