Sampling bias minimization in disease frequency estimates

Oshrit Shtossel; Yoram Louzoun

doi:10.1016/j.jtbi.2021.110972

Sampling bias minimization in disease frequency estimates

J Theor Biol. 2022 Feb 7:534:110972. doi: 10.1016/j.jtbi.2021.110972. Epub 2021 Nov 29.

Authors

Oshrit Shtossel¹, Yoram Louzoun²

Affiliations

¹ Department of Mathematics, Bar-Ilan University, Ramat Gan 52900, Israel.
² Department of Mathematics, Bar-Ilan University, Ramat Gan 52900, Israel; Gonda Brain Research Center, Bar-Ilan University, Ramat Gan 52900, Israel. Electronic address: louzouy@math.biu.ac.il.

PMID: 34856201
DOI: 10.1016/j.jtbi.2021.110972

Abstract

An accurate estimate of the number of infected individuals in any disease is crucial. Current estimates are mainly based on the fraction of positive samples or the total number of positive samples. However, both methods are biased and sensitive to the sampling depth. We here propose an alternative method to use the attributes of each sample to estimate the change in the total number of positive patients in the total population. We present a Bayesian estimator assuming a combination of condition and time-dependent probability of being positive, and mixed implicit-explicit solution for the probability of a person with conditions i at time t of being positive. We use this estimate to predict the total probability of being positive at a given day t. We show that these estimate results are smooth and not sensitive to the properties of the samples. Moreover, these results are a better predictor of future mortality.

Keywords: Bayesian statistics; COVID-19; Data analysis; Mathematical modeling.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Bayes Theorem*
Bias
Forecasting
Humans
Probability
Selection Bias