As COVID-19 sweeps across the globe, tools are being developed to effectively curb the spread of the disease. However, most of these tools are suitable only in industrialised countries where smartphone penetration is high. Much less emphasis has been placed on developing tools suitable for emerging markets. In this article, I discuss five such tools, symptom screening, contact tracing, exposure notification, hotspot detection, and health certificates.
TLDR;
- Emerging markets face a unique challenge when it comes to deploying technology to fight COVID-19. Each of the five tools that are commonly used, symptom screening, contact tracing, exposure notification, hotspot detection, and health certificates have specific disadvantages.
- Symptom screening obstructs a clear view on how the progression of the COVID-19 and introduces a selection bias which is impossible to correct without testing a representative sample of the population.
- Manual contact tracing is a slow process — likely too slow to be effective based on the latest research — and needs to be augmented with other technologies. However, data cellphone triangulation does not have sufficient spatial resolution to be effective in emerging market countries.
- Exposure notification via smartphone apps using Bluetooth cannot achieve sufficient adoption in a country with low smartphone penetration. For example, if only 50% of the population have a smartphone and 50% of smartphone users would adopt a nationally promoted app (which would be a lot!) only a fraction of the necessary contacts could be traced. Furthermore, there are serious ethical considerations around stigma that need to be addressed.
- Hotspot detection and health credentials suffer from moral hazard problems due to issues associated with self-reporting of test results. There is not yet sufficient evidence to support immunity passports and they have serious ethical issues.
Policy Recommendation 1: Cut all unnecessary testing and as soon as possible test a representative sample of the population to get a clear picture of how much COVID-19 has progressed. Repeat the testing regularly to understand the basic disease dynamics and effectiveness of different policy responses.
Policy Recommendation 2: Instead of relying on a single method of contact tracing, adopt a technology-augmented approach where various data sources are jointly used. Here I outline one such system using Covi-ID and PathCheck.
Policy Recommendation 3: Personal data regarding health status should not be collected by employers. Provide clear regulatory guidance for the collection, processing, sharing, and storage of such data to private sector participants.
Policy Recommendation 4: Create an interface for applications to query test data collected e.g. at the NICD. This will spur private sector innovation in order to provide public health officials with the best available tools. However, access to this interface should only be granted to apps officially endorsed by the NDoH.
Symptom Screening
Symptom screening is the process of deploying healthcare workers to conduct large scale screening of symptoms, for example by going door to door in an informal settlement and asking each inhabitant about their health status. Symptom screening is widely deployed in South Africa with thousands of health workers operating across the country. The National Department of Health outlines the process: “The health worker will ask if you: 1. have travelled to a high risk country in the last 14 days; 2. have had contact with anyone with confirmed COVID-19 in the last 14 days; 3. have symptoms such as fever, cough and difficulty in breathing.” and goes on to explain that health workers will refer patients with symptoms to a testing facility.
What makes symptom screening necessary is insufficient testing capacity. The NDOH writes: “There is a limited number of test kits in South Africa. To decide who should be tested, health workers will use a formula by looking at how serious your symptoms are and what high risk factors you have.” Symptom screening is also part of the South African Department of Labour’s measures on “COVID-19 Occupational Health and Safety Measures in Workplaces”, gazetted 29 April 2020. The measure prescribes that employees need to complete a self-screening of their health symptoms and report these to their employer.
The major downside of symptom screening is that it obstructs a clear view on how the progression of the disease. If symptom screening is used to determine who should be tested, a self-selection bias afflicts important epidemiological data. To date, it is not clear how large this error is and consequently we do not know which part of the population has already contracted and recovered from COVID-19. This makes both model validation and assessing policy responses extremely challenging.
Policy Recommendation 1: Cut all unnecessary testing and as soon as possible test a representative sample of the population to get a clear picture of how much COVID-19 has progressed. Repeat the testing regularly to understand the basic disease dynamics and effectiveness of different policy responses.
Contact Tracing
During the contact tracing process, the contacts of a patient who tests positive for COVID-19 are identified and informed about their exposure so that they can self-isolate and watch their symptoms. The hope to curb the spread of the disease by isolating infected patients and their exposed contacts makes contact tracing one of the key tools to contain limited outbreaks of COVID-19 and is the reason why it is recommended for example by the US Center for Disease Control. Contact tracing can be either largely manual, largely digital, or a hybrid of the two.
In a largely manual contact tracing process, a manual contact tracer receives a patient’s test information from a test lab, contacts the patient, and if the test is positive, asks the patient about who they have been in contact with. Then the manual contact tracer informs the contacts as well. In a largely digital process, a patient collects contact information through technological means, usually through a smartphone app that uses bluetooth low energy (BLE) technology (see e.g. PathCheck, DP³T). More and more dedicated wearable devices are being developed for this purpose, but it is unclear whether those would gain sufficient traction.
Another approach is the use of geolocation data obtained from cell phone triangulation. The advantage of this approach is that it can be applied to smartphones and feature phones (only using USSD) alike. The downside of this method is that it is afflicted with large localization errors, often only localizing a phone within a 300–400m radius. However, in an informal settlement like Khayelitsha in Cape Town, the density is about 10,000 inhabitants per square kilometer as of 2011, which means that there are about 5,000 people in a 400 meter radius of a phone located in this way. Thus, geo-location through cell phone triangulation is likely to be ineffective for contact tracing.
There are other issues with manual contact tracing, too, as it is a slow process. We know that efficient containment of COVID-19 requires almost immediate notification of contacts when a patient tests positive for COVID-19. Their paper shows that even a three-day delay in identification of contacts makes it impossible to contain the spread of the virus. For an immediate notification of contacts to be efficient, about sixty percent of the contacts of about half of all positive cases need to be quarantined. If testing capacity remains constrained and we can only identify around twenty percent of positive cases, we need to be able to quarantine more than seventy percent of contacts immediately. This number increases to over eighty percent if it takes two days to patient isolation and contact quarantine.
Exposure Notification
Several initiatives, including DP³T, Covid Watch, and Bluetrace have launched projects to notify users who might have been exposed to someone with COVID-19. The goal is that these users self-isolate and thus curb the spread of the virus. Where manual contact tracing relies largely on the intervention of a third party, the contact tracer, exposure notification works largely autonomously and relies on the self-isolation of exposed users as a preventive measure.
The most prominent implementation of exposure notification is the Google / Apple exposure notification (GAEN) which will be included in various applications with the idea being that each country has only one “national” app in use. It uses Bluetooth low energy (BLE) to detect proximity of two smartphones. If a user tests positive for COVID-19 she can self-attest to this using an official health authority app. Upon self-attestation the user can then upload her broadcast beacons to a cloud storage. The GAEN API regularly downloads the list of the latest broadcast beacons and locally checks if a user has been exposed to a COVID-19 positive patient. If so, an exposure notification is triggered on the user’s phone.
There are two critical issues with BLE-based exposure notification, though: (i) smartphone penetration in resource-constrained countries is much lower than in industrialised countries, rendering the method ineffective as a reliable means of contact tracing; (ii) most existing platforms do not carefully consider ethical implications of exposure notification.
The most pressing problem of BLE based exposure notification is that it is difficult to reach critical adoption to contact a sufficient number of users who have been exposed. If we want to reach about sixty percent of all contacts of fifty percent of COVID-19 patients immediately — which is necessary for contact tracing to be effective — we have to identify sixty percent of all contacts. A contact is recorded only if both users have a BLE-enabled application installed. In South Africa, only about 50% of citizens have a smartphone, which puts an upper limit on the number of possible users that is below the threshold at which contact tracing can be effective. But this is a best case scenario. In reality, only a fraction of all smartphone owners have activated BLE.
In Singapore, for example, only 450,000 users or about 8% of the population have updated their app when it became necessary to register with a phone number. Even if we assume an incredibly optimistic scenario where GAEN leads to a 50% adoption among smartphone users, only 25% of all South Africans Could be included in exposure notification. This is exemplified in the picture below. There are eight users with four interactions each. Four of the users do not have smart phones (red squares). Of the four that have a smartphone two do not have activated the app (yellow squares) and two have an activated app (green squares). What is the probability that a contact between two users with activated app is being recorded? In the below setting there are 16 connections in total, but only one of them can be recorded, so 1/16 = 6.25%.
In reality the number might be a little higher because citizens with smartphones might be more likely to be in contact with other citizens with smartphones. However, this coverage of contacts is still about an order of magnitude below what is necessary.
Aside from these practical considerations, fundamental ethical issues arise. What, for example, if a user receives an exposure notification while being in a taxi or their place of work? Can we reasonably assume that they are not being stigmatised because of an exposure? What if they are asked to leave the taxi in the middle of nowhere, or are being forcefully removed? What if they are being fired from their job because their boss must fear losing business if there is a COVID-19 outbreak in the workplace? None of these issues have been addressed by GAEN, but are pertinent in South Africa and any resource-constrained country, where stigma around COVID-19 is prevalent.
Policy Recommendation 2: Instead of relying on a single method of contact tracing, adopt a technology-augmented approach where various data sources are jointly used. Here I outline one such system using Covi-ID and PathCheck.
Hotspot Detection
Hotspot detection can be a helpful tool to issue warnings regarding specific hotspots which act as super spreaders. This is particularly useful when directing cleaning and sanitising efforts.
A hotspot map can be created in various ways, for example when manual contact tracers ask COVID-19 positive patients where they have been in the past two weeks and if they e.g. have attended a gathering or spent a prolonged amount of time in an office or other place of work. Tools like PathCheck use geolocation data from GPS to help contact tracers create a hotspot map in the Safe Places platform.
Identifying hotspots can be difficult, however, because the exact disease transmission vector might be difficult to identify even if a public health official is reasonably sure that a certain location is a hotspot of infections. There also is the potential to abuse a system that sees extra requirements for COVID-19 hotspots, e.g. when false positives are reported. This last point hints towards an existing open problem in the ecosystem for COVID-19 tools: the lack of integration with test laboratories. In many countries, such an integration would be necessary for private innovation to thrive, but is closely guarded — not least because of privacy concerns — by a public health authority like the NDOH or NICD in South Africa.
Policy Recommendation 3: Personal data regarding health status should not be collected by employers. Provide clear regulatory guidance for the collection, processing, sharing, and storage of such data to private sector participants.
Policy Recommendation 4: Create an interface for applications to query test data collected e.g. at the NICD. This will spur private sector innovation in order to provide public health officials with the best available tools. However, access to this interface should only be granted to apps officially endorsed by the NDoH.
Health certificates
Some environments pose a higher risk of COVID-19 transmission than others. In order to enable workplaces, public transport, schools, etc. to minimise the risk of transmission, health certificates are being used in some countries and considered in others. This would allow employers, taxi operators, teachers, etc. to verify the health status of employees, customers, contractors, and vendors and set individual limits e.g. for persons who have recently shown symptoms of COVID-19.
Health certificates can come in various forms; The most stringent one is an immunity certificate that would be issued once a patient has recovered from COVID-19 — and thus developed immunity — or received a vaccine. This would allow those with immunity to have their civil liberties restored sooner. But health certificates can also be based on self-reported symptoms or other risk metrics (e.g. age, pre-existing conditions). There are various approaches to recording health certificates, including in a self-sovereign identity (SSI) paradigm (see below), but none of these have gained mainstream adoption yet.
There are three principal issues with health certificates, which might explain the lack of traction. First, the WHO has warned against immunity certificates because it is not yet clear if patients who have recovered from COVID-19 indeed have developed immunity. Second, risk assessment based on self-reported data is prone to errors and manipulation (e.g. customers who have a score to settle with a company might report having had symptoms in order to have the company shut down; NB: That’s one of the reason why we need an API for test results for authorised apps). And third, there are serious ethical concerns about a health certificate, not least because of stigma. If it is economically advantageous to develop immunity because access to work is easier, or even if it is only more convenient because civil liberties like freedom of movement are being restored, there is an incentive for citizens to get infected with COVID-19. This perverse incentive can offset all efforts to flatten the curve and even result in an increased caseload.
A Privacy-Preserving Toolkit to Effectively Fight COVID-19 in Resource-Constrained Countries
The above-mentioned tools can be implemented in a variety of ways. What they have in common is that each tool is a socio-technical system that combines technological and process innovation, although not all implementations emphasise both aspects equally. Given the urgency of an efficient crisis response, some countries have been arguing in favour of process innovation above all else, even at the expense of civil liberties.
We believe that this approach is not sustainable, not least because there is an increasing awareness of the importance of privacy since the Facebook / Cambridge Analytica and NSA surveillance scandals.
The Value of Privacy
Various definitions of privacy exist. We follow Westin (1967) and define:
“Privacy is the claim of individuals, groups, or institutions to determine for themselves when, how, and to what extent information about them is communicated to others.”
In the context of COVID-19, protecting privacy is more difficult than ever. Among the tools described above, contact tracing, exposure notification, and hotspot maps all require user data to be shared. Most implementations of these tools therefore resort to some sort of obfuscation of users’ identity (e.g. by using rotating beacons, sanitising GPS data) to provide users with a notion of privacy through anonymity. This notion of privacy through anonymity ensures that no information is produced that can be linked to a user. However, this notion of privacy is not quite what Westin (1967) and other scholars have in mind when they speak about the claim to determine for themselves how information about them is communicated to others. Such a strong notion of privacy can only be implemented when users can control how information about them is communicated to others. There is one obvious benefit that this strong notion of privacy provides — the ability to combine various forms of data about a user (e.g health status, geolocation history, demographic information) and the associated increase in analytics capabilities.
Aside from this, there are practical concerns: Without privacy, users will be less likely to adopt a particular app. Furthermore, legal challenges are likely to be mounted if e.g. privacy of lawyers, doctors, priests is violated by a COVID-19 toolkit. This could delay an efficient response and jeopardise a country’s economic recovery.
In part two of this article, we introduce Covi-ID, a privacy-preserving integrated toolkit to fight COVID-19 in emerging markets.