Immunity score calibration based on vaccination rebalancingDownload PDF

Published: 02 Mar 2023, Last Modified: 25 Apr 20232023 ICLR - MLGH PosterReaders: Everyone
Keywords: Vaccination, Normalization, SARS-CoV-2, Optimization
TL;DR: ICLR 2023, MLGH Workshop - Balancing vaccination data allows better global health strategies
Abstract: Having fine-grained vaccination data enables authorities to discover and target deficiencies in vaccination endeavors more easily and focused, leading in the best case to better containment of outbreaks. In many countries there are two factors that hinder more detailed breakdowns of vaccination data: Firstly, vaccinations are registered solely by place of vaccination, instead of the actual residency of the vaccinated person. Additionally, significant proportions of the population in rural areas travel to neighboring, more urban counties, to receive vaccinations. For related vaccination efforts for SARS-CoV-2, these factors resulted in very distorted vaccination numbers, with some counties having theoretical vaccination rates exceeding 100%. Furthermore, a lack of vaccination registry which records exact days and vaccine types for each person decreases retractability. It is well established, though, that for SARS-CoV-2 the effectiveness of vaccinations decreases over time. Therefore, exact calculations of actual effectiveness depending on the day and type of the last vaccination are not possible, either. Our work presents a two-part approach to tackling this problem and delivering more reliable county-wide data, using Germany as an example. In the first step, we reverse the aforementioned effect of vaccination tourism using a flow-based linear programming model. In the second step, we approximate a vaccination registry per county using daily vaccination data of each county’s health authority. Combining recent insights in the effectiveness of SARS-CoV-2 vaccines regarding contagiousness over time, we calculate lower and upper bounds of the current effectiveness of vaccinations per county via linear programming and show that these are in fact close to each other.
0 Replies