Department of Statistics, Federal University of Technology, P.M.B. 704, Akure, Nigeria.
Research and Development Department, South African Weather Service, Private Bag X097, Pretoria 0001, South Africa.
J Epidemiol Glob Health. 2021 Jun;11(2):200-207. doi: 10.2991/jegh.k.210215.001. Epub 2021 Feb 22.
The novel Coronavirus Disease 2019 (COVID-19) remains a worldwide threat to community health, social stability, and economic development. Since the first case was recorded on December 29, 2019, in Wuhan of China, the disease has rapidly extended to other nations of the world to claim many lives, especially in the USA, the United Kingdom, and Western Europe. To stay ahead of the curve consequent of the continued increase in case and mortality, predictive tools are needed to guide adequate response. Therefore, this study aims to determine the best predictive models and investigate the impact of lockdown policy on the USA' COVID-19 incidence and mortality. This study focuses on the statistical modelling of the USA daily COVID-19 incidence and mortality cases based on some intuitive properties of the data such as overdispersion and autoregressive conditional heteroscedasticity. The impact of the lockdown policy on cases and mortality was assessed by comparing the USA incidence case with that of Sweden where there is no strict lockdown. Stochastic models based on negative binomial autoregressive conditional heteroscedasticity [NB INGARCH (,)], the negative binomial regression, the autoregressive integrated moving average model with exogenous variables (ARIMAX) and without exogenous variables (ARIMA) models of several orders are presented, to identify the best fitting model for the USA daily incidence cases. The performance of the optimal NB INGARCH model on daily incidence cases was compared with the optimal ARIMA model in terms of their Akaike Information Criteria (AIC). Also, the NB model, ARIMA model and without exogenous variables are formulated for USA daily COVID-19 death cases. It was observed that the incidence and mortality cases show statistically significant increasing trends over the study period. The USA daily COVID-19 incidence is autocorrelated, linear and contains a structural break but exhibits autoregressive conditional heteroscedasticity. Observed data are compared with the fitted data from the optimal models. The results further indicate that the NB INGARCH fits the observed incidence better than ARIMA while the NB models perform better than the optimal ARIMA and ARIMAX models for death counts in terms of AIC and root mean square error (RMSE). The results show a statistically significant relationship between the lockdown policy in the USA and incidence and death counts. This suggests the efficacy of the lockdown policy in the USA.
新型冠状病毒病 2019（COVID-19）仍然是对社区健康、社会稳定和经济发展的全球性威胁。自 2019 年 12 月 29 日在中国武汉首次记录病例以来，该疾病迅速蔓延到世界其他国家，导致许多人死亡，尤其是在美国、英国和西欧。为了应对病例和死亡率持续增加的情况，需要预测工具来指导做出充分的应对。因此，本研究旨在确定最佳预测模型，并研究封锁政策对美国 COVID-19 发病率和死亡率的影响。本研究专注于基于数据的一些直观特性（如过分散和自回归条件异方差）对美国每日 COVID-19 发病率和死亡率病例进行统计建模。通过比较美国的发病率与没有严格封锁的瑞典的发病率，评估了封锁政策对病例和死亡率的影响。提出了基于负二项式自回归条件异方差[NB INGARCH（，）]、负二项式回归、带外生变量的自回归综合移动平均模型（ARIMAX）和不带外生变量的自回归综合移动平均模型（ARIMA）的几个阶数的随机模型，以确定最适合美国每日发病率病例的拟合模型。根据赤池信息量准则（AIC），比较最优 NB INGARCH 模型在每日发病率病例上的性能与最优 ARIMA 模型。此外，还为美国每日 COVID-19 死亡病例建立了 NB 模型、ARIMA 模型和不带外生变量的模型。观察到发病率和死亡率病例在研究期间呈统计上显著的上升趋势。美国每日 COVID-19 发病率具有自相关性、线性且包含结构断裂，但表现出自回归条件异方差。将观测数据与最优模型的拟合数据进行比较。结果进一步表明，NB INGARCH 比 ARIMA 更适合拟合观测到的发病率，而 NB 模型在 AIC 和均方根误差（RMSE）方面比最优的 ARIMA 和 ARIMAX 模型更适合死亡人数。结果表明，美国的封锁政策与发病率和死亡率之间存在统计学上的显著关系。这表明美国的封锁政策是有效的。
