Calculating the mortality rate of a pandemic, e.g. coronavirus 2019-nCov?

Question

When calculating how lethal a virus outbreak is, I've noticed that most sources use

CONFIRMED_DEATHS/CONFIRMED_INFECTIONS

However, given that there are only 2 possible outcomes; death or recovery, wouldn't it be more logical to use:

CONFIRMED_DEATHS/(CONFIRMED_DEATHS + CONFIRMED_RECOVERIES)

Which formula is applicable to which situations?

Answer 1

This may not strike most as a bioinformatics, but getting the key clinical outcome is essential in understanding the molecular basis of pathogenicity.

I think the mortality rate is over-reported. This is not to say the situation of 2019-nCov is not serious - it is very serious.

The two essential factors missing in your equation for 2019-nCov are:

• Age. Mortality is likely skewed towards the old and very young
• Seroprevalence. To identify the number of asymptomatic patients.

Old people have a weakened immune system (e.g. reduced T-cell response). The very young have an underdeveloped immune system. The problem is hospital cases are skewed towards those with severe infection, which is likely the old and very young. In 2019-nCov the first infecteds were middle aged men, but likely to be restricted to those exposed to the "Wuhan meat market".

Problem Asymptomatic "spreaders" are known in influenza and are suspected in 2019-nCov. Asymptomatics will inflate the mortality rate (because we're not detecting them).

Solution(s)

1. One solution is to conduct seroprevalence surveys. Serological surveys screening for antibodies (IgG) against 2019-nCov will identify asymptomatic patients, using the first formula you stated. However, would you seriously go around the general population in a 2019-nCov epidemic taking blood samples? Errhhh no.
2. The bio-statistics approach is to use a catalytic model and critical to this is the age of the patient. Catalytic models are a priori models that incorporates the age-prevelance profile of infection. By introducing the incidence and age of hospitalisation of e.g. the old and the young, the rest of the distribution, i.e. asymptomatics can be calculated by regression. This estimates the total population of infecteds and again you use equation 1 to calculate the mortality rate recorded by the hospital.
3. Mathematical models based on transmission dynamics. The problem here is the rate of transmission is unknown.

I would assume the hospital is the place virus-induced mortality is reasonably accurate.

The second formula you described works very well for Ebolavirus, no-one has ever heard of asymptomatic Ebolavirus. It would also work well for another betacoronavirus Middle Eastern respiratory syndrome. The key problem with 2019-nCov is we don't know the number of asymptomatic patients or patients with such mild infections they never report illness.

Summary In summary, I think the mortality is being over-reported if it is based on your equation 2. Equation 1, however, will also over-report the mortality rate.

Answer 2

To add to Michael G.'s answer, there is a further wrinkle in the case of 2019-nCov that is due to the exponential growth in confirmed infections daily.

That growth affects your second formula CONFIRMED_DEATHS/(CONFIRMED_DEATHS + CONFIRMED_RECOVERIES since deaths probably occur earlier on in the disease trajectory than recoveries do. Since the number of confirmed cases is growing exponentially on a daily basis, the number of deaths will presumably be inflated relative to the number of recoveries at this point.

This feature may have the opposite effect on your first formula CONFIRMED DEATHS/CONFIRMED INFECTIONS, depending on when the infections are confirmed. If the infections are generally confirmed several days before death occurs, then a large portion of confirmed infections are still too early to have ended in death and CONFIRMED DEATHS/CONFIRMED INFECTIONS actually underestimates the actual number of fatalities.

Answer 3

I was facing this problem in January when coming up with a methodology to forecast on the UrbanSurvival.com site.

My solution - using low data collection was to use a daily rolling rate of infections and divide that into the 5-day previous reported death counts.

This was based on the assumption that, for a given cohort, the people who get sick today will "get dead" in about 10-days time.
So half of this (5 days) ought to get me into the ballpark.

In Feb, when CDC was reporting 3.4% mort. rate, we'd been reporting that for a couple of weeks.

As a formula it's not unlike deriving an exponent.

That is: Cases Today (T1) / Deaths 5-Days Ago (T2)

Unfortunately, this is presently running a bit higher mortality rate now (4.5%) but I would expect on the backside of the expansion of cases it will decline.

Here's hopin'...