When new diseases appear, the most serious cases and the majority of deaths tend to take place in hospital. Early data records the number of patients admitted, the treatments that they are given, those that have recovered and those that die. This gives a hospital case fatality rate (CFR). In a study published in the Lancet of the novel Coronavirus, now known as COVID-19, in the first cohort in Wuhan the CFR was 11%1.
There are two main reasons why this data can be unreliable when projected forwards to different populations.
1. The numerator and denominator issue
Mortality rate = Number of deaths/number of cases of infection
Death is the numerator. Death is hard data. With established and known medical conditions, it is easy to measure the number of people that have died and relatively easy to know the number of people that have been diagnosed and admitted to hospital.
One dilemma in new infectious illnesses is the lack of available and reliable tests. It takes time for these to be developed. If we only make a diagnosis on cases with a positive test we may miss some people who have died without having the diagnosis confirmed. Death is also lagging data, 10-12 days in the China studies so early mortality rates may also underestimate cases which will die later. Both of these factors could make early mortality rates underestimate the real mortality from newly evolving disease.
How about the number of total infections? This is the denominator. This is soft data. Because it takes time to develop cheap and widely available tests there is an inherent bias in the people that are tested. People with less severe symptoms and those who do not go to hospital are less likely to be diagnosed. The HKU study suggested that there may be many thousands of people in the community in Wuhan with some degree of illness2. Obviously the hospital mortality in this situation will significantly overestimate the true severity of the disease.
China has changed the criteria by which they report this disease. They have moved away from reporting only positive tests and are now using clinical diagnosis. This makes sense. It recognises the reality on the ground, that universal testing will not be possible until a more readily available test is developed. The sudden surge in cases must be interpreted in this context. The death rate has doubled but the incidence has increased by a factor of 10. Counterintuitive but this reduces the case fatality rate by a factor of five and may explain some of the higher mortality in Hubei. There may also be other factors which need to be considered.
The crude fatality rate Worldwide excluding China is 0.9%. Although this dataset is small, it is evolving and because it includes contact cases it is likely to give a better indication of the severity in population terms.
2. Viral Load
A second problem with projections based on early cohorts is the assumption that the disease will always behave in the same way in future clusters. One of the reasons that the AIDS specialist was so wrong in the 1980s related to the issue of viral load. We now know that during the AIDS epidemic, the first clusters were infected with a large amount of virus. Haemophiliacs had high levels of virus in blood transfusions and the bathhouses of the LGBT community at that time led to repetitive exposure to infected individuals. The result was that people died very quickly and this data was projected forward.
It took time to appreciate that there were factors in the first population which were different to future clusters. The early clusters basically had more serious disease because they were infected with more virus. We know that during SARS viral load was associated with severe disease and this is the case in many other infectious illness.
We have no evidence yet that viral load has had an impact on COVID-19 infection. The principle is that we must always be cautious in projecting the behaviour of a disease in one population into subsequent clusters. Measuring mortality early in epidemics is notoriously difficult. As a general rule mortality rates early in epidemics tend to overestimate ultimate mortality and it is common to see the severity of the illness down regulated over time. Furthermore, whilst it is possible for viruses to mutate and become either more infectious or more severe in general it is statistically more likely for coronavirus infections to become less severe over time.
As always better data at least gives the option to make better decisions and the nature of this epidemic (epidemic doubling time approximately 7 days) means that we will get a doubling of our data every week.
- Chen, N., Zhou, M., Dong, X., Qu, J., Gong, F., Han, Y., Qiu, Y., Wang, J., Liu, Y., Wei, Y., Xia, J., Yu, T., Zhang, X. and Zhang, L. (2020). Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study. The Lancet.
Wu, Joseph T, et al. “Nowcasting and Forecasting the Potential Domestic and International Spread of the 2019-NCoV Outbreak Originating in Wuhan, China: a Modelling Study.” The Lancet, 31 Jan. 2020, www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)30260-9/fulltext.