一种用于预测SARS-CoV-2危重症监护病房占用情况的基于点数的预测工具的推导与验证：一项基于人群的建模研究。
Derivation and validation of a point-based forecasting tool for SARS-CoV-2 critical care occupancy: a population-based modeling study.
作者信息
Grima Alicia A, Lee Clara Eunyoung, Tuite Ashleigh R, Wilson Natalie J, Simmons Alison E, Fisman David N
机构信息
Dalla Lana School of Public Health, University of Toronto, Toronto, Canada.
出版信息
Lancet Reg Health Am. 2025 Jun 5;47:101143. doi: 10.1016/j.lana.2025.101143. eCollection 2025 Jul.
BACKGROUND
The requirement for critical care in even a modest fraction of SARS-CoV-2-infected individuals made critical care resources a key societal chokepoint during the COVID-19 pandemic. We previously developed a simple regression-based point score to forecast critical care occupancy in Ontario, Canada, using case numbers, mean age of cases, and testing volume. In this study, we aimed to validate and update this forecasting model to account for evolving population immunity, including the effects of widespread vaccination.
METHODS
We obtained complete provincial SARS-CoV-2 case, testing, and vaccination data from March 2020 to September 2022, subdividing the pandemic into six waves. Our initial model was fitted using data from the first two waves; an updated model included wave 3, which was dominated by N501Y+ variants. We validated the models by comparing projections to waves not used for fitting. Predictive validity was assessed using Spearman's rho. Counterfactual scenarios without vaccination were modeled to estimate vaccine-attributable reductions in critical care admissions.
FINDINGS
The initial model (waves 1-2) was well calibrated (rho = 0.85) but had modest predictive validity (rho = 0.46). Predictive validity improved with models fitted to waves 1-3, both without (rho = 0.60) and with vaccination (rho = 0.68); model fit improved significantly with vaccination (p = 0.013). Averted admissions attributable to vaccination were estimated at 144% (22,017 expected vs. 9020 observed).
INTERPRETATION
Simple regression-based forecasting tools remain valuable for predicting SARS-CoV-2 critical care occupancy. However, models developed early in the pandemic should be recalibrated to account for evolving immunity, including widespread vaccination.
FUNDING
Canadian Institutes of Health Research (OV4-170360); R. Howard Webster Foundation (via the University of Toronto Institute for Pandemics).
背景
即使只有一小部分感染严重急性呼吸综合征冠状病毒2（SARS-CoV-2）的患者需要重症监护，在2019冠状病毒病大流行期间，重症监护资源也成为了社会的一个关键瓶颈。我们之前开发了一种基于简单回归的评分系统，利用病例数、病例平均年龄和检测量来预测加拿大安大略省的重症监护床位占用情况。在本研究中，我们旨在验证并更新此预测模型，以考虑不断变化的人群免疫力，包括广泛接种疫苗的影响。
方法
我们获取了2020年3月至2022年9月全省完整的SARS-CoV-2病例、检测和疫苗接种数据，将大流行细分为六个波次。我们的初始模型使用前两个波次的数据进行拟合；更新后的模型纳入了第三波次的数据，该波次以N501Y+变体为主。我们通过将预测结果与未用于拟合的波次进行比较来验证模型。使用斯皮尔曼等级相关系数（Spearman's rho）评估预测效度。对未接种疫苗的反事实情景进行建模，以估计疫苗接种可归因于重症监护入院人数的减少。
结果
初始模型（第1 - 2波次）校准良好（rho = 0.85），但预测效度一般（rho = 0.46）。使用第1 - 3波次数据拟合的模型，无论是否纳入疫苗接种因素，预测效度均有所提高（未纳入疫苗接种时rho = 0.60，纳入疫苗接种时rho = 0.68）；纳入疫苗接种后模型拟合显著改善（p = 0.013）。估计疫苗接种可避免的入院人数为144%（预期2,2017例，实际观察到9,020例）。
解读
基于简单回归的预测工具对于预测SARS-CoV-2重症监护床位占用情况仍然很有价值。然而，在大流行早期开发的模型应重新校准，以考虑不断变化的免疫力，包括广泛接种疫苗的情况。
资金来源
加拿大卫生研究院（OV4 - 170360）；R. 霍华德·韦伯斯特基金会（通过多伦多大学大流行研究所）。
Zotero 插件