Skip to content

Latest commit

 

History

History
145 lines (104 loc) · 9.63 KB

README.en.md

File metadata and controls

145 lines (104 loc) · 9.63 KB
Raw

Nowcasting and Forecasting the 2019-nCoV Outbreak

简体中文 | English

Date: Since Jan 26 2020

Contents：

  1. Nowcasting and Forecasting the 2019-nCoV Outbreak size in Wuhan

    MSE, basic SEIR model, sentiment analysis Overview of SEIR model

    • Model 1: Estimating the potential number of cases in Wuhan until Jan 23
    • Model 2: Simulating Peak of 2019-nCoV in Wuhan after 23 Jan

  2. 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

    • Prediction for China total trend
    • Prediction for Hubei and ex-hubei trend

Key limitation within models below:

  1. My models conclusions are critically dependent on the assumptions underpinning models.
    • Adjustments are considered by sensitivity analysis (but not enough)
  2. The model s' structures are quite simple, haven't combine enough information, so cannot get really good or robust result.
    • But for the prediction of Wuhan City only, maybe enough.
    • Will keep updating model based on the latest information.

Main Challenges for all predictions:

  1. We have limited understanding of this new disease
    • For example, we did not test all people with 2019-nCoV correctly. (unclear symptoms: whether people with 2019-nCoV who do not have symptoms can transmit an infection)
  2. It is hard to get the real-time correct information. (official Chinese data is under-report 100% ), especially for Wuhan.
    • For example, we do not know how many people infected when wuhan shut down.
  3. The prediction is highly sensitive to the policy
    • (for example, travel restriction, force people stay in home, wuhan build 3 new hospital for 2019-nCoV etc. ), all of these policies influenced a lot on the time-line. When we do prediction, our key assumption is no new policy in the future.

Data

Real-time data query and save to csv

Nowcasting and Forecasting the 2019-nCoV Outbreak size in Wuhan (Model 1 and 2)

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.

Model 1: Estimating the potential number of cases in Wuhan until Jan 23😷

  • Author: Yiran Jing
  • Main Conclusion (within Wuhan City only): There are more than 38500 cases 95% CI(30000, 48470) until Jan 23, based on 29 Jan data.

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)

Model 2: Simulating Peak of 2019-nCoV in Wuhan after 23 Jan📈

  • Author: Yiran Jing

Method: Deterministic SEIR (susceptible-exposed-infectious- recovered) model and Sensitivity analysis

Reference: Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak (Jan 31)

  • Main Conclusion (within Wuhan City only): (using Chinese official data between 2019-12-08 and 2020-02-02)

    • 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.

Model 3:Real-Time forecasting of the confirmed cases in China in the next 2 months📉

Method: Dynamic SEIR (susceptible-exposed-infectious- recovered) model, Gradient Descent 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 2020-02-13)
    • The number of net confirmed cases will exceed 60000, and the peak can be reach before 20 Feb.
    • The transmissibility has been controlled from initial 3.2(R0) to less than 0.5.
    • The estimated number of infected case will be less than 4000 in early April
  • Model assumptions: Overview of SEIR model
    • Constant (closed) population size: Due to the international travel ban, strict home quarantine rules in China and the low death rate of COVID-19 (less than 2%), we can assume the China population is constant.
    • In SEIR models, the exposed individuals is infected but not yet infectious, and the first transmission can only happen after symptoms appear. However, InCOVID-19 case, we know that individuals are infectious during the whole incubation period. Assume latent period is the same as incubation.
    • Suppose the average duration of recovery is 14 days, which is similar with SARS
    • Suppose the total number of individuals within incubation period is 4 time of susceptible case reported by CCDI.
    • Assume the died people is around 2%, belonging to removed individuals (R).

The red line shows the trend of net confirmed cases in the next 50 days. Note:

  • Removed: heal or death
  • Death: Removed group * death_rate
  • Exposed: individuals during incubation period
  • Susceptible: Healthy people
  • Infected: Confirmed cases

Model performance

The mean absolute percentage error (MAPE) is a measure of prediction accuracy of a forecasting method in statistics. The MAPE of confirmed cases using data between 2020- 2-14 to 2020-02-22 is 0.0066. The figure below visualizes the real observation and the SEIR model predictions for the next 9 days. Overall, SEIR model predicts well for the peaking time and the general trend.

Dynamic contact rate β as a function of time t

Optimization algorithm Gradient Descent

Real-Time forecasting of the confirmed cases in China

Estimating the confirmed cases of China in the next few days based on the latest data from DingXiangYuan

Real-time data query Step:

  1. Query the latest data from DingXiangYuan
## Update data from DXY
$ cd data_processing && python DXY_AreaData_query.py # save data out to data folder.

Visualization

The best visualization of 2019-nCoV in China

image

Dashboard overseas

CoronaTracker Analytics Dashboard

My current study and tasks are updated here: Project

Welcome to connect with me if you are interested in this project!