Covid Undercounting in Africa?

It would seem that Africa has been spared the worst of the pandemic, with most countries on the continent reporting much lower case counts and fatalities than Europe and the Americas. Why? Maybe it’s the climate — the virus seems to spread fastest in cold dry conditions. Or maybe the continent’s relative isolation has buffered it from travelers carrying the virus with them from harder-hit areas of the world.

Now a couple of studies suggest that the situation in Africa is worse than the numbers would indicate.

The other day a friend sent me the draft of an article summarizing a seroprevalence survey, conducted in Addis Ababa Ethiopia in late April, that found 8% covid prevalence. The official count as of late April showed only 125 test-positives cumulative and 1 covid death in the whole country, which has a population of 112 million. Could the antibody tests the researchers used have been grossly inaccurate? Not according to the validation data reported in the abstract. There certainly were places in the world that had prevalence levels of 8% or higher in late April, and Addis Ababa’s crowded conditions could certainly have accelerated contagion. So it’s not out of the question that the findings reported in the abstract accurately reflect the situation on the ground at the time.

A few days later I came across another study, published in the 1 January online issue of Science. This Kenya study is based on analysis of more than 3,000 blood donor samples collected between 30 April and 16 June 2020. They found a seroprevalence of 5.6% — comparable to the Ethiopian seroprevalence survey. The much lower official tallies for Kenya as of mid-June are very similar to those of Ethiopia. This Kenyan study also found higher seroprevalence in the big cities, again comparable to the Addis Ababa study.

The Kenya study authors point out that covid mortality is much lower than might be expected for that level of infection because covid is fatal mostly for older people, and older people comprise a much smaller percentage in Kenya than is the case in the US and Europe. In my most recent post I updated my age-adjustment algorithm to take into account countries with very young populations. For Ethiopia and Kenya, both with median population age of 20 years, the age-adjusted covid fatality rate is 0.06 percent — a fraction of the US’s 0.65 percent and lower than the influenza fatality rate. 

It’s certainly possible that in the early days of the pandemic nobody in these countries was recognizing covid infections. The official counts in both countries started climbing in late June — maybe that jump is attributable not to a spike in contagion but to the onset of more accurate diagnosis and reporting. Alternatively, the official case and death counts from Ethiopia and Kenya might continue to be grossly underreported, in which case the infection rates for those countries might be among the highest in the world rather than among the lowest.

For comparison’s sake, the IHME’s estimated infection rate for Ethiopia is very close to my own, suggesting a comparable age-adjustment factor. However, they rely on the official death tallies in estimating the cumulative infection percentage, which might represent a significant undercount.

So it’s a matter of gradually closing in on better estimates and reducing the uncertainties, to which both of these seroprevalence studies contribute significant information. Hopefully vaccination will start drastically reducing the counts everywhere.

International Covid Fatality Rate Age Adjustment Algorithm

In estimating national infection rates, I’ve been using death rates as a lagging indicator, adjusted based on the median age of the national population. Compiling available published evidence, I’ve estimated that the covid fatality rate for the US is 0.65%. The median age of the US population is 38 years. Based on several seroprevalence studies conducted in the US and Europe, I’ve been age-adjusting other countries’ estimated fatality rates according to this formula:

  • For countries with a median age higher than the US, multiply 0.65% by 1.1 raised to the Nth power, where N is the number of years that the country’s median age is above 38 years.
  • For countries with a median age lower than the US, divide 0.65% by 1.1 raised to the Nth power, where N is the number of years that the country’s median age is below 38 years. 

This age adjustment factor works well enough within the relatively narrow median age range of economically developed countries. However, many countries in Africa have much younger populations; e.g., the median age is 20 in both Ethiopia and Kenya.

Using the CDC’s demographic data on covid-related deaths, it’s possible to calculate more accurate age adjusters for fatality rates. A couple of simplifying assumptions:

Using the CDC data, here are the age-adjusted fatality rate “power multipliers” for each ten-year age interval:

  • 15 – 24 years = 1.25
  • 25 – 34 years = 1.1
  • 35 – 44 years = 1.1
  • 45 – 54 years = 1.1

So the geometric increase in fatality rate by age is 1.1 per year across the 30-year range from age 25 through 54 — consistent with my algorithm. In the 15 to 24 year age range, however, the rate of increase is faster. Therefore I need to adjust my power multiplier accordingly:

  • For countries with a median age lower than the US, divide 0.65% by 1.1 raised to the Nth power, where N is the number of years that the country’s median age is between 38 years and 25 years. If the country’s median age is lower than 25 years, (a) divide 0.65% by 1.1(38-25)=13 or 3.45 –> 0.19%, then (b) further divide that number by 1.25 raised to the Yth power, where Y is the number of years that the country’s median age is below 25 years.

Based on this revision, the age-adjusted covid fatality rate for Ethiopia and Kenya, with median population ages of 20 years, would be 0.19/1.255 = 0.06%. That’s less than one-tenth the covid fatality rate of the US, and half the fatality rate calculated using the original simplified power multiplier.