In the Netherlands, SARS-CoV-2 tests are performed at publicly funded testing facilities of the PHS and hospitals, free of charge, or by commercial providers for a fee. SARS-CoV-2 is a notifiable infection, which means that all confirmed SARS-CoV-2 cases must be reported to the PHS regardless of where the testing took place. In this analysis, we included all adults (≥18 years old) who live in the Amsterdam region and were approached by the PHS of Amsterdam between October 28, 2020, and February 26, 2021, for contact tracing after a SARS-CoV-2 diagnosis.

Using data directly from the CoronaMelder app itself was not possible due to the anonymous nature of those data and privacy regulations. Therefore, we used data collected by PHS staff during routine contact tracing by phone and stored in HPZone (inFact UK Ltd). Routine procedure stipulates that PHS staff call persons diagnosed with SARS-CoV-2 (ie, cases) in the Netherlands to inform them about the diagnosis and isolation measures, and to initiate contact tracing. The case and a PHS staff member together systematically make an inventory of all identifiable persons that the case had been in contact with, 2 days prior to the date of symptom onset (if symptomatic) or positive test result (if asymptomatic). Moreover, PHS staff members are instructed to ask if the case used the CoronaMelder app, to note the answer in a standard format in a text field template in HPZone, and to activate the contact notification function of the app.

Data for this study were extracted from HPZone and anonymized before analysis. We extracted age in years at symptom onset (for symptomatic cases) or at the time of initiating contact tracing (for asymptomatic individuals) as a continuous variable. Other variables of interest were gender, categorized as male and female (other or nonbinary was not available in the system, was regarded as missing, and was therefore excluded from the analyses), and the municipality of residence (Aalsmeer, Amstelveen, Ouder-Amstel, Diemen, Uithoorn, or Amsterdam). During contact tracing, contacts were categorized into household contacts, close contacts, or other contacts. For this study, we extracted the number of close contacts, defined as contacts with whom a case had been within 1.5 meters for more than 15 minutes, excluding household contacts. Self-reported country of birth was recorded and later categorized as the Netherlands, other Western country, or non-Western country, in accordance with the definition used by Statistics Netherlands [24]. Employment in health care was categorized as “not,” “hospital,” “nursing home for elderly,” “other 24-hour care home,” “in-home care,” and “other health care.” Employment in education was categorized as “not,” “day care,” “elementary school,” and “secondary or higher education.” Data on CoronaMelder app use was extracted using a regular expression (“Gebruik coronamelder:”) from the free text notes. For those who used the app, we also extracted data on the reason of requesting a SARS-CoV-2 test.

The medical ethics committee of the Amsterdam University Medical Centers deemed it not necessary to fully review the study, because the study does not fall under the scope of the Medical Research Involving Human Subjects Act (W20_432#20.479). No data from the app are used in this paper; therefore, the privacy regulations of the app were not reviewed for the purposes of this study, though they can be found on the web [25].

Differences in characteristics between individuals who reported to use the mobile app and individuals who did not were assessed with chi-squared tests for categorical variables and Kruskal-Wallis tests for continuous variables. Trends over time in data availability on app use, as well as app use itself, were tested with the Pettitt test. Logistic regression analyses were performed to identify determinants of mobile app use. First, in univariable models, we tested for each independent variable (age, gender, country of birth, municipality, number of close contacts, employment in health care, and employment in education) whether they were associated with the dependent variable—self-reported use of the CoronaMelder app. Second, we combined all aforementioned independent variables and the dependent variable in 1 multivariable model. Age in years and the number of close contacts were added as continuous variables. As these variables were found to have a nonlinear relation to the outcome variable in exploratory analysis and regressions, B-splines were used with respectively 4 and 2 knots and a degree of 2. Gender and self-reported use of the CoronaMelder app were added as dichotomous variables, and all other variables were added as categorical variables. A complete cases analysis was performed; cases with missing data were excluded from the analysis. Outliers in the continuous variables age in years (above 100 years old) and number of close contacts (more than 12 close contacts, 99th percentile) were removed. In sensitivity analysis, multiple imputation using Multivariate Imputation by Chained Equations was carried out to impute missing outcomes and independent variables [26,27]. Analysis was performed using the statsmodels library in Python3 (Python Software Foundation) [28].

In this study, we found that fewer than 1 in 6 individuals diagnosed with SARS-CoV-2 in the region of Amsterdam reported using the CoronaMelder contact-tracing app. As 24,942 (84%) out of 29,766 cases were not using the app, their close contacts could never receive a notification through the app, even though they might have installed it themselves. Only 6.5% (1935/29,766) of the positive cases with the app had received an app notification themselves. Reporting app use was associated with being middle-aged, having a few (ie, 3-8) close contacts during the infectious period, living in municipalities surrounding Amsterdam (rather than the city itself), and being born in the Netherlands. App use was less often reported by individuals with more than 8 close contacts and individuals who are born outside the Netherlands.

Caution is warranted when interpreting these results and the potential explanations and implications. Moreover, these results cannot be directly extrapolated to the general population, including those who did not test positive for SARS-CoV-2. According to national data, over 4.5 million people have downloaded the app during the study period [29], which is approximately 26% of the Dutch population. In our study population of cases, however, this percentage was only 16% (4824/29,766). It is possible that the cases included in our sample represent a population that is less likely to take any preventive measures. It is also possible that PHS staff did not consistently ask cases about app use, as Amsterdam has been a region with high infection numbers, leading to high work pressure for the contact-tracing team. On the other hand, the national number is a cumulative number that does not account for app removals or inactivation, multiple app downloads by one person, or underreporting of app use during contact tracing, while the number in our study represents prevalent use. A second limitation is that the routine PHS data were not collected for the purpose of scientific research. This limits the number of variables and thus the potential to explain our observations. Furthermore, ascertainment bias may have been introduced because data may not have been collected consistently and uniformly. However, sensitivity analysis showed that bias caused by missing data was very small.

In Dutch acceptability studies performed prior to the introduction of the app, in April 2020, younger individuals reported to be more willing to download the app once available [20,21]. A survey performed in France, Germany, Italy, the United Kingdom, and the United States showed the same age trend [22]. Willingness to download the app was associated with positive attitudes toward technology and with fear for COVID-19 [21]. However, in our study, middle-aged individuals were more likely to use the app compared to younger individuals. Fear for COVID-19 might have played a role in these older age groups, in line with their higher risk of more serious disease once infected. Conversely, the absence of fear, privacy concerns, and a lower willingness to obey COVID-19 control measures might have been more important among younger individuals. The oldest individuals in our study were less likely to use the app, which may relate to lower smartphone and app usage among elderly people in general [18]. This is supported by another evaluation of the same CoronaMelder app, which showed that elderly people had problems with understanding why, when, and how to use the app [30].

Cases living in the municipalities of Amsterdam or Diemen, the latter being geographically strongly connected to Amsterdam, were less likely to use the app compared with cases in the surrounding municipalities. Improving app use in more densely populated urban settings might be worthwhile because the app is especially useful to identify anonymous close contacts who cannot be traced otherwise. Additionally, we found strong associations with being born outside the Netherlands and not using the app. If national app usage trends reflect those found in our sample, this would be worrisome given that previous studies in Amsterdam and internationally have shown that some ethnic minority groups are disproportionately affected by SARS-CoV-2 [31-36]. Cultural differences or distrust in the authorities may underlie this observation, but other more practical issues might be important as well. Even though the app itself and information on the CoronaMelder app website are available in 10 different languages, language barriers might still exist, and communication about the app might not reach all groups. Unfortunately, the routine data used in this study do not contain information on language skills or parental birth country, and thus we cannot investigate app use among second-generation immigrants. Further research in this group is therefore needed to reduce health inequalities between ethnic groups [35,36].

While the likelihood of using the app increased with the number of close contacts in the range of 0 to 8 close contacts, it decreased with higher number of contacts. The advantages of using an app for contact tracing include speed, the fact that anonymous close contacts can be reached, and that there is no recall bias, which is especially beneficial among individuals with many contacts. Thus, it might be worthwhile to study barriers for use and promote app use among individuals with many close contacts.

Lastly, we saw moderate differences in app use among people working in high-risk professions during which many contacts may be unavoidable. In the health care sector, precautions are taken to prevent infection (eg, use of personal protective equipment). Using the app during working hours may result in false notifications that are indistinguishable from notifications after real risk contacts. This might explain why people working in nursing homes and home care were using the app less often compared to individuals not working in health care. However, a pause button was introduced to the app (to be used in situations such as when the phone is left in a locker) to allow people to keep the app but reduce the chance of receiving false notifications [37]. Place of work as a reason for not downloading the app was mentioned in a survey in the United Kingdom [38]. Individuals working in day care centers were also less likely to have the app. For this group, the app could be of added value because they encounter parents of children without full protection.

If contact-tracing apps are used efficiently and uptake is high, they have the potential to speed up contact tracing, identify contacts that would otherwise go unnoticed, and prevent infections. For instance, the app of the National Health Service in the United Kingdom has been downloaded by 49% of the eligible population with compatible smartphones [39], which is >30% of the total population. A modelling study showed that this app averted about one case per index case willing to send the notification to their contacts [39]. The percentage of the population who downloaded tracing apps was also high in countries such as Germany, Switzerland, and Finland (ie, 32%-45%), but much lower in Spain, Italy, and France (ie, 15%-19%) [40]. This high level of adoption in some countries could be driven by the fact that the National Health Service app and the German app are among several apps that combine the tracing function with other features such as local area risk indicators and a link for booking a test, or by differences in promotion efforts. The Dutch Ministry of Health decided not to equip the CoronaMelder app with such additional features that might appeal to users, and soon after introduction, it stopped actively promoting app use. Instead, the Dutch Ministry of Health developed a second national app to function as a COVID-19 passport, registering vaccinations, recovery from infection, and negative test results. Combining the contact-tracing app and the corona passport app might have increased use of the tracing app. This knowledge, combined with the observed low uptake of the app in our sample, suggests that, to yield its potential effect on the control of the COVID-19 epidemic, the app needs to be used by more people. Based on our findings, app promotion efforts should particularly target younger individuals, individuals with >8 close contacts, and individuals who are not born in the Netherlands.

This study shows that app use is low; only 4824 (16.2%) out of 29,766 individuals who tested positive for SARS-CoV-2 in the Amsterdam region. Moreover, we observed significant differences in app uptake by sociodemographic factors. Elderly persons, women, people not born in the Netherlands, and those either reporting none or many close contacts were less likely to have installed the CoronaMelder app. If confirmed in a nationally representative sample, this would mean the app is unlikely to have the impact on SARS-CoV-2 spread it could potentially have. Moreover, app uptake seems to be lower in certain subgroups of the population, indicating that more targeted efforts to improve uptake are necessary.