Examine the margin of error in the number of deaths from pneumonia
If Covid19's deaths were counted for pneumonia, we would like to know how much it can be said to be an error. Since there are clearly seasonal fluctuations and trends, we will look at the variation in the residuals after removing each.
Time series analysis
Data from vital statistics. We will confirm the monthly changes in the number of deaths from 2,015 to 2,018. The vertical axis is the number of deaths, and the horizontal axis is the month.
12-month cycle without looking at partial autocorrelation. Every year, there is a trend of rising overall.
This time, the death toll from pneumonia is taken as 9% of the total (from the material at the end of the post), and it is decomposed into trends, seasonal fluctuations, and residuals using an additive model. Use seasonal_decompose from StatsModels in an additive model (it should be repeating the moving average to calculate seasonal variation)
Y[t] = T[t] + S[t] + e[t]
T is the trend, S is the seasonal variation, e is the residual
import statsmodels.api as sm
#
#death is data that contains only the number of deaths in one column
#
dec = sm.tsa.seasonal_decompose(death,period=12)
trend = dec.trend
seasonal = dec.seasonal
residual = dec.resid
Data plot
trend
Seasonal variation
Residual error
Conclusion
The minimum residual value is -409.1 and the maximum value is 282.8. The standard deviation is 136.5. The average is -4.59. When it comes to an increase of 1,000 or more, the impression is that it probably contains a lot of Covid19.
Assuming a case fatality rate of 2%, if the number of infected people is about 50,000, the line of 1,000 dead.
With Covid19 patients sparsely distributed throughout the country, it is difficult to see the increase in deaths from pneumonia from hospitals (the number of hospitals without mental hospitals should be more than 7,000, about one more. Not even).
Data by cause of death nationwide will be delayed by 5 months, so In order to grasp the current situation, it is necessary to look for a fast place in the bulletin of the local government. It may not exist.
About the current situation
Currently, two weeks have passed since the expert meeting designated the remaining 1-2 weeks in Japan. A state of emergency has been declared in NY, and the damage in the EU is becoming more serious.
In Japan, no damage has been reported as much as in the EU and South Korea, The characteristics of the virus that is prevalent in the main are different, the medical system is different, etc. There are various possible factors, and there is not enough data to link to a specific scenario.
If the environment is equivalent to that of Wuhan virus and similar to that of Europe and the United States, The death toll numbers are lower than expected, so we examined whether the data could detect whether Covid 19 deaths were among the pneumonia deaths.
(I'm just fixing the virus and environment as variables, I'm not judging that, and the verification of other scenarios is a different story. I confirmed this much this time)
data
Monthly death data (annual deaths and mortality rate by month of death (per 1,000 population)): https://www.e-stat.go.jp/stat-search/files?page=1&layout=datalist&toukei=00450011&tstat=000001028897&cycle=7&year=20180&month=0&tclass1=000001053058&tclass2=000001053061&tclass3=000001053065&result_back=1
There is a preliminary report on the total number of deaths by month, and the latest is 2 months ago (Vital Statistics Bulletin for November 19): https://www.mhlw.go.jp/toukei/saikin/hw/jinkou/geppo/s2019/11.html
Monthly approximate number by cause of death (latest 5 months ago) https://www.e-stat.go.jp/stat-search/files?page=1&layout=datalist&toukei=00450011&tstat=000001028897&cycle=1&year=20190&month=23070909&tclass1=000001053058&tclass2=000001053060
Pneumonia (Death due to pneumonia in Japan from the viewpoint of vital statistics): http://www.tokyo-eiken.go.jp/sage/sage2018/
Recommended Posts