- HOME
- Reports
- Simulation of infection spread and control
- Factors behind the infection decrease in Tokyo: quantitative analysis.
Factors behind the infection decrease in Tokyo: quantitative analysis.
/
-
Graduate School of Economics, Faculty of Economics, The University of Tokyo
Background
■Claims heard in late July and early August
■"Without a lockdown, infection will not be suppressed"
■"Infections will not decrease until we achieve a 50% reduction in human flow"
■The analysis presented by the various teams at the time was somewhat consistent with these claims
■In Tokyo, infections declined rapidly in the second half of August, even though a variety of human flow data subsequently turned upward or stopped downward trend
■In this document, the term "human flow data" is used in a broad sense
■Here, we analyze the quantitative importance of several factors that may have contributed to the decline in infections since late August in Tokyo
■Beppu et al. (2021): Sources of the Sharp Decline in COVID-19 in Tokyo: Summer 2021
Key human flow data increased or stopped decreasing from mid-August


*The right figure shows the same trend for all time periods (https://www.mhlw.go.jp/content/10900000/000845988.pdf)
Factors Contributing to the Decline in Infections
■Here we list the following six factors
■Hypothesis 1: Weather
■Hypothesis 2: Basic reproduction number is lower than expected
■Delta variant is actually not that infectious
■Factors that may reduce the threshold for herd immunity acquisition, such as immunity heterogeneity among individuals and fragmented communities
■Example: Keio University Kurihara Laboratory Model, University of Tokyo Osawa Laboratory Model, etc.
■It is possible to actually solve such a model to test its quantitative importance, but we will not deal with that here
■Hypothesis 3: Cycles
■Hypothesis 4: The actual cumulative number of infected people is higher than the cumulative number of PCR-positive people, which is close to herd immunity
■Hypothesis 5: Effects of vaccination on infection prevention is higher than expected in people other than the elderly who are active
■Hypothesis 6: Risk-averse behavior due to tightness of medical capacity
Analysis
■Step 1: If we had done a hypothetical "human flow data-focused" analysis just before the infection decline began (mid-August), we would have presented a hypothetical outlook in mid-August
■Epidemiological macro-model was used (Fujii and Nakata (2021))
■Google Mobility (entertainment and retail) was used / assuming realistic vaccination trends
■In this hypothetical basic outlook, the number of new infections in the first week of October is about 7,000 per day
■The actual figure was about 200 people
■Consistent with the view that "even with increased vaccination rates, infections will not decline without a reduction in human flow"
■Note that this is different from the outlook presented by the Fujii-Nakata team at the time
■(August 10) "Scenario for controlling the spread of infection through voluntary behavior change" (https://covid19outputjapan.github.io/JP/tokyo_20210810.html)
■Step 2: Calculate how much the outlook for the number of new infections would have fallen (or come closer to reality) if the mentioned factors had been taken into account
Simulated basic outlook

"Even taking into account the rise in vaccination rate"


■Vaccination has been a force in controlling the spread of infection since late July and has been working continuously
■However, the changes in infection are not significantly affected by the presence or absence of vaccination
■Vaccination pace has been continuous, so that alone is unlikely to explain the timing and rapidity of the decline in infections since late August
"Even taking into account the rise in vaccination rate"


Summary

*Note that this hypothesis is unlikely to explain the continued decline of infections since October
Important Points
■Vaccination has and will continue to greatly reduce infections. However, it is unlikely to be the main factor for the rapid decline in infections from the second half of August
■Without additional factors, it is difficult to explain the timing and rapidity of the decline in infections from late August
■Lower-than-expected basic reproduction number, cycles, and risk aversion behavior due to tightness of medical capacity can quantitatively explain some of the infection decline
■That does not necessarily mean these were important in the real world
■If the cycle is a real factor, it is important to understand why in order to be able to predict the future
■If behavioral change caused by tightness of medical capacity is a real factor, the outlook for the future will not necessarily improve
■Lower-than-expected basic reproduction number will improve the outlook
■There are a wide range of reasons why the basic reproduction number that should be used for a standard model may be low
■Delta variant is less infectious than expected / various heterogeneity among individuals / existence of fragmented communities
Comment
■Here we analyze the quantitative importance of each hypothesis by one particular method
■A different method of analysis could lead to a completely different conclusion
■The results of analysis by one method should not be interpreted as absolute truth
■It is desirable for more researchers to analyze past phenomena and policy effects and to reflect the results in future analyses
■Hirata Laboratory, Nagoya Institute of Technology: "Changes in new positive cases in Jul-Sep period"
■https://www.covid19-ai.jp/ja-jp/researcher/akimasa-hirata/
■It is important to analyze which hypothesis is correct, and at the same time to consider how the future outlook will change if the hypothesis is correct, and what the policy implications of the hypothesis are
Lessons
■Infection may decline rapidly without lockdown/additional restriction on human flow
■An observation independent of "what are the factors for the rapid decline in infection"
■Lessons
■Going forward, we should be more cautious than ever about lockdowns and human flow policies
■Why? (1) Because the uncertainty of the effects of human flow policies has increased, and (2) Because these policies have significant costs (negative impact on society, economy, culture, and education)
■A classic of decision making in the face of uncertainty: W. Brainard (1967): Uncertainty and the Effectiveness of Policy
■This does not necessarily mean that they should be excluded from policy options altogether
■"Infection can be reduced without additional restriction on human flow" is not the same as "restriction on human flow is not always helpful in reducing infection"
■If it is thought that reducing human flow is effective in reducing infection, and if measures can be taken to mitigate the negative effects of this, they can be considered
Hypothesis 1: Weather
Background
■Late August and early September were wet, humid, and cool
■Document 1 of Cabinet Office Advisory Board on Artificial Intelligence (Sept. 15) ("Why did the number of new positives decline so rapidly?") last page
■https://corona.go.jp/prevention/pdf/advisory_kaigou_20210916.pdf
■https://corona.go.jp/prevention/#ai-cont

In this section, we estimate the impact of taking these weather factors into account in the infection outlook









Summary of Hypothesis 1
■The difference between the simulated outlook and reality cannot really be explained by this hypothesis
■Note that this is only the result of one analysis
■Some analyses suggest that clusters are more likely to occur during periods of reduced rainfall
■Osawa and Maekawa (October 5, 2021)
■ https://www.covid19-ai.jp/ja-jp/presentation/2021_rq3_countermeasures_simulation/articles/article164/
Hypothesis 2: Lower than expected basic reproduction number
Background
■After the rapid spread of infections in the last week of July, the Fujii-Nakata team revised the relative infectivity of the Delta variant (compared to the Alpha variant) from 1.3x to 1.5x
■However, the effective reproduction number outside the last week of July was not so high
■It is also reasonable to conclude that the sharp increase in the last week of July was due to other special reasons
■Here, we estimate what the impact on the outlook would have been if we had adopted the assumption of Delta variant infectivity of less than 1.5x
■More broadly, please interpret this as an estimate of the impact of "if the basic reproduction number to be used in this model were lower" for a variety of reasons
■Heterogeneity of immunity and importance of segmented communities (interaction between them and vaccination coverage)

Summary of Hypothesis 2
■Some of the differences between the simulated outlook and reality could be explained
■Whether or not this hypothesis is correct will have a significant impact on the medium- to long-term outlook
■If correct, it suggests a low threshold for acquiring herd immunity
Hypothesis 3: Cycles
Background
■Cyclicality can arise for a variety of reasons
■Examples of exogenous reasons
■A contingent 120-day cycle has been observed with "dry winter, Alpha variant spread in spring, and Delta variant spread in summer"
■Examples of endogenous reasons
■When infections go down, vigilance goes down, and when infections go up, vigilance goes up. Cyclicality could occur with this repetition
■Here, we estimate what the hypothetical outlook would have been if we had factored in such a cycle



Summary of Hypothesis 3
■Many of the differences between the simulated outlook and reality could be explained
■The impact on the future outlook depends on "why the cycle is created"
■When a cycle is produced for exogenous reasons
■Case 1: Regardless of people's behavior, the forces of increased infection are stronger in winter
■Case 2: If past cycles were created by chance due to new mutated variants, they would not be a factor in increasing the force of infection in winter, unless variants more infectious than the Delta variant were reported
■When a cycle is produced for endogenous reasons
■As people become less vigilant (or for other reasons), there is the possibility of another wave
■See "Hypothesis 6: People's risk aversion due to tightness of medical capacity"
Hypothesis 4: Capture rate
Background
■If there have been other infected people besides PCR-positive people, herd immunity can be said to be closer than otherwise.
■Here, we estimate the impact on the outlook of assuming that the actual number of infected people is X times the number of PCR-positive people
■Assume "actual infected = PCR positive * X"
■Historical parameter estimates are also adjusted to be consistent with this assumption

Summary of Hypothesis 4
■Some of the differences between the simulated outlook and reality could be explained
■Not necessarily a significant improvement on the medium- to long-term outlook
■It cannot be said that "X more infected people than PCR positive people = X closer to gaining herd immunity"
■Why? Because infected people get vaccinated too
■Example: "80% vaccinated. 10% of the population is infected"
■Not "80 + 10 = 90," but "80 + 10 * 0.2 = 82"
Hypothesis 5: Higher than expected effects of vaccination on infection prevention
Background
■Vaccinations for non-elderly people began in earnest in early July
■The infection-preventing effect of the 2nd vaccination showed up about six weeks later (mid-August)
■If the effect of vaccination on infection prevention among non-elderly people was higher than expected, it could have contributed to the decrease in infections from the second half of August
■Estimated using a four-generation epidemiological macro model

Summary of Hypothesis 5
■The difference between the simulated outlook and reality can be explained only a few by this hypothesis
■Short-term prospects are not significantly affected by the effectiveness of infection prevention
■Whether or not this hypothesis is correct could have a significant impact on the medium- to long-term outlook
■Even a difference of just 10% in infection prevention has a significant impact on the calculation of the number of immunized individuals needed to achieve herd immunity
■Note that the pace at which the effectiveness of infection prevention declines in the future is equally important
Hypothesis 6: People's risk aversion due to tightness of medical capacity / thorough countermeasures against infectious disease at the individual level
Background
■The hypothesis that people avoided risky behavior because of the fear that "if they got infected and became seriously ill, they might not be able to find a hospital bed"
■Anecdotal Evidence
■Behavioral changes in the first half of August for oneself personally, the people around oneself, and some people observed on social networking sites
■An empirical study of the "information effect": Watanabe and Yabu (2021), Goolsbee and Syverson (2021)
■The Fujii-Nakata team has long emphasized this hypothesis
■Example: August 10, "Scenario for controlling the spread of infection through voluntary behavior change"
■This scenario presents the prospect of infection peaking out and then declining without additional human flow control
■(Although the peak level is slightly higher and the pace of decline is slower than in reality)
■Science and Art. Judgment is important in policy analysis
■A good example of incorporating Anecdotal Evidence as part of policy analysis: Beige Book
■https://www.federalreserve.gov/monetarypolicy/beige-book-default.htm
August 10 (just before the infections began to decline in Tokyo)

Analysis
■1. Estimate how the outlook would have changed if we had taken into account the risky behavior that can be read on Twitter
■Tweets from users who have actually performed actions such as karaoke and attending drinking parties
■Created by Toyoda Laboratory, Institute of Industrial Science, the University of Tokyo, using Twitter data provided by NTT Data
■Due to the sample size, not only the Tokyo index but also the national index was considered
■2. Estimate if the impact of the utilization rate of hospital beds for seriously ill patients on the effective reproduction number had been taken into account
■Current high utilization rate of hospital beds for seriously ill patients predicts lower effective reproduction numbers in the future
■3. Estimate if the impact of the number of newly infected people on the effective reproduction number had been taken into account
■Current high number of newly infected people predicts lower effective reproduction numbers in the future
Twitter: (Karaoke [Tokyo])
![Twitter: (Karaoke [Tokyo])](https://www.covid19-ai.jp/wp-content/uploads/2021/10/article177e-19-1.png)
![Twitter: (Karaoke [Tokyo])](https://www.covid19-ai.jp/wp-content/uploads/2021/10/article177e-19-2.png)
Note: Created by Toyoda Laboratory, Institute of Industrial Science, the University of Tokyo, using Twitter data provided by NTT Data
Twitter: (Drinking parties [Tokyo])
![Twitter: (Drinking parties [Tokyo])](https://www.covid19-ai.jp/wp-content/uploads/2021/10/article177e-20-1.png)
![Twitter: (Drinking parties [Tokyo])](https://www.covid19-ai.jp/wp-content/uploads/2021/10/article177e-20-2.png)
Note: Created by Toyoda Laboratory, Institute of Industrial Science, the University of Tokyo, using Twitter data provided by NTT Data
Hospital bed utilization rate for seriously ill patients


Number of newly infected people


Tokyo Tweet Data Use

National Tweet Data Use

Tokyo Tweet Data Use

National Tweet Data Use



Summary of Hypothesis 6
■Depending on the variables used, some of the differences between the simulated outlook and reality could be explained
■If this hypothesis is correct, it is a factor that worsens the outlook
■Why? Because the current low number of new infections and hospital bed utilization rates suggest that infections will increase in the future
■Note that the fact that the number of infected people continued to fall even in October, when the medical crunch was resolved, suggests that the quantitative importance of this hypothesis is limited
■Analysis update and Zoom briefing on Tuesdays:https://Covid19OutputJapan.github.io/JP/
■Reference materials:https://covid19outputjapan.github.io/JP/resources.html
■Zoom briefing video:https://covid19outputjapan.github.io/JP/recording.html
■Economic Seminar Series
■https://note.com/keisemi/n/n9d8f9c9b72af
■https://note.com/keisemi/n/n7f38099d0fa2
■https://note.com/keisemi/n/nd1a6da98f00e
■Papers available at:https://link.springer.com/article/10.1007%2Fs42973-021-00098-4
■Twitter: https://twitter.com/NakataTaisuke
■Questions, requests for analysis, etc.
■dfujii@e.u-tokyo.ac.jp
■taisuke.nakata@e.u-tokyo.ac.jp