Using a Phenomenological Approach to Forecast Tuberculosis Cases in the USA

Abstract

Background: Measurement errors produce bias and uncertainty. These are transferred to the outcomes of fitted models that describe an observed phenomenon. The models describing tuberculosis epidemics often include a large number of variables and parameters whose values are measured. In such cases, the width of the confidence intervals of the outcomes may be so large that the model renders useless. In contrast, straightforward phenomenological models can provide, theoretically, very precise outputs. Method: We use a phenomenological approach to describe and forecast TB incidence numbers in the US as a function of time. The tested models have only two variables whose values are measured. We use a standard method to measure the size of the observed variability and a heuristic method to evaluate the magnitude of the variability which cannot be explained as random variability. Findings: We find numerical regularities that provide trustworthy tools to make a prediction (provided there is no change in current trends) and forecast the tuberculosis incidence trends with great precision. However, after assessing CDC measurement errors -CDC surveillance reports do not reveal the size of measurement errors, we conclude that CDC measurement errors are probably larger than the magnitude of the unexplained variability and that, therefore, our predictions are provided with an overestimated precision. Interpretation: The uncertainty of collected epidemiological data seems to be typically large. This means that not only highly elaborate epidemic models can produce large uncertainties, but less elaborate models also have the risk of producing them. We propose to use epidemic models with caution. A general strategy of implementation could consist of first modeling of the data using a straightforward phenomenological approach and add variables and parameters one by one until the model explains all the variability unless it is well justified. Funding Statement: This study was not funded. Declaration of Interests: None.