A response to some questions about our COVID-19 model

A response to some questions about our COVID-19 model

On the 2nd April we published the results of a projection exercise to estimate the potential level of healthcare resources needed to deal with the Novel Coronavirus (COVID-19) in Kenya. This model was primarily created to help those at the frontline of the public health response to COVID-19 in Kenya estimate and seek resourcing for extra hospital beds, ICU facilities, ventilators and EMS assets. We have been extremely excited by the level of interest in our model and wanted to take a few moments to address some of the questions, concerns and potential alternative inputs that you have raised.

Click here for the full predictions.

If you have any questions about this model or would like to access the raw logs, please email info@rescue.co


Why have you chosen these scenarios, what do they mean in real terms?

It is hard to accurately determine the effect of quarantining efforts put in place on epidemiological outcomes. Our model quantifies these outcomes in terms of the % reduction in social contact in several different scenarios. This is based on the model built by researchers at the Penn Medical Predictive Healthcare Unit called the CHIME model.

This article attempts to quantify the % reduction in social contact in terms of specific measures. These reductions are on par with research by Imperial College which produced specific scenarios of social contact reduction, these results can be seen below and are roughly similar to the levels presented within our scenarios.


Kenya has responded faster than most countries to the virus, how is that integrated into your model?

So far the Kenyan response to COVID-19 has been extensive. As of the 7th April (at the time of writing) Kenyan airspace was severely restricted, travel across county lines illegal and curfews preventing intra-county travel implemented between certain times. The country is yet to impose a complete quarantine or ‘lockdown’ restricting movement to all but essential travel, but it will likely do so within days. This is despite the country having significantly lower mortality figures than comparable countries when they imposed their lockdowns. Given this it is clear that the Kenyan government has acted significantly faster and with more severity than countries such as the UK or US.

Days since first confirmed case reported

CountryDate of first caseBan on large gatheringsInternational travel control/banInternational travel banEnforced mask useTotal lockdown
Kenya13 Mar0112523N/A
Italy31 Jan22020N/A39
Spain1 Feb343537N/A37
UK31 Jan43N/A53N/A53
US15 Jan6165N/AN/AN/A
South Korea20 Jan810303030
Wuhan, China31 Dec2024242324

For more detailed breakdown of the speed of government preventive measures around the world, please consult this blog at Brookings.

This excellent and widely cited blog demonstrates the need for early, draconian social distancing measures akin to those undertaken in China or Singapore. It clearly shows the impact of immediate social distancing measures which significantly reduce the peak of the viral curve. If Kenya is successful in enforcing this, not only should the country track against our 80% reduction in social contact scenario, but will be potentially manageable from a health response perspective.  

As we have mentioned above there are significant challenges for Kenya to successfully enforce it’s quarantining policies, not least the need to subsidise a significant proportion of the populations salaries for at least 1-2 months, but it is possible with adequate planning and creative solutions for basic service distribution and access.

Our model does not use guessed R rates, but instead uses real data about the spread of the virus and recorded cases/mortality within Kenya to constantly update it’s R factor. We believe this is a better system than estimating the potential R factor based on predicted protective measures. We will continue to review the R factor based on the strength and effectiveness of Kenya’s protective measures.


What % reduction in social contact is actually happening in Kenya?

There are various interesting data-sets being developed to help quantify the reduction in social contact in contained environments around the world. Perhaps the most often cited currently is the Google Global mobility Index, a rich data-set that uses Google Maps data to determine the reduction in ‘normal’ levels of mobility post COVID-19 protective measures.

This index currently shows the following reductions:

  • Retail & recreation: – 45% on baseline levels
  • Grocery & pharmacy: – 33% on baseline levels
  • Parks: – 20% on baseline levels
  • Transit stations: – 39% on baseline levels
  • Workplaces: – 22% on baseline levels

This data is hard to interpret in terms of overall reduction in social contact. Perhaps a good indication of overall contact reduction is the reduction in transit stations, which are used by a cross section of society. When comparing the reduction in this metric with other countries experiencing COVID-19 it seems that Kenya has not had significant success in reducing social contact. By comparison the UK has reduced transit station mobility by 75%. 

Based on this we estimate the current % reduction in social contact to be tracking between 20 and 50% in Kenya.


Kenya has a warm climate, will COVID-19 react similarly to countries with colder climates? Is the reproductive factor the same?

A common question amongst readers has been whether the warm year-round weather in Kenya will play a role in slowing the reproductive rate of the virus. Many of the largest outbreaks have been in regions where the weather is cooler, leading to speculation that the disease might begin to tail off with the arrival of summer. 

The simple truth is that because COVID-19 has not been present for a full seasonal passage we do not yet know the endemic qualities of the virus. The closely related Sars virus that spread in 2003 was contained quickly, meaning there is little information about how it was affected by the seasons.

There are some early hints that Covid-19 may also vary with the seasons. An unpublished analysis comparing the weather in 500 locations around the world where there have been Covid-19 cases seems to suggest a link between the spread of the virus and temperature, wind speed and relative humidity. The study described how the virus seemed to spread fastest at an ‘optimal’ temperature range of 8.07 degrees centigrade, this is similar to that found by Wang et al. which is 8.72 degrees centigrade. In comparison Kenya’s daily temperature in April ranges from 13.4 to 25 degrees centigrade. This is outside of the optimal range proposed by the researchers. Another unpublished study has also shown higher temperatures are linked to lower incidence of Covid-19. However, the effect of temperature on COVID-19 reproductive rate was meagre. The R values with and without temperature were 0.44 and 0.39, respectively, indicating that the inclusion of the temperature effect only provided a relatively modest slowing of transmission rate. This study also notes a notable outlier in the Daegu province of South Korea which has a high average temperature and has experienced an extremely high rate of transmission

However, it is important to be cautious about this research. Endemic viruses are seasonal for a number of reasons that might not currently apply to the Covid-19 pandemic. Pandemics often don’t follow the same seasonal patterns seen in more normal outbreaks.

There is also research to the contrary. A recent analysis of the spread of the virus in Asia by researchers at Harvard Medical School suggests that this pandemic coronavirus will be less sensitive to the weather than many hope. The reality of all this research is that COVID-19 is so recent that there is not a substantive enough research base to definitively prove a relationship between environmental factors and transmission rate. Once such information becomes available we will adjust the inputs in our model to reflect it. 


What is the level of testing in Kenya and what does that mean for the model?

As of Monday 6th April there have been 4,277 tests taken place in Kenya. Of these 158 people have tested positive. It is clear that there is a substantial challenge for testing for COVID-19 in Kenya. Kenya has performed one of the lowest levels of tests in the world, a result partly attributable to the late on-set of the virus in the country, but also due to the limited number of labs currently approved to test. The Kenyan Government is currently estimating it has the total capacity to do 300 tests per day at the national lab and 5000 per day at the KEMRI labs via a Cobas 880 automated testing machine. To improve this rate the government must urgently construct additional labs with PCRs and introduce RDTs.

Kenya is currently testing patients within official quarantine facilities, at designated facilities such as Mbagathi and Aga Khan, this means the country is probably missing a significant % of infections occurring in the wider community.

It is never possible to measure the exact number of infections within an area, but accuracy increases with increased testing. Researchers can use data from places that conduct a significant amount of testing to estimate the total number of infections based on confirmed COVID-19 cases, deaths, and recoveries. Researchers from Imperial College have produced perhaps the most comprehensive analysis of infection rates in Europe. They estimate that on March 28th, 5.9 million people in Italy had been infected with the virus and 16,523 had died, meaning that 0.28% of infections resulted in death. Using this methodology, and assuming the mortality rate is the same in Kenya as in Italy, we can extrapolate that on April 6th, when 4 people had died in Kenya from the infection, there would have been a total of 2,150 infections in the country. Limited testing and asymptomatic carriers would account for the fact that only 158 cases had been officially confirmed on that date.


Kenya has a high disease profile, how will that impact fatalities?

A very important factor in the Case Fatality Rate (CFR) with COVID-19 and other coronaviruses is the comorbidities in the underlying population. Researchers in China have found a significant adjustment, after adjusting for age and smoking status for a variety of comorbidities including diabetes, hypertension and malignancy. Comobitieis of specific concern in Kenya include:

  1. Conditions that can cause a person to be immunocompromised, including cancer treatment, smoking, bone marrow or organ transplantation, immune deficiencies, poorly controlled HIV or AIDS

Kenya has an average HIV prevalence rate of 6% and with about 1.6 million people living with HIV infection, it is one of the six HIV ‘high burden’ countries in Africa. This will put significant and specific stress on Kenya’s healthcare system during the COVID-19 crisis.

Along with other low-middle income countries the prevalence of cancers is lower than in higher income countries as a proportion of the population. According to the WHO there are currently 86,592 cancer patients in Kenya with 19,199 added each year. This population is at significant risk of morbidity/complications.

  1. Chronic kidney disease and who are undergoing dialysis

Amid rapid urbanisation, the HIV epidemic, and increasing rates of non-communicable diseases, people in sub-Saharan Africa are especially vulnerable to kidney disease. 2017 estimates show that 4 million Kenyans have chronic kidney disease with a significant proportion of this population progressing to kidney failure. Out of these, about 10,000 people have end stage renal disease and require dialysis.

  1. Liver disease

According to the latest WHO data published in 2017 Liver Disease Deaths in Kenya reached 5,214 or 1.85% of total deaths. The age adjusted Death Rate is 23.04 per 100,000 of population ranks Kenya #62 in the world, putting it significantly above countries like the US and UK currently experiencing the virus at its peak. 

  1. Respiratory diseases and illnesses

Pneumonia is a serious public health problem in Kenya. Although the percentage of children routinely vaccinated against some strains of the illness is increasing, reaching 84 percent by 2018, it still poses a significant risk in the country. In 2017, pneumonia was responsible for 21,584 deaths according to the Economic Survey 2018, accounting for 22% of deaths, and standing as the leading cause of the death for the third year in a row. In Kenya, about 700,000 cases of pneumonia in children under the age of five are treated every year. Although children are significantly less susceptible to the health implications of COVID-19 this level of pneumonia in the baseline population is concerning.

Tuberculosis remains high in Kenya, and experts say the country lags in the fight against the disease. According to a 2018 TB study there are 558 per 100,000 adult population. While experience on COVID-19 infection in TB patients remains limited, it is anticipated that people ill with both TB and COVID-19 may have poorer treatment outcomes, especially if TB treatment is interrupted.

It is difficult to accurately bake these comorbidity rates into our SIR model, therefore we have excluded at the moment. However, we would recommend that you consider our estimates to exclude, rather than include, this increased risk.


Kenya has a small eldery population, how does that affect the hospitalisation rate?

First, it is worth explaining the basic epidemiological model we have used for this exercise. There are several models to choose from, but we opted for the Discrete Time-Based SIR model of virus transmission in a closed population over time. This is the orthodox epidemiological model and is being used by the majority of modellers around the world to assess the spread of COVID-19 within closed populations. You can read more about the SIR model here and for full documentation of the exact SIR model we used for our projections you can find out more here.

One challenge with a SIR model is that it treats the susceptibility and recovery rate of people with the virus the same across the population. As such it does not take the demographics of a population into account within the dynamics of the model. Instead users of SIR models bake these factors about the population into the inputs of the model: e.g. the % of infections that result in hospitalisation, the % of infections that result in ICU usage and the % of infections that result in the need for mechanical ventilation.

There is significant evidence that the older a population is the higher the mortality rate and utilisation rate of health assets will be. We now have data on over 1.2m confirmed COVID-19 cases and the result is conclusive. The overall death rate from COVID-19 has been estimated at 0.66%, rising sharply to 7.8% in people aged over 80 and declining to 0.0016% in children aged 9 and under. Other studies, such as that focused on the UK by Imperial College researchers have found the rate rising to 27% amongst over 80 year olds.

Many of the countries currently most affected by COVID-19 have eldery populations. Around 23% of Italy’s population is over 65 years of age. In comparison Kenya, as with most other Sub-Saharan countries has a proportionally smaller ageing population, at around 3.4% of the population. This should mean that the overall % of infections requiring hospitalisation will be significantly smaller in Kenya than Italy. 

We have attempted to take this into account in our model. In a comprehensive assessment of the infection-hospitalisation rate in China researchers found that the rate ranged from 0.04% to 18.4% depending on the age of the patient (Verity et al). This study is rigorous and we have used it to weight for Kenya’s small eldery population and have aggregated to assume that 5% of infections will require hospitalisation. Other averaged infection-hospitalisation rates (IHR) include:

  • Imperial college analysis of UK cases and projections: 9.78% IHR
  • Verity et al analysis of China cases and projections: 7.96% IHR
  • Rescue.co analysis of Kenya risk profile and demographics: 5% IHR

One interesting point is that the study by Verity et al, assumes that only ‘severe’ COVID-19 cases will require hospitalisation. We do not yet know whether this will be the case in Kenya or how COVID-19 will react with coexisting medical conditions within the population. Life expectancy at birth in Italy is 83.1 years (2017) whilst Kenya’s is 66 years. Our model does not take these health factors into account and should be viewed with a +/- 5% infection-to-hospitalisation rate variability as a result.


What other factors might influence the R factor in Kenya?

The reproductive or R factor of COVID-19 is determined on far more factors than the virus’ ability to survive different environments. Probably of far greater significance is the level of social contact and density of individuals within a closed environment. On these factors Kenya is potentially in a challenging position. Whilst Kenya has a small population density overall, in it’s cities it is a different matter. 60% of Nairobi’s residents live in informal settlements, with population density often as much as 300,000 people per square kilometer. We believe these conditions will influence the reproductive rate of the virus far faster than any possible temperature factors.

It is also important to note that the SIR model we have used to compute the R factor of the disease computes based on all known data on different reproductive rates and compares that against the known history and growth of cases within Kenya. Because of this we will continue to adjust the model’s outcomes over time, but believe this offers us a better estimation than manual input. The current doubling time of confirmed cases in Kenya compared with other countries in the region is as follows:

  1. Guinea: 2 days
  2. Cameroon: 3 days
  3. Congo: 4 days
  4. Kenya: 5 days
  5. Ethiopia: 7 days
  6. Nigeria: 7 days
  7. South Africa: 11 days
  8. Tanzania: 15 days

Source (Our world in Data)


Is this model conservative or overly fatalistic?

As discussed above there are several reasons why our model might be overly fatalistic. It is possible that the disease will indeed react to warm climates significantly differently, it is also possible that Kenya’s small eldery population will have a more significant effect on hospitalisation than expected. 

But, there are also several reasons why we should view these figures as the best case situation. Containment and quarantine in Kenya will be extremely challenging. To properly slow a virus people must either be forced or highly incentivised to self-isolate. In Kenya neither of these options are possible:

To force those to stay inside people must have accessible basic services. 41% of Kenyans still rely on unimproved water sources, such as ponds, shallow wells and rivers. Simply put they must leave the home for essential services. To incentivize Kenyans to stay inside the government would need to pay the wages of furrowed workers on an unprecedented scale for several months. Estimates of the UK scheme to pay 85% of furrowed workers estimates costs for that scheme could be as high as $96 billion dollars. Assuming Kenya were to furlough workers and pay 85% of monthly wages akin to the UK and assuming that 60% of workers are furloughed under the scheme we believe that Kenya will need a KSH 2,248,763,060,000 ($21.2bn) support package for workers. 


What does your model say about mortality?

Our model does not compute mortality rates. This is deliberate. Instead, the model shows the peak hospitalisations, infections, ICU admissions and ventilations required to stabilize the infected population on a given day. This helps public health agencies and hospitals to plan under several case-load scenarios. 

Our model therefore is not measuring the TOTAL number of people that need ventilation or ICU facilities over time (from which one could infer mortality), but rather the PEAK number of people who will require ventilation on a certain day. This is a more useful number when predicting the number of health assets that need to be purchased, but clearly prevents us from making an inference about mortality.

At a very crude level you can infer from this information the potential mortality based on the number of health assets within the health system. For example, if Y people need mechanical ventilation on a given day, and only X ventilators exist within the population, it is reasonable to suggest that Y-X might die. But, this isn’t correct, as many people require ventilation for multiple to multiple subsequent days (the COVID-19 average being 9 days). The Y people requiring ventilation on any given day includes those who are already on ventilators from previous days and need continued ventilation. Because of this, many or all of the ventilators may be continuously occupied by existing patients for several days.

When estimating the potential mortality rate in Kenya during COVID-19 we recommend referring to other models such as this extremely helpful Case Fatality Rate calculator.


What’s next?

We will continue to review our model and will continue to input real data about cases/mortality in Kenya to maintain its accuracy. Any major revisions or outcomes will be posted on this blog and disseminated to Rescue.co members.

We are working to ensure Kenya’s EMS system has capacity and is prepared for the week’s ahead. We are purchasing PPE equipment in bulk ensuring our front-line health workers are protected and are able to do their job. If you are interested in donating to these efforts please click here.

As the largest digitally connected ambulance network in sub-saharan Africa we have the best and most extensive real-time data on emergency response resources, the widest network of first responders (private and public) on a single hotline and a comprehensive set of COVID-19 training materials and programs in Kenya. We are looking to contribute as much as we can to national planning and response efforts. Please get in touch with any relevant connections or interest at info@rescue.co

1 COMMENT
  • Rob
    Reply

    Great updates to the model. It’s great to see how quickly Kenya has reacted compared to other countries.

What do you think?

0

Your Cart