Skip to main content
SpringerLink
Log in

G-Estimation of Structural Nested Models: Recent Applications in Two Subfields of Epidemiology

  • Epidemiologic Methods (D Westreich, Section Editor)
  • Published:
Current Epidemiology Reports Aims and scope Submit manuscript

Abstract

Correct adjustment for time-varying confounding affected by prior exposure is often not straightforward. G-estimation of structural nested models is a method of data analysis that allows for estimation of the combined effects of exposures that vary over time in a longitudinal cohort study. The method has not been widely adopted, but its use has increased in recent years, particularly in two subfields of epidemiology. Pharmacoepidemiologists have explored its applications to randomized trials with non-adherence or treatment switching and to finding optimal dynamic treatment regimens. Occupational epidemiologists have used it to correct for healthy worker survivor bias. Pharmacoepidemiologists have used simulations to illustrate extensions and novel applications, while occupational epidemiologists have focused on practical applications to observational data, often with careful attention to the handling of exposures. In theory, g-estimation of an appropriate structural nested model should be considered in the context of any longitudinal cohort in which at least one time-varying confounder is affected by prior exposure.

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

Similar content being viewed by others

References

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

  1. Robins JM. The analysis of randomized and non-randomized AIDS treatment trials using a new approach to causal inference in longitudinal studies. In: Sechrest L, Freeman H, Mulley A, editors. Health service research methodology: a focus on AIDS. Washington: U.S. Public Health Service, National Center for Health Services Research; 1989. p. 113–59.

    Google Scholar 

  2. Robins J. Estimation of the time-dependent accelerated failure time model in the presence of confounding factors. Biometrika. 1992;79(2):321–34.

    Article  Google Scholar 

  3. Robins J, Tsiatis AA. Semiparametric estimation of an accelerated failure time model with time-dependent covariates. Biometrika. 1992;79(2):311–9.

    Google Scholar 

  4. Witteman JC et al. G-estimation of causal effects: isolated systolic hypertension and cardiovascular death in the Framingham Heart Study. Am J Epidemiol. 1998;148(4):390–401.

    Article  CAS  PubMed  Google Scholar 

  5. Hernan MA et al. Structural accelerated failure time models for survival analysis in studies with time-varying treatments. Pharmacoepidemiol Drug Saf. 2005;14(7):477–91.

    Article  PubMed  Google Scholar 

  6. Robins JM. Testing and estimation of direct effects by reparameterizing directed acyclic graphs with structural nested models. In: Glymour C, Cooper G, editors. Computation, causation, and discovery. Menlo Park, CA, Cambridge, MA: AAAI Press/The MIT Press; 1999. p. 349–405.

    Google Scholar 

  7. Joffe MM, Greene T. Related causal frameworks for surrogate outcomes. Biometrics. 2009;65(2):530–8.

    Article  PubMed  Google Scholar 

  8. Vansteelandt S. Estimating direct effects in cohort and case-control studies. Epidemiology. 2009;20(6):851–60.

    Article  PubMed  Google Scholar 

  9. Vansteelandt S, Joffe M. Structural nested models and g-estimation: the partially realized promise. Stat Sci. 2014;4:707–31. A comprehensive, detailed, technical summary of g-estimation of structural nested models. The authors give a concise, readable explanation of the reasons to choose it over other methods, including a description of ways the assumption of no unmeasured confounders can be relaxed without compromising the consistency of the g-estimator. They conclude with a call for statisticians to produce algorithms and software that might enable epidemiologists to make greater use of it.

  10. Robins JM, Hernán MA. Estimation of the causal effects of time-varying exposures. In: Fitzmaurice G et al., editors. Longitudinal data analysis. New York: Chapman & Hall/CRC; 2009. p. 553–99.

    Google Scholar 

  11. Cole SR, Frangakis CE. The consistency statement in causal inference: a definition or an assumption? Epidemiology. 2009;20(1):3–5.

    Article  PubMed  Google Scholar 

  12. VanderWeele TJ. Concerning the consistency assumption in causal inference. Epidemiology. 2009;20(6):880–3.

    Article  PubMed  Google Scholar 

  13. Pearl J. On the consistency rule in causal inference: axiom, definition, assumption, or theorem? Epidemiology. 2010;21(6):872–5.

    Article  PubMed  Google Scholar 

  14. Hernan MA, Robins JM. Estimating causal effects from epidemiological data. J Epidemiol Community Health. 2006;60(7):578–86.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Picciotto S et al. Structural nested cumulative failure time models to estimate the effects of interventions. J Am Stat Assoc. 2012;107(499):886–900.

    Article  Google Scholar 

  16. Robins JM et al. G-estimation of the effect of prophylaxis therapy for Pneumocystis carinii pneumonia on the survival of AIDS patients. Epidemiology. 1992;3(4):319–36.

    Article  CAS  PubMed  Google Scholar 

  17. Neophytou AM et al. A structural approach to address the healthy-worker survivor effect in occupational cohorts: an application in the trucking industry cohort. Occup Environ Med. 2014;71(6):442–7. Compares results from structural nested accelerated and cumulative failure time models in an examination of heart disease in relation to duration of employment as a long-haul trucker. This is the first application of any SNCFTM to an occupational cohort and a useful illustration of the importance of choosing the correct class of models for the rarity or commonness of the outcome.

  18. Joffe MM, Yang WP, Feldman H. G-estimation and artificial censoring: problems, challenges, and applications. Biometrics. 2012;68(1):275–86.

    Article  PubMed  Google Scholar 

  19. Tilling K, Sterne JA, Szklo M. Estimating the effect of cardiovascular risk factors on all-cause mortality and incidence of coronary heart disease using g-estimation: the atherosclerosis risk in communities study. Am J Epidemiol. 2002;155(8):710–8.

    Article  PubMed  Google Scholar 

  20. Chevrier J, Picciotto S, Eisen EA. A comparison of standard methods with g-estimation of accelerated failure-time models to address the healthy-worker survivor effect: application in a cohort of autoworkers exposed to metalworking fluids. Epidemiology. 2012;23(2):212–9.

    Article  PubMed  Google Scholar 

  21. Picciotto S et al. Healthy worker survivor bias: implications of truncating follow-up at employment termination. Occup Environ Med. 2013;70(10):736–42.

    Article  PubMed  Google Scholar 

  22. Latimer NR et al. Adjusting for treatment switching in randomised controlled trials—a simulation study and a simplified two-stage method. Stat Methods Med Res. 2014

  23. Wu Y et al. Correcting for non-compliance in randomized non-inferiority trials with active and placebo control using structural models. Stat Med. 2015;34(6):950–65.

    Article  PubMed  Google Scholar 

  24. Naimi AI et al. Estimating the effect of cumulative occupational asbestos exposure on time to lung cancer mortality: using structural nested failure-time models to account for healthy-worker survivor bias. Epidemiology. 2014;25(2):246–54. Compares results from a fully parametric accelerated failure time model and from g-estimation of a SNAFTM in a study of cumulative exposure to asbestos and lung cancer mortality in a cohort of textile factory workers. Estimates were stronger using g-estimation, suggesting that healthy worker survivor bias was present in the estimate when g-estimation was not used. This is the first occupational epidemiology application of the SNAFTM to use a quantitative exposure variable.

  25. Picciotto S et al. Hypothetical interventions to limit metalworking fluid exposures and their effects on COPD mortality: g-estimation within a public health framework. Epidemiology. 2014;25(3):436–43. Introduces a public health framework for indirect consideration of quantitative exposure via separate analyses of binary exposure variables defined by a series of cutoffs. The authors use a SNAFTM with several possible exposure limits for each of three classes of metalworking fluids to analyze the potential number of years of life lost due to COPD that could have been saved if exposures were held below each of the hypothetical limits.

  26. Kubo Y et al. Assessing the treatment effect in a randomized controlled trial with extensive non-adherence: the EVOLVE trial. Pharm Stat. 2015;14(3):242–51.

    Article  PubMed  Google Scholar 

  27. Björ O et al. A comparison between standard methods and structural nested modelling when bias from a healthy worker survivor effect is suspected: an iron-ore mining cohort study. Occup Environ Med. 2015;72(7):536–42. Compares results from SNAFTM and traditional Cox regression of the relationship between mortality and cancer incidence among Swedish iron-ore miners exposed to respirable dust. The authors also consider a composite outcome consisting of mortality after censoring deaths from causes already known to be related to respirable dust in this cohort. This idea helps avoid studying a rare outcome using a model ill-suited for that purpose, while still examining the possibility that exposure causes additional diseases not known to be related.

  28. Costello S et al. Social disparities in heart disease risk and survivor bias among autoworkers: an examination based on survival models and g-estimation. Occup Environ Med. 2015;72(2):138–44. Uses SNAFTM and traditional Cox regression to analyze the relationship between water-based metalworking fluids and heart disease in different strata of race and sex within a cohort of automobile manufacturing workers. The authors raise the question of whether social disparities lead not only to disparities in health, but also to differential presence of healthy worker survivor bias in different strata.

  29. Picciotto S, Peters A, Eisen EA. Hypothetical exposure limits for oil-based metalworking fluids and cardiovascular mortality in a cohort of autoworkers: structural accelerated failure time models in a public health framework. Am J Epidemiol. 2015;181(8):563–70.

    Article  PubMed  PubMed Central  Google Scholar 

  30. Picciotto S, Ljungman PL, Eisen EA. Straight metalworking fluids and all-cause and cardiovascular mortality analyzed by using g-estimation of an accelerated failure time model with quantitative exposure: methods and interpretations. Am J Epidemiol. 2016;183(7):680–8. Compares results from SNAFTMs for all-cause and ischemic heart disease mortality using a binary annual exposure variable vs. using quantitative exposure to straight metalworking fluids in each year. This is the first occupational epidemiology application of the SNAFTM to consider quantitative exposure without summing these exposures over time to create cumulative exposure; the way the model works, the entire history of exposure is taken into account, even though cumulative exposure (the sum of past history of exposure) is not analyzed.

  31. Wallace MP, Moodie EE, Stephens DA. Model assessment in dynamic treatment regimen estimation via double robustness. Biometrics. 2016. Proposes a diagnostic approach for model validation using double robustness when estimating an optimal dynamic treatment regimen via SNMM. The principle underlying the approach is that, assuming the blip function is correctly specified, only one of the two “nuisance” models need be correct to produce a consistent estimator. Thus, varying the treatment-free model and the treatment model and comparing standard deviations of bootstrap estimates can help identify the correct models (and thus the correct estimates). The authors illustrate the technique using a simulation and then apply it to an observational study of HIV-1-infected men to estimate the optimal time to instigate zidovudine treatment in order to maximize CD4 count one year after baseline.

  32. Sung M, Erkanli A, Costello EJ. Estimating the causal effect of conduct disorder on the time from first substance use to substance use disorders using g-estimation. Subst Abus. 2014;35(2):141–6.

    Article  PubMed  PubMed Central  Google Scholar 

  33. Rich B, Moodie EE, Stephens DA. Influence re-weighted g-estimation. Int J Biostat. 2015

  34. He J, Stephens-Shields A, Joffe M. Structural nested mean models to estimate the effects of time-varying treatments on clustered outcomes. Int J Biostat. 2015;11(2):203–22. Develops a method to use SNMMs to adjust for confounding even when outcomes are clustered (i.e., when the stable unit treatment value assumption is not met within clusters), presents a simulation to evaluate the method, and then applies it to data from a randomized trial evaluating topical ocular hypertensive medication for prevention of glaucoma.

  35. Rich B, Moodie EE, Stephens DA. Simulating sequential multiple assignment randomized trials to generate optimal personalized warfarin dosing strategies. Clin Trials. 2014;11(4):435–44.

    Article  PubMed  Google Scholar 

  36. Rich B, Moodie EE, Stephens DA. Optimal individualized dosing strategies: a pharmacologic approach to developing dynamic treatment regimens for continuous-valued treatments. Biom J. 2015

  37. Shinozaki T, Matsuyama Y, Ohashi Y. Estimation of controlled direct effects in time-varying treatments using structural nested mean models: application to a primary prevention trial for coronary events with pravastatin. Stat Med. 2014;33(18):3214–28. Develops a method for mediation analysis using SNMMs. The proposed technique requires modeling both treatment and intermediate and then performing g-estimation sequentially, using the first g-estimate (of the effect of the intermediate) in the second g-estimation in order to estimate the controlled direct effect of time-varying treatment. The authors illustrate the method using data from a randomized trial to evaluate the controlled direct effects of pravastatin on CHD occurrence within 5 years, controlling cholesterol levels at a fixed value, with adjustment for treatment switching. An appendix extends the method to SNCFTMs.

  38. Checkoway H, Pearce N, Kriebel D. Research methods in occupational epidemiology. In: Monographs in epidemiology and biostatistics v 34. New York ; Oxford: Oxford University Press; 2004. p. xiv. 372 p.

    Google Scholar 

  39. Eisen EA, Picciotto S, Robins JM. Healthy worker effect in occupational studies. In: El-Shaarawi AH, Piegorsch WW, editors. Encyclopedia of environmetrics, second edition. Chichester: John Wiley & Sons; 2012. p. 1269–72.

    Google Scholar 

  40. Buckley JP et al. Evolving methods for inference in the presence of healthy worker survivor bias. Epidemiology. 2015;26(2):204–12.

    Article  PubMed  Google Scholar 

  41. Picciotto S, Hertz-Picciotto I. Commentary: healthy worker survivor bias: a still-evolving concept. Epidemiology. 2015;26(2):213–5.

    Article  PubMed  Google Scholar 

  42. Keil AP, Richardson DB, Troester MA. Healthy worker survivor bias in the Colorado Plateau uranium miners cohort. Am J Epidemiol. 2015;181(10):762–70. Proposes and applies a linear SNAFTM to examine the relationship between cumulative exposure to radon and lung cancer mortality in uranium miners. The authors also analyze separate time windows in a single multi-parameter linear SNAFTM; this cannot be done correctly in a traditional regression. This is also the first occupational epidemiology application to vary the form of the model in an SNAFTM.

  43. Westreich D, Greenland S. The table 2 fallacy: presenting and interpreting confounder and modifier coefficients. Am J Epidemiol. 2013;177(4):292–8.

    Article  PubMed  PubMed Central  Google Scholar 

  44. Richardson DB, Cole SR, Langholz B. Regression models for the effects of exposure rate and cumulative exposure. Epidemiology. 2012;23(6):892–9.

    Article  PubMed  Google Scholar 

  45. Hubbard AE, Laan MJ. Population intervention models in causal inference. Biometrika. 2008;95(1):35–47.

    Article  PubMed  PubMed Central  Google Scholar 

  46. Naimi AI, Tchetgen Tchetgen EJ. Invited commentary: estimating population impact in the presence of competing events. Am J Epidemiol. 2015;181(8):571–4.

    Article  PubMed  PubMed Central  Google Scholar 

  47. Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology. 1999;10(1):37–48.

    Article  CAS  PubMed  Google Scholar 

  48. Greenland S. Quantifying biases in causal models: classical confounding vs collider-stratification bias. Epidemiology. 2003;14(3):300–6.

    PubMed  Google Scholar 

  49. Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004;15(5):615–25.

    Article  PubMed  Google Scholar 

Download references

Author information

Authors and Affiliations

  1. UC Berkeley School of Public Health, Environmental Health Sciences, 50 University Hall #7360, Berkeley, CA, 94720-7360, USA

    Sally Picciotto & Andreas M. Neophytou

Authors
  1. Sally Picciotto
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Andreas M. Neophytou
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Sally Picciotto.

Ethics declarations

Conflict of Interest

The authors declare that they have no conflict of interest.

Human Rights and Informed Consent

This article does not contain any studies with animals performed by any of the authors. All studies by S. Picciotto and/or A.M. Neophytou involving human subjects were performed after approval by the appropriate institutional review boards. When required, written informed consent was obtained from all participants.

Additional information

This article is part of the Topical Collection on Epidemiologic Methods

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Picciotto, S., Neophytou, A.M. G-Estimation of Structural Nested Models: Recent Applications in Two Subfields of Epidemiology. Curr Epidemiol Rep 3, 242–251 (2016). https://doi.org/10.1007/s40471-016-0081-9

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s40471-016-0081-9

Keywords

Search

Navigation