Category: models

As we suggested earlier, unresolved (existing) cases (E) tend to resolve into recovery rather than death. The table below shows this clearly.  Between 28 February and 1 March, there were 79 deaths in Hubei, but 5,133 recoveries. By the same token, the apparent fatality rate also falls.

Place	Date	C	D	R	E	D/(D+R)
Hubei	28-Feb-20	65,914	2,682	26,403	36,829	9.22%
Hubei	29-Feb-20	66,337	2,727	28,930	34,680	8.61%
Hubei	01-Mar-20	66,907	2,761	31,536	32,610	8.05%

Note again that the percentage of cases is small compared to the population. The population of Hubei is about the same as the UK. So about 1 in 1,000 have been diagnosed with the disease since it began.  Only because of draconian isolation measures, however, which is why the impact on the economy is so severe.

The table below shows, for 28 February 2020, the diagnosed cases (C), deaths (D), and recoveries (R) from 25 parts of China. Data from John Hopkins University.

The usual caveats apply. 

The last three columns are a function of the primary data. Existing cases (E) is C-(D+R), i.e. diagnosed cases less resolved cases. D/(D+R) is one method for estimating the case fatality ratio. E/C is the unresolved cases divided by diagnosed cases. This ratio will fall to zero over time given that all cases will resolve into recoveries or deaths. The table is sorted by this number.

As is evident, the fatality ratio varies wildly, from Jiangxi, where 790 out of 935 cases have already been resolved, with only one death, to Hubei where little more than half the 65,914 cases have been resolved, with a 9.22% apparent fatality ratio.

Time series analysis (not shown here) suggests that unresolved cases (E) tend to resolve into recovery rather than death, hence there is a moderately strong correlation (0.6) between E and apparent fatality.

There is no explanation yet of why cases should have taken so long to resolve in Hubei, which the media call the epicentre of the outbreak. Note also, as before, that the cases in Hubei are a tiny fraction of its population.

In other news, manufacturing activity in China in February plunged faster than during the 2008 financial crisis.

Place	Date	C	D	R	E	D/(D+R)	E/C
Qinghai	28-Feb-20	18	0	18	0	0.00%	0%
Gansu	28-Feb-20	91	2	82	7	2.38%	8%
Yunnan	28-Feb-20	174	2	156	16	1.27%	9%
Henan	28-Feb-20	1,272	20	1,112	140	1.77%	11%
Hebei	28-Feb-20	318	6	277	35	2.12%	11%
Jiangxi	28-Feb-20	935	1	790	144	0.13%	15%
Shanghai	28-Feb-20	337	3	279	55	1.06%	16%
Anhui	28-Feb-20	990	6	821	163	0.73%	16%
Hunan	28-Feb-20	1,017	4	830	183	0.48%	18%
Shanxi	28-Feb-20	133	0	109	24	0.00%	18%
Shaanxi	28-Feb-20	245	1	199	45	0.50%	18%
Jiangsu	28-Feb-20	631	0	515	116	0.00%	18%
Zhejiang	28-Feb-20	1,205	1	975	229	0.10%	19%
Fujian	28-Feb-20	296	1	235	60	0.42%	20%
Jilin	28-Feb-20	93	1	73	19	1.35%	20%
Guizhou	28-Feb-20	146	2	112	32	1.75%	22%
Liaoning	28-Feb-20	121	1	93	27	1.06%	22%
Tianjin	28-Feb-20	136	3	102	31	2.86%	23%
Xinjiang	28-Feb-20	76	3	52	21	5.45%	28%
Chongqing	28-Feb-20	576	6	402	168	1.47%	29%
Beijing	28-Feb-20	410	7	257	146	2.65%	36%
Sichuan	28-Feb-20	538	3	338	197	0.88%	37%
Shandong	28-Feb-20	756	6	405	345	1.46%	46%
Hubei	28-Feb-20	65,914	2,682	26,403	36,829	9.22%	56%

 

The markets have been in a tailspin this week due to coronavirus and Eumaeus is a bit glum.  On the other hand, is the apocalypse really upon us?

Continue reading “Perhaps not so grim”

The table below is from Worldometers, based on a paper by the Chinese CCDC released on February 17 and published in the Chinese Journal of Epidemiology¹

I would treat all such statistics with caution, but here they are anyway.

AGE	DEATH RATE
80+ years old	14.8%
70-79 years old	8.0%
60-69 years old	3.6%
50-59 years old	1.3%
40-49 years old	0.4%
30-39 years old	0.2%
20-29 years old	0.2%
10-19 years old	0.2%
0-9 years old	0.0%

Death Rate is number of deaths divided by number of known cases of Coronavirus, but we still don’t know the ratio of known to unknown cases.

The reporting of the corona virus outbreak makes almost no sense to Eumaeus – so he won’t comment.

But this site has some helpful statistics and explanations, and this paper published 7 February outlines the methodology for estimating case fatality rate, as well as (v important) the flaws inherent in the models used to estimate it.  See also this paper on a much earlier epidemic, which discusses many of the same problems.

One problem is to estimate the number of cases, particularly difficult when the disease may never show any symptoms. Dividing the number of deaths by the number of reported cases, i.e. those where the patient had symptoms, reported them and was correctly diagnosed, may grossly overestimate the fatality rate. General insurance actuaries may compare this to IBNR.

There is also the problem that if the disease is prolonged, there may be many cases where the outcome is not known, hence the correct method is to divide deaths by the number of cases reported days or weeks ago, where the ‘days or weeks’ is given by some estimate (again, another estimate) of the period from diagnosis to outcome.

[edit] See also the DXY site , a platform run by members of the Chinese medical community, aggregating media and government reports giving COVID-19 cumulative cases in near real-time.