Overcoming COVID-19 in China despite shortcomings of the public health system: what can we learn?

Phase I

Between December 2019 and 19th of January 2020, the number of patients grew exponentially, but it was too small that officials did not recognize it as a public health threat. When WHO was informed about the new disease of unknown etiology on 31st of December 2019, 27 patients had been diagnosed with viral pneumonia of unknown cause. On the 19th of January, already 198 patients had been recorded which is an increase of 11% per day (geometric mean).

In this phase, most official activities were confidential and not transparent, but the National Health Commission of China started with the pathogen detection and epidemiological investigation, issued technical guidelines to Wuhan, and notified other regions. However, it seems that Wuhan and other parts of China did not implement any significant epidemic prevention actions except for case detection and disease surveillance (Table 1). For instance, a community in Wuhan held the annual “10,000 Family Banquet” as scheduled on 18th of January.

Phase II

Phase II was the phase of high incidence from 20th of January to 20th of February 2020. The emergence of COVID-19 coincided with the world’s largest annual human migration, the Chinese Spring Festival travel season. Under normal circumstances, China could have expected some 3 billion trips in China during the 40-day holiday period from 15 days before the Spring Festival to 25 days after the Spring Festival. For this “spring transit”, the Chinese language has even a special word called “Chunyun” where almost the entire population moves back to their place of origin. As the largest transportation hub in central China, Wuhan would have expected to transfer at least 15 million passengers during the Spring Festival holiday. According to Wuhan Transport Bureau, Wuhan counted a total of 15,223,900 passengers leaving Wuhan and 14,662,200 passengers arriving in Wuhan during the 2018 Spring Festival with a total of 271,339,500 travel activities during this period in Wuhan. There is no reason to assume that the situation would have been different in 2020 without COVID-19 and the Spring Festival 2020 could have become the super-spreader event for the entire country. In fact, it was reported that about 5 million people had already left Wuhan before the city was locked down on January 23, 2020 [18].

After it had been detected that the new disease was contagious [19, 20], the Government of China took emergency measures to reduce the risk of spreading the disease all over China, such as cancelling the Spring Festival, postponing work and school opening, restricting travel, closing entertainment venues and banning public gatherings (Table 2) [27]. However, during this period the number of cases and deaths strongly increased reaching a peak of 32,994 cases on 12th of February (Fig. 1). However, during this period the statistical detection of cases changed so that the jump from 11th to 12th of February is mainly due to different reporting than to a real increase of cases.

A crucial component of the intervention portfolio became the “Health Code Model”. On February 11, 2020, the Health Code model was first launched in Hangzhou, Zhejiang Province, China [29], and has been implemented in other provinces and cities in China since then. The smartphone-based application (app) registers the visited locations and all encounters and transfers it to a national database (whereby – from a western perspective – little value is placed on data protection rights). It gives the users an individual “health code” according to the traffic light scheme. The color green means that users can move around freely. Those who receive the yellow warning signal have to stay at home for 7 days, and red means a two-week quarantine [30]. These measures enable a relatively “normal” life, but they interfere with the privacy of citizens.

Phase III

The third phase concentrated on stabilization of the epidemic (21 February - 28 April). As shown in Fig. 1 the incidence remained at a very low level since mid-March, and since April 1, there were no new infections. Consequently, the number of deaths also strongly declined. However, a new number of 1290 deaths was added in Wuhan on April 17, which was mainly a recalculation of the number of unreported deaths in the early stage of the epidemic, rather than the current number of deaths [31]. On April 26, all COVID-19 hospitalized cases in Wuhan had died or were discharged from the hospital. The epidemic situation was generally stable; the Chinese government began to coordinate the epidemic prevention and control with economic and social development, and to resume work and production.

Phase IV

The interventions against COVID-19 in Wuhan are still in the fourth phase concentrating on early detection of cases and in particular efforts of epidemic prevention and control. However, Wuhan is not seen an exceptionality any longer but the countrywide measures to control the epidemic in the country are applied in Hubei Province as well. At this stage, China mainly focuses on “preventing import from outside and preventing rebound from inside” [22], so as to comprehensively promote the resumption of work, production and school, and restore the normal economic and social order. On April 30, the level of emergency response to public health emergencies in the Beijing-Tianjin-Hebei region was adjusted from level 1 to level 2, on June 13; the adjustment was lowered to level 3. On May 2, the emergency response level of public health emergencies in Hubei province was adjusted from level 1 to level 2. From May 14 to June 1, nearly 9.9 million nucleic acid tests were conducted in Wuhan, and no confirmed cases were found.

Meanwhile, according to the severity of the epidemic, the Chinese government splits up their counties (cities, districts) into low-risk areas, medium-risk areas and high-risk areas. High risk or medium risk areas can be very small units, such as a community, a street and a housing estate. The number is updated every day. As of February 17, China has dropped to 1 high-risk area and 5 medium risk areas.

As shown in Table 3, the specific prevention and control measures are adjusted according to the situation of provinces, counties and even communities, environmental capacity and the nature of the spread of novel coronavirus. “Low-risk areas” are areas where no new cases are confirmed for 14 consecutive days. “Medium-risk areas” have newly confirmed cases within 14 days, but the cumulative number of confirmed cases does not exceed 50, or the cumulative number of confirmed cases exceeds 50, and no cluster epidemic occurs within 14 days. “High-risk areas” refers to more than 50 cumulative cases and a cluster of epidemics occurred within 14 days. People can enter the “State Council Client” applet and perform the “Inquiry on Epidemic Risk Level” to show which epidemic risk level each person is in. For instance, on February 17, 2021, the data base showed 1 high-risk region and 5 medium-risk regions [34]. People are still required to go outside with a mask, take their temperature, and show a health code and a negative certificate of nucleic acid.

It has been questioned whether China’s official statistics represent the real situation. Some argue that the number of cases and deaths during the peak of the epidemic must have been much higher than presented in the official statistics [35], while others question that the disease could disappear completely from Wuhan [36]. While these arguments might be true, it is a matter of fact that China managed to bring life in Wuhan back almost to normal within a short period. A second wave with tremendous consequences for the public health situation could not have been hidden. Thus, irrespective of the reliability of the statistics basis, the measures taken by the Chinese government must have been quite successful in containing the outbreak and in preventing a second or third wave of outbreaks. In other words: The measures shown in Table 3 must have made it possible to keep the basic reproductive rate low.

Consequently, we have to analyze the parameters determining the dynamics of the spread of COVID-19 in order to understand how the second wave was prevented in Wuhan. For this purpose, we develop a simple and basic model predicting the spread of the disease in the region in order to analyze the influence of different parameters on it.

Modelling COVID-19 – an overview

An epidemic is terminated if the net reproductive rate (N_t) at a point of time t is lower than one, i.e., if every newly infected will infect less than one other person. N_t is the product of the basic reproductive rate (R₀, under the condition that nobody is immune) with the likelihood that the contact partner is not immune, i.e., at a given share of immune population (x_t) at a point of time t, N_t can be calculated as

If N_t is less than 1, a population has reached herd immunity (e.g. 60% for R₀ = 2.5).

Consequently, we have to analyze the dynamics of the diffusion of COVID-19 and estimate R₀ in order to assess the factors contributing to the success of interventions against the diseases in Wuhan. A huge variety of mathematical models has been developed to forecast the spread of a disease. The simplest approach calculates the basic reproductive rate as a function of some of variables (analytical models). As early as 1889, this model type was developed to calculate R₀ for malaria (reprint in English in 1989 [37]) and became the foundation of the well-known Ross-MacDonald model [38]. The disadvantage of these simple models is that they cannot cover interdependencies and changes of variables.

Homogenous Markov models are also widely used to forecast the spread of a disease [39]. They are capable to estimate the number of individuals in different health states as long as the transition probabilities are constant, i.e., if they do not depend on the number of individuals in the compartments [40]. This is the case for chronic-degenerative diseases, but the probability of being infected depends on the infectious population. Thus, traditional (homogeneous) Markov models are not applicable for infectious diseases such as COVID-19.

An inhomogeneous Markov chain implies that the transition probabilities can change. It is, in principle, a system dynamics model. This type of model was developed by Forrester in 1964 [41] in order to account for feed-back loops (e.g. number of infectious population determining the risk of being infected). The principle has been applied in many fields, such as “Industrial Dynamics “[41], “World Dynamics “[42], “Urban Dynamics” [43] or “Disease Dynamics” [44]. The simplest system dynamics model of a disease is the so-called SI-model where S denotes the population susceptible to a disease and I the infectious population. Figure 2 shows that the infection rate depends on the susceptible population (S), the infectious population (I), the contact rate (c), the total population (N) and the infectivity (i) of the infectious disease. The model can be easily enhanced to include the recovered population (SIR-model), exposed population (SEIR-model), different age-sets, re-infections, vaccinations, locations etc. The model has been applied to many infectious diseases, such as HIV/AIDS, malaria and cervical cancer [46–48].

Discrete Event Simulations (DES) and Agent-Based Simulations have also been used to predict the spread of infectious diseases [49, 50]. The advantage of these models is that they do not simulate compartments but individuals allowing to attach personal characteristics (e.g. being a child of an individual mother or having certain comorbidities) to each person. Thus, they are more precise, but require many input variables frequently unknown. In addition, designing and validating these models is much more effort than for the other model types.

In principle, a model should not be more complex than necessary to give an answer to the specific research question. For the target of this paper of determining the reasons for the successful fight against COVID-19 in Wuhan, a rather simple system dynamics model seems appropriate. There is a tremendous number of COVID-19 models available. Stegmaier lists 53 different models of COVID-19 relevant to German public health research, the majority of them system dynamics [51]. However, he also makes aware of the weaknesses of these models, such as poor data input, wrong assumptions, poor transparency, selective reporting etc. Ioannidis, Cripps and Tanner even state that “Forecasting for COVID-19 has failed” [52] because the results were frequently unreliable and of limited value for decision-makers.

The objective of the model presented in the next section is not to forecast the future development of COVID-19 in Wuhan. Instead, we would like to focus on very few parameters influencing the spread of the disease and analyse how they must have developed in order to allow the epidemic dynamics of COVID-19 in Wuhan. While many of the models presented by Stegmaier [51] are much more complex than ours, we also do not pretend to give a precise forecast. Our intention is a “modelling for insights, not for numbers” [53], i.e., we want to learn more about the prerequisites of the real spread of the disease than about the future dynamics.

Modelling COVID-19 – a basic model for Wuhan

For this purpose, we develop a generic COVID-19 model [54] in order to analyze the factors determining the spread of the disease in Wuhan in the first year. The model does not consider age-sets, locations or social differentiations (e.g. schools, universities, traditional markets) as this is not necessary to answer the question how China managed to avoid a second wave. Instead, we focus on the determinants of the basic reproductive rate R₀. The infection life cycle is presented in Fig. 3 and modelled as a System Dynamics Models [55, 56].

The system dynamics model defines difference equations for the healthy, infected, sick and immune population:

With the following variables and constants:

In comparison to other models [51], the infection life cycle and the number of compartments is rather simple, but we focus much more on the impact of contact rates on the basic reproductive rate. As (8) shows, R₀ depends on the infectivity (i.e. probability that one contact of an infectious person with a healthy person leads to an infection), the number of people an infectious person meets within the infectious period and number of contacts the infectious person has with each of the healthy persons.

The probability that an infectious person infects a healthy person when meeting once is p. The probability that an infectious person does not infect a healthy person when meeting this person n₁ times is ${\left(1-p\right)}^{n_1}$. Thus, if an infectious person meets m healthy people during the infectious period and has n_i contacts with each of them during this time, is the basic reproductive rate and can be calculated [57] as

For our analysis of the COVID-19 diffusion in Wuhan, we assume that the disease spread without interventions for a certain time. After d₁, the public health care system started interventions resulting in a reduction of R₀. However, it took some time until the rate had declined strongly. At d₂ all measures reached their maximum effectiveness and this condition was sustained until d₃. Afterwards the interventions were relaxed until the old situation was reached again in d₄. The respective development is presented in Fig. 4. This can be presented in the formula:

Consequently, the net reproductive rate (N_t) can be calculated as

Based on this model, we can simulate the diffusion of COVID-19 in a generic region with many characteristics of Wuhan. We simulate under the assumption that no intervention had been taken (scenario I), that the reduction of the basic reproductive rate was sustained (scenario II) and that a successful intervention was relaxed too early so that basic reproductive rate returns to its original value (scenario III). The last scenario assumes that some measures are sustained but R₀ will be above 1 (scenario IV). Furthermore, we can model the impact of different rates of infectivity (p), number of different contact partners (m) and contacts per partner (n_i). Under the assumption that an infectious person has the same number of contacts with each person (for n₁ = n₂ = … = n_m = n), we receive

or

Finally, we can calculate the basic reproductive rate under the assumption that the total number of contacts as the product of people met and contacts per person is constant (for m*n = k = const) as

Relevance of simulation results

Based on these calculations we can state that the public health care system in Wuhan managed to reduce the risky contacts strongly. The success of keeping R₀ under 1 for several months can only be explained by effective efforts to exclude infectious people from contacts.

The results also indicate that it was very difficult in the beginning of the epidemic to see its pandemic potential. Our simulations show that it took almost 3 months after the first case until 1000 patients were sick at the same time. It is obvious that the traditional routines of case detection (focusing on the number of cases) could not work with COVID-19. Once the figures are visible, it is already too late and the exponential growth has started. Only an excellent public health system could have determined the pandemic potential early enough.

Our results also indicate that it is likely that Wuhan had much more cases and deaths in the first wave than reported. Even scenario II results in 8269 death cases in the first year while Wuhan reported “only” 2997. The statistics of Wuhan have been questioned elsewhere [63, 64] and our computations show equal results.

The simulations also show that wearing surgical masks is highly effective to reduce the basic reproductive rate and the spread of the epidemic. Our simulation results are highly sensitive to changes of the infectivity, which can be strongly influenced by wearing masks. This instrument of protecting oneself and others has been quite common in China before, but it has become almost a universal habit since the pandemic started. Chinese citizens wear surgical masks, not only in public transport but almost everywhere. It has become a common habit as a population response to air pollution [65] and hardly anybody would see it as an insult to their liberty rights as citizens.

Finally, the simulation results show that a second wave can only be avoided if interventions are sustained. The reduction of the medical infectivity (p), the number of contact persons (m) and the number of contacts per contact person (_n) is the key to control the pandemic. It seems that China managed well to sustain a low R₀ by controlling these variables.

Geography

There was some discussion in the beginning whether Wuhan managed to control the pandemic because of the geographical location and the respective climate [66]. However, while other states located at the same altitude (e.G. Florida, Louisiana, Texas, Egypt) are facing a second wave, Wuhan has not reported Corona cases since March, i.e., the geographical location cannot explain the difference. Without doubt, spring and summer helped to control COVID-19 in Wuhan in 2020. There is a clear negative correlation between temperature and COVID-19 incidence, but for other parameters (e.g. humidity, wind speed, rain fall) the results are not significant [66]. It is likely that temperature does not have a direct impact on the transmission of the virus but increases the parameter m and n, i.e., during the cold season people have more and closer contact in rooms. However, this argument is true for all cities on the same latitude and does not explain the successful avoidance of a second wave in Wuhan. Geography does not explain this success.

Public health system

The simulation results also show that an early detection of cases and the implementation of early and effective control measures would require an excellent public health system. However, this does not seem to be the case. Instead, a number of shortcomings of China’s CDC have become visible during the epidemic [67]. Firstly, the communication between the national the local CDCs as well as with the healthcare facilities did not work well. Although an infectious disease information system had been developed after SARS-1, it did not work properly during phase I of COVID-19. Secondly, the CDC of China had a very limited influence on the Government. As early as January 6, 2020, the Chinese CDC initiated the second-level response to the epidemic, which was upgraded to a first-level response on January 15. However, these emergency responses were almost ignored by the Government [1].

Thirdly, the public health system of the middle-income country China suffers from low resources. As shown in Table 5, financial (health expenditure p.c.) and personnel resources of the system are much lower than in high income countries. In particular, the funds allocated to primary services have been declining for years (Fig. 10). The absolute amount of public health expenditure in China increased tremendously from 14.3 billion yuan (2.99 billion US$ or 2.63 US$ p.c.) in 1990 to 860 billion yuan (130 billion US$ or 617 US$ p.c.) in 2018. However, the proportion of preventive and promotive public services in the total public health expenditure decreased from 76.3% in 1990 to 52.5% in 2018. It seems that the Government of China puts less emphasis on prevention than treatment.

During the COVID-19 outbreak, the Chinese government has borne the cost of all confirmed and suspected COVID-19 patients. It is estimated that the respective costs amounted to 15.696 billion US$, mainly on treatment of patients, subsidies for epidemic prevention and control personnel, and purchase of equipment and protective materials. In comparison to the total health expenditure, the cost of the epidemic amounted to 1.65% of total public health expenditure (11.21 US$ per capita resp. 0.1% of GNP p.c.), i.e., a rather small amount. Consequently, neither a brilliant, well-financed and well-staffed public health system nor tremendously high health care expenditure are the key to understand the effective control of the Wuhan epidemic.

Portfolio of interventions

A number of analyses have been published that provide a taxonomy of different interventions against the diffusion of COVID-19 and assess their effectiveness. For instance, Baker et al. [71] listed the components of pandemic control of COVID-19. A comparison with the interventions of Table 3 shows that China has not implemented any measures that are not practiced elsewhere as well. Improvement of hygiene (e.g. hand washing, surgical masks), contact tracing, quarantine of sick and suspected, high volume testing, physical distancing, movement restrictions, and border management (incl. Exclusion and quarantine) are the international standards to fight COVID-19 [71].

Other studies analyzed the effectiveness of interventions in 40 countries. They record the strongest reduction of R₀ if gatherings of more than 5 people are banned followed by closing stores, restaurants, bars and schools [61, 72–74]. China implemented all of these intervention measures – so as many other countries that experienced a second wave. Consequently, it seems that there is no “magic bullet” against the pandemic; China has not implemented different measures, but it seems that the timing and intensiveness was different.

China follows a “zero-COVID” strategy. For instance, a recent breakout in Shijiazhuang (10.9 Mio. inhabitants) in Hebei province exemplifies this “no-tolerance against COVID-19”. After the public health system recorded 300 cases (i.e. 2.75 cases per 100,000), the full program shown in Table 3 started. The objective is clearly described by „zero-COVID “[75]. While European countries discuss whether interventions should be relaxed at a rate of 50:100,000, China implements its full portfolio at a rate of 2.75:100,000.

Without doubt, this is only possible with strong limitations of citizen rights. In particular, the Chinese intervention system builds on the App-based location analyses (see section 2.2). Every contact is recorded and access to gatherings is only permitted if the smart phone gives green light. This seems quite restrictive for Western societies. However, China is not alone in its “Zero-COVID” paradigm [76]. For instance, Australia [77], New Zealand [71] and Southern Korea [78] were quite successful in their eradication campaigns. New Zealand, for instance, never wanted to live with COVID-19, but eradicate it. When it started its campaign on March 23, 2020, the country just had about 100 COVID-19 cases and no deaths. As Philippe and Marques have shown for 11 G10 counties, countries following this strategy of early elimination are epidemiologically and economically more successful that countries pursuing a mitigation or suppression strategy [78]. This “go early go hard’ approach is exactly what China is doing – it seems to work even in a liberal Western society like New Zealand [79].

Finally, China invests efforts to vaccinate its population against SARS-Cov-2 [80]. However, there is evidence that the combination of limited coverage (i.e. share of population able and willing to be vaccinated) and effectiveness of the vaccine will now allow to reduce completely the other interventions [81]. A certain part of the population will not be vaccinated because they will refuse or because age and/or health conditions do not allow [82, 83]. Moreover, the effectiveness of the vaccine to prevent the spread of the disease might be less than 90%. Consequently, there will be (smaller) waves of COVID-19 after the vaccination program will have been completed. Consequently, the instruments described in Table 3 will still have to be employed for a longer time.

Limitations

The results presented in this paper are subject to a number of limitations. Firstly, we did not model and simulate the precise reality of Wuhan. For a detailed analysis we would have to distinguish age-sets, locations (e.g. city quarters) and social interaction (e.g. schools, work place etc.). Our model is generic, but it permits the conclusion that the public health care system of China managed to control the most important parameters (number of persons contacted and number of contacts per person).

Secondly, some of the data applied to the simulation are uncertain. For instance, as the real number of infections in Wuhan is unknown (and will most likely remain unknown for political reasons) it is difficult to assess the infection fatality rate (f). As Meyerowitz-Katz & Merone show [58], the parameter f strongly differs from place to place with an average of 0.68% and a highest estimate of 1.7%. We assume that the case and consequently the infection fatality rate was towards the higher end in Wuhan in January and February 2020 as no diagnostic and treatment standards had been developed for COVID-19 patients. However, we are aware of the fact that this is an assumption.

For scenario II, an f of 0.015 (see Table 4) results in 8269 death cases within the first year, an f of 0.02 in 10,745, an f of 0.01 in 5656 and an f of 0.005 in 2901 death cases. Consequently, the results react on changes on the parameters, but it is difficult to believe that medical care in Wuhan in the first months of the unknown diseases was as effective as health care systems that had months to learn how to diagnose and treat COVID-19 patients. Therefore, the simulation results might be challenged because of the uncertainty of input data, but the general finding that the number of death cases must be higher than reported is still valid.

Finally, the model presented in this paper does only present the situation in Wuhan in the first year of the epidemic. Consequently, we did not consider vaccination programs, temporary immunity or re-infections. As our objective was the analysis of the public health response in Wuhan in 2020, there was no need to include these aspects. Further research will have to focus on these issues much more.

Summarizing we can state that the model presented in this paper must not be used to predict the future spread of the disease. Instead, it is “modelling for insights, not for numbers” [53].

Authors’ information

None

Ethics approval and consent to participate

Not relevant

Consent for publication

Not relevant

Competing interests

None