⇓ More from ICTworks

4 Ways to Use National Data in African COVID-19 Digital Response

By Guest Writer on April 23, 2020

covid 19 data malawi

At Cooper/Smith, we are working with Malawi’s Ministry of Health to help them prepare for and fight COVID-19. Here is what we did – other countries might find this useful as well:

1. Identify the most important information

As COVID-19 began to spread in neighboring countries, officials at the Ministry of Health wanted to know:

  • What will COVID-19 look like in Malawi (i.e., what is the epidemiological model for Malawi)?
  • What areas of Malawi are most at risk (down to the highest geographical resolution possible)?
  • When and where will we need to make resources (PPE, testing) available?

2. Review data already available

Every country can access these publicly available datasets:

The Ministry of Health in Malawi also has access to, or the ability to access, certain country-specific and non-public data sets. This is likely the case for most countries in Sub-Saharan Africa. These include:

Country-specific data:

  • Latest census data (population, age, sex)
  • Master Health Facility Registry (data about Malawi’s health facilities)
  • Surveillance data (for COVID-19, as it becomes available)
  • Food insecurity, poverty, weather, etc.

Non-public data sets:

  • Aggregated, anonymized Call Detail Record data (through existing, government-approved, public health research initiatives)
  • Disease burden & health outcome data (usually housed in DHIS2)
  • Case-based surveillance
  • Commodity data (OpenLMIS)

3. Consider new sources of data

We also looked at previously untapped sources of data that can be used to predict the spread of COVID-19. For example, Google Community Mobility Reports chart movement trends over time across different categories of places, such as retail and recreation, groceries and pharmacies, transit stations, and workplaces.

With this data, we can gain insights into the effects of policies implemented to fight COVID-19.

To help create a model for Malawi, we examined the effects in other African countries of three recently recommended COVID-19 prevention policies currently in use around the world (these categories correspond to those being considered by the government of Malawi):

  1. Social distancing
  2. Social distancing plus additional limited-movement guidelines (e.g., work from home, limiting gatherings to 10 or fewer attendees);
  3. Enforced population restrictions (e.g., shelter-in-place, closing non-essential businesses and reducing public transit, prohibiting virtually all gatherings).

First, we identified other African countries implementing policies in each of these categories (as of March 29).

We then used Google Mobility Reports to see how much the restrictions in force in each country reduced interpersonal contact. Google Mobility Reports breaks down its data by categories: Retail and Recreation, Grocery & Pharmacy, Parks, Transit Stations, and Workplaces. But we wanted to give policymakers an overall estimate of how much these restrictions would reduce interpersonal contact, across the population.

To turn that category-specific data into overall estimates, we made some simple assumptions about how often a person would go to each of these places each week, under each set of restrictions.

Combining Google’s category-specific data with those assumptions, we calculated a weighted, overall “reduction-in-contact rate” for each country. For example:

  • Zambia, employing social distancing, reduced contact by 15%.
  • Kenya, using social distancing plus additional restrictions, reduced contact by 37%
  • Rwanda, with enforced population restrictions, reduced contact by 57%.

By doing this across multiple countries, we were able to estimate a “contact-reduction range” to help policymakers anticipate the likely effect of each category of restrictions:

  • Social distancing would be expected to reduce contact by 10–20%.
  • Social distancing plus additional guidelines would be expected to reduce contact by 30–40%.
  •  Enforced population restrictions, the strictest category, would be expected to reduce contact by 40–60%.

These are rough estimates, based on assumptions, and many caveats apply. But these estimates, however imperfect, have a foundation in observed behavior, making them a reasonable basis to inform policy-making. Here is more information on our approach.

More data will help make these models more precise. Next steps could include high frequency data collection from health facilities and feedback on behavioral interventions KAP (using tools such as the UNICEF rapid survey: UReport).

4. Pull the information together

Now that we’ve identified our “use cases” (most important questions for our decision-makers) and our available data , we can start to bring it together to answer our three questions.

  • What will COVID-19 look like in Malawi?

The next step is to use these estimates to model the spread of COVID-19 in Malawi under different sets of restrictions. To estimate the incidence and mortality of COVID-19, we used closed, deterministic and compartmental susceptible-exposed-infected-recovered (SEIR) model with mild, hospitalized, and critical care sub-states within the infected state. (More methodological detail is included at the end, and in a forthcoming paper.)

First, we projected infections and deaths from COVID-19 in each scenario:

covid19 infection malawi

Then we projected how many people would be hospitalized and require critical care:

covid19 hospitalizations malawi

We can also project how much each mitigation scenario would reduce COVID-19 infections:

covid19 mitigation malawi

Now to answer the next two questions:

  • What areas of Malawi are most at risk (down to the highest geographical resolution possible)?
  • When and where will we need to make resources (PPE, testing) available?

We modeled the effects of these scenarios within Malawi. (Again, more information on our methodology is at the end, and in a forthcoming paper.) First, we used data about population density, percentage of residents over 65, access to health facilities, and current infections to identify the areas of Malawi at greatest risk:

covid19 risk model malawi

Then we used knowledge about population movements, drawn from aggregated cell-site location information, to anticipate how the end of the rainy season in May might affect the spread of the disease within Malawi:

covid19 migrqation risk malawi

5 Guiding Factors for Success

In closing, we offer some thoughts for those conducting similar analysis in other countries:

  • Know and use your publicly available data. Most of the data sets that we used, or their local analogues, should be available for every country.
  • Present the data in a way that is useful to decision-makers. That means finding out what questions they want answered, and adapting the model to the parameters or categories that are being considered for policy. (For example, here we adopted the Malawian government’s trichotomy of potential restrictions on movements.)
  • Incorporate local conditions. In Malawi, the onset of the rainy season in May produces major shifts in population. In other countries, recurring climatic, cultural, or political events may have comparable effects. Historical cell-site data, securely aggregated to protect privacy, can be used to build these events into COVID-19 modeling.
  • Refresh as often as possible. We plan to rapidly and regularly update our models as more information becomes available.
  • Document and share as much as you can. Share on github, write up results, and document and publish approaches, so that other can replicate, improve, and share.

For the epidemiological model

We simulated an outbreak in each of Malawi’s Traditional Authorities (a sub-district, or TA) as the result of a single-seed infection, with individuals progressing through states as determined by a series of epidemiological and behavioral parameters estimated daily using a series of ordinary differential equations.

We calculated the age distribution in each TA in ten-year age bands up to 79, and then 80 and greater using World Pop data. We then estimated age-standardized hospitalization, critical care, and fatality rates using age-stratified estimates of severity of disease from China, Hong Kong, and Macau from the literature, which were adjusted for censoring, demographics, under-ascertainment, and prevalence.

We parameterized the model using other model inputs from peer-reviewed literature, as well as estimates of population mobility from Google Community Mobility Reports.

We then summarize the expected number of infections, peak and cumulative hospitalization and critical care need, and total deaths.

We’ve also explored the potential impact of mitigation scenarios on the epidemic outlook, similarly drawing on data from neighboring countries in sub-Saharan Africa who have implemented social distancing or lockdown policies.

For the risk model

We generated a risk model using publicly available data in the form of World Pop & humanitarian data exchange shapefiles to identify base level vulnerability (i.e. population density & population > 65 years of age).

We used country specific data to further inform the model and built-in current infection location and an associated proximity effect (distance decay model) along with overarching health facility information (availability within a 5km radius) and more detailed readiness assessments (availability of commodities & services at a given site).

Then we used novel data sources to help identify where the population is currently residing using the call record data to understand migration patterns given time of year. This data was used in conjunction with the population estimates to develop a more informed density calculation. In addition, this data will give us insights into travel and movement.

Originally published by Cooper/Smith as How to use your data to fight COVID-19: A roadmap for countries in Sub-Saharan Africa

Filed Under: Data, Healthcare
More About: , , , , , ,

Written by
This Guest Post is an ICTworks community knowledge-sharing effort. We actively solicit original content and search for and re-publish quality ICT-related posts we find online. Please suggest a post (even your own) to add to our collective insight.
Stay Current with ICTworksGet Regular Updates via Email

One Comment to “4 Ways to Use National Data in African COVID-19 Digital Response”

  1. Pastor PRINCEMOSES says:

    I would Loves us to Help the Deafs people most especially in the rural Community ,they don’t have full information about how Dangerous COVID19 is. Most especially in my Country Zambia were Technology is poor in Rural communities. I love the Deafs and I am a Deafs specialist ,as well as a preacher. If you would Love to help Deafs ,kindly Contact me. Am ready to educate them with your support.