Jump to comment:

• Re:Adjustment or over-adjustment for 'TB incidence in the country of birth'?
  Guy B. Marks
  Published on: 11 May 2012
• Adjustment or over-adjustment for 'TB incidence in the country of birth'?
  Sooyeon Min
  Published on: 28 March 2012

• Published on: 11 May 2012

  Re:Adjustment or over-adjustment for 'TB incidence in the country of birth'?

  • Guy B. Marks, Head
  • Other Contributors:
    • Claudia C. Dobler

  We thank the correspondents Min and Gyorkos for their comments on our paper, which focus on the appropriateness of adjustment for confounding in the analysis. We do not agree that the analysis suffers from over- adjustment for confounding. Over-adjustment occurs when a factor that is on the casual pathway between the main effect (exposure of interest) and the outcome is included as covariate. The decision about whether a fa...

  Show More

  We thank the correspondents Min and Gyorkos for their comments on our paper, which focus on the appropriateness of adjustment for confounding in the analysis. We do not agree that the analysis suffers from over- adjustment for confounding. Over-adjustment occurs when a factor that is on the casual pathway between the main effect (exposure of interest) and the outcome is included as covariate. The decision about whether a factor is on the causal pathway is based on the background knowledge and a theoretical model of the causal pathway [1]. In this case there is no theoretical basis for regarding country of birth as being on the causal pathway between diabetes and tuberculosis. Hence, there is no issue of over-adjustment here. There is a strong theoretical basis for believing that country of birth is related to the risk of tuberculosis (85% of cases of TB in Australia occur in foreign born persons). Hence, it is a strong potential confounder and even slight imbalance in the distribution of country of birth between those with and without diabetes would cause actual confounding. For this reason we chose to include it in the group of potential confounders that we adjusted for. As it turns out, the effect estimate (relative risk) was changed by adjusting for the covariates as a group (Table 3) and this is evidence that the potential confounders were actual confounders.

  The fact that other investigators have not adjusted for country of birth is not, as Min and Gyorkos claim, evidence that adjustment for this covariate is inappropriate. It almost certainly reflects the fact that they did not have the relevant information for cases or for non-cases and hence could not adjust for it. Adjustment of country of birth among the potential confounders is one of the methodological strengths of this paper.

  We used the most recent WHO estimates for TB incidence in country of birth. We acknowledge that this may not always be relevant to the incidence at the time the person emigrated from that country. However, it does allow approximate categorisation and provides some basis for adjusting for this confounder.

  Reference: 1. Hernan MA, Hernandez-Diaz S, Werler MM, Mitchell AA (2002) Causal knowledge as a prerequisite for confounding evaluation: an application to birth defects epidemiology. Am J Epidemiol 155: 176-184.

  Conflict of Interest:

  None declared

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top
• Published on: 28 March 2012

  Adjustment or over-adjustment for 'TB incidence in the country of birth'?

  • Sooyeon Min, MSc Student
  • Other Contributors:
    • Theresa Gyorkos

  The recent paper by Dobler et al. (2012) highlights the challenge of variable selection and adjustment in multivariate analyses [1]. The authors state that including the potentially confounding variable 'TB incidence in the country of birth' was one of the major strengths of their study. TB incidence in the country of birth is indeed an important predictor of the risk of developing TB - but it may not be a true confounde...

  Show More

  The recent paper by Dobler et al. (2012) highlights the challenge of variable selection and adjustment in multivariate analyses [1]. The authors state that including the potentially confounding variable 'TB incidence in the country of birth' was one of the major strengths of their study. TB incidence in the country of birth is indeed an important predictor of the risk of developing TB - but it may not be a true confounder.

  The general rule for defining a confounding variable in epidemiological research is the following: 1. The confounding variable is causally associated with the outcome. 2. The confounding variable is non- causally or causally associated with the exposure. 3. The confounding variable is not an intermediate variable in the causal pathway between exposure and outcome [2,3]. From the Dobler et al. paper, where diabetes is the exposure variable and TB is the outcome, 'TB incidence in the country of birth' is causally associated with the outcome, but it is not associated with the exposure. Adjustment for a strong predictor of an outcome that is not a confounder results in overadjustment [4], which distorts the true relationship between the exposure and outcome [3]. As the authors of the paper noticed, none of the previously published studies conducted in the US or Canada, countries with a large number of immigrants, have adjusted for TB incidence in the country of birth [5]. And this seems to be well justified due to the above reason.

  A further concern relates to the measurement of the 'TB incidence in the country of birth' variable. For example, was current or past TB incidence used? What is 'current' and what is 'past'? Was TB incidence obtained from the region of birth, or was it a national incidence estimate? Was TB incidence available for all countries of origin? As TB incidence greatly varies according to the time period and geographical region - even within the same country - clarification on how this variable was measured is important. Also, if a single estimate was used for multiple individuals, the true heterogeneity in TB incidence rates experienced by these individuals will be missed, leading to measurement error [6]. This calls into question the validity of the measurement [7].

  These concerns warrant a re-analysis of the data.

  References: 1. Dobler CC, Flack JR, Marks GB. Risk of tuberculosis among people with diabetes mellitus: an Australian nationwide cohort study. BMJ Open 2012;2:e00066. doi:10.1136/bmjopen-2011-000666. 2. Rothman KJ. Epidemiology: An Introduction. New York: Oxford University Press; 2002. 3. Szklo M, Nieto J. Epidemiology: Beyond the Basics. 2nd ed. Massachusetts: Jones and Bartlett Publishers; 2006. 4. Porta M. A Dictionary of Epidemiology. 5th edition. New York: Oxford University Press; 2008. 5. Jeon CY, Murray MB. Diabetes mellitus increases the risk of active tuberculosis: a systematic review of 13 observational studies. PLoS Med 2008;5:e152 6. Diez Roux AV. The study of group-level factors in epidemiology: rethinking variables, study designs, and analytical approaches. Epidemiol Rev 2004;26:104-11 7. Schsterman EF, Cole SR, Platt RW. Overadjustment bias and unnecessary adjustment in epidemiologic studies. Epidemiology 2009;20(4):488-95

  Conflict of Interest:

  None declared

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top