Back to Science Blog
This page is optimized for mobile devices, i.e. graphics are shown as static images.

SB04: Who is using the Corona-Warn-App, who is warned - and how fast?

CWA Team, on December 2, 2021, from German version (last updated October 15, 2021)
  1. First things first: a thank you
  2. The most important findings in a nutshell: numbers, data and facts
  3. Privacy-preserving Analytics
    1. The purpose of data donation
    2. Data collected
  4. Data about the participants and their devices
    1. Number of donations over time (by OS)
    2. Number of donations over time (by OS version)
    3. Number of donations over time (by CWA version)
    4. Participant distribution by age group
    5. Geographical distribution
    6. Differences between urban and rural districts
  5. Receipt of a warning
    1. Warnings and the people who received them
    2. Warned individuals over time
    3. Individuals warned during check-ins
  6. Test results
    1. Delay between test registration and receipt of a test result
    2. Delay between test registration and receipt of a warning
    3. Delay between risk exposure and test registration
    4. Delay between risk exposure and receipt of a warning
    5. Association between risk assessment and infection
    6. The percentage of users who donate data
  7. What to expect in the future

1 First things first: a thank you

We would like to begin by thanking the Corona-Warn-App’s users, especially those who donate their data to us every day. These donations provide evidence of the app’s effectiveness, help us improve its functionality and deliver important information about the course of the spread of COVID-19 in Germany. Furthermore, they supply the figures for epidemiological indicators that could not be depicted in the same quantity or quality without them. Therefore, this article is also aimed at settling a small part of the debt that we owe the people who use the app. Once again, we’d like to thank the app’s users for providing us with their data.

2 The most important findings in a nutshell: numbers, data and facts

  • Around 12 million app users donate their data every day (this has resulted in 9,638,099,766 records).
  • The percentage of data donors is about 49%.
  • There are differences between the number of app users in urban and rural districts, and between eastern and western Germany.
  • On average, 5 people who have been exposed to an increased risk and 11 people who have been exposed to a low-level risk are warned for each positive test result registered in the app.
  • Each check-in results in warnings about increased risk being sent out to around 23 people.
  • On average, PCR test results are delivered 21 hours after test registration (half of the test results are received within 12 hours).
  • On average, tests are registered in the app 3.6 days after a user has received a warning (half of tests are registered within 1.8 days).
  • On average, users receive warnings 3.7 days after having been exposed to a risk (half of warnings were received after about 3 days).
  • About one in five people who underwent a PCR test and who were warned about an increased risk tested positive. In contrast, less than half as many app users who did not receive a warning and still underwent a PCR test tested positive. Therefore, the CWA is warning the right people.

In the following, we focus on the purpose of data donation and describe the data that are being collected.

3 Privacy-Preserving Analytics

Privacy-preserving analytics or PPA enables CWA users to provide data held on their smartphone to an evaluating entity without having to reveal their identity.

Since 5 March 2021, CWA users have been able to donate operational data on a daily basis in an anonymous, yet authenticated, manner. Part of this process involves checking that the device sending the data and its CWA installation are genuine (for detailed documentation, see github.com/corona-warn-app/cwa-ppa-server).

CWA data donation should not be confused with Corona data donation. Corona data donation involves around half a million users donating data on body temperature, resting heart rate and other vital information collected using fitness bracelets.

3.1 The purpose of data donation

Data donation is aimed at gaining a better understanding of the app’s processes and usage. In turn, this enables its functions and user-friendliness to be continuously improved. As the app is based on a decentralised approach, a lot of important data about the CWA’s functionality is only available on end devices. Therefore, data donation is the only way of gaining the relevant data.

CWA data donation helps provide an understanding of:

  1. the events that occur in the app and how often they take place,
  2. when and the sequence in which these events occur and the interval (delay) between them,
  3. the selections that users make in response to particular events,
  4. where and when users cancel their input,
  5. and whether cancellation is due to the technical properties of a particular device or aspects such as demography (e.g. age group, residential area).

This information enables:

  1. risk assessment algorithms to be improved,
  2. parameters to be configured so as to ensure events occur at reasonable intervals,
  3. user navigation to be optimised,
  4. targeted communication measures to be targeted even more appropriately.

In addition, the data collected can be used to evaluate the temporal and spatial distribution of certain events in order to provide information about the course of the pandemic in real time and to identify local peculiarities. In turn, this enables targeted measures to be put in place at an early stage so as to counteract unfavourable developments.

3.2 Data collected

The following data are collected (see github.com/corona-warn-app/cwa-ppa-server/../PPAC):

3.2.1 Technical metadata

All data sets contain the following information:

  • Date of donation (submitted_at),
  • Authorisation flags to prevent abuse (android_ppac_basic_integrity, android_ppac_cts_profile_match, android_ppac_evaluation_type_basic, android_ppac_evaluation_type_hardware_backed).

3.2.2 Metadata from users

Users can provide the following optional information:

  • Federal state and district (federal_state, administrative_unit),
  • Age group (<30, 30-59, 60+) (age_group).

1,862,487,469 records contain this information (as of 7 October 2021).

3.2.3 Metadata from devices

The following data about end devices is available:

  • CWA Version (cwa_version_major, cwa_version_minor, cwa_version_patch),
  • Configuration token (app_config_etag),
  • OS Version (ios_version_major, ios_version_minor, ios_version_patch or android_api_level).

1,862,486,604 records contain this information (as of 7 October 2021).

3.2.4 People who received a warning (exposure risk)

The following data can be collected for proximity tracing (BLE, ENF) and presence tracing (event check-ins) (pt_..):

  • Risk level (red or green warnings); are there any differences between the current data and those that were available yesterday? (risk_level, risk_level_changed),
  • If a person has been exposed to a risk, the date when the warning was displayed on the user’s screen (see illustration). Difference between today and yesterday (most_recent_date_at_risk_level, most_recent_date_changed),
  • User metadata,
  • Technical metadata.

1,852,088,985 records contain this information (as of 7 October 2021).

Indicates exposure to an increased risk

Indicates exposure to an increased risk.

3.2.5 Test results

The following data are collected about all tests that are registered in the app, either when a result comes available or after a specific period (currently 7 days):

  • Test result (test_result) (positive, negative, indeterminate, pending) [A],
  • Hours since test registration (hours_since_test_registration),
  • Risk level (displayed) at test registration (risk_level_at_test_registration) [B],
  • days_since_most_recent_date_at_risk_level_at_test_registration [B] = Delay between the last time a user was exposed to a risk (most recent warning) and test registration (in days),
  • hours_since_high_risk_warning_at_test_registration [B] = Delay between warning and test registration (in hours),
  • User metadata,
  • Technical metadata.

The data are:

  • [A] available separately for PCR and rapid antigen tests (RAT) (values of test_result),
  • [B] and for proximity tracing (BLE, ENF) and presence tracing (event check-ins) (pt_..).

4,313,299 records contain this information (as of 7 October 2021).

3.2.6 Key submission

The following data are transmitted about the submission of diagnosis keys either as soon as a key is shared or after a specified period (currently 36 hours):

  • [B] for both proximity tracing (BLE, ENF) and presence tracing (event check-ins) (pt_..)
  • Key submitted with user metadata
    • Key submitted? (submitted)
    • Key submitted after running through the symptoms flow chart? (submitted_after_symptom_flow)
    • Key submitted with a teleTAN? (submitted_with_teletan)
    • Key submitted after a rapid antigen test? (submitted_after_rapid_antigen_test)
    • Hours since receiving test results (hours_since_reception_of_test_result)
    • Hours since test registration (hours_since_test_registration)
    • days_since_most_recent_date_at_risk_level_at_test_registration [B] = delay between the last time the user was exposed to a risk (most recent warning) and test registration (in days)
    • hours_since_high_risk_warning_at_test_registration [B] = Delay between warning and test registration (in hours)
    • User metadata,
    • Technical metadata.
  • Key submission with metadata from end devices
    • Key submitted? (submitted)
    • Advanced consent granted? (advanced_consent_given)
    • Key submitted in the background? (submitted_in_background)
    • Key submitted after running through the symptoms flow chart? (submitted_after_symptom_flow)
    • Key submitted after cancelling the key submission process? (submitted_after_cancel)
    • Last event displayed as part of the key submission process(last_submission_flow_screen)
      • Test result (2)
      • Warning others (3)
      • Symptoms (4)
      • Onset of symptoms (5)
    • Device metadata,
    • Technical metadata.

251,161 records contain this information (as of 7 October 2021).

3.2.7 Exposure Windows and Scan Instances

The following information is available from data recorded locally using the Google or Apple Exposure Notification Framework (ENF; see Google Exposure Notifications API, Apple Exposure Notifications API):

1,036,601,141 records contain information about exposure windows and 3,019,619,946 contain information about scan instances (as of 7 October 2021).

4 Data about the participants and their devices

In the next section, we take a closer look at the data donors themselves (or, more specifically, at the devices they use). Since 7 October 2021, data has been donated around 1.86 billion times. During the past few weeks, more than 11 million records have been donated every day.

4.1 Number of donations over time (by OS)

The following chart demonstrates trends over time and the types of operating system used. Out of the 12,159,958 records donated on 7 October 2021, 6,101,821 were received from devices using Apple’s iOS and 6,058,137 from Android-based devices.

Fluctuations in the number of daily donations particularly result from measures and restrictions put in place to secure the network against DDoS attacks . This mainly affects devices running Android, because the authenticity of these devices is reviewed to ensure data protection laws are upheld (for more detailed information, see the PPA documentation).

Number of donations (total and by operating system)

Figure 1: Number of donations (total and by operating system).

4.2 Number of donations over time (by OS version)

A breakdown of the data by operating system demonstrates the typical homogeneity associated with iOS and the heterogeneity associated with Android-based devices.

Number of donations (by operating system version)

Figure 2: Number of donations (by operating system version).

4.3 Number of donations over time (by CWA version)

The following graph depicts the data by CWA version. The graph demonstrates a clear exponential switch to a new CWA version.

Number of donations (by CWA version)

Figure 3: Number of donations (by CWA version).

4.4 Participant distribution by age group

The following graph depicts the distribution of the participants by age group. 34% of participants provided an indication of their age. Of these, 18.3% stated that they were under 30, 62% that they were aged between 30 and 59 and 19.7% that they were 60 or older. These figures are within expected ranges (see EDUS participants by age and sex).

Number of CWA donations (by age group)

Figure 4: Number of CWA donations (by age group).

4.5 Geographical distribution

The following graph depicts the geographical distribution of the participants and uses data from 31 August 2021. On 31 August 2021, 12,109,252 records were donated, and 3,991,351 of them (33%) include information about the user’s district.

Donations by district

Figure 5: Donations by district.

The clear differences in this chart between urban and rural areas and between eastern and western districts were also identified by the EDUS.

4.6 Differences between urban and rural districts

The following graph depicts the relationship between population density and the number of donations per 100,000 inhabitants. The representation of each district on the graph is proportional to its number of inhabitants.

Participants by population density in urban and rural districts.

The differences between urban and rural districts were strongest in eastern districts. In order to make the graph a bit easier to understand, these data are also displayed in an interactive graphic.

Participants by population density in urban and rural districts

Figure 7: Participants by population density in urban and rural districts.

The “incidence” of people providing their data (the number of participants per 100,000 inhabitants) rises with increasing population density. On average, 4,191 participants (per 100,000 inhabitants) live in rural areas (median: 4,246), with 4,875 living in urban areas (median: 4,780). 5,154 participants (per 100,000 inhabitants) live in western districts and 3,331 live in eastern districts. This gives an average of 4,388 users. Once the percentage of those who stated their district is taken into account, the figure rises to 13,312. Later on, we will see that around half of all CWA users donate their data, which means that about 27% of the population actively uses the CWA.

5 Receipt of a warning

5.1 Warnings and the people who received them

Figure 7 shows trends in people who provided or received a warning. Whereas the number of people who provided a warning is known from the CWA ecosystem, the number of people who subsequently received a warning can only be calculated using data donation. Since 5 March 2021, the CWA has sent out 619,189 red warnings (increased risk) and 1,518,663 green warnings (low risk).

During this period 273,953 people warned others through the CWA, which allows us to infer about 5 (red) warned people per warning person, considering the estimated proportion of data donors. For warnings without an increased risk, the mean value is 11 (green) warned persons per warning person.

People who provided a warning and those who were warned by the CWA

Figure 8: People who provided a warning and those who were warned by the CWA.

The large number of people who received a warning on 19 and 23 March is particularly notable. These differences can be explained by the adjustments made to the parameters used by the CWA to calculate risk . These adjustments were accompanied by a reassessment of the risk that users had been exposed to over the preceding 14 days. Any warnings that this resulted in were then sent out on the same day instead of over a two-week period. On 17 April, further adjustments were made to the lower threshold and this resulted in red and green warnings being issued at a similar frequency. Since the end of the lockdown (30 June 2021) and the start of the school holidays, more green warnings have been issued than red warnings.

5.2 Warned individuals over time

Figure 8 shows trends in the number of people who received a red or green warning. The graph is based on the daily estimate of the percentage of users donating data, which was calculated using the number of tests registered in the app. Since the end of June, it is noticeable that green warnings have been sent out once again more frequently than red warnings (this still holds true when they are put in relation to one another). Increased mobility and the associated increase in contact with other people since the end of the “Bundesnotbremse” (the “federal emergency brake”) certainly plays a significant role here. Finally, the graph also demonstrates that, on average, each exposure registered in the app resulted in 8 people receiving a warning of an increased risk and 26 people receiving a warning of a low risk (with risk exposure).

Number of people who received a red or green warning for each person who provided a warning, over time

5.3 Individuals warned during check-ins

A total of 9,075 people received a warning from a check-in. 4,505 were notified about an increased risk (red warning) and 4,570 were notified about a low risk (green warning).

During the same period data were donated on 398 check-in events, which allows us to infer approximately 46.5 (of which 23.1 were red and 23.4 were green) warnings per event, considering the estimated proportion of data donors. Due to the fact that the CWA is not covered by the Corona-Schutzverordnungen (the Corona Protection Ordinance), the app cannot always be used for official check-ins. Saxony is the only federal state to have anchored the CWA as a check-in tool into its corona protection regulations.

People who received a warning (PPA) and people who provided a warning (backend)

Figure 10: People who received a warning (PPA) and people who provided a warning (backend).

6 Test results

We begin by analysing delays that can be estimated using data on test results received via data donation.

6.1 Delay between test registration and receipt of a test result

Information taken from the 1,567,815 donated records on PCR test results demonstrates that 395,157 tests (25.2%) were received within one hour of test registration. As such, these tests were not registered at the time they were carried out. Results from 1,172,212 tests (74.8%) were received within 24 hours. The following graph shows the hourly distribution of the delay between test registration and receipt of a test result. On average, test results were received 21 hours after test registration (median: 12 hours).

Delay between test registration and receipt of a test result (PCR).

The 2,373,969 donated records that provide a rapid antigen test result and information about the delay between test registration and receipt of a result demonstrate that 1,532,490 (64.6%) test results were received within one hour of test registration. In total, 2,175,533 (91.6%) tests were received within 24 hours. The following graph depicts the hourly distribution of the delay between test registration and receipt of a test result. On average, results were received 10.5 hours after registration (median: 0 hours).

One reason for the relatively long delay between registration of a rapid antigen test and receipt of a test result is that the CWA does not continuously download test results. In order to receive a result as soon as possible, users need to call up their result manually. Although the CWA does attempt to do so automatically, the time needed to gain a result depends on the operating system in question and Internet availability. As such, these aspects can sometimes lead to long delays.

Delay between test registration and receipt of a test result (rapid antigen test).

6.2 Delay between test registration and receipt of a warning

Data on the delay between receipt of a warning and test registration are available for 97,322 PCR tests.

The following graph depicts the hourly distribution of the delay between receipt of a warning and the registration of a PCR test. On average, tests were registered 3.6 days after a warning (median: 1.8 days).

52,559 PCR tests (54%) were registered within 48 hours of a user having received a warning.

Delay between receipt of a warning and test registration (by test result) (PCR).

Data on the delay between receipt of a warning and test registration are available for 42,611 rapid antigen tests.

The following graph demonstrates the hourly distribution of the delay between receipt of a warning and registration of a rapid antigen test. On average, tests were registered 3.2 days after a user had received a warning (median: 1.2 days).

25,518 rapid antigen tests (59.9%) were registered within 48 hours of a user having received a warning.

Delay between receipt of a warning and test registration (by test result) (rapid antigen test).

6.3 Delay between risk exposure and test registration

The following graph depicts the daily distribution of the delay between risk exposure and registration of a PCR test. On average, PCR tests were registered 6 days after risk exposure (median: 6 days).

Delay between risk exposure and test registration (in days before test registration) (PCR).

The following graph depicts the daily distribution of the delay between risk exposure and the registration of a rapid antigen test. On average, tests were registered 7.4 days after risk exposure (median: 7 days).

Delay between risk exposure and test registration (in days before test registration) (rapid antigen test).

6.4 Delay between risk exposure and receipt of a warning

The daily distribution of the delay between risk exposure and receipt of a warning results from the difference between the distribution of the delay between risk exposure and test registration, on the one hand, and receipt of a warning and registration of a PCR test, on the other. This is depicted in the following graph. On average, warnings were received 3.8 days after risk exposure (median: 4 days).

As the following example demonstrates, the average delay between risk exposure and receipt of a warning (3.8 days) is quite short. Let us assume that person A meets person B on day 0 for approximately 2 hours, and that person A displays typical SARS-CoV-2 symptoms on day 1. If person A were to be tested on day 2, receive a positive PCR test result on day 3, and register the result immediately in the app, person B’s smartphone would receive a push notification on day 3 informing them that they had been exposed to an increased risk on day 0. This leads to a 3-day delay between risk exposure and receipt of a warning.

Delay between risk exposure and receipt of a warning (in days before the warning) (PCR).

The daily distribution of the delay between risk exposure and receipt of a warning results from the difference between the distribution of the delay between risk exposure and registration, on the one hand, and receipt of a warning and registration of a rapid antigen test, on the other. The following graph depicts the daily distribution of the delay between risk exposure and receipt of a warning. On average, warnings were received 5.4 days after risk exposure (median: 5 days).

Delay between risk exposure and receipt of a warning (in days before the warning) (Rapid antigen test).

6.5 Association between risk assessment and infection

The data also provide information about whether people who received a warning from the CWA are more likely to test positive: in short, they are. More than one in five people who received a red warning before having registered a test went on to test positive for SARS-CoV-2. In contrast, the number of people who tested positive but did not receive a warning was less than half as high (see Fig. 18).

Variable Test result
pos (PCR) neg (PCR) total
Risk assessment
Increased risk 20,655 (21%) 77,090 (79%) 97,745 (100%)
Low risk (with risk exposure) 1,930 (14%) 11,919 (86%) 13,849 (100%)
Low risk (no risk exposure) 140,128 (9.7%) 1,306,963 (90%) 1,447,091 (100%)
Total 162,713 (10%) 1,395,972 (90%) 1,558,685 (100%)
Association between risk notification and test result (PCR).

The following graph demonstrates the percentage of positive test results by risk notification over time. The percentage of people who tested positive after having received a red warning remains relatively constant. A comparison with the RKI test numbers shows similar figures in terms of the percentage of people who did not receive a warning but who still tested positive. The slightly higher proportion of people who tested positive is, on the one hand, due to differences in the underlying survey population. For example, children under the age of 16 rarely use the CWA. On the other hand, the RKI figures increasingly include routine and pooled test results that are probably rarely if ever submitted to the CWA.

Proportion of positive tests by risk notification over time (PCR).

This same association is also clear for rapid antigen tests, even if the data on this are sparser (see Fig. 19).

Variable Test result
pos (RAT) neg (RAT) Total
Risk assessment
Increased risk 245 (0.6%) 42,325 (99%) 45,570 (100%)
Low risk (with risk exposure) 31 (0.3%) 11,731 (100%) 11,762 (100%)
Low risk (no risk exposure) 1,922 (<0.1%) 2,304,896 (100%) 2,306,818 (100%)
Total 2,198 (<0.1%) 2,358,952 (100%) 2,361,150 (100%)

<img src="./ppa_risk_infection_ct_mosaic_rat_en.png" class="figure-img img-fluid" alt="Association between risk notification and test result (RAT)"./>

The following graph demonstrates the proportion of positive tests by warning over time.

Proportion of positive tests by warning over time (PCR).

6.6 The percentage of users who donate data

The data on the number of test results registered via the CWA backend and those available from data donation can be used to estimate the percentage of users who donate their data.

The following graph depicts the percentage of donated test results (positive or negative) by test type (PCR or RAT (rapid antigen test)). The graph demonstrates the relationship between the test results donated on one particular day and the test results made available on the CWA Lab result server, either on the day before (in the case of PCR tests) or on the same day (in the case of RAT).

Relationship between the number of test results provided via CWA data donation and the number of test results registered via the backend (PCR/RAT and pos/neg).

The percentage of PCR test results that have been retrieved and transmitted via CWA data donation can also be related to the number of tests that were registered the day before. Doing so provides us with an estimate of the percentage of participants who donate their data: it averages 49% over the last 30 days.

Because the percentage of people who donate their data is estimated using information about donated test results, it is impossible to use this figure to draw robust conclusions about the number of active users and/or devices on which the CWA is installed. However, we intend to devote a separate article to this issue.

Daily estimate of the proportion of data donors – based on the number of test results (CWA data donation) per test registration (backend).

7 What to expect in the future

In the coming weeks we will continue to focus on data donation via the Corona-Warn-App. We will be describing the information that can be gained from analysing key submission and BLE measurement using ENF (exposure windows and scan instances).

However, before we do so, we intend to take a short detour to analyse the number of active apps/devices and users in more detail. In order to do so, we will be using data from Apple App Store and Google Play Store. We aim to compare the various key figures that they make available and to provide additional estimates.