Quantitative Bias Analysis in Dental Research

Data used in epidemiology usually are either from experiments (randomised controlled trials) or observational/surveillance data. These data are used to estimate a parameter of interest. It must be kept in mind that this is a predicted value of the true parameter that exists out there in the real world, which can be either unknown or at times known. If the predicted parameter is the same as the true parameter of interest, then we say that the estimate is unbiased. However this is never a possibility without making assumptions about both the data collected and the modelling approach used. It is for the reason that the data will have both systematic and random errors. Apart from this, we also do not know the data-generating mechanism. The systematic error includes aspects such as selection bias, measurement error, confounding bias and unmeasured confounding. In a perfect randomised control trial, we can attempt to remove the biases such as measured and unmeasured confounding. Unfortunately this is not the case with observational data; hence it is important to describe how these errors are handled so that one can place more confidence in the estimate. Three important questions to ask before conducting quantitative bias analysis are as follows: (1) when should we conduct this, (2) how do we select which bias to address and (3) how do we select a method to model bias? The other question about conducting the bias analysis is how do we interpret and present these results. The aim of this chapter is to answer these questions, as well as provide an example and illustrate the software that can be used for conducting quantitative bias analysis.