top of page
The Impact of Measurement Error in Epidemiology
Measurement error occurs when either health status or the determinants of health status are measured with tools that are not perfect. This is the norm in medicine, not the exception, and is especially true for the measurement of neglected diseases for which little investment has been made to develop good and affordable tests. As mentioned above, this is a major issue for the study and treatment of central nervous system infections, but also when it comes to measuring the role that infection in animals may play in human infection (or vice versa).
Our team used Bayesian statistical methods to adjust for measurement error and thus demonstrate that dogs appeared to play a role in the infection of humans with Schistosoma japonicum in the Philippines. We also showed that ignoring this type of error can mask an association between two infectious agents, both measured with non-perfect tests. These methods have also been used by our research group to obtain more correct estimates of the prevalence of various outcomes such as epilepsy, severe headaches, injection drug use, performance of family physicians in controlling cardiovascular disease, etc. We have demonstrated many times that measurement error of one or more variables can mask or falsely suggest important associations between variables.
Our team was the first to obtain estimates of the sensitivity and specificity of the Kato Katz method repeated once, twice or three times for the detection of three helminths without assuming the accuracy of any test. We were also the first to describe the limitations of screening for neurological diseases such as epilepsy and severe headaches in an African community.
bottom of page