In studies that measure or assess exposure for everyone, between- and within-persons variability in either exposure or biomarker of exposure play an important role in influencing power and bias. However, there is a competing approach (group-based exposure assessment, semi-ecological design) in which persons are grouped based on similar exposures and exposure typical of the group is assigned to all members of that group (e.g. based on residency, occupation). In this approach, between-group variability is involved in affecting power and bias. How important is it that researchers assess the power and bias of this group-based approach to help guide design of epidemiology studies that rely on biomarkers? (please explain)

Answer Explanations

• Very important

  Expert 3
  The statement is somewhat misleading. Similar exposure groups need to be validated using statistical tests, as grouping categories such as occupation alone are not always reliable predictors. The statement also underestimates the purpose of group-based exposure assessment, which is not a competing approach but rather an extension of exposure assessment strategies. The key idea behind grouping is that more samples/measurements can be combined (and assigned to a subject), increasing the sample size (N) and thereby reducing noise while increasing the signal-to-noise ratio (SNR), as described by the formula SNR=mean/(SD/sqrt(N)) ​, where N is the sample size and SD is the standard deviation of noise.
  
  

  Grouping strategies are effective when within-group variability is small and between-group variability is large (which creates more contrast). For example, a geometric standard deviation (GSD) of 2 (though some research suggests GSDs of up to 4 are acceptable) is typically considered low enough variance to justify grouping subjects. If the GSD is small, a grouping strategy can help reduce attenuation in the association between exposure and outcome.  In all strategies, the attenuation of the association needs to be calculated to compare strategies.
  lambda = vb/(vb+vw/n)
  where lambda = attenuation, vb = between-subject OR between-group vw =  within-subject OR within-group AND n = number of repeated measurements per subject or subjects per group.

  
  Therefore, both between- and within-group variance must be assessed for the exposure, the biomarker, and the exposure assessment strategies.
  
  

  Once this assessment is complete, the best exposure strategy can be used to calculate the study's power and bias, and to determine the appropriate sample size.
  
  

  
  
  
• Very important

  Expert 2
  It is very important for researchers to assess the power and bias of the group-based exposure approach in epidemiology studies, especially when individual-level variability is unavailable. In this approach, individuals are assigned a common exposure level, which simplifies assessment but presents significant challenges by overlooking individual variability. This can lead to exposure misclassification, resulting in biased or attenuated effect estimates. The success of the group-based approach relies on sufficient between-group variability; without well-differentiated groups, statistical power is reduced, making it harder to detect true associations. Additionally, ecological bias may arise, where group-level exposure does not accurately reflect individual outcomes. When using individual-level biomarker data, further complications can occur, as biomarkers reflect internal doses that may vary within groups and be missed by group-based assessments. Therefore, evaluating power and bias is critical, particularly in the absence of individual variability, to ensure the study design is robust and produces reliable, valid results. This assessment ensures researchers sufficiently account for variability in exposure, optimizing the ability to detect meaningful associations while minimizing bias.
• Important

  Expert 6
  This approach can be very useful if repeated exposure measurements in individual participants is infeasible and within-person variability in such measurements is known to be high.  Comparing outcomes in groups known to be at high environmental exposure (e.g., by virtue of occupation, residence, or accidental contamination) vs no or very low exposure can provide an efficient and unbiased design.  Dose-response relationships will not be possible with such a design, however, unless one or more intermediate exposure groups are also compared.
• Very important

  Expert 7
  It is important to evaluate potential bias in all studies (not just one that include biomarkers.
  I do not agree with a premise that exposure based on residency or occupation are nesseserily of semi-ecological design and/or that such studies are more prone to bias than biomonitoring studies.
  Biomonitoring studies rely on a different set of assumptions and can be biased as well.  For example, missing data can be selective and lead to bias, or timing of monitoring can lead to bias, or be not representative of exposures of interest for a variety of reasons. 
• Important

  Expert 1
  There is likely to be confounding factors between groups and these should be determined in as much detail as possible so that any potential bias can be accounted for in each group. When all things have  been considered and variation exists, it can be seen to be a true variation between groups. The assessment of potential confounding factors is complex and difficult and also limited by other factors such as sample size when conducting analysis. In reality it is not always possible to account for all bias factors in a study, but the study that attempts to do this will have the better study and outcomes.
• Important

  Expert 9
  This is important if the group-based approach is both relevant to the study question and valid, and if the groups can be reliably distinguished.  Also, sometimes it might be the most practical solution if individual-level data are not available or impractical to obtain.  However, the lack of data on individuals may lead to misclassification and bias if there is significant variability within the groups, so it would be helpful to assess the power and bias of this approach.  In searching for groups with sufficiently contrasting exposures, it could also be helpful to consider selection bias and representativeness of the resulting study population.
• Important

  Expert 8
  It's important to take into account as this may indeed affect power and thereby decisions on sample size. Group-level exposure = lower variability: this may either increase power by reducing noise, but also decrease it by reducing overall contrasts. 
  In terms of bias, there is a risk of ecological fallacy and of course exposure misclassification. Confounding factors may also be different, they won't all be similar at the same grouping level, it may be good to also measure and analyze these at an individual level. 
• Equivocal

  Expert 5
  I wouldn't necessarily call this a semi-ecological design.  In occupational studies, workers may be classified by occupation, industry, job title, tasks performed etc. but we recognize that their exposures within the group will be different.  We assume that the exposures between the groups will show more variability.  This is often not the case (and why we don't often find significant results), but I'm not aware of many studies that can quantify these relationships (variations within groups/between groups) expect for some early work done by Kromhout et al.  
  
  For some basic info see: Mannetje A', Kromhout H. The use of occupation and industry classifications in general population studies. Int J Epidemiol. 2003 Jun;32(3):419-28. doi: 10.1093/ije/dyg080. PMID: 12777430.
• Important

  Expert 4
  If group-based approach is used, the between-individual variability would be ignored.