How Kenya is tracking against our COVID-19 model

How Kenya is tracking against our COVID-19 model

Update 5th May 2020

We believe that the infection rate is tracking at worst against our ‘70% reduction in social contact’ scenario, and at best against our ‘90% reduction in social contact’ scenario. This means the peak will hit sometime between 1st August 2020 and 1st October with somewhere between 416k – 8.72m Kenyans infected. However, the virus seems to be less fatal than in other countries. Due to this we are lowering our estimation of health assets needed to overcome the virus.

Best case outcome
90% reduction in social contact
Worst case outcome
70% reduction in social contact
Infections as of 5th May3,70079,181
Deaths as of 5th May24288
Infections at peak416k8.75m
ICU beds needed6169,285
Ventilators needed4576,866
rescue.co dispatches1,727114

Predicting which scenario Kenya is tracking against

Key terms used

1. Cases: This is what is confirmed as a positive COVID19 case through PCR testing.

2. Infections: The number of total people who have the virus (symptomatic & asymptomatic). This number will always be much bigger than cases. 

3. Case Fatality Rate (CFR): The % of confirmed cases who die.

4. Infection Fatality Rate (IFR): The % of total people infected who die.


Part 1: A look at confirmed cases and how we could estimate infections

As of 5th May 2020, Kenya had 490 confirmed cases of COVID-19. Cases are only a small subset of total infections. This is for two reasons:

  1. Many infections of COVID-19 are mild or even asymptomatic and unlikely to be detected. This rate seems to be extremely high in Kenya. According to the daily situation report (May 2nd) from the Ministry of Health of 435 cases, 71% were asymptomatic.
  2. Lack of testing is also a challenge. Testing is only taking place for those with heightened symptoms, within quarantine and isolation facilities, or those who have come into contact with positive cases. As of 5th May 2020, Kenya has tested 22,897 people, an extremely low per-capita rate. KEMRI, the principal test laboratory facility has warned that it lacks the technical personnel and resources to continue testing and has asked for KSH 790 million in immediate financial support to restock.
CountryApproximate tests per million people (end of April)
Portugal35,321
Italy29,637
Germany24,738
USA17,211
Brazil1,597
India519
Kenya387
Nigeria24

Due to the high level of asymptomatic cases and low level of testing in Kenya we are not able to infer much from the confirmed cases as to how many actual infections exist in Kenya. It is very likely the real number of infections in Kenya may be significantly higher than the confirmed number of cases. 

Here are two alternative ways to try to estimate total number of infections: 

  1. Kennedy Odede, a grassroots organiser in Kibera, Nairobi recently undertook a controlled randomised testing of residents. Of 400 results, 3 tested positive for COVID-19 (0.75%). If we believe this is true for the whole of Kibera – a settlement with a population of around 2 million people – we can estimate that 15,000 people are currently infected within this specific area. Projected for Mombasa and Nairobi at the same rate, we can estimate that 42,037 have been infected. This research was conducted around 28th April 2020 and we project total infection rates have since risen by 88% based on the growth in reported cases at a national level. This means that total current infections could be around 79,181 in Kenya. Please note: As has been pointed out by our readers, there is high potential for ‘false-positives’ in this study due to the low results size, therefore we would encourage caution when extrapolating these results.
  1. According to Verity et al., the infection-fatality-rates (IFR) of observed COVID-19 infections in China was 0.66%. This means that 0.66% of infections in China resulted in death. Projecting to Kenya this means the country should have approximately 3,700 COVID-19 infections as of 5th May 2020. 
CountryInfection-fatality-rateRecorded cases% of pop. infectedIf applied to Kenya on May 5th. No. infected
1 case = 1 infection490490
Italy1.29% (Rinaldi et al.)4901,900
UK1.20% (Imperial College)4902,000
Germany1.17% (Imperial College)4902,200
China0.66% Verity et al.4903,700
Kibera randomised study4900.75%79,181

Given this we believe that infections range from 490 (confirmed cases) to 79,181 (infected infections from Kibera study). Clearly there is a wide discrepancy here. Our best estimate is that Kenya’s infection rate will be in-line with China’s and the infection-fatality-rate proposed by Verity et al. However, we cannot discount the possibility that the Kibera study’s findings are in-line with the current status, this therefore forms our outlier result.

Best case estimate no. current infections: 3,700 / Worst case estimate no. current infections: 79,181

Please note: At this stage this IFR is just a hypothesis. Early data from South Africa suggests that natural cause deaths are level with the same period in 2019. But, this data is rudimentary and does not account for specific mortalities within respiratory diseases and we are choosing not to over utilize this data at this time. Data on total mortality rates in Kenya has not been made available to us. If you have this information and would like to help us in refining these assumptions please get in touch.


Part 2: What we know about case fatalities and how we can better estimate the likely current and future fatality rates in Kenya

As of 5th May 2020 there were 24 confirmed COVID-19 deaths in Kenya. In our best estimate explained above we take this number as 100% correct. However, there is a strong possibility that this number omits some of the deaths that are actually occurring from COVID-19. It is hard to reliably track mortality rates in Kenya, particularly during periods of heightened mortality. According to the World Economic Forum, as many as 500 million people living in Sub-Saharan Africa have no official documentation. This makes tracking official mortality rates challenging. 

There are a few ways we can estimate deaths in Kenya: 

  1. Look at other countries and the rate of under-reporting: Many analysts in Nigeria believe the mortality rate in that country is significantly underestimated. Whilst the country has only recorded 1,182 cases and a national death toll of 35, local media reports from Kano province of hundreds of deaths in recent days have raised the possibility of major outbreaks of the lethal disease going undetected for weeks across Africa. Brazil, is one of the first countries to publish a peer reviewed analysis of the extent of COVID-19 mortality underreporting. Like Kenya it too has struggled with testing and has similar issues of mortality registration. Researchers there are arguing the true number of COVID-19 deaths may be 1200% higher than that being officially reported. As Kenya has performed 4x fewer tests than Brazil it seems likely that Kenya is underreporting the true number of deaths to a roughly similar extent. Given this Kenya should have 288 deaths by now (reported cases of 24 x underreporting rate of 1200%).

2. Look at other countries infection-fatality rates: 

CountryInfection-fatality-ratioIf applied to Kenya (no. deaths) on May 5th
Italy1.29% (Rinaldi et al.)1,021
UK1.20% (Imperial College)950
Germany1.17% (Imperial College)926
China0.66% Verity et al.522

The discrepancy between reported fatalities and expected fatalities based on other country’s Infection-fatality-rates is extremely large. Because of this we will use peer reviewed analysis of mortality underreporting to hypothesise that potentially up to 288 people have died in Kenya from COVID-19 in our worst case scenario.

Best case estimate no. current deaths: 24 / Worst case estimate no. current deaths: 288

Please note: It is hard to say with certainty why this rate is so much lower than that oversever in European and Asian countries. As we previously discussed it may be that Kenya has a significantly lower mortality risk profile for the virus due to its small ageing population when compared with countries who have already experienced the peak of the virus. Kenya 6.7x fewer people aged over 65 than Italy. Our projected infection-fatality-rate is around 4x times smaller than that recorded in Italy. It may also be that the high level of malnutrition and underlying health issues in Kenya such as existent levels of pneumonia do not seem to be correlating with higher COVID-19 mortalities. Both of these ideas are hypotheses and peer reviewed analysis is necessary.

Which scenarios are we currently tracking against?

Now, back to our model. How do we use these ranges of cases and deaths to best understand what scenario we’re tracking against? 

First, let’s make sure to review how our model works. In early April we published a projection for how many health assets would be needed to combat a potential outbreak of COVID-19 in Kenya. This projection looks at the potential needs of hospital beds, ICU beds and ventilators and was based on a Discrete Time SIR model of infection based on best known inputs for Kenya at the time. The model is projected on different scenarios which describe how much the country is able to reduce social contact through preventative measures.

We can now compare the real infections and expected infections at these different scenarios.

ScenarioNo. infections as of 5th May
90% reduction in social contact (NEW)3,000
Best case estimate no. current infections3,700
80% reduction in social contact15,022
70% reduction in social contact79,000
Worst case estimate no. current infections79,181
60% reduction in social contact384,636
50% reduction in social contact1,696,428

Clearly our worst case scenario is tracking against our original 70% reduction in social contact scenario. However, our best case scenario is tracking far lower than our original 80% reduction in social contact scenario, what we considered our best possible outcome at the time of drawing up the model. To cater for this we have created a new scenario that correlates to 90% reduction in social contact. 

Our ‘70% reduction in social contact’ scenario is one of the least severe of all possible scenarios for Kenya. It shows a peak in those infected with the virus on 1st August 2020 before a rapid decline. This scenario shows a severely delayed and diminished spike of the virus compared with scenarios where the Government puts in place fewer and less restrictive preventive measures. In total 8.72 million Kenyans will be infected with the virus at its peak.

Please note: We have adjusted this model based on how many deaths we now project to have happened (288) and how many should have been observed and which we previously based our model on (522).

Our ‘90% reduction in social contact’ scenario is extremely reduced . It shows a peak in those infected with the virus on 1st October 2020 before a slow decline. This scenario shows a severely delayed and diminished spike of the virus. In total 416,390 Kenyans will be infected with the virus at its peak.


Get in touch

For those tackling the outbreak

rescue.co is capturing important proprietary information during the coronavirus (COVID-19) outbreak in Kenya. We are keen to support public health organisations and companies tackling this outbreak. If you are interested in learning more about this model and/or our coronavirus cases and data, please contact info@rescue.co

Cover you and your family now

We are expecting and projecting a major outbreak of coronavirus (COVID-19) in Kenya over the coming weeks. To ensure you and your family are covered for free quality ambulance response (24/7) please visit rescue.co to sign-up now.

Donate to help us fight COVID-19

Help us make sure that all EMTs in Kenya have proper PPE (Personal Protective Equipment) and training. 100% of your gift will go directly toward training and equipping first responders or toward a rescue or safe transfer for somebody who couldn’t otherwise afford it. Click here.

2 COMMENTS
  • Rob
    Reply

    Outstanding update from the Rescue team! Thank you so much for this work. A few thoughts:

    1) I agree that one thing we’ve learned is that the true infection fatality rate is lower than originally thought. I believe that is true worldwide, as we have discovered more about asymptomatic infection, and I believe it is particularly true in Kenya where we have the vast majority of the population under 50 years old.

    2) I agree the true infection rate is likely much higher than the number of confirmed cases. The ratio you propose is ~160x (79,181/490). That is quite a substantial ratio, which I believe you have come up with based on prevalence from the Kibera study. The trouble with estimating prevalence from such studies is the problem with test characteristics (sensitivity and specificity).

    To give a concrete example, if the Covid-19 PCR test had a specificity of 99% — which would be extremely good — then in a sample of 400 people, by definition, we would expect 4 False Positives. In the study you cited, they only had 3 positives, which we would actually expect, according to statistics, would all be false positives. In other words, a biostats analysis of that Kibera study, assuming a specificity of 99%, would place the true number of infections in Kibera at 0. There are error bars of course, but the central point around the error bars would be 0.

    The trouble with estimating prevalence from these studies is that extremely small numbers (like 3 positive tests, which we actually would expect would happen given a very highly specific test), end up having outsize effects on prevalence estimates. There is a terrific blog post that discusses this concept: https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaws-in-stanford-study-of-coronavirus-prevalence/

    3) I also agree that deaths are likely under-reported, not necessarily through anything malicious, but because it is difficult to ascertain the cause of death, especially if there is insufficient testing or people are avoiding medical facilities for stigma, fear, or economic reasons.

    Glad you are continuing the conversation, and hope to hear any additional thoughts you have on the matter!

    1. Hugo Winn
      Reply

      Hi Rob, sorry for the late reply. You make some fantastic points and we have adjusted the assumptions in this update to note the potential error in the Kibera study. Thankyou for sharing your thoughts and we look forward to sharing new developments with you soon.

Leave a Reply to Hugo Winn Cancel reply

Your email address will not be published. Required fields are marked *

0

Your Cart