How important would you consider the following practices for the analysis of epidemiology studies that rely on biomonitoring for exposure assessments? (rate from not at all important (1) to very important (5) on visual analogue scale)

Answer Explanations

• View answer

  Expert 3

  	1	2	3	4	5
  Adjust for bias due to measurement error in exposure	0	0	0	0	1
  Adjust for bias due to measurement error in covariates	0	0	0	0	1
  Adjust for bias due to measurement error in outcome	0	0	0	0	1
  Adjust for multiple comparisons	0	0	0	0	1
  Quantitatively account for unmeasured confounding	0	0	0	0	1
  Employ Bayesian methods that quantitatively integrate prior knowledge on the topic	0	0	0	0	1
  Other ______________(please specify)

  All of these practices are very important and should be standard in epidemiological studies. Measurement bias, covariate bias, and measurement error bias can significantly alter (e.g., attenuate or distort) the associations between exposure and outcome. Unmeasured confounding can bias results, and Bayesian methods can enhance the robustness of the assessment by integrating prior knowledge, thereby strengthening the analysis of the associations between outcome and predictor. 
• View answer

  Expert 2

  	1	2	3	4	5
  Adjust for bias due to measurement error in exposure	0	0	0	0	1
  Adjust for bias due to measurement error in covariates	0	0	0	1	0
  Adjust for bias due to measurement error in outcome	0	0	0	0	1
  Adjust for multiple comparisons	0	0	1	0	0
  Quantitatively account for unmeasured confounding	0	0	0	1	0
  Employ Bayesian methods that quantitatively integrate prior knowledge on the topic	0	0	0	1	0
  Other ______________(please specify)	0	0	1	0	0

  In the analysis of epidemiology studies using biomonitoring for exposure assessments, adjusting for bias due to measurement error in both exposure and outcome is critically important, as misclassification in these areas can significantly distort study results. Measurement error in covariates is also important to address, though its impact is generally less severe. Adjusting for multiple comparisons is necessary to prevent false positives, especially in studies with numerous tests. Quantitatively accounting for unmeasured confounding is crucial for producing valid causal inferences. While Bayesian methods that integrate prior knowledge can be beneficial, their importance varies depending on the study context and availability of reliable prior data. 
  It's also important to address several additional issues below:

  • Handling missing data properly through methods like multiple imputation helps prevent bias.
  • Selection bias must be considered to ensure the study population is representative. 
  • Non-linear relationships and interaction effects between exposures and covariates should be explored for a more accurate understanding of the data.
  • Accounting for multiple sources of exposure and performing sensitivity analyses to test the robustness of assumptions are also essential for producing reliable and comprehensive results. 
• View answer

  Expert 6

  	1	2	3	4	5
  Adjust for bias due to measurement error in exposure	0	0	0	1	0
  Adjust for bias due to measurement error in covariates	0	0	0	1	0
  Adjust for bias due to measurement error in outcome	0	0	0	1	0
  Adjust for multiple comparisons	0	0	0	1	0
  Quantitatively account for unmeasured confounding	0	0	1	0	0
  Employ Bayesian methods that quantitatively integrate prior knowledge on the topic	0	0	1	0	0
  Other ______________(please specify)	0	0	0	1	0

  Sensitivity analyses can be helpful to assess effects of metabolism, timing of sample, and other sources of within-person variability.
• View answer

  Expert 9

  	1	2	3	4	5
  Adjust for bias due to measurement error in exposure	0	0	0	0	1
  Adjust for bias due to measurement error in covariates	0	0	0	1	0
  Adjust for bias due to measurement error in outcome	0	0	0	0	1
  Adjust for multiple comparisons	0	0	1	0	0
  Quantitatively account for unmeasured confounding	0	0	1	0	0
  Employ Bayesian methods that quantitatively integrate prior knowledge on the topic	0	0	0	1	0
  Other ______________(please specify)	0	0	0	1	0

  Adjusting for bias due to measurement error in exposure and outcome seems to more directly address the primary study question(s), and probably more critical than measurement error in covariates.    
  Most of the time it is helpful to adjust for multiple comparisons.  But researchers may find it helpful to consider other aspects in evaluating measures of association besides just statistical significance.  For example, are there exciting new leads that just had too small a number of people exposed to be highly significant?  Are there consistent findings between related biomarkers that might indicate something important is going on, or that might support interpretation of a common pathway?
  Sensitivity analyses can be useful in interpretation, for example, by giving upper and lower bounds of likely results.
• View answer

  Expert 8

  	1	2	3	4	5
  Adjust for bias due to measurement error in exposure	0	0	0	0	1
  Adjust for bias due to measurement error in covariates	0	0	1	0	0
  Adjust for bias due to measurement error in outcome	0	0	1	0	0
  Adjust for multiple comparisons	0	0	0	1	0
  Quantitatively account for unmeasured confounding	0	0	0	1	0
  Employ Bayesian methods that quantitatively integrate prior knowledge on the topic	0	0	1	0	0
  Other ______________(please specify)

  Adjusting for bias in exposure assessment - where possible - is very important to reduce bias in effect estimates.  If non-differential, these errors typically result in attenuation bias and null findings where there may be true effects. 
  
  Adjusting for bias due to measurement errors in covariates and outcomes is also important but in my opinion to a lesser extent. For potential confounders there is of course a risk is residual confounding by poorly measured covariates. For outcomes, a main risk is reduced precision in effect estimates. But there are differences for binary versus continuous outcome variables. 
  
  Adjusting for multiple testing is important in any epidemiological study design to reduce the risk for false positive (chance) findings. Bonferroni correcation may be too strict/conservative, but other methods are available. 
• View answer

  Expert 5

  	1	2	3	4	5
  Adjust for bias due to measurement error in exposure	0	0	0	0	1
  Adjust for bias due to measurement error in covariates	0	0	0	1	0
  Adjust for bias due to measurement error in outcome	0	0	0	0	1
  Adjust for multiple comparisons	0	1	0	0	0
  Quantitatively account for unmeasured confounding	1	0	0	0	0
  Employ Bayesian methods that quantitatively integrate prior knowledge on the topic	0	0	1	0	0
  Other ______________(please specify)

  I think the adjustment for measurement error is always important regardless of whether the variable is an exposure, confounder, effect modifier or outcome.  However, we generally don't have parameter estimates for these associations (i.e. correlations, ICCs etc.).  More efforts in pilot studies and exposure validation studies would help to improve this situation, but generally these types of studies are difficult to get funded.   
  
  I'm not clear if you mean unmeasured or unknown confounding.  If unmeasured (i.e. a smoking variable is not measured in an occupational lung cancer study), it would be very important to account for a smoking confounder by using a proxy (such as SES) or other techniques.