IRL3189 Environment Health and Society, Public Health Department, Faculty of Medicine, Pharmacy and Dentistry of Cheikh Anta Diop University, Dakar, Senegal.
University of Ottawa, Faculty of Law, Civil Law Section Coordinator of the Research Chair on Accountable AI in a Global Context, Ottawa, Ontario, Canada.
Int J Health Plann Manage. 2022 Jul;37(4):2468-2473. doi: 10.1002/hpm.3459. Epub 2022 Mar 20.
Since the outbreak of the SARS-COV2 epidemic turned into a COVID-19 pandemic, international bodies such as the WHO as well as governments have announced projections for morbidity and mortality indicators related to COVID-19. Most of them indicated that the health situation would be worrying. Although using artificial intelligence with mathematical algorithms and/or neural networks, the results of the SIR models were poorly performing and not very accurate in relation to the observed reality in the African states in general and in Senegal in particular. Hence the imperative need to configure the modelling process and approach considering local contexts.
The model implemented is a mixed prediction model based on the Bucky model developed by OCHA and adapted to the context. The construction of the mixed model was done in two steps (basic model with publicly available data, such as those from United Nations-like organisations such as OCHA or WHO for Senegal), (adding more specific data collected through the mixed epidemiological survey). This survey was conducted in Senegal in six localities (Dakar, Thies, Diourbel, Kedougou, Saint-Louis and Ziguinchor) chosen according to the number of confirmed cases of COVID-19. In total, 1000 individuals distributed in proportion to the size of the regions were interviewed in April 2021.
The projected cases in the baseline model were already considerably higher than the cases reported in April. This may be plausible, given the low detection rates throughout Senegal during this period. However, the hybrid model predicted an even higher infection rate than the baseline, perhaps mainly due to vulnerability related to food insecurity and solid cooking fuels. This may mean that there would be more unreported cases than reported. Overall, the mortality rate of both models would be considerably lower than the government-reported mortality rate, even though the number of confirmed cases remains high. This may be an underestimate of the death rate.
An accurate and reliable prediction in times of epidemics and/or pandemics, such as COVID-19, should be based on mixed or hybrid data integrating a quantitative and qualitative approach to enable better policymaking. The projections resulting from this approach would still be effective and would take better account of local realities and contexts, especially for developing countries.
自 SARS-COV2 疫情爆发演变为 COVID-19 大流行以来，世界卫生组织（WHO）等国际机构以及各国政府都公布了与 COVID-19 相关的发病率和死亡率指标预测。大多数预测都表明，卫生形势将令人担忧。尽管使用带有数学算法和/或神经网络的人工智能，但 SIR 模型的结果在非洲国家，特别是塞内加尔的实际情况方面表现不佳，且不够准确。因此，迫切需要根据当地情况配置建模过程和方法。
实施的模型是一种混合预测模型，基于人道协调厅开发的 Bucky 模型，并针对当地情况进行了调整。混合模型的构建分两步进行（基本模型使用公开数据，例如人道协调厅或世卫组织等类似联合国组织提供的数据，用于塞内加尔），（添加通过混合流行病学调查收集的更具体数据）。该调查在塞内加尔六个地区（达喀尔、捷斯、迪奥尔贝、凯杜古、圣路易和济金绍尔）进行，选择这些地区是根据 COVID-19 确诊病例的数量。2021 年 4 月，总共对 1000 名按地区规模比例分布的个人进行了采访。
基线模型预测的病例数已经明显高于 4 月报告的病例数。鉴于整个塞内加尔在此期间的低检测率，这可能是合理的。然而，混合模型预测的感染率甚至高于基线模型，这可能主要是由于与粮食不安全和固体烹饪燃料相关的脆弱性所致。这可能意味着，未报告的病例数可能比报告的病例数多。总体而言，即使确诊病例数量仍然很高，两种模型的死亡率都将明显低于政府报告的死亡率。这可能是对死亡率的低估。
在 COVID-19 等大流行或流行病期间，准确和可靠的预测应该基于混合或混合数据，整合定量和定性方法，以便更好地制定政策。这种方法得出的预测仍然有效，并且会更好地考虑到当地现实和情况，特别是对于发展中国家。
