A phenomenological approach to predicting tuberculosis cases with an assessment of measurement errors

Measurement errors produce bias and uncertainty. They affect the outcomes of data-fitted models. Unfortunately, tuberculosis incidence data providers do not appear to be interested in this topic.

We use a phenomenological approach to describe and forecast tuberculosis incidence numbers in the US, and England and Wales as a function of time. We use a heuristic method to evaluate if the lack of fit of our descriptive models to the data could be explained by measurement errors.

We find that if the lack of fit of our proposed models to the data is due to measurement errors, these are not so large as to make the models useless. We find numerical regularities that provide trustworthy tools to make a description and forecast the tuberculosis incidence trends.

Measurement errors of epidemiological data collected "in the field" are probably large and could justify the lack of fit of an epidemic model to the data. Thus, to choose an appropriate model to describe an epidemic, we propose to assess the magnitude of measurement errors and apply the epidemic theory.