How useful are the calculators for estimating the bias in odds ratio of epidemiological studies under varying levels of measurement error and sample sizes in the presence of classical measurement error in exposure?

Answer Explanations

• Not useful

  Expert 3
  In my field, odds ratios are not recommended because they are non-directional and tend to inflate the effect. The relative risk would be a more appropriate and reliable measure. 
• Somewhat useful

  Expert 4
  Somewhat useful because epidemiological studies usually have broad research questions and we would not estimate the sample size base on one single outcome. However it will still be useful to specify the minimal detectable effect under the planned sample size. 
• Somewhat useful

  Expert 6
  They would be far more useful if the OR varied more widely than 1.1 or 1.2.  Observational studies, no matter how well designed, cannot reliably identify causal effects of environmental exposures with such small increases in risk (odds).  Such studies usually aim to detect larger increases in risk, with OR in the 1.5-2.0 range.
• Extremely useful

  Expert 7
  Bias is often discussed but not estimated.  Ability to postulate its potential magnitude is very useful.
• Somewhat useful

  Expert 8
  More tricky, as associations with dichotomous outcomes can be either over- or underestimated with random error in exposure assessment, but definitely helpful!  
  
  Note here: odds ratios from logistic regression models are often analyzed and presented, but please consider other types of common outcomes and models, especially models taking into account repeated/longitudinal data or other aspects of time. E.g. survival models (cox models); for which effects of measurement errors in the exposure are somewhat similar to logistic models but a bit more complex with a time component. 
• I cannot tell

  Expert 5
  I'm not sure how useful the program is - I note that the highest OR in the calculator is 1.2.  This is a very low OR that one would typically see in an environmental epidemiology or nutrition study.  It would be helpful if the calculator allowed for a higher OR to be set.
  
  Also, it would be a good idea to have someone run additional simulations for the logistic regression analysis to confirm the unexpected finding of increased bias. Perhaps expand the discussion in this section with some referenced papers so we can more effectively evaluate this unexpected result.
• Somewhat useful

  Expert 2
   The calculators specifically address the challenge of measurement error, which is known to bias estimates like odds ratios in epidemiological studies. By simulating the effects of varying levels of measurement error, the tools help researchers understand how such errors might distort their results.  While the calculators are useful, as I mentioned above, they do have limitations, such as their focus on classical measurement error models and the exclusion of confounders or complex study designs. Despite these limitations, the calculators offer valuable insights into how measurement error can impact odds ratios, helping researchers design studies that mitigate bias.