How important would you consider the following practices for the design 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
  Conduct a pilot study to estimate variance components of the biomarker of exposure (when not already known)	0	0	0	0	1
  Identify a hypothesized minimal effect size (in the outcomes measure) that the researcher intends to be able to identify	0	0	0	0	1
  Estimate the impact on bias and power of measurement error in outcome, i.e. within-person variability in outcome	0	0	0	0	1
  For a given exposure of interest, select biomarkers with the largest intraclass correlation coefficient (ICC) (e.g., select biomarkers with the longest half-life)	0	0	0	0	1
  Consider study designs that do not rely only on contrasts in exposures among individuals but instead seek groups of individuals with very different exposures	0	0	0	0	1
  Articulate the hypothesized causal diagram (DAG) that includes variability in biomarkers of exposure between- and within-persons	0	0	0	0	1
  Calculate the efficiency, in terms of costs, of investing into more measurements within person versus more persons to minimize bias (as was done by Armstrong (1996))	0	1	0	0	0
  Utilize pools of biofluids from each individual to help minimize within-person variability and increase ICC	1	0	0	0	0
  Other ______________(please specify)	0	0	0	0	1

  Calculate the attenuation for different exposure strategies to identify and select the most effective exposure strategy. 
  
  Utilizing pooled biofluids from each individual to minimize within-person variability and increase the ICC sounds like a poor strategy, as important signals could be lost. Instead, the variability should be calculated and used in the assessment. An epidemiological study should be optimized for the strength of the association between the predictor and the outcome. Focusing on financial efficiency too early could lead to poor decisions. Whether more samples per person or more participants are needed should be based solely on variability. Once the optimal strategy is selected, cost efficiency can be calculated, but not beforehand. 
• View answer

  Expert 2

  	1	2	3	4	5
  Conduct a pilot study to estimate variance components of the biomarker of exposure (when not already known)	0	0	0	0	1
  Identify a hypothesized minimal effect size (in the outcomes measure) that the researcher intends to be able to identify	0	0	0	0	1
  Estimate the impact on bias and power of measurement error in outcome, i.e. within-person variability in outcome	0	0	0	0	1
  For a given exposure of interest, select biomarkers with the largest intraclass correlation coefficient (ICC) (e.g., select biomarkers with the longest half-life)	0	0	0	1	0
  Consider study designs that do not rely only on contrasts in exposures among individuals but instead seek groups of individuals with very different exposures	0	0	0	1	0
  Articulate the hypothesized causal diagram (DAG) that includes variability in biomarkers of exposure between- and within-persons	0	0	1	0	0
  Calculate the efficiency, in terms of costs, of investing into more measurements within person versus more persons to minimize bias (as was done by Armstrong (1996))	0	0	1	0	0
  Utilize pools of biofluids from each individual to help minimize within-person variability and increase ICC	0	1	0	0	0
  Other ______________(please specify)

  The most important practices in designing epidemiology studies that rely on biomonitoring for exposure assessments include conducting a pilot study to estimate variance components of the biomarker, identifying a hypothesized minimal effect size, and estimating the impact of measurement error, as these directly influence the study's reliability, power, and ability to detect meaningful associations. Selecting biomarkers with high intraclass correlation coefficients (ICC) and considering study designs that focus on groups with highly contrasting exposures are also important for reducing variability and increasing the robustness of findings. Articulating a hypothesized causal diagram (DAG) is useful for clarifying assumptions and identifying confounders. Calculating the efficiency of additional within-person versus more-person measurements is important for cost-effective study design, while utilizing pooled biofluids, though beneficial for reducing variability, is less universally applicable but still useful in certain contexts.
  
  
• View answer

  Expert 6

  	1	2	3	4	5
  Conduct a pilot study to estimate variance components of the biomarker of exposure (when not already known)	0	0	0	0	1
  Identify a hypothesized minimal effect size (in the outcomes measure) that the researcher intends to be able to identify	0	0	0	0	1
  Estimate the impact on bias and power of measurement error in outcome, i.e. within-person variability in outcome	0	0	0	1	0
  For a given exposure of interest, select biomarkers with the largest intraclass correlation coefficient (ICC) (e.g., select biomarkers with the longest half-life)	0	0	0	1	0
  Consider study designs that do not rely only on contrasts in exposures among individuals but instead seek groups of individuals with very different exposures	0	0	0	1	0
  Articulate the hypothesized causal diagram (DAG) that includes variability in biomarkers of exposure between- and within-persons	0	1	0	0	0
  Calculate the efficiency, in terms of costs, of investing into more measurements within person versus more persons to minimize bias (as was done by Armstrong (1996))	0	0	0	1	0
  Utilize pools of biofluids from each individual to help minimize within-person variability and increase ICC	0	0	0	1	0
  Other ______________(please specify)	0	0	0	0	1

  Using biosamples that reflect long-term exposure, such as urine, hair, or nails can provide time-averaged exposures that can either replace or supplement single-sample blood or saliva samples.  Another strategy is to include standardized timing of sampling, if single.
• View answer

  Expert 9

  	1	2	3	4	5
  Conduct a pilot study to estimate variance components of the biomarker of exposure (when not already known)	0	0	0	1	0
  Identify a hypothesized minimal effect size (in the outcomes measure) that the researcher intends to be able to identify	0	0	0	0	1
  Estimate the impact on bias and power of measurement error in outcome, i.e. within-person variability in outcome	0	0	0	1	0
  For a given exposure of interest, select biomarkers with the largest intraclass correlation coefficient (ICC) (e.g., select biomarkers with the longest half-life)	0	0	0	1	0
  Consider study designs that do not rely only on contrasts in exposures among individuals but instead seek groups of individuals with very different exposures	0	0	0	1	0
  Articulate the hypothesized causal diagram (DAG) that includes variability in biomarkers of exposure between- and within-persons	0	0	1	0	0
  Calculate the efficiency, in terms of costs, of investing into more measurements within person versus more persons to minimize bias (as was done by Armstrong (1996))	0	0	0	1	0
  Utilize pools of biofluids from each individual to help minimize within-person variability and increase ICC	0	0	1	0	0
  Other ______________(please specify)	0	0	0	0	1

  A pilot study is recommended if it is feasible to conduct.  Not only will it give an indication of between- and within-person variability, but will also help train staff in sample collection and analysis (particularly if they have little experience with the samples involved).    
  It's helpful to select biomarkers with the largest ICC, but balanced with considerations if those biomarkers are relevant to the study question and if they are practical to collect.  For example, blood samples might have the largest ICC, but it might be more practical to collect saliva or urine samples.    
  Groups of individuals with very different exposures can be helpful when individual measurements are less feasible.  However, (a) the researcher must be aware that there may be variability between individuals within the group that will be hidden, and (b) selection bias and representativeness of groups for the study question should be considered.    
  DAGs can be a useful thought experiment.  However, sometimes I wonder if they might give us a false sense of security; do we truly know that level of detail regarding how all the  variables relate to and interact with each other?
  As well as the statistical considerations focused on in this project, the biomonitored samples must be valid measures of the exposure of interest; they must actually measure what the researcher wants them to.  This includes considering biological aspects such as physiology and toxicology, and practical aspects like timing of collection, ease of collection, and likelihood of contamination.   
• View answer

  Expert 8

  	1	2	3	4	5
  Conduct a pilot study to estimate variance components of the biomarker of exposure (when not already known)	0	0	0	0	1
  Identify a hypothesized minimal effect size (in the outcomes measure) that the researcher intends to be able to identify	0	0	0	0	1
  Estimate the impact on bias and power of measurement error in outcome, i.e. within-person variability in outcome	0	0	0	1	0
  For a given exposure of interest, select biomarkers with the largest intraclass correlation coefficient (ICC) (e.g., select biomarkers with the longest half-life)	0	0	0	1	0
  Consider study designs that do not rely only on contrasts in exposures among individuals but instead seek groups of individuals with very different exposures	0	0	0	1	0
  Articulate the hypothesized causal diagram (DAG) that includes variability in biomarkers of exposure between- and within-persons	0	0	0	0	1
  Calculate the efficiency, in terms of costs, of investing into more measurements within person versus more persons to minimize bias (as was done by Armstrong (1996))	0	0	0	1	0
  Utilize pools of biofluids from each individual to help minimize within-person variability and increase ICC	0	0	1	0	0
  Other ______________(please specify)

  I think it's crucial to know the variability in biomarkers, so when not known from the literature, pilot studies are very important. Similarly, determining the minimal effect size is also very important. Both are essential for power calculations and determining appropriate sample sizes and sampling strategies - to avoid underpowered studies and/or a waste of resources of participant burden. 
  
  Bias in outcome measurement is of course also important, for choosing better methods where possible, and/or to take into account in adjusting analyses and interpreting results. 
  
  I'm a big fan of DAGs for understanding and communicating complex relationships between many variables (exposures, outcomes, covariates) and help guide analyses and interpretation. 
  
  Regarding pools of biofluids, I'm not sure. This may reduce withinperson variability but could also obscure e.g. differences in time or other changes?
  
   
  
  
• View answer

  Expert 5

  	1	2	3	4	5
  Conduct a pilot study to estimate variance components of the biomarker of exposure (when not already known)	0	0	0	0	1
  Identify a hypothesized minimal effect size (in the outcomes measure) that the researcher intends to be able to identify	0	0	0	0	1
  Estimate the impact on bias and power of measurement error in outcome, i.e. within-person variability in outcome	0	0	0	0	1
  For a given exposure of interest, select biomarkers with the largest intraclass correlation coefficient (ICC) (e.g., select biomarkers with the longest half-life)	0	0	1	0	0
  Consider study designs that do not rely only on contrasts in exposures among individuals but instead seek groups of individuals with very different exposures	0	0	1	0	0
  Articulate the hypothesized causal diagram (DAG) that includes variability in biomarkers of exposure between- and within-persons	0	0	1	0	0
  Calculate the efficiency, in terms of costs, of investing into more measurements within person versus more persons to minimize bias (as was done by Armstrong (1996))	0	0	0	0	1
  Utilize pools of biofluids from each individual to help minimize within-person variability and increase ICC	0	1	0	0	0
  Other ______________(please specify)

  I would only recommend using pools of biofluids when it is necessary to combine samples because of laboratory analytical capabilities (i.e. they need a larger volume for appropriate limits of detection), or if multiple samples are collected from an individual and the resources ($) are not available to analyze the individual samples.  It's better to understand both the intra and inter-individual variability when the samples are available. 
  
  Please also note that biomarkers with the longest half life may not be those with the largest ICC.  They just reflect a subject's longer term exposure which is particularly useful for case-control studies.   However, if all subjects are similarity exposed to a chemical with a long half-life, their between-subject variation may be minimal as well as the within.
• View answer

  Expert 4

  	1	2	3	4	5
  Conduct a pilot study to estimate variance components of the biomarker of exposure (when not already known)	0	0	0	1	0
  Identify a hypothesized minimal effect size (in the outcomes measure) that the researcher intends to be able to identify	0	0	0	0	1
  Estimate the impact on bias and power of measurement error in outcome, i.e. within-person variability in outcome	0	0	0	1	0
  For a given exposure of interest, select biomarkers with the largest intraclass correlation coefficient (ICC) (e.g., select biomarkers with the longest half-life)	0	0	0	1	0
  Consider study designs that do not rely only on contrasts in exposures among individuals but instead seek groups of individuals with very different exposures	0	0	0	1	0
  Articulate the hypothesized causal diagram (DAG) that includes variability in biomarkers of exposure between- and within-persons	0	0	0	1	0
  Calculate the efficiency, in terms of costs, of investing into more measurements within person versus more persons to minimize bias (as was done by Armstrong (1996))	0	0	0	0	1
  Utilize pools of biofluids from each individual to help minimize within-person variability and increase ICC	0	0	0	1	0
  Other ______________(please specify)

  The idea of conducting a pilot study to estimate the variance components is interesting, but often involve a large cost as the variance components have a higher standard error than a mean / proportion.