China Covid Big Bang final death toll : we’re getting there, finally

Solidarity Covid – Expats in China, the volunteer project which has fostered unique expertise on the dynamics of the pandemic in China is closing its research with an analysis of the various publications post re-opening about the death toll of the very intense and concentrated wave that hit the Chinese na?ve population infecting about 90% of the population in 3 weeks.

Models, hints and estimates

The final death toll of the China Reopening tsunami Covid wave in December 2022 - January 2023 may remain a mystery forever, as it is globally considering the many inconsistencies in definitions, tracking methods, and of fatigue of the general public. For China, a totally na?ve population hit overnight by a most contagious variant, Omicron BA.5, we were left in January 2023 with a re-re-adjusted count of 83000 deaths measured in the hospitals recovering from a daunting wave of saturation, leaving about 3/4th of severe patients unattended and most staff sick. A simulation released on May 10 2022, in the thick of the Shanghai lockdown, by researchers from Fudan University had already forecasted a high toll of 1.6 million deaths, should reopening occur, even after ramping up vaccination. High toll, because concentrated infections on na?ve population can only generate huge saturation, they knew it and they kept the country locked as long as they could.

After the authorities gave up on Zero Covid, following a multiplication of outbreaks throughout the country and huge fatigue of the population, they had promised they would run the excess death analysis the same way most developed countries have been doing. They didn’t say though that they wouldn’t carry out additional early analysis ahead in time of their usual statistical demographic publications and conceal all information on cremation volume until then… So, we had to wait much longer than we had hoped.

Our volunteer team of Solidarité Covid - Expats in China was more than relieved to end our daily curation task on the Covid statistics of China. We decided we would keep tracking the dynamics of reopening (international flight schedules and visas), vaccination policies, subsequent waves with weekly posts until we could get an estimate of this reopening death toll through the yearly bulletin of the National Statistics Bureau, released every year since 2021 on January 17, 10.00 am.

Such a major event wouldn’t wait one year without undergoing several attempts to find out the excess deaths through alternative sources. The first attempt was published on February 5 2023 in the New York Times, after hearing about many celebrities, scholars, not always very old, had passed away without further information on the cause of death. The 3 journalists reported the obituaries from two academic institutions from October to End January showing the names and faces of each deceased scholar. From a baseline of 4-5 deaths per month in October-November, the toll sky-rocketed to up to 23 in?December, hence a multiple of 4.5 per month.

The second hint came on July 19, through a Caixin newsbrief which was captured again by the New York Times before it got censored. The Caixin article was reporting a leak on the number of cremations recorded in Q1 2023 in the Eastern province of Zhejiang which were increased by 70% vs Q1 2022. What is interesting with this second hint is not the surly titles of the New York Times articles, but the increments on 2 distinct sub-territories (Beijing Universities and Zhejiang) which come up to the same multiple, because x4.5 in one month, extended to 3 months (considering the excess is concentrated in December / January) comes up to +73%... a good hint indeed.

Baidu vs Baidu

Then came on August 24 2023 the first scientific article post reopening addressing the topic of the death toll. It explored the same source as New York Times of the obituaries of 2 Universities in China (Beijing and Harbin), but went back since 2016. It also used another non-official yet public source such as the Baidu Index, a free service provided by the Chinese Internet giant (which captures 90% of Internet searches in China) on key words around mourning : you prompt on the choice of words, the time and geographic frames, the granularity on these, and you get back (through Python) a weighted index (not absolute number of searches) on the relative share of each of the chosen words. A service fairly similar to Google trends, except that the exact formula of the weighting is not disclosed. In China, next to Baidu, you also have a lot of wechat talk exchanging contacts and tips, those threads are not captured in Baidu Index…

It’s important to use those two sources to come up with a good estimate, because Baidu Index can provide full geographic coverage and granularity but it’s only weighted and if other topics become popular on Baidu, then you get a drop in the index (see figure above) of those 4 words around funeral, which doesn’t mean there are less deaths…

The good thing with these sources is that they are not official sources and hence do not require official validation for publication (an HKU paper using the National Health Commission data over 18 months submitted to Lancet on January 12 2023, is still not peer reviewed over one year later…). And the most interesting result with this article is that we get a similar multiple as the 2 NYT articles : 2.44 at national level in two months, which converts into 4.88 in one month, very close to the University obituaries’ increment… and in absolute deaths, it converts into 1.87 million incremental deaths, not so far from the Fudan prediction in March 2022 on Covid deaths only…

We felt very confident having three sources come up with the same increment, but those didn’t come from Chinese official sources. A fully Chinese paper was released in Nature less than a month after the Jama Network article. It used the same source ‘Baidu Index’ on a slightly different list of words, and strangely, the same Baidu Index in a different article comes out with only 710 000 incremental deaths.

There’s only one word in common between the two prompts, and the word “crematorium” is not selected in the Nature exercise. Based on the reports on saturation of the crematorium facilities in December 2022 – January 2023, it’s likely that “crematorium” is a critical word to be included. Before a bereaved family looks for a “casket” or “wreath and elegiac couplets”, they need to find a crematorium, it’s the first step in planning a funeral. Also, doing a search on "obituary" doesn't mean that it's a close parent you have just lost, it can be any acquaintance or celebrity. One death can generate many searches on obituaries, hence the higher index reached by this word (above 12000), higher than any of the other words selected by Jama Network (max below 3500).

Common sense vs Python

Another caveat in the Nature exercise is the baseline used : it goes back only three years and the baseline is stable, not taking into account any seasonality of deaths or the growing trend in deaths reported by the National Statistics Bureau.

Just a week before the publication of Jama Network article, Solidarity Covid had worked on evaluating this baseline, and we came across two abnormalities :

?1) a surge of deaths between 2005 and 2008 (+10.1% in annual deaths in 3 years) for which we have no sound explanation. Should there be some sampling re-design to explain, we would get similar growths in nb of births and total population, yet those indicators are very linear between 2005 and 2008.

2) the year 2021 is already atypical with a 1.60% growth rate in deaths. That year was the height of the Zero Covid era, and China recorded only about 22 000 new cases (a majority imported) and 3 deaths.

There remain three hypotheses for this 2021 increase :

1) the acceleration of ageing population. It’s a fact for sure (+45% of 65+ Years between 2010 and 2020 according to the latest Census), but why such a surge after years of linear growth ?

2) the side effects of Covid vaccination with inactivated virus (83.4% of the double vaccinated population as of 12/31/2021). Impossible to rule because no matter what anti-vax mRNA / AstraZeneca or detractors of Chinese vaccines may think, there is no evidence of excess mortality following the 2021 vaccination campaigns in the many countries that have administered Chinese vaccines (Middle East, Africa, South America, Asia, etc.).

3) a catch-up of 2020 : during two months between end January and end March 2020 (and beyond for schools, indoor entertainment, travel to Beijing…), Wuhan was isolated, there was a soft lockdown in all of China, inducing slow-down in economic activity and travel. These reduce deaths (stable in 2020 while the increase had been sustained for years), the fear of going to the hospital remained stubborn for a very long time in China, and there’s nowhere else to go to when you’re sick in China. Therefore, there were delays in diagnosis of chronic diseases, which generate a surplus of deaths the following year. This phenomenon was observed in 2022 also in Western countries. So it remains the most plausible explanation.

On the D-day

As planned, on the morning of January 17 2024, we checked the National Statistics Bureau website, and here it was, the final piece of the puzzle : 11.1 million deaths in 2023 through official sources, a +6.6% increase vs 2022, +11.2% in 3 years, slightly above the 2005-2008 jump. The bulletin also reports the continued decline in births to 9.02 million, and therefore a decline in the total population, but the fall in births is much softer than in previous years.

To calculate our baseline, the growth in 2021 being abnormal and unexplained, we applied an average growth over 3 years, i.e. 0.7% annual growth in deaths over 2022 and 2023. On this hypothesis, the incremental number of deaths linked to reopening stands at 1 million. This compares to 1.87 million estimated by the Fred Hutchinson Center in Seattle in Jama Network over the period alone from December 2022 to the end of January 2023. This gives us a retraction, or “backlash” effect of 0.87 million deaths of vulnerable people which occurred in the Covid wave rather than in the following months.

In annual results, 1 million additional deaths represent 10% excess mortality, which compares to 9% in 2020 in France; reopened China would therefore have suffered an excess comparable to France in 2020 (2020 only, then it repeated again in 2021 and 2022 outside of China). And over four years, even assuming all incremental deaths are Covid, China is doing pretty well in number of deaths/1M pop.

2020-2023 cumulative data of declared Covid deaths with their multiple biases widely debated in China and elsewhere. Across all countries, China's reconstituted 1M/pop deaths place it in 88th position. The vaccination deficits of seniors in China and Hong Kong nevertheless weigh on the result, because among the “late Zero Covid exiters”, Singapore, Australia and New Zealand are doing much better.

This first article in English in one year is a combination of two articles in French published in September 2023 and January 2024. It is complemented with the insights collected from the Hutchinson cancer Care center in Seattle, thanks for your input. Please check our exclusive analytics on all the phases of the pandemic in China on the final version of our book (all illustrations in English) “Planète Chine Zéro Covid, trois ans sur orbite”.