Skip to main content
SpringerLink
Log in

A Review of Causal Inference Methods for Estimating the Effects of Exposure Change when Incident Exposure Is Unobservable

  • Published:
Current Epidemiology Reports Aims and scope Submit manuscript

Abstract

Purpose of Review

Research questions on exposure change and health outcomes are both relevant to clinical and policy decision making for public health. Causal inference methods can help investigators answer questions about exposure change when the first or incident exposure is unobserved or not well defined. This review aims to help researchers conceive of helpful causal research questions about exposure change and understand various statistical methods for answering these questions to promote wider adoption of causal inference methods in research on exposure change outside the field of pharmacoepidemiology.

Recent Findings

Epidemiologic studies examining exposure changes face challenges that can be addressed by causal inference methods, including target trial emulation. However, their application outside the field of pharmacoepidemiology is limited.

Summary

In this review, we (a) illustrate considerations in defining an exposure change and defining the total and joint effects of an exposure change, (b) provide practical guidance on trial emulation design and data set-up for statistical analysis, (c) demonstrate four statistical methods that can estimate total and/or joint effects (structural conditional mean models, time-dependent matching, inverse probability weighting, and the parametric g-formula), and (d) compare the advantages and limitations of these statistical methods.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3

Similar content being viewed by others

References

Papers of particular interest, published recently, have been highlighted as: • Of importance •• Of major importance

  1. Jefferis BJ, Sartini C, Lee I-M, et al. Adherence to physical activity guidelines in older adults, using objectively measured physical activity in a population-based study. BMC Public Health. 2014;14:382.

    Article  PubMed  PubMed Central  Google Scholar 

  2. •• Lund JL, Richardson DB, Stürmer T. The active comparator, new user study design in pharmacoepidemiology: historical foundations and contemporary application. Curr Epidemiol Rep. 2015;2:221–8. (This paper summarized three biases that could result from not properly defining a comparison group and a treatment initiation time in pharmacoepidemiology. Among the three biases, the "heathy adherer bias" due to ill-defined treatment initiation was extended to a non-pharmacoepidemiology context in the current paper. This paper also described the "active comparator, new user" design that can mitigate the three biases, some of the principles and practices can be applied to other observational studies in non-pharmacoepidemiology.)

    Article  PubMed  PubMed Central  Google Scholar 

  3. •• Brookhart MA. Counterpoint: the treatment decision design. Am J Epidemiol. 2015;182:840–5. (In a series of short papers discussing the benefits and challenges of only including participants whose first exposure were observed under study, this paper proposed an alternative approach that anchors the start of follow-up to the time when medical decisions are made instead of the first exposure. Though the paper focused on pharmacoepidemiology, the principle was extended to non-pharmacoepidemiology context in our paper.)

    Article  PubMed  PubMed Central  Google Scholar 

  4. •• Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183:758–64. (This paper introduced the general framework of target trial emulation for causal analysis of observational data. Although the concept of a target trial has been around for some time, the extension to time-varying treatments expressed in this paper enables a general framework for static and dynamic treatment strategies.)

    Article  PubMed  PubMed Central  Google Scholar 

  5. Hernán MA, Sauer BC, Hernández-Díaz S, Platt R, Shrier I. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016;79:70–5.

    Article  PubMed  PubMed Central  Google Scholar 

  6. Hernán MA, Robins JM. Causal inference: what if. Boca Raton: Chapman & Hall/CRC; 2020.

    Google Scholar 

  7. Wanis KN, Sarvet AL, Wen L, Block JP, Rifas-Shiman SL, Robins JM, Young JG. The role of grace periods in comparative effectiveness studies of different medications. 2022. https://doi.org/10.48550/arXiv.2212.11398.

  8. Emilsson L, García-Albéniz X, Logan RW, Caniglia EC, Kalager M, Hernán MA. Examining bias in studies of statin treatment and survival in patients with cancer. JAMA Oncol. 2018;4:63–70.

    Article  PubMed  Google Scholar 

  9. Cain LE, Robins JM, Lanoy E, Logan R, Costagliola D, Hernán MA. When to start treatment? A systematic approach to the comparison of dynamic regimes using observational data. Int J Biostat. 2010;6:18.

    Article  MathSciNet  PubMed  PubMed Central  Google Scholar 

  10. • Wen L, Young JG, Robins JM, Hernán MA. Parametric g-formula implementations for causal survival analyses. Biometrics. 2021;77:740–53. (This paper proposed a new iterative conditional expectation (ICE) estimator for both static and dynamic treatment strategies of the joint effects. The paper compared the difference between the noniteratrive conditional expectoration (NICE) vs. ICE approaches using simulation.)

    Article  MathSciNet  PubMed  Google Scholar 

  11. Bonvini M, McClean A, Branson Z, Kennedy EH. Incremental causal effects: an introduction and review. In: Handbook of Matching and Weighting Adjustments for Causal Inference. Chapman and Hall/CRC; 2023.

    Google Scholar 

  12. VanderWeele TJ. Controlled direct and mediated effects: definition, identification and bounds. Scand Stat Theory Appl. 2011;38:551–63.

    Article  MathSciNet  PubMed  PubMed Central  Google Scholar 

  13. •• Keogh RH, Daniel RM, VanderWeele TJ, Vansteelandt S. Analysis of longitudinal studies with repeated outcome measures: adjusting for time-dependent confounding using conventional methods. Am J Epidemiol. 2018;187:1085–92. (This paper introduced the sequential conditional mean model (SCMM) for the causal estimation of total effect of an exposure. Though not discussed in our paper, they also described using sequential conditional mean model to estimate the "long-term direct effect", the effect of an exposure directly on the outcome not via intermediate exposures. This paper also demonstrated the use of inverse probability weighting (IPW) for total effect. The paper is important because it demonstrates that standard regression methods can be used to emulate trials after careful study design.)

    Article  PubMed  Google Scholar 

  14. VanderWeele TJ. On the distinction between interaction and effect modification. Epidemiology. 2009;20:863–71.

    Article  PubMed  Google Scholar 

  15. VanderWeele TJ, Jackson JW, Li S. Causal inference and longitudinal data: a case study of religion and mental health. Soc Psychiatry Psychiatr Epidemiol. 2016;51:1457–66.

    Article  PubMed  Google Scholar 

  16. •• Thomas LE, Yang S, Wojdyla D, Schaubel DE. Matching with time-dependent treatments: a review and look forward. Stat Med. 2020;39:2350–70. (This paper reviewed the study design considerations, matching approaches, outcome analysis and interpretations of the time-dependent matching method. It showed how several seemingly disconnected statistical methods actually turned on a similar underlying framework, including that of trial emulation, time-dependent propensity scores, and the sequential Cox model.)

    Article  MathSciNet  PubMed  PubMed Central  Google Scholar 

  17. • Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000;11:550–60. (This paper introduced marginal structural models via inverse probability weighting (MSM via IPW) as a novel class of causal models to address time-varying confounders affected by prior exposure.)

    Article  CAS  PubMed  Google Scholar 

  18. Taubman SL, Robins JM, Mittleman MA, Hernán MA. Intervening on risk factors for coronary heart disease: an application of the parametric g-formula. Int J Epidemiol. 2009;38:1599–611.

    Article  PubMed  PubMed Central  Google Scholar 

  19. VanderWeele TJ, Hernán MA. Causal inference under multiple versions of treatment. J Causal Inference. 2013;1:1–20.

    Article  MathSciNet  PubMed  PubMed Central  Google Scholar 

  20. Bartlett JW, Morris TP, Stensrud MJ, Daniel RM, Vansteelandt SK, Burman C-F. The hazards of period specific and weighted hazard ratios. Stat Biopharm Res. 2020;12:518–9.

    Article  PubMed  PubMed Central  Google Scholar 

  21. Hernán MA. The hazards of hazard ratios. Epidemiology. 2010;21:13–5.

    Article  PubMed  PubMed Central  Google Scholar 

  22. Sjölander A, Greenland S. Ignoring the matching variables in cohort studies - when is it valid and why? Stat Med. 2013;32:4696–708.

    Article  MathSciNet  PubMed  Google Scholar 

  23. Austin PC, Small DS. The use of bootstrapping when using propensity-score matching without replacement: a simulation study. Stat Med. 2014;33:4306–19.

    Article  MathSciNet  PubMed  PubMed Central  Google Scholar 

  24. • Cole SR, Hernán MA. Constructing inverse probability weights for marginal structural models. Am J Epidemiol. 2008;168:656–64. (This paper outlined the key steps and assumptions (exchangeability, positivity, correct model specification) required to estimate causal joint effects using IPW via MSM in the presence of time-varying confounding. They noted the trade-offs when using this method between bias and precision and discuss how weight truncation and sensitivity analyses could help address the bias-variance trade-offs.)

    Article  PubMed  PubMed Central  Google Scholar 

  25. Austin PC, Stuart EA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med. 2015;34:3661–79.

    Article  MathSciNet  PubMed  PubMed Central  Google Scholar 

  26. Williamson EJ, Morley R, Lucas A, Carpenter JR. Variance estimation for stratified propensity score estimators. Stat Med. 2012;31:1617–32.

    Article  MathSciNet  CAS  PubMed  Google Scholar 

  27. Young JG, Logan RW, Robins JM, Hernán MA. Inverse probability weighted estimation of risk under representative interventions in observational studies. J Am Stat Assoc. 2019;114:938–47.

    Article  MathSciNet  CAS  PubMed  Google Scholar 

  28. Jackson JW. Diagnosing covariate balance across levels of right-censoring before and after application of inverse-probability-of-censoring weights. Am J Epidemiol. 2019;188:2213–21.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Gooley TA, Leisenring W, Crowley J, Storer BE. Estimation of failure probabilities in the presence of competing risks: new representations of old estimators. Stat Med. 1999;18:695–706.

    Article  CAS  PubMed  Google Scholar 

  30. • Keil AP, Edwards JK, Richardson DR, Naimi AI, Cole SR. The parametric G-formula for time-to-event data: towards intuition with a worked example. Epidemiology. 2014;25:889–97. (This tutorial paper demonstrated how the parametric g-formula using NICE estimator can help address time-varying confounding for the estimation of the joint effect that traditional regression models cannot address. They discussed the key assumptions for the parametric g-formula, including conditional exchangeability, treatment version irrelevance, and correct model specification. The authors included SAS code in an appendix so researchers can implement the method in their own work.)

    Article  PubMed  PubMed Central  Google Scholar 

  31. • Wang A, Nianogo RA, Arah OA. G-computation of average treatment effects on the treated and the untreated. BMC Med Res Methodol. 2017;17:3. (This paper demonstrated the procedure of implement parametric g-formula using the noniterative conditional expectation for a total effect.)

    Article  PubMed  PubMed Central  Google Scholar 

  32. Petersen ML. Applying a causal road map in settings with time-dependent confounding. Epidemiology. 2014;25:898–901.

    Article  PubMed  PubMed Central  Google Scholar 

  33. • Snowden JM, Rose S, Mortimer KM. Implementation of G-computation on a simulated data set: demonstration of a causal inference technique. Am J Epidemiol. 2011;173:731–8. (This is a tutorial paper that used simulation to demonstrate the difference between standard regression techniques and g-formula. The paper walked readers through the steps of implementing the ICE approach of g-formula on a simple simulated dataset and interpreting the results. Importantly, the paper explained the difference in interpretation between marginal versus conditional effects.)

    Article  PubMed  PubMed Central  Google Scholar 

  34. McGrath S, Lin V, Zhang Z, Petito LC, Logan RW, Hernán MA, Young JG. gfoRmula: an R package for estimating the effects of sustained treatment strategies via the parametric g-formula. Patterns (N Y). 2020;1: 100008.

    Article  CAS  PubMed  Google Scholar 

  35. Daniel RM, De Stavola BL, Cousens SN. Gformula: estimating causal effects in the presence of time-varying confounding or mediation using the g-computation formula. Stand Genomic Sci. 2011;11:479–517.

    Google Scholar 

  36. Logan RW, Young JG, Taubman S, et al. GFORMULA-SAS MACRO. 2022. https://github.com/CausalInference/GFORMULA-SAS.

  37. Lendle SD, Schwab J, Petersen ML, van der Laan MJ. ltmle: an R package implementing targeted minimum loss-based estimation for longitudinal data. J Stat Softw. 2017;81:1–21.

    Article  Google Scholar 

  38. Mohammad K, Hashemi-Nazari SS, Mansournia N, Mansournia M. Marginal versus conditional causal effects. J Biostat Epidemiol. 2015;1:121–8.

    Google Scholar 

  39. Sjölander A, Dahlqwist E, Zetterqvist J. A note on the noncollapsibility of rate differences and rate ratios. Epidemiology. 2016;27:356–9.

    Article  PubMed  Google Scholar 

  40. Bang H, Robins JM. Doubly robust estimation in missing data and causal inference models. Biometrics. 2005;61:962–73.

    Article  MathSciNet  PubMed  Google Scholar 

  41. Vansteelandt S, Keiding N. Invited commentary: G-computation–lost in translation? Am J Epidemiol. 2011;173:739–42.

    Article  PubMed  Google Scholar 

  42. Funk MJ, Westreich D, Wiesen C, Stürmer T, Brookhart MA, Davidian M. Doubly robust estimation of causal effects. Am J Epidemiol. 2011;173:761–7.

    Article  PubMed  PubMed Central  Google Scholar 

  43. Wen L, Hernán MA, Robins JM. Multiply robust estimators of causal effects for survival outcomes. Scand J Stat. 2022;49:1304–28.

    Article  MathSciNet  Google Scholar 

  44. Schomaker M, Luque-Fernandez MA, Leroy V, Davies MA. Using longitudinal targeted maximum likelihood estimation in complex settings with dynamic interventions. Stat Med. 2019;38:4888–911.

    Article  MathSciNet  CAS  PubMed  PubMed Central  Google Scholar 

  45. Wu X, Mealli F, Kioumourtzoglou MA, Dominici F, Braun D. Matching on generalized propensity scores with continuous exposures. J Am Stat Assoc. 2022;1–29. https://doi.org/10.1080/01621459.2022.2144737.

  46. •• Daniel RM, Cousens SN, De Stavola BL, Kenward MG, Sterne JAC. Methods for dealing with time-dependent confounding. Stat Med. 2013;32:1584–618. (This is a tutorial paper that explained the time-dependent confounding and demonstrated the implementation of g-formula and inverse probability weighting. They provided examples using nonparametric estimation before extending to parametric estimation, which are helpful to understand the basic ideas underlie the two methods. The paper also discussed the strength and limitations of g-formula and inverse probability weighting methods.)

    Article  MathSciNet  CAS  PubMed  Google Scholar 

  47. Robins JM, Wasserman LA. Estimation of effects of sequential treatments by reparameterizing directed acyclic graphs. In: arXiv.org. 2013. https://arxiv.org/abs/1302.1566v1. Accessed 10 Apr 2023

  48. McGrath S, Young JG, Hernán MA. Revisiting the g-null paradox. Epidemiology. 2022;33:114–20.

    Article  PubMed  PubMed Central  Google Scholar 

  49. Imbens G. The role of the propensity score in estimating dose-response functions. Biometrika. 2000;87:706–10.

    Article  MathSciNet  Google Scholar 

  50. Hirano K, Imbens GW. The propensity score with continuous treatments. In: Applied Bayesian modeling and causal inference from incomplete-data perspectives. John Wiley & Sons, Ltd, pp 73–84. 2004. https://doi.org/10.1002/0470090456.ch7.

  51. Naimi AI, Moodie EEM, Auger N, Kaufman JS. Constructing inverse probability weights for continuous exposures: a comparison of methods. Epidemiology. 2014;25:292–9.

    Article  PubMed  Google Scholar 

  52. Petersen ML, Deeks SG, Martin JN, van der Laan MJ. History-adjusted marginal structural models for estimating time-varying effect modification. Am J Epidemiol. 2007;166:985–93.

    Article  PubMed  Google Scholar 

  53. García-Albéniz X, Hsu J, Hernán MA. The value of explicitly emulating a target trial when using real world evidence: an application to colorectal cancer screening. Eur J Epidemiol. 2017;32:495–500.

    Article  PubMed  PubMed Central  Google Scholar 

  54. Keogh RH, Gran JM, Seaman SR, Davies G, Vansteelandt S. Causal inference in survival analysis using longitudinal observational data: sequential trials and marginal structural models. Stat Med. 2023;42:2191–225.

    Article  MathSciNet  PubMed  PubMed Central  Google Scholar 

  55. Lu B. Propensity score matching with time-dependent covariates. Biometrics. 2005;61:721–8.

    Article  MathSciNet  PubMed  Google Scholar 

  56. Pearce N, Vandenbroucke JP. Are target trial emulations the gold standard for observational studies? Epidemiology. 2023;34:614.

    Article  PubMed  Google Scholar 

Download references

Funding

J.W.J. was supported by a grant from the National Heart Lung and Blood Institute K01HL145320. E.D.D. was supported by the Cancer Care Quality Training Program at the Lineberger Comprehensive Cancer Center, University of North Carolina at Chapel Hill (grant T32CA116339) at the time of this work.

Author information

Authors and Affiliations

  1. Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA

    Fangyu Liu & John W. Jackson

  2. Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA

    Fangyu Liu & John W. Jackson

  3. Department of Epidemiology and Prevention, Division of Public Health Sciences, Wake Forest School of Medicine, Winston-Salem, NC, USA

    Emilie D. Duchesneau

  4. Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA

    Jennifer L. Lund

  5. Center for Drug Safety and Effectiveness, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA

    John W. Jackson

Authors
  1. Fangyu Liu
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Emilie D. Duchesneau
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Jennifer L. Lund
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. John W. Jackson
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

The content outline of this review was conceived jointly by all four authors. F.L. drafted the manuscript, tables, figures, and web appendix. E.D.D., J.L.L, and J.W.J reviewed, revised the manuscript, tables, figures, and web appendix, and approved the version for submission.

Corresponding author

Correspondence to Fangyu Liu.

Ethics declarations

Competing interests

F.L. has nothing to declare. E.D.D. was supported by the Cancer Care Quality Training Program at the Lineberger Comprehensive Cancer Center, University of North Carolina at Chapel Hill (grant T32CA116339) at the time of this work. E.D.D. previously received salary support from AbbVie Inc. for unrelated work. J.L.L. reports prior stock ownership that was sold approximately 24 months ago. J.W.J. was supported by a grant from the National Heart Lung and Blood Institute K01HL145320.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary file1 (DOCX 261 KB)

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Liu, F., Duchesneau, E.D., Lund, J.L. et al. A Review of Causal Inference Methods for Estimating the Effects of Exposure Change when Incident Exposure Is Unobservable. Curr Epidemiol Rep (2024). https://doi.org/10.1007/s40471-024-00343-5

Download citation

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1007/s40471-024-00343-5

Keywords

Search

Navigation