简体中文 | English
Nowcasting and Forecasting the 2019-nCoV Outbreak size in Wuhan
MSE, basic SEIR model, sentiment analysis Overview of SEIR model
Model 3: Real-Time forecasting of the confirmed cases in China in the next 2 months
Baseline: Ridge regression, improved by Dynamic SEIR model
Author: Shih Heng Lo; Yiran Jing
On January 23, authorities in Wuhan shut down the city’s public transportation, including buses, trains, ferries, and the airport. There are 9 million people stay in Wuhan after 23 Jan. And official reported that 5 million people travel out Wuhan for Chinese Spring Festival. The effective catchment population of Wuhan international airport is around 19 million.
Consider the transmissibility and population of Wuhan changed a lot before and after Jan 23, 2020, I choice different methods to nowcasting and forecasting the potential outbreak size in Wuhan city referencing by published papers.
Method: Considering Wuhan is the major air and train transportation hub of China, we use the number of cases exported from Wuhan internationally as the sample, assuming the infected people follow a Possion distribution, then calculate the 95% confidence interval by profile likelihood method. Sensitivity analysis followed by.
Reference: report2 (Jan 21)
Method: Deterministic SEIR (susceptible-exposed-infectious- recovered) model and Sensitivity analysis
Estimated initial transmissibility R0 (the basic reproduction number) of 2019-nCoV: 2.9
Under the most optimistic estimate, the maximum infected case in Wuhan: more than 14000 (peak, not cumulative) (the peak of red line of the plot below.) And the cumulative number of cases in the whole period is around 50 thousand (the green line).
Truth 1: Consider inadequate medical resources and under-reported official data, Maximum infected case (peak, not cumulative) in Wuhan might between 16000 and 25000
Truth 2: Risk of transmission is still high between 23 Jan and 04 Feb, and begin to decrease after 5 Feb.
Based on official news on 2 Feb, cases cannot be detected immediately, also not perfect isolation. Under this situation, Maximum infected case (peak, not cumulative) in Wuhan can more than 100 thousand or even 150 thousand. (suppose k=2)
Update: 3 new hospitals begin to accept patents after 5 Feb.(can accept around 6 thousand patients total). Now the risk of transmission is decrease, since more patients can be in hospitals and isolated.
Consider truth 1 and 2, the maximum infected case (peak, not cumulative) in Wuhan maybe between 25 thousand and 100 thousand.
The peak will appear after 22 Feb, 2020
Close City policy has significant control for 2019-nCoV, otherwise, the peak of infected cases may up to 200 thousand.
Method: Dynamic SEIR (susceptible-exposed-infectious- recovered) model, estimate contact rate per day Model comparison based on the test score (MAPE) of last 5 days, baseline is ridge Ridge regression Reference: Dynamic SIR model
Main Conclusion (China TOtal): (using Chinese official data between 2019-12-08 and 2020-02-14)
Model assumptions: Overview of SEIR model
The red line shows the trend of net confirmed cases in the next 50 days. Note:
Estimating the confirmed cases of China in the next few days based on the latest data from DingXiangYuan
## Update data from DXY $ cd data_processing && python DXY_AreaData_query.py # save data out to data folder.
My current study and tasks are updated here: Project
Welcome to connect with me if you are interested in this project!