シミュレーションモデルを通じたCOVID-19予測の理解

タイトル: Interpreting epidemiological surveillance data: A modelling study from Pune City

概要: Routine epidemiological surveillance data represents one of the most continuous and current sources of data during the course of an epidemic. This data is used to calibrate epidemiological forecasting models as well as for public health decision making such as imposition and lifting of lockdowns and quarantine measures. However, such data is generated during testing and contact tracing and not through randomized sampling. Furthermore, since the process of generating this data affects the epidemic trajectory itself - identification of infected persons might lead to them being quarantined, for instance - it is unclear how representative such data is of the actual epidemic itself. For example, will the observed rise in infections correspond well with the actual rise in infections? To answer such questions, we employ epidemiological simulations not to study the effectiveness of different public health strategies in controlling the spread of the epidemic, but to study the quality of the resulting surveillance data and derived metrics and their utility for decision making. Using the BharatSim simulation framework, we build an agent-based epidemiological model with a detailed representation of testing and contact tracing strategies based on those employed in Pune city during the COVID-19 pandemic. Infected persons are identified, quarantined and/or hospitalized based on these strategies, and to generate synthetic surveillance data as well. We perform extensive simulations to study the impact of different public health strategies and availability of tests and contact tracing efficiencies on the resulting surveillance data as well as on the course of the epidemic. The fidelity of the resulting surveillance data in representing the real-time state of the epidemic and in decision-making is explored in the context of Pune city. Author SummaryThrough this study, we evaluate the effectiveness of different public health metrics in guiding decision-making during epidemics, using the COVID-19 pandemic in Pune, India, as a case study. We analysed key public health metrics including the test positivity rate (TPR), case fatality rate (CFR), and reproduction number (Rt). Through simulations of the epidemic and the public health response and by varying levels of testing and contact tracing, we assess how these metrics are related to epidemic curves such as infections and deaths. The results show that the rate of change of TPR can help estimate the severity of the outbreak and predict when it will peak. Rt is a strong predictor of the infection peak, but large computed confidence intervals can place strong caveats on its use in decision-making. In contrast, CFR is not useful for predicting the epidemics severity or peak,as it tends to peak when the infection curve is on a decline and scales non-linearly with the severity of the epidemic. Overall, our findings highlight that TPR and Rt are valuable tools for real-time epidemic management, while CFR may have limited utility. Through this study, we provide modelling evidence to support the use of some metrics for public-health decision making during epidemics.