Disclaimer: This manuscript has not been peer‐reviewed.€¦ · Disclaimer: This manuscript has not been peer‐reviewed. Research in context Evidence before this study To estimate

Disclaimer: This manuscript has not been peer‐reviewed.

Prevalence and Risk Factors of Infection in the Representative COVID-19 Cohort Munich

Michael Pritsch1,2*, MD; Katja Radon3,4,5*, PhD; Abhishek Bakuli1*, PhD; Ronan Le Gleut6,7*, MSc; Laura Olbrich1,2, MD; Jessica M. Guggenbuehl Noller1, MSc; Elmar Saathoff

1,2, PhD; Noemi Castelletti1, Dr; Mercè Garí6, PhD; Peter Pütz6,8, PhD; Yannik Schälte6,10, MSc; Turid Frahnow6,8, PhD; Roman Wölfel2,9, MD; Camilla Rothe1, MD; Michel

Pletschette1, MD; Dafni Metaxa1, MB BChir; Felix Forster3,5, MSc; Verena Thiel1, MD; Friedrich Rieß 1,2, Dr; Maximilian N. Diefenbach1, Dr; Günter Fröschl1,4, MD; Jan Bruger1; Simon Winter1; Jonathan Frese1; Kerstin Puchinger1; Isabel Brand1; Inge Kroidl1,2, MD; Jan

Hasenauer6,10,11+, Dr; Christiane Fuchs6,7,8,10+, PhD; Andreas Wieser1,2+, MD; Michael Hoelscher1,2,4+§, MD; on behalf of the KoCo19 study group#

1Division of Infectious Diseases and Tropical Medicine, University Hospital, LMU Munich, 80802 Munich, Germany 2German Center for Infection Research (DZIF), partner site Munich, Germany 3Institute and Outpatient Clinic for Occupational, Social and Environmental Medicine, University Hospital, LMU Munich, 80336 Munich, Germany 4Center for International Health (CIH), University Hospital, LMU Munich, 80336 Munich, Germany 5Comprehensive Pneumology Center (CPC) Munich, German Center for Lung Research (DZL), 89337 Munich, Germany 6Institute of Computational Biology, Helmholtz Zentrum München - German Research Center for Environmental Health, 85764 Neuherberg, Germany 7Core Facility Statistical Consulting, Helmholtz Zentrum München - German Research Center for Environmental Health, 85764 Neuherberg, Germany 8Faculty of Business Administration and Economics, Bielefeld University, 33615 Bielefeld, Germany 9Bundeswehr Institute of Microbiology, 80937 Munich, Germany

10Center for Mathematics, Technische Universität München, 85748 Garching, Germany 11Interdisciplinary Research Unit Mathematics and Life Sciences, University of Bonn, 53113 Bonn, Germany

*Those authors contributed equally +Those authors contributed equally

§Corresponding author: Michael Hoelscher, MD Division of Infectious Diseases and Tropical Medicine University Hospital, LMU Munich Leopoldstr. 5, 80802 Munich, Germany Email: [email protected] Phone: +4989440059801 #KoCo19 study group: Emad Alamoudi, Jared Anderson, Valeria Baldassare, Maximilian Baumann, Marc Becker, Marieke Behlen, Jessica Beyerl, Rebecca Böhnlein, Anna Brauer, Vera Britz, Friedrich Caroli, Lorenzo Contento, Alina Czwienzek, Flora Deák, Emma Dech, Laura Dech, Jana Diekmannshemke, Anna Do, Gerhard Dobler, Jürgen Durner, Ute Eberle,


Judith Eckstein, Tabea M. Eser, Philine Falk, Volker Fingerle, Stefanie Fischer, Marius Gasser, Sonja Gauder, Otto Geisenberger, Christof Geldmacher, Leonard Gilberg, Kristina Gillig, Philipp Girl, Elias Golschan, Vitus Grauvogl, Celina Halfmann, Tim Haselwarter, Arlett Heiber, Matthias Herrmann, Stefan Hillmann, Christian Hinske, Janna Hoefflin, Tim Hofberger, Michael Höfinger, Larissa Hofmann, Sacha Horn, Kristina Huber, Christian Janke, Ursula Kappl, Charlotte Kiani, Isabel Klugherz, Norah Kreider, Arne Kroidl, Magdalena Lang, Clemens Lang, Silvan Lange, Ekaterina Lapteva, Michael Laxy, Reiner Leidl, Felix Lindner, Alexander Maczka, Alisa Markgraf, Paula Matcau, Rebecca Mayrhofer, Anna-Maria Mekota, Hannah Müller, Katharina Müller, Leonie Pattard, Claire Pleimelding, Stephan Prückner, Konstantin Pusl, Elba Raimúndez, Julius Raschka, Jakob Reich, Raquel Rubio-Acero, Nicole Schäfer, Paul Schandelmaier, Lara Schneider, Sophie Schultz, Mirjam Schunk, Lars Schwettmann, Heidi Seibold, Paul Stapor, Jeni Tang, Fabian Theis, Sophie Thiesbrummel, Eva Thumser, Niklas Thur, Julian Ullrich, Julia Waibel, Claudia Wallrauch, Julia Wolff, Pia Wullinger, Tobias Würfel, Patrick Wustrow, Houda Yaqine, Sabine Zange, Eleftheria Zeggini, Thorbjörn Zimmer, Thomas Zimmermann, Lea Zuche Word count text body: 3496 Word count abstract: 233 Number of references: 30 Number of tables: 1 Number of figures: 6


Abstract

Background. Population-based studies investigating the dynamics of the SARS-CoV-2 pandemic are needed. Here, we report on baseline findings from April through June 2020 of a prospective cohort study in Munich, Germany.

Methods. We drew a representative sample of 2994 private households. The 5313 participating household members 14 years and older completed questionnaires and provided blood samples. SARS-CoV-2 seropositivity was defined as Roche N pan-Ig ≥ 0ꞏ4218. We adjusted the prevalence for the sampling design, sensitivity, and specificity. We investigated risk factors for SARS-CoV-2 seropositivity and geospatial transmission patterns by generalized linear mixed models and permutation tests.

Findings. Seropositivity for SARS-CoV-2 specific antibodies was 1ꞏ82% (95% confidence interval (CI) 1ꞏ28-2ꞏ37%) as compared to 0ꞏ46% PCR-positive cases officially registered in Munich. Loss of the sense of smell or taste was associated with seropositivity (odds ratio (OR) 47ꞏ4; 95% CI 7ꞏ2-307ꞏ0) and infections clustered within households. Participants suffering from respiratory allergies (OR 2ꞏ7; 95% CI 0ꞏ9-8ꞏ6) and working in high-risk jobs (OR 2ꞏ0; 95% CI 0ꞏ5-6ꞏ7) showed non-significantly increased odds for SARS-CoV-2 seropositivity.

Interpretation. Applying a validated assay, we demonstrate a low SARS-CoV-2 seroprevalence in the Munich population 14 years and older towards the end of the first pandemic wave. However, we noted official sub-registration at this early stage of the pandemic.

Funding. Bavarian State Ministry of Science and the Arts, University Hospital of LMU Munich, Helmholtz Centre Munich, University of Bonn, and University of Bielefeld. Keywords. COVID-19; SARS-CoV-2; population-based cohort study; seroprevalence; seroepidemiology; serosurvey; infection fatality ratio; underreporting; risk factors; cluster; Germany


Research in context

Evidence before this study

To estimate the burden of disease, population-based seroprevalence studies are considered as gold standard. Yet, for COVID-19, only a limited number of peer-reviewed articles have been published. On November 6, 2020, we found fifteen papers in PubMed using the terms “seroprevalence” or “seroepidemiology” and “SARS-CoV-2” and “population-based” without restrictions. Screening the abstracts, only five of them actually included population-representative samples. More studies were published on preprint servers without peer review, one of which was a systematic review summarising five additional population-based studies. Among other limitations, the low specificity of the used test methods reduced validity of the seroprevalence estimates.

Added value of this study

Here, we present cross-sectional results of a representative sample of the Munich population 14 years and older living in private households. In this population, we measured the seroprevalence of SARS-CoV-2 specific antibodies towards the end of the first wave of infections using a previously validated test strategy. Our findings provide evidence that during the first wave of the pandemic, only about one in four to five SARS-CoV-2 infections were officially registered by the authorities. No major risk factors were identified. We confirm household clustering of SARS-CoV-2 and provide indications for clustering of the disease within apartment buildings.

Implications of all the available evidence

Our data and previous findings indicate that after the first wave of infections most European populations are still far from a level of immunity considered protective in transmission models. Risk of death due to SARS-CoV-2 infection seems to be high. Follow-ups of the study population will help to determine disease dynamics.


Introduction

The COVID-19 pandemic has changed life across the globe. The global case number, based on positive PCR results, is ever increasing. In Germany, the first COVID-19 case was diagnosed in Munich on January 27, 2020.1 While further spread could be limited to a cluster of 15 co-workers and their families by thorough contact tracing and quarantine, the next cases were detected only five weeks later. Following this, the number of SARS-CoV-2 infected individuals in Munich increased exponentially from 44 cases diagnosed by March 7, to 3304 cases diagnosed by April 3, 2020.2 To control the pandemic, an eight-week long lockdown phase began on March 16 including restrictions such as school closures.

So far, the population seroprevalence of SARS-CoV-2, including asymptomatic individuals or patients with mild symptoms not presenting to the health system, is largely unknown. This knowledge is crucial to estimate the burden of disease in a population, including true reproduction numbers and attack rates.3,4 Up to now, only few peer-reviewed data from population-based cross-sectional serosurveys on SARS-CoV-2 antibody prevalence at different time points are available. A recent meta-analysis additionally evaluated pre-prints of seroprevalence studies.5 The reported seroprevalences ranged from below 1% in Iceland to more than 30% in Guilan province, Iran.6,7

In Germany, one population-based study was conducted after a superspreading event in a smaller town, showing an adjusted Immunoglobulin G (IgG) seroprevalence close to 20%.8 These previous serosurveys used diverse sampling as well as testing methods with varying validity, which affects comparability as well as interpretability of results.3 Although testing for SARS-CoV-2 specific antibodies seems better for defining the real dimension of past infections than PCR testing, high sensitivity and especially specificity are crucial due to the low prevalence at the population level.9

Here, we report baseline data from the ongoing population-based prospective cohort study (KoCo19) in Munich, Germany.10 Aims of the present analyses were to (i) estimate the seroprevalence of SARS-CoV-2 specific antibodies in the Munich population living in private households after the first pandemic wave, (ii) identify risk factors for SARS-CoV-2 seropositivity, and to (iii) assess household and neighbourhood transmission patterns.

We established and validated our test algorithm in parallel to the recruitment of study participants.11


Methods

Study design, setting and population. A detailed description of the study design, setting and population was previously published.10 In short, we carried out the fieldwork between April 5 and June 12, 2020, in which we selected a random sample of 100 out of 755 Munich constituencies as starting points to represent the Munich population (Figure 1A). Using random route methodology, fieldworkers selected approximately 30 households per constituency starting from its geographic centre. They partly crossed the borders of constituencies, resulting in a selection of 2994 households in 368 of the 755 constituencies (Figure 1B). If multi-party houses were selected, we aimed to include one household per floor to study transmission dynamics within buildings. The mean number of households recruited per house varied between one and seven across constituencies (Figure 1C). In each selected household, all members were invited to participate in the study to assess within-household transmissions (Figure 1D). More details on ethical as well as statistical considerations and on sampling design are given in the appendix (pp 4-6).

Specimen collection and laboratory analyses. During study visits at each household, fieldworkers collected venous blood samples using EDTA tubes from each consenting study participant 14 years and older. For ethical reasons, younger children could not provide venous blood samples at this stage. All laboratory methods are described in detail in the adjacent manuscript.11 In brief, we determined antibody reactivity using Anti-SARS-CoV-2-ELISA for IgG / IgA (Euroimmun Anti-S1-SARS-CoV-2-ELISA-IgG hereafter EI-S1-IgG / Euroimmun Anti-S1-SARS-CoV-2-ELISA-IgA hereafter EI-S1-IgA), and the Elecsys® Anti-SARS-CoV-2 Roche anti-N pan-Ig (hereafter Ro-N-Ig). We also used the GenScript® cPass assay. For serological confirmation, we used a virus micro-neutralisation test as described previously.12 To obtain accurate seroprevalences, we performed validation studies using a panel of 991 truly SARS-CoV-2 negative plasma samples from the pre-COVID-19 era and 193 samples from PCR confirmed COVID-19 patients.11

While the agreement between EI-S1-IgG and Ro-N-Ig assays was generally high (appendix p 10), the latter gave more valid results. For this assay an optimised cut-off of 0ꞏ4218 (instead of 1ꞏ0) yielded a sensitivity of 88ꞏ60% and a specificity of 99ꞏ72%.11 We therefore used the Ro-N-Ig assay with this optimised cut-off to determine seropositivity for SARS-CoV-2 in our analyses. Robustness of the prevalence estimates was tested with EI-S1-IgG, EI-S1-IgA, and combinations of different assays as well as different cut-offs and measures of test performance to predict seropositivity (appendix pp 10-11, 23).

Household and personal data collection. During household visits, field workers used the mobile data collection tool OpenDataKit (ODK) to capture contact details of household members on Android smartphones. Participants completed a household form and a personal questionnaire online using a newly developed web-based application. Non-responders were reminded by email no later than two weeks after the household visit, followed by telephone reminders. Telephone interviews were offered to those who felt unable to complete the questionnaires online.

Statistical analyses. All statistical analyses were performed using the statistical software R (version 4.0.2, R Development Core Team, 2020).


We calculated absolute and relative frequencies of sociodemographic and household variables and compared them to data of the general Munich population. Online and telephone responses were compared using Fisher's exact test if there were only two categories and chi-squared test if there were more than two categories.

To assess the seroprevalence, we defined SARS-CoV-2 seropositivity based on the Ro-N-Ig test result applying the optimised cut-off of 0ꞏ4218 as described above. To account for the sampling design, the sampling weights were computed and calibrated such that the sample structure mirrored the Munich population (regarding age, sex, migration background, presence of children in the households, single member households). Prevalence estimates were calculated using these calibrated weights and 95% confidence intervals were computed based on the variability associated to the sampling design. These prevalence estimates were additionally adjusted for sensitivity and specificity of the test as described by Sempos and colleagues (for details on sampling design and sensitivity/specificity adjustments see appendix pp 5-6).9 We calculated the infection fatality ratio (IFR) for the population aged 14 years and older using the seroprevalence estimates for Munich as described above and officially reported numbers of COVID-19 deaths. Fatality counts follow a binomial distribution with small success probability parameter and relatively large number of trials, thus can be approximately described by a Poisson distribution. This assumption leads to estimates of seroprevalence and corresponding 95% confidence intervals (CI), both for the entire period of our study as well as for weekly incidences. Official numbers include both subjects living in private households and institutions (e.g., homes for the elderly). To the best of our knowledge, there exists no data for Munich on the percentages of SARS-CoV-2 infections and deaths occurring in institutions. For Germany, the Robert-Koch-Institute (RKI) reported that up to the end of the study period, 13% of infections and 46% of deaths occurred in institutions.13 Given the lack of data for Munich, we estimated the IFR for Munich assuming that the percentage of deaths occurring in members of private households lies in the range between 20 and 100. Likewise, we calculated the factor of underreported infections assuming that between 20 and 100% of reported infections occurred outside institutions (appendix pp 5-6).

In the risk factor analyses, associations between the personal and household level covariates and seropositivity were evaluated using logistic regression, i.e. generalized linear models (GLM). We adjusted for age and sex and assumed item-nonresponse to be missing at random. The computed odds ratios (OR) and 95% CIs were compared to the results of generalised linear mixed models (GLMM) which allowed us to consider the effect of household clustering during the estimation process.14 For sensitivity analysis, we imputed missing values (of covariates) under the (Bayesian) Joint Analysis and Imputation of Incomplete Data (JointAI) framework, which allowed us to avoid pooling.15 The identified important risk factors for seropositivity were included in multiple regression models. Specifically, we compared the frequentist setup of GLM and GLMM as well as a GLMM using simultaneous imputation of multiple missing covariates under the Bayesian framework (appendix p 8).

To analyse whether SARS-CoV-2 infections were spatially clustered, we used the similarity of seropositivity levels within spatial clusters of different sizes, i.e. households, buildings, and geospatial clusters of various distances. As a test statistic, we employed the average within-cluster variance. To assess significance, we performed a non-parametric approximate permutation test with n=10000 randomly permuted measurement assignments.16 To account


for household clustering, when analysing buildings and geospatial clusters, we only permuted households of the same size (appendix p 9).

For official incidence and mortality, as well as for data on the general population, we used data provided by the Statistical Office of the City of Munich.

Role of the funding sources. The funders had no role in study design, data collection, data analyses, data interpretation, writing or submission of this manuscript.

Ethics committee approval. Prior to study initiation, this study had been approved by the respective Institutional Review Board (appendix p 4).


Results

Description of the study samples and population. Of the 6896 households identified, 4903 were eligible and 2994 were included in the analyses. Within these households, fieldworkers invited 6117 persons to participate, of which 5313 agreed and provided blood samples (Figure 2).

The study population was comparable to the Munich population in many aspects (Table 1). However, it contained less children and adolescents (5%) than the general population (17%) as children younger than 14 years had to be excluded. In addition, persons born outside Germany were underrepresented (18% in the study population vs. 31% in the general population). Regarding household characteristics, the sampling design resulted in a preference for larger apartment buildings with 71% of the study population living in apartment houses with five or more apartments compared to 34% of the Munich population.

Some groups were more likely to participate in telephone interviews than to complete the online questionnaire: Women (59% of telephone interviewees vs. 52% of online interviewees), participants aged 65+ years (66% vs. 12%), and subjects with a lower level of education (72% vs. 26%). These marked differences were also reflected in other personal characteristics like employment status, smoking behaviour, as well as in health-related-parameters (appendix p 18).

Prevalence of SARS-CoV-2 specific antibodies and mortality. The highest number of PCR-positive individuals reported by the Statistical Office of the City of Munich was registered in March 2020 before the start of the fieldwork, reaching a maximum in week 14 (Figure 3A and 3D). The overall number of PCR-positive individuals 14 years and older from the beginning of March until the end of the fieldwork was 6293. This corresponded to a positivity rate of 0ꞏ46%. We began the fieldwork in week 15 and ended in week 24 (Figure 3B). Over these weeks, official case numbers went markedly down. In contrast, Ro-N-Ig seroprevalence stayed stable throughout the study period (Figure 3E and appendix p 15). Using the optimised cut-off, the crude Ro-N-Ig seropositivity for the whole study population was 1ꞏ75% (95% CI 1ꞏ28-2ꞏ22%), with similar results when accounting for the sampling design using calibrated weights (1ꞏ89%; 95% CI 1ꞏ41-2ꞏ37%). Adjusting for test sensitivity and specificity slightly lowered the prevalence estimate (unweighted: 1ꞏ67%; CI 1ꞏ13-2ꞏ20%; weighted: 1ꞏ82%; CI 1ꞏ28-2ꞏ37%) (appendix p 12). The estimated number of cases was about four fold the official number (Figure 3C). Assuming (based on data for whole Germany) that 87% of infections occurred in private households, 4.5 times (95% CI 3.2-5.9) more infections than reported occurred (for details, see appendix p 6).

Based on official mortality data for the Munich population, we observed an excess mortality in weeks 14 to 19, 2020 compared to the previous four years (Figure 3G and 3H). The resulting overall number of deaths over the whole period was similar to the 216 COVID-19 deaths reported by the Statistical Office of Munich (Figure 3I). Assuming the weighted and adjusted Ro-N-Ig prevalence of 1ꞏ82% for the Munich population 14 years and older (1369444 inhabitants), we estimated that until the end of the field work 24990 individuals developed SARS-CoV-2 antibodies (for details on this and the following calculations see appendix p 6). Up to the end of the fieldwork, the 216 registered COVID-19 related deaths yielded an IFR of


0ꞏ86% (95% CI: 0ꞏ67-1ꞏ23%) (Figure 3F). Estimating that only 54% of them occurred in private households, IFR lowered to 0ꞏ47% (95% CI: 0ꞏ36-0ꞏ67%).

Associations with SARS-CoV-2 seropositivity. Bivariate analyses suggested that Ro-N-Ig seropositivity depended only weakly on most surveyed factors. Loss of the sense of smell or taste at the time of the study was associated with the outcome, however, the confidence interval was wide (OR 41ꞏ3; 95% CI 6ꞏ7-231ꞏ0) using a classical GLMM. In addition, respiratory allergies (OR 3ꞏ3; 95% CI 1ꞏ1-10ꞏ3) were statistically significantly associated with Ro-N-Ig seropositivity. Only weakly related to the outcome were working in a high-risk job, household type, and living area per inhabitant (Figure 4 and appendix pp 19-21). Besides the loss of the sense of smell or taste, which was considered a symptom of the outcome rather than a risk factor, we included these variables in the final GLMM in which none of the associations was statistically significant. When we applied Bayesian GLMM with imputation of missing values in the sensitivity analyses, we obtained similar results (Figure 5 and appendix p 22).

SARS-CoV-2 transmission within households, buildings and neighbourhoods. Ro-N-Ig test outcomes of participants had a significantly lower variance within households than among the entire population (Figure 6 and appendix p 9). In contrast, we did not find a statistically lower variance within buildings (p=0ꞏ26) nor within neighbourhoods applying radii from 50m (p=0ꞏ16) to 4000m (p=0ꞏ78). Yet, a lower than expected mean variance was seen up to a distance of 200m.


Discussion

We present an estimate for the SARS-CoV-2 seroprevalence in the Munich general population 14 years and older which is still low towards the end of the first pandemic wave (1ꞏ82%). However, our results indicate that the seroprevalence was substantially higher than official numbers based on registered PCR-positive cases. We could only identify weak risk factors for SARS-CoV-2 seropositivity. Finally, our data confirms household clustering of infection.17

As study participants were enrolled at a time when the newly released serological assays were not fully validated, we carefully evaluated three primary serological assays.11 As similarly reported by Gudbjartsson et al., the specificity and sensitivity of Roche anti-N pan-Ig was superior to Euroimmun.6 By using the cut-off index raw values, we were able to optimise the cut-off based on our study population.

Our prevalence estimates are in line with findings from Gudbjartsson et al. who estimated the SARS-CoV-2 antibody seropositivity for the general population of Iceland at around 1% and thus, slightly lower than in our population.6 Other studies estimating seroprevalence for European general populations reported results between 2% in Luxembourg,18 4% in Spain,19 and 11% in Geneva, Switzerland.20 The proportion of officially registered cases vs. the number of serologically positive cases also varied significantly between the studies. In our study, about one in four to five of seropositive cases were officially registered. One must consider that we did not have data on the seroprevalence among children younger than 14 years. For them, only one out of six infections might have been officially registered as recently shown by Hippich et al.21 for children living in Bavaria, Germany. In addition, our study population did not include institutions in which residents might be more frequently tested than in other settings. Therefore, our estimate might even be conservative for the Munich overall population. In other population-based European studies, 10-56% of infections were detected by the healthcare system.6,19,20 We believe that the proportion of registered vs. seropositive but not registered individuals could be used to measure the efficiency of a public health testing system.

While only a small proportion of the general population in Germany live in institutions, official data of the Robert-Koch-Institute (RKI) report that during the study period 13% of reported SARS-CoV-2 infections and 46% of COVID-19 related deaths occurred in institutionalised persons.13 Therefore, we report a range of IFR depending on the percentage of deaths occurring in private households. Translating the RKI numbers to Munich, the IFR resulting from our study is 0.47% (95% CI 0.36-0.67%). IFR calculations are difficult to compare for the reasons mentioned above and e.g. due to different approaches or unequal case ascertainment. The IFR resulting from our study is in line with recent data from Geneva (Switzerland; 0ꞏ64; 95% CI 0ꞏ38-0ꞏ98%) but slightly lower than the one reported for Spain (0.83; 95% 0ꞏ78-0ꞏ89).22,23

Among the risk factors identified, albeit not statistically significant, the increased risk of working in a job with a high potential of contact to COVID-19 cases appears plausible and is in line with other studies.6,19,24 Among the COVID-19 cases reported globally to the World Health Organization, 14% belong to the group of healthcare workers whereas this group represents less than 3% of the general population in most countries.25 Special preventive efforts should therefore be targeting the group of workers occupationally exposed to SARS-CoV-2. We saw slightly increased odds of seropositivity among participants with respiratory co-


morbidity, especially patients with allergies. Angiotensin‐converting enzyme 2 (ACE2) receptors are thought to be main cell entries for SARS-CoV-2. Differences in expression levels of ACE2 in patients with allergic asthma might be one reason for the increased risk of SARS-CoV-2 seropositivity in our study population.26 However, our data on respiratory allergies were self-reported and therefore, specificity might be low.

Participants with SARS-CoV-2 antibody responses more frequently reported the loss of the sense of smell or taste even though the questionnaire was answered up to several weeks after the blood samples, which is consistent with other studies.27 The underlying pathological mechanism might again be explained by the high concentration of ACE2 in olfactory cells.27

Household clustering of SARS-CoV-2 was described before and is well known from other respiratory infections.17,24,28,29 Apart from transmission from one person to the other within one household, similar behaviour and contacts outside the own household might explain this. Our results weakly suggest clustering within the same house and eventually the close neighbourhood. However, these associations were not statistically significant and need to be confirmed. With respect to within-household transmission, our seroprevalence observations are limited by the fact that we did not take any blood samples from children below 14 years. Since the number of children in a household is presumably positively correlated with household size, this biases our data, and within-household transmission may even be more pronounced than reported here.

The major strengths of this study are its population-based approach, weighting of results for the general Munich population, the high number of participants, the thorough validation of the assays used, and (where available) the use of validated questionnaire items. The overall response to the study was high compared to other population-based epidemiologic studies in Germany.30 While most participants completed the questionnaire online, we also provided the alternative of telephone interviews increasing participation and thus, making our study population more representative for the target population. Our study started towards the end of the first pandemic wave in Munich as illustrated by the incidence of SARS-CoV-2 registered cases and mortality. This resulted in a relatively stable antibody prevalence over the course of the fieldwork.

As mentioned, a relevant limitation of our study is the exclusion of children and residents not living in private households. While in general, people with migration background are less likely to participate in population-based studies, the lack of translated questionnaires further limited the number of migrants participating in our study.10 To increase response, blood samples were collected at participants' homes and not at a centralised testing facility. However, we could not interview participants during home visits. This led to lower questionnaire response and a time interval of up to six weeks between blood sample collection and completion of the questionnaire. Item non-response in online completed questionnaires was higher compared to personal interviews. As in other studies, questions on income (21% missing responses) as well as weight and height (73% missing responses) were especially prone to non-response. They were thus not included in the final models. Despite the high number of participants, the power for our risk factor analyses was relatively low due to the low seroprevalence, resulting in large confidence intervals and potentially negative findings. However, one may assume that any strong risk factors would have been identified if included.


Although commercial and scalable testing methods for SARS-CoV-2 seem robust, cut-offs should be adapted to specific populations to improve their validity. However, the relevance of antibody-based testing to assess previous infection remains partly unclear. To learn more about the meaning of detected SARS-CoV-2 specific antibodies, correlates of protection, and testing methodologies, this as well as other longitudinal cohort studies need to be continued.

Even when considering all these uncertainties, our results indicate that only a very small proportion of the Munich population encountered SARS-CoV-2 during spring 2020, most citizens stay vulnerable to infection, and the associated risk of death among those infected was high.


Contributors

MH is the principal investigator of this study and obtained most necessary funds. MH, KR, and ES conceived the study with input from MND, FF, JMGN, IK, LO, MPl, MPr, FR, VT, and AW. Sample collection was led by JMGN, IK, DM, LO, MPr, and VT with assistance from IB, JB, JF, GF, KP, CR, and SW. The laboratory set-up and sample processing were led by AW. RW was responsible for virus micro-neutralisation tests. MND coordinated the development of the web-based study application with input by ES. MND, FF, KR, FR, ES, and SW were coordinating data acquisition and data management. Data was cleaned and prepared by NC, MND, FF, and SW. Statistical analyses and data visualization were done by AB, NC, TF, MG, RLG, PP, and YS. CF and JH lead the statistical analyses. CF, JH, KR, FR, ES, and SW conceptualized the result presentation with input from MH, IK, LO, MPr, and AW. The manuscript was primarily written by MPr and KR with significant contributions from AB, NC, TF, CF, MG, RLG, JH, MH, IK, LO, PP, ES, YS, and AW. All authors revised and approved the final version of this manuscript. Declaration of interests

FF, TF, DM, LO, and VT report grants from the Bavarian State Ministry of Science and the Arts during the conduct oft he study. TF reports grants from the University Hospital of LMU Munich, Helmholtz Center Munich, University of Bonn, University of Bielefeld, and German Ministry for Education and Research during the conduct of the study. JH reports grants from the German Federal Ministry of Education and Research during the conduct of the study. MH and AW report personal fees and non-financial support, LO and MP report non-financial support from Roche Diagnostics. MH, LO, MP, and AW report non-financial support from Euroimmun, Viramed, and Mikrogen. MH, MP, and AW report grants, non-financial support, and other from German Center for Infection Research (DZIF). FF, MH, LO, MP, VT, and AW report grants and non-financial support from the Government of Bavaria. MH, LO, MP, and AW report non-financial support from BMW, Mercedes Benz, Munich Police, and Accenture. MH and AW report personal fees and non-financial support from Dr. Box Betrobox during the conduct of the study. LO and MP report non-financial support from Dr. Box Betrobox. MH and AW have a patent Sample System for Sputum Diagnostics of SARS-CoV-2 pending. DM reports to be a a sub-investigator on a Phase I SARS-CoV-2 vaccine trial and on a Phase I rabies vaccine trial, both sponsored by CureVac AG. MP and AW report non-financial support from Dr. Becker MVZ. VT reports support from CureVac AG outside the submitted work. AW reports personal fees and other from Haeraeus Sensors. AW reports non-financial support from Bruker Daltronics outside the submitted work. AW is involved in other different patents and companies not in relation with the serology of SARS-CoV-2. All other authors report nothing to disclose.

Data Sharing

Our data are accessible to researchers upon reasonable request to the corresponding author taking data protection laws and privacy of study participants into account. Acknowledgments


We gratefully thank all study participants for their trust, time, data, and specimen. We are grateful for the financial support of the Bavarian State Ministry of Science and the Arts, the University Hospital of LMU Munich, the Helmholtz Centre Munich, the University of Bonn, and the University of Bielefeld. JH acknowledges financial support by the German Ministry for Education and Research via the project MoKoCo19 (reference number: 01KI20271). MG acknowledges support from the Joachim Herz Foundation through the Add-on Fellowship for Interdisciplinary Science. This study would also not have been possible without the staff of the Division of Infectious Diseases and Tropical Medicine at the University Hospital of LMU Munich, Helmholtz Centre Munich, Bundeswehr Institute of Microbiology, as well as all medical students involved. We thank Judith Eckstein for outstanding support regarding public relations. We thank the teams from the press offices of LMU, University Hospital of LMU Munich, and of Helmholtz Centre Munich. We thank the KoCo19 advisory board members Stefan Endres, Stephanie Jacobs, Bernhard Liebl, Michael Mihatsch, Matthias Tschöp, Manfred Wildner, and Andreas Zapf. We thank Accenture for the development of the KoCo19 web-based survey application. We are grateful to the Statistical Office of the City of Munich, Germany, for providing statistical data on the Munich general population. We thank Helmut Küchenhoff for a critical review of an earlier version of the manuscript and Jared Anderson for English language corrections. We also thank Guillaume Chauvet for advice on the sampling design and variance estimation. We are grateful to the Munich police for their support in the fieldwork. The Munich Surgical Imaging GmbH, Cisco Systems, and the graphic/photo/IT infrastructure departments at the University Hospital of LMU Munich provided support during video production and online events. For fieldwork, BMW Group as part of their campaign “BMW hilft Helfenden” provided free cars. Mercedes-Benz Munich provided support with Mercedes-Benz Rent in the project infrastructure. Funding

This study was funded by the Bavarian State Ministry of Science and the Arts, the University Hospital of LMU Munich, the Helmholtz Centre Munich, the University of Bonn, the University of Bielefeld, and the German Ministry for Education and Research (MoKoCo19, reference number 01KI20271). Euroimmun, Mikrogen, Roche, and Viramed provided kits and machines for analyses at discounted rates.


References

1. Rothe C, Schunk M, Sothmann P, Bretzel G, Froeschl G, Wallrauch C, et al. Transmission of 2019-nCoV Infection from an Asymptomatic Contact in Germany. N Engl J Med. 2020 Mar 5;382(10):970–1.

2. Coronavirus in München: Die aktuelle Entwicklung [Internet]. 2020 [cited 2020 Oct 9]. Available from: https://www.muenchen.de/aktuell/2020-03/coronavirus-muenchen-infektion-aktueller-stand.html

3. Pearce N, Vandenbroucke JP, VanderWeele TJ, Greenland S. Accurate Statistics on COVID-19 Are Essential for Policy Guidance and Decisions. Am J Public Health. 2020 Jul;110(7):949–51.

4. Lipsitch M, Swerdlow DL, Finelli L. Defining the Epidemiology of Covid-19 — Studies Needed. N Engl J Med. 2020 Mar 26;382(13):1194–6.

5. Meyerowitz-Katz G, Merone L. A systematic review and meta-analysis of published research data on COVID-19 infection fatality rates. International Journal of Infectious Diseases. 2020 Dec;101:138–48.

6. Gudbjartsson DF, Norddahl GL, Melsted P, Gunnarsdottir K, Holm H, Eythorsson E, et al. Humoral Immune Response to SARS-CoV-2 in Iceland. N Engl J Med. 2020 Sep 1;NEJMoa2026116.

7. Shakiba M, Hashemi Nazari SS, Mehrabian F, Rezvani SM, Ghasempour Z, Heidarzadeh A. Seroprevalence of COVID-19 virus infection in Guilan province, Iran [Internet]. Infectious Diseases (except HIV/AIDS); 2020 May [cited 2020 Aug 5]. Available from: http://medrxiv.org/lookup/doi/10.1101/2020.04.26.20079244

8. Streeck H, Schulte B, Kümmerer BM, Richter E, Höller T, Fuhrmann C, et al. Infection fatality rate of SARS-CoV2 in a super-spreading event in Germany. Nat Commun. 2020 Dec;11(1):5829.

9. Sempos CT, Tian L. Adjusting Coronavirus Prevalence Estimates for Laboratory Test Kit Error. American Journal of Epidemiology. 2020 Aug 17;kwaa174.

10. the KoCo19 collaboration group, Radon K, Saathoff E, Pritsch M, Guggenbühl Noller JM, Kroidl I, et al. Protocol of a population-based prospective COVID-19 cohort study Munich, Germany (KoCo19). BMC Public Health. 2020 Dec;20(1):1036.

11. Olbrich L, et al. A Serology Strategy for Epidemiological Studies Based on the Comparison of the Performance of Seven Different Test Systems - The Representative COVID-19 Cohort Munich. Adjacent manuscript.

12. Haselmann V, Özçürümez MK, Klawonn F, Ast V, Gerhards C, Eichner R, et al. Results of the first pilot external quality assessment (EQA) scheme for anti-SARS-CoV2-antibody testing. Clin Chem Lab Med. 2020 Aug 27;58(12):2121-2130.

13. Robert Koch Institut. 10. Täglicher Lagebericht (13.06.2020) des RKI zur Coronavirus-Krankheit-2019 (COVID-19) – Aktualisierter Stand für Deutschland [Internet]. 2020 [cited 2020 Dec 2]. Available from: https://www.rki.de/DE/Content/InfAZ/N/Neuartiges_Coronavirus/Situationsberichte/2020-06-13-de.pdf;jsessionid=2A2033699409CB8BA030EB0F22EA0BD4.internet062?__blob=publicationFile

14. Molenberghs G, Verbeke G. Models for discrete longitudinal data. New York ; London: Springer; 2005. 683 p. (Springer series in statistics).


15. Erler NS, Rizopoulos D, Lesaffre EMEH. JointAI: Joint Analysis and Imputation of Incomplete Data in R. arXiv:190710867 [stat] [Internet]. 2020 Sep 1 [cited 2020 Dec 2]; Available from: http://arxiv.org/abs/1907.10867

16. Nichols TE, Holmes AP. Nonparametric permutation tests for functional neuroimaging: A primer with examples. Hum Brain Mapp. 2002 Jan;15(1):1–25. 1.

17. Lee EC, Wada NI, Grabowski MK, Gurley ES, Lessler J. The engines of SARS-CoV-2 spread. Science. 2020 Oct 23;370(6515):406–7.

18. Snoeck CJ, Vaillant M, Abdelrahman T, Satagopam VP, Turner JD, Beaumont K, et al. Prevalence of SARS-CoV-2 infection in the Luxembourgish population: the CON-VINCE study. [Internet]. Infectious Diseases (except HIV/AIDS); 2020 May [cited 2020 Aug 5]. Available from: http://medrxiv.org/lookup/doi/10.1101/2020.05.11.20092916

19. Pollán M, Pérez-Gómez B, Pastor-Barriuso R, Oteo J, Hernán MA, Pérez-Olmeda M, et al. Prevalence of SARS-CoV-2 in Spain (ENE-COVID): a nationwide, population-based seroepidemiological study. The Lancet. 2020 Aug;396(10250):535–44.

20. Stringhini S, Wisniak A, Piumatti G, Azman AS, Lauer SA, Baysson H, et al. Seroprevalence of anti-SARS-CoV-2 IgG antibodies in Geneva, Switzerland (SEROCoV-POP): a population-based study. The Lancet. 2020 Jun;S0140673620313040.

21. Hippich M, Holthaus L, Assfalg R, Zapardiel-Gonzalo J, Kapfelsperger H, Heigermoser M, et al. A Public Health Antibody Screening Indicates a 6-Fold Higher SARS-CoV-2 Exposure Rate than Reported Cases in Children. Med. 2020 Oct;S2666634020300209.

22. Pastor-Barriuso R, Perez-Gomez B, Hernan MA, Perez-Olmeda M, Yotti R, Oteo J, et al. Infection fatality risk for SARS-CoV-2: a nationwide seroepidemiological study in the non-institutionalized population of Spain [Internet]. Epidemiology; 2020 Aug [cited 2020 Dec 2]. Available from: http://medrxiv.org/lookup/doi/10.1101/2020.08.06.20169722

23. Perez-Saez J, Lauer SA, Kaiser L, Regard S, Delaporte E, Guessous I, et al. Serology-informed estimates of SARS-CoV-2 infection fatality risk in Geneva, Switzerland. The Lancet Infectious Diseases. 2020 Jul;S1473309920305843.

24. McDade TW, McNally EM, Zelikovich AS, D’Aquila R, Mustanski B, Miller A, et al. High seroprevalence for SARS-CoV-2 among household members of essential workers detected using a dried blood spot assay. Khudyakov YE, editor. PLoS ONE. 2020 Aug 14;15(8):e0237833.

25. WHO. Keep health workers safe to keep patients safe: WHO [Internet]. 2020. Available from: https://www.who.int/news/item/17-09-2020-keep-health-workers-safe-to-keep-patients-safe-who

26. Yao Y, Wang H, Liu Z. Expression of ACE2 in airways: Implication for COVID‐19 risk and disease management in patients with chronic inflammatory respiratory diseases. Clin Exp Allergy. 2020 Dec;50(12):1313–24.

27. Ibekwe TS, Fasunla AJ, Orimadegun AE. Systematic Review and Meta-analysis of Smell and Taste Disorders in COVID-19. OTO Open. 2020 Jul;4(3):2473974X2095797.

28. Tsang TK, Lau LLH, Cauchemez S, Cowling BJ. Household Transmission of Influenza Virus. Trends in Microbiology. 2016 Feb;24(2):123–33.


29. Laxminarayan R, Wahl B, Dudala SR, Gopal K, Mohan B C, Neelima S, et al. Epidemiology and transmission dynamics of COVID-19 in two Indian states. Science. 2020 06;370(6517):691–7.

30. Rach S, Günther K, Hadeler B. Participants who were difficult to recruit at baseline are less likely to complete a follow-up questionnaire – results from the German National Cohort. BMC Med Res Methodol. 2020 Dec;20(1):187.


Table 1: Individual and household characteristics of the KoCo19-study participants compared to the Munich population.

Characteristics Munich population

KoCo19 Study Participants

Individual Characteristics

N 1,561,720 5313 n % nmissing n %

Sex 0 Female 789,437 50∙1 2766 52∙1

Age (years) 0 0‐19 263,053 16∙8 267 5∙0

20‐34 390,382 25∙0 1346 25∙3 35‐49 348,651 22∙3 1542 29∙0 50‐64 291,562 18∙7 1306 24∙6 65‐79 184,764 11∙8 676 12∙7 80+ 83,308 5∙3 176 3∙3

Country of Birth 465 Outside Germany 476,575 30∙5 849 17∙5

Level of education NA NA 701 Still in school 100 2∙2

<12 yrs 1386 30∙1 ≥12 yrs 3126 67∙8

Employment Status NA NA 576 Employed 2911 61∙5

Self‐Employed 471 9∙9 Not working1 1258 26∙6

Others2 97 2∙0 Risk Employment3 NA NA 470

Yes 851 17∙6 Household characteristics

N 2994

Housing Type: Building with

148,607

100 0

1‐2 apartments 82,119 55∙3 661 22∙1 3‐4 apartments 10,938 7∙4 192 6∙4 ≥5 apartments 50,339 33∙9 2137 71∙4

Others4 5211 3∙5 4 0∙1 Household type 854,288 100 307

Single 468,937 54∙9 680 25∙3 Couple 160,339 18∙8 922 34∙3 Family 185,752 21∙7 875 32∙6 Others5 39,260 4∙6 210 7∙8

No. of Household Members

854,288

100 1

1 468,937 54∙9 784 26∙2

2 193,376 22∙6 1171 39∙1

3‐4 106,074 12∙4 880 29∙4

5+ 85,901 10∙1 158 5∙3


Characteristics Munich population

KoCo19 Study Participants

Living area per Inhabitant NA NA 319 ≤ 30 m2 800 29∙9

30‐40 m2 634 23∙7 40‐55 m2 579 21∙6 >55 m2 662 24∙7

Net Family Income NA NA 924

≤ 2500 € 445 21∙5 2500‐4000 € 502 24∙3 4000‐6000 € 607 29∙3

>6000+ € 516 24∙9 1 “Not working” includes unemployed, retired, parental leave, sabbatical, students 2 “Others” includes voluntary social year, military service, part‐time jobber, reduced working hours 3 Considered as “Risk employment” for COVID‐19 infections were employees in the: healthcare sector, emergency

service, senior homes, airport, public transport, education, sales, social work and other risk jobs. 4 Other types of housing include tents, caravans, or the like 5 Other household types include shared apartments by e.g., students, subleasing, and assisted accommodation


Figure 1: Selection procedure and geospatial distribution of the study population. (A) The municipality of Munich together with its districts (distinguished by colours). The 100 selected start constituencies for the random walks are marked by dark colour. (B) All 2994 included households and their respective 368 constituencies. (C) Average number of recruited households per building by constituency. (D) Average number of members per recruited household by constituency.


Figure 2: Flow chart on participant selection for the KoCo19 baseline survey

6896 households identified

5320 households contacted and reached

1576 contacted but not reached

417 households not eligible 252 replacement households 96 language barrier incidents 28 wrong phone number 10 apartments no longer occupied 5 moving within one month 4 unable to give informed consent 22 other reasons (e.g. illness)

4903 households eligible

1909 households not included 1813 refusals 83 no further response 8 no blood sample of any member 5 locations outside city border

2994 households included

6117 household members invited

5313 household members 14 years and older included

804 no blood sample provided


Figure 3: Dynamics of the COVID-19 pandemic and course of the KoCo19 study in Munich since the beginning of the pandemic to the end of the KoCo19 study period. (A) Official weekly absolute number of newly diagnosed COVID-19 cases based on positive PCR tests; (B) Weekly number of participants recruited to the KoCo19 study; (C) Estimated underreporting factor depending on the percentage of reported cases in private households with respect to all reported cases in Munich; (D) Cumulative weekly number of officially registered COVID-19 infections in Munich; (E) Numbers of Ro-N-Ig seropositive samples per week (blue) divided by the number of blood draws in the respective time frame. 95% CIs (blue dashed lines) are based on an approximate Poisson assumption. Black line and shaded area indicate the weighted and adjusted prevalence estimate with 95% CI. Due to low recruitment numbers in the last week, in (D) and (E) the data from the last week are integrated with the pre-last week; (F) Estimated infection fatality ratio depending on the percentage of reported COVID-19 related deaths in private households with respect to all reported COVID-19 related deaths in Munich; (G) Weekly number of deaths in Munich for 2016-2020 based on official numbers; (H) Weekly excess mortality in 2020 compared to 2016-2019 based on official death counts and official SARS-CoV-2 related deaths; (I) Comparison of total number of deaths based on excess mortality and registered SARS-CoV-2 related deaths.


Figure 4: Risk factor analysis for SARS-CoV-2 seropositivity in the KoCo19 study population comparing crude, adjusted for clustering and Bayesian (after imputation and adjusted for clustering) estimates. All ORs and 95% CIs are adjusted for age (continuous scale) and sex. OR odds ratio; 95% CI: 95% confidence interval (frequentist GLMM) / 95% credible interval (Bayesian analyses).


Figure 5: Multivariate risk factor analysis for SARS-CoV-2 seropositivity mutually adjusted for all variables in the figure. OR odds ratio; 95% CI: 95% credible interval (Bayesian analyses) / 95% confidence interval (frequentist GLMM).


Figure 6: Proximity clustering of Ro-N-Ig test outcomes. We subdivide the participants into disjoint clusters, according to various cluster definitions: households, buildings, and spatial clusters of various diameters (x-axis). For each cluster, we calculate the within-cluster variance of observed Ro-N-Ig test outcomes of all participants in the cluster. Their means over all clusters are marked by green horizontal lines for each cluster size. We then perform 10000 random permutations of measurements assignments. The black dots show the respective mean within-cluster variances; along with density estimates as grey curves. For buildings and spatial clusters, measurements of a household were only permuted with measurements of a household of the same size. p-values indicate the one-sided probability of a random value being smaller than or equal to the observed one.


1

Supplementary materials for

Prevalence and Risk Factors of Infection in the Representative COVID-19 Cohort Munich

Michael Pritsch1,2*, MD; Katja Radon3,4,5*, PhD; Abhishek Bakuli1*, PhD; Ronan Le Gleut6,7*, MSc; Laura Olbrich1, MD; Jessica Michelle Guggenbuehl Noller1, MSc; Elmar Saathoff 1,2,

PhD; Noemi Castelletti1, Dr; Mercè Garí6, PhD; Peter Pütz6,8, PhD; Yannik Schälte6,10, MSc; Turid Frahnow6,8, PhD; Roman Wölfel2,9, MD; Camilla Rothe1, MD; Michel Pletschette1,

MD; Dafni Metaxa1, MB BChir; Felix Forster3,5, MSc; Verena Thiel1, MD; Friedrich Rieß 1,2, Dr; Maximilian N. Diefenbach1, Dr; Günter Fröschl1,4, MD; Jan Bruger1; Simon Winter1;

Jonathan Frese1; Kerstin Puchinger1; Isabel Brand1; Inge Kroidl1,2, MD; Jan Hasenauer6,10,11+, Dr; Christiane Fuchs6,7,8,10+, PhD; Andreas Wieser1,2+, MD; Michael Hoelscher1,2,4+§, MD; on

behalf of the KoCo19 study group#

1Division of Infectious Diseases and Tropical Medicine, University Hospital, LMU Munich, 80802 Munich, Germany 2German Center for Infection Research (DZIF), partner site Munich, Germany 3Institute and Outpatient Clinic for Occupational, Social and Environmental Medicine, University Hospital, LMU Munich, 80336 Munich, Germany 4Center for International Health (CIH), University Hospital, LMU Munich, 80336 Munich, Germany 5Comprehensive Pneumology Center (CPC) Munich, German Center for Lung Research (DZL), 89337 Munich, Germany 6Institute of Computational Biology, Helmholtz Zentrum München - German Research Center for Environmental Health, 85764 Neuherberg, Germany 7Core Facility Statistical Consulting, Helmholtz Zentrum München - German Research Center for Environmental Health, 85764 Neuherberg, Germany 8Faculty of Business Administration and Economics, Bielefeld University, 33615 Bielefeld, Germany 9Bundeswehr Institute of Microbiology, 80937 Munich, Germany

10Center for Mathematics, Technische Universität München, 85748 Garching, Germany 11Interdisciplinary Research Unit Mathematics and Life Sciences, University of Bonn, 53113 Bonn, Germany

*Those authors contributed equally +Those authors contributed equally

§Corresponding author: Michael Hoelscher, MD Division of Infectious Diseases and Tropical Medicine University Hospital, LMU Munich Leopoldstr. 5, 80802 Munich, Germany Email: [email protected] Phone: +4989440059801


2

#KoCo19 study group: Emad Alamoudi, Jared Anderson, Valeria Baldassare, Maximilian Baumann, Marc Becker, Marieke Behlen, Jessica Beyerl, Rebecca Böhnlein, Anna Brauer, Vera Britz, Friedrich Caroli, Lorenzo Contento, Alina Czwienzek, Flora Deák, Emma Dech, Laura Dech, Jana Diekmannshemke, Anna Do, Gerhard Dobler, Jürgen Durner, Ute Eberle, Judith Eckstein, Tabea M. Eser, Philine Falk, Volker Fingerle, Stefanie Fischer, Marius Gasser, Sonja Gauder, Otto Geisenberger, Christof Geldmacher, Leonard Gilberg, Kristina Gillig, Philipp Girl, Elias Golschan, Vitus Grauvogl, Celina Halfmann, Tim Haselwarter, Arlett Heiber, Matthias Herrmann, Stefan Hillmann, Christian Hinske, Janna Hoefflin, Tim Hofberger, Michael Höfinger, Larissa Hofmann, Sacha Horn, Kristina Huber, Christian Janke, Ursula Kappl, Charlotte Kiani, Isabel Klugherz, Norah Kreider, Arne Kroidl, Magdalena Lang, Clemens Lang, Silvan Lange, Ekaterina Lapteva, Michael Laxy, Reiner Leidl, Felix Lindner, Alexander Maczka, Alisa Markgraf, Paula Matcau, Rebecca Mayrhofer, Anna-Maria Mekota, Hannah Müller, Katharina Müller, Leonie Pattard, Claire Pleimelding, Stephan Prückner, Konstantin Pusl, Elba Raimúndez, Julius Raschka, Jakob Reich, Raquel Rubio-Acero, Nicole Schäfer, Paul Schandelmaier, Lara Schneider, Sophie Schultz, Mirjam Schunk, Lars Schwettmann, Heidi Seibold, Paul Stapor, Jeni Tang, Fabian Theis, Sophie Thiesbrummel, Eva Thumser, Niklas Thur, Julian Ullrich, Julia Waibel, Claudia Wallrauch, Julia Wolff, Pia Wullinger, Tobias Würfel, Patrick Wustrow, Houda Yaqine, Sabine Zange, Eleftheria Zeggini, Thorbjörn Zimmer, Thomas Zimmermann, Lea Zuche


3

Appendix contents Text S1, p 4 Text S2, p 5 Text S3, p 7 Text S4, p 8 Text S5, p 9 Figure S1, p 10 Figure S2, p 11 Figure S3, p 12 Figure S4, p 13 Figure S5, p 14 Figure S6, p 15 Figure S7, p 16 Figure S8, p 17 Table S1, p 18 Table S2, p 19 Table S3, p 22 Table S4, p 23 Table S5, p 24 References, p 25


4

Text S1 Ethical considerations. The study protocol was approved by the Institutional Review Board of the Medical Faculty at Ludwig Maximilian University Munich, Germany (opinion dated March 31, 2020; number 20-275) prior to study initiation. All participants provided oral as well as written informed consent. The study was conducted in accordance with good clinical (GCP) and epidemiological practice (GEP) standards as well as the Declaration of Helsinki in its most recent form (as amended by the 64th WMA General Assembly, Fortaleza, Brazil, in October 2013).


5

Text S2 Sampling design and corrections of seroprevalence. To obtain reliable and representative estimates of SARS-CoV-2 seroprevalence in the Munich population, we corrected the observed (crude) seroprevalence in our data, i.e. the number of positive cases divided by the sample size, for two characteristics: the sampling design underlying our cohort, and the sensitivity as well as specificity of tests used. The population-based prospective cohort study KoCo19 is located in the city of Munich, Germany. From April 5 to June 12, 2020, we carried out the initial fieldwork in randomly selected households, using a two-stage sampling design. The initial study population of KoCo19 consists of the Munich general population 14 years and older. The first stage of the sampling procedure refers to the selection of the 100 out of 755 constituencies. This selection was done via a rejective sampling design,1 initially with equal probabilities for each constituency to be included in the sample (~13%). The sample of 100 constituencies was checked to be a representative sample of the Munich population regarding the age structure, the percentages of the population with migration background, of households with children and households with only one member. A sample was considered representative if the respective mean fractions in the sample differed from the mean fractions across all 755 Munich constituencies by less than 10 percent points. Only samples of 100 constituencies that fulfilled these requirements had a non-zero probability to be selected. A Monte Carlo simulation using 5000 iterations for random samples of 100 constituencies showed inclusion probabilities at the constituency level ranging from 12%-15% (Figure S7), indicating that the rejection step does not induce considerable bias. Moreover, the regression estimator (which is equivalent to the calibrated estimator, see below) for the rejective sample (with inclusion probabilities between 12% and 15%) has similar properties to the regression estimator for the original selection procedure (simple random sampling with inclusion probabilities ~13%),2 which facilitates the calculation of the variance and the associated confidence interval. The second stage of the sampling procedure consists in the selection of approximately 30 households for each of the 100 drawn constituencies, totalling around 3000 households in the sample. These households were obtained via random routes starting in each selected constituency (which could be assimilated to a systematic sampling with equal probabilities3 or to a simple random sampling). The random routes often crossed the borders of the constituencies, which means that a household could be included in the sample via its own constituency or via a neighbouring one. To account for these multiple ways to be included in the sample, we considered the first and second order neighbours (neighbours of a neighbour) for each selected constituency and applied a generalized weight share method4 for the weights of the households. Finally, all members 14 years and older were asked to donate blood samples. If participants refused to give blood samples, the sampling weights of the consenting participants within the same household were increased to represent the other members. In case the age of one or more members in the household was missing, we accounted for household structure (number of members, age of the members, etc.) to impute the missing age(s) by its mode (i.e. by the case with the most occurrences). This results in considering that in households with two members, the missing age is almost always one of another adult in the household, while in households with three or more members, the missing ages are usually concerning children in the household. Once the sampling weights for all participants were computed, they were calibrated in order to fit to the Munich population regarding the sex and age structure, the percentages of the population with migration background, of households with children and single-person households (auxiliary information). Here, we employed the calibration technique by Deville et al.5 The resulting calibrated estimator (e.g. the seroprevalence for the Munich population using calibrated weights) is asymptotically equivalent to the regression estimator. The variance associated to this calibrated estimator can then be derived thanks to linearisation6 and residual5,6 techniques. In short, the variance of the calibrated estimator is asymptotically equivalent to the variance of the total of the residuals of a linear regression using the linearized variable as response and the auxiliary variables used in the calibration process as covariates. This variance estimation accounts for the different stages of the sampling design (inference on finite population), i.e. 𝑉 𝑉 𝑉 with 𝑉 the variance associated to the selection of the constituencies and 𝑉 the one associated to the selection of the households. 95% confidence intervals could then be calculated. The normality of the Horvitz-Thompson estimator in case of a two-stage sampling design has been proven by Chauvet et al.7 In Figure S6, we compare the estimated SARS-CoV-2 seroprevalence (and the associated 95% confidence interval) obtained according to the sampling design (using calibrated weights as described above) to the one without considering any weighting strategy. The variance associated to the weighted estimator was calculated as described above. For the variance of the unweighted estimator we used the classical formula 𝑃 1 𝑃 /𝑛 (inference on infinite population). To account for the fact that the target variable within households is correlated, the population size used for this variance was n = 2994, that is the number of households in the sample rather than n = 5313, the number of participants. In addition to accounting for the sampling design, we also consider the probabilities of the laboratory tests to yield false negatives or false positive results. Following Sempos and Tian,8 we calculate the adjusted seroprevalence as �̂� 𝑠𝑝 1 / 𝑠𝑒𝑛 𝑠𝑝 1 , where �̂� is the crude seroprevalence (fraction of positive cases among the observed


6

population), 𝑠𝑝 the estimated specificity and 𝑠𝑒𝑛 the estimated sensitivity. Whilst (𝑝 𝑠𝑝 1 / 𝑠𝑒𝑛 𝑠𝑝 1 with true probability 𝑝 for a positive test result and true sensitivity and specificity 𝑠𝑒𝑛 and 𝑠𝑝 is an exact formula for the true seroprevalence, it is only approximate if �̂�, 𝑠𝑝 are 𝑠𝑒𝑛 are calculated and plugged in independently. For that reason, the case �̂� 𝑠𝑝 1 can occur. In this case, the above formula would yield a negative adjusted seroprevalence estimate. For our data, this happened for EI-S1-IgG. Instead of reporting a negative adjusted seroprevalence, we set the value to zero although positive test results have been observed. This estimate has to be considered with care since the standard adjustment method is not fully applicable.9 To summarize, we report SARS-CoV-2 seroprevalence in this manuscript calculated in four different ways:

- weighted: correcting for the sampling design for our study cohort using calibrated weights, - unweighted: not correcting for the sampling design, - adjusted: correcting for false positives and false negatives based on a test’s sensitivity and specificity, - unadjusted: not correcting for false test results.

Applying the weighted and adjusted Ro-N-Ig seroprevalence (1ꞏ82%; 95% CI: 1ꞏ28-2ꞏ37%) to the number of inhabitants aged 14 and older living in Munich (1,369,444) allows us to estimate the number of individuals who developed SARS-CoV-2 antibodies (24,990; 95% CI 17,584-32,396; using more than two digits for seroprevalence). The underreporting factor from Figure 3C varies depending on the share of the 6,293 officially registered PCR-positive cases in Munich who are living in private households. The calculation of the infection fatality ratio is similar, using the estimated number of infections and assuming a share of the 216 registered COVID-19 related deaths in Munich occurring in private households (see Figure 3F). According to RKI, around 13% of the reported cases for COVID-19 in Germany occurred in institutions until the end of the study period.10 We can thus consider that 87% of the infections occurred in private households. Assuming that this percentage is the same for Munich leads to an underreporting factor of 4.5. In this calculation we use the entire population size as an approximation for the number of people living in private households since only a small number lives in institutions and an adjustment does not change the rough estimate of 4.5. If all cases were registered in private households, this factor would be approximately equal to 4. The same calculation was done for the number of registered deaths: 46% of COVID-19 related deaths in Germany occurred in institutions as reported by RKI.10 Assuming that 54% of the registered deaths in Munich occurred in private households leads to an IFR of 0.47%. If all deaths had occurred in private households, the IFR would be approximately equal to 0.86%. In our survey, we asked participants about already known (and thus registered) SARS-CoV-2 infection status. Based on this information, we estimate 4,367 (95% CI 1,952-6,783) officially reported cases in private households in Munich. Using these numbers allows us to estimate the share of the 6,293 officially reported cases that occurred in private households (69%; 95% CI 31-100%) for Munich. Assuming that 69% of the registered cases occurred in private households leads to an underreporting factor of 5.7. To facilitate reproducibility and reuse, the code used to perform the analyses and generate the figures was made available on GitHub (https://github.com/koco19/lab_epi) and has been uploaded to ZENODO (http://doi.org/10.5281/zenodo.4300922) for long-term storage.


7

Text S3 Details on cut-offs and validation. To assess the robustness of the seropositivity estimate obtained based on the Ro-N-Ig assay with an optimised cut-off, we compared it to seropositivity estimates obtained using other classifiers: Ro-N-Ig assay with the manufacturer’s cut-off, EI-S1-IgG and EI-S1-IgA each with an optimised and the manufacturer’s cut-off, and two machine learning techniques – namely a support vector machine and a random forest – using information from Ro-N-Ig, EI-S1-IgG and EI-S1-IgA. Detailed descriptions of the cut-off optimisation for the three assays, the training of the support vector machine and the random forest, as well as the estimation of the specificities and sensitivities of all classifiers are provided by Olbrich et al.11 In short, an optimised cut-off for a single assay is the median of the cut-offs that classify best in 10000 bootstrap samples, each sampled with replacement from our set of 1266 individuals with known seropositivity status. In each bootstrap replication, the respective best cut-off is used to predict the serological status of the observations not included in the bootstrap sample (out-of-bag observations). After averaging the predictions for each observation over all bootstrap replications, the majority votes are chosen and compared to the true serological status to calculate specificity and sensitivity. For the support vector machine and the random forest, a similar approach is employed: After having optimised the tuning parameters, 2000 bootstraps and respective out-of-bag estimations are used to calculate specificity and sensitivity. The manufacturer’s cut-off for a single assay is determined by the manufacturer based on its own data, and specificity and sensitivity are computed by us by comparing the true serological status to the predictions based on the cut-off for the entire sample of 1266 individuals. Bootstrapping this sample and predicting out-of-bag observations is also used to optimise the tuning parameters and estimate specificities and sensitivities of the support vector machine and the random forest. Classification of test results according to optimized cut-offs are reported in Figures S1 and S2. Table S4 reports resulting test-sensitivity and -specificity for all classifiers. For the classifiers based on a single assay and the cut-off provided by the manufacturer, also one specificity and two sensitivities using manufacturer’s information are given. In particular, the specificity chosen is the average specificity over all cohorts investigated by the manufacturer, while for the sensitivity scenarios of low sensitivity (assuming the time elapsed from a positive PCR test to the antibody test to be intermediate) and high sensitivity (using the highest category of elapsed time reported) are considered. This allows the assessment of the variability of the seroprevalence estimates when different measures of test performance are applied. As it is typical for low-prevalence settings, even small differences in specificity lead to fundamentally different results, as can be seen for both Euroimmun assays. Despite the variability in the results and independent of the application of weights, all classifiers indicate a low seropositivity with all point estimates and most of the upper bounds of the 95% bootstrap percentile confidence intervals being below 2ꞏ5% (Figure S3).


8

Text S4 Handling of missing data and risk factor analysis. The data collected in this study contained gaps. Overall, the data table was filled to 94ꞏ1%. Most missing values (5ꞏ9%) were due to non-response on the personal questionnaires. The serological data were complete for Ro-N-Ig, and there was only one missing observation for E1-S1-IgA/G. The missingness pattern and frequencies are illustrated in Figure S4 and S5. To handle the missing data, we performed multiple imputation as a way of sensitivity analysis and compare it to scenarios where imputation is not performed. The standard method for analysis of binary outcomes for risk factor analysis is the framework of generalized linear models (GLM), in particular logistic regression analyses. However, in our study, the participants were clustered within the households as a sampled unit, and thus, we use the generalized linear mixed models (GLMM).12,13 We assume the missingness in our data to be missing at random (MAR), where the specification of a dropout model is not necessary. This allows us to make valid inference which is based on the likelihood function conditional on the observed data alone (complete case analyses). The GLMM is valid under the above conditions since the estimates are obtained by maximum likelihood. In the case of non-likelihood marginal models, the semi-parametric method of generalized estimating equations (GEE) is popularly used to model correlated non-Gaussian outcomes including missing outcomes. However, GEE would require the stronger condition of missingness completely at random (MCAR) and is not performed here.12,14 We performed the imputation of the missing covariates to compare our estimates after imputation with the GLMM estimates. For this imputation of covariates under the assumption of the base model, we used the Joint Analysis and Imputation of Incomplete Data Framework (JointAI) in R to obtain the updated OR estimates and standard errors.12,14,15 Under this framework, the estimation process is sequentially fully Bayesian, by modelling the GLMM of interest jointly with the incomplete covariates information. Joint analysis and imputation are performed simultaneously to allow for consistency in the parameter estimation process without the requirement for any pooling steps for the imputed data. One of the major advantages of the Bayesian approach is that the joint distribution of all data allows for the use of all available information of the outcome in the imputation of the incomplete covariates particularly in complex setups of clustered data analysis where covariates can be present in multiple levels of hierarchy. For Table S2 we allowed for 100 burn-in/adaption samples followed by 500 MCMC iterations for each of the models considered. For Table S3 we allowed for 3000 burn-in/adaption samples and 12000 MCMC iterations, respectively. Within each iteration, there is an imputed data set created and the regression coefficient(s) for the corresponding risk factor(s) of interest are estimated. The hyperparameters (non- informative) for the fixed effect regression coefficients are assumed to be Gaussian with mean 0 and standard deviation 0.01 whereas the variance components (1/sigma) are assumed to be Gamma distributed with shape and rate parameter being equal to 0.001.


9

Text S5 Influence of household size and household clustering. In total 93 participants of the KoCo19 study were tested positive. These individuals lived in households of different sizes, and there were 82 households in which at least one infection was observed. As household members below 14 years of age were excluded and as some household members did not agree to participate, only 76% of the members of all households were tested. While we had a large number of households with more than two household members, in most households only two persons were tested. Interestingly, we observed that members of 1-person households were more likely to be tested positive than members of households of size 2, 3 and 4, while for large households the fraction of positive tests was again increasing (Tables S2 and S3, Figure S8). This is in agreement with the OR of 1ꞏ54 computed by the risk factor analysis for 1-persons households, and might point towards different habits. Yet, despite the higher infection risk in 1-person households, out of the 93 individuals with a positive test result, 22 lived in a household with another individual which was tested positive. Furthermore, out of the 66 households with more than one inhabitant in which at least one individual was tested positive, 33ꞏ33% had a second individual who was tested positive. Due to the low overall prevalence, both is unlikely in the absence of transmission within households. Accordingly, it indicates a substantial fraction of within-household transmissions.

To further study the household clustering, we performed a risk factor analysis using the GLMM framework (which used a logistic model). Using a likelihood ratio test (LRT), we investigated whether the additional variance component (random effect) due to the clustering within households in the GLMM compared to the GLM was significantly different from zero. We considered the null hypothesis 𝐻 ∶ 𝐷 0 versus the alternative 𝐻 ∶ 𝐷 𝑑 for some non-negative scalar 𝑑 , where 𝐷 is the variance of the random effect. The hypotheses were evaluated using the asymptotic null distribution for the difference between the likelihoods of the two models,

2 ln 𝜆 ln 𝜆 → 𝜒 : (mixture of 𝜒 and 𝜒 with equal weights 0.5) .12,16,17 𝜆 is the likelihood under the GLM without any random effect, whereas 𝜆 is the corresponding likelihood under the GLMM with one random effect for the intra-household clustering. For the analysis described in Table S3 under the frequentist framework we obtained the value for 2 ln 𝜆 ln 𝜆 278 85, and when compared to the mixture of 𝜒 and 𝜒 , we obtain a p-value less than 0ꞏ001. This implied again that we have a significant clustering within households. In addition to this, in the Bayesian framework we estimated the random effect variance to be 7ꞏ13 with 95% Credible Interval from 3ꞏ4 to 12ꞏ4. All this evidence favoured the presence of positive clustering within households.


10

Figure S1 Validation of classifier based on Ro-N-Ig results. The subplots depict the results of the Ro-N-Ig and EI-S1-IgG results for 5313 participants. A) Prediction of classifier based on Ro-N-Ig results using optimised threshold derived by Olbrich et al., B) results using cPass test, C) results of the neutralisation test, and D) results of a PCR test.11


11

Figure S2 Comparison of primary test results (Ro-N-Ig vs EI-S1-IgG) with other tests. All plots show the same measurements of Ro-N-Ig values vs. EI-S1-IgG values on a continuous scale. Lines indicate whether they lie below or above the optimized (dashed) or the manufacturer’s (dotted) cut-off for Ro-N-Ig (vertical) and EI-S1-IgG (horizontal). In each plot, colors represent the binary result of the test indicated in the respective title: blue for negative, orange for positive and grey for non-performed analysis of the respective test.


12

Figure S3 SARS-CoV-2 seroprevalence obtained by different classifiers for different specificities and sensitivities. Seroprevalence estimates accounting for the sampling design (right) and not accounting for sampling design (left). Values are adjusted (orange, blue and grey) or unadjusted (black) for the tests’ sensitivities and specificities, which are specified in Table S4. In the main manuscript, we report weighted and adjusted Ro-N-Ig seroprevalence with optimised cut-off.


13

Figure S4 Missing patterns observed in our complete analysis data. We observe 4081 out of 5313 individuals have complete information. Our primary serological test (Ro-N-Ig) along with age, sex and the type of housing has complete information for 5313 observations and one individual has missing information on E1-S1-IgG/IgA. Number of positive symptoms is a derived variable based on selection among a list of symptoms mentioned in the questionnaire. Majority of the non-response can be attributed to non-answering of the questionnaire for household and individual characteristics. Missing education level of the participants was the biggest contributor to the missing data.


14

Figure S5 Missing patterns observed in our multiple regression analysis data with important covariates only. For this subset of data, we observe 4460 out of 5313 observations have complete information, which is used for obtaining the regression estimates without any imputation (Table S3).


15

Figure S6 Roche N pan-Ig seroprevalence in the KoCo19 study population unweighted (top) and weighted (bottom). The values are either unadjusted (left) or adjusted (right) for sensitivity and specificity of the test method. The weekly seroprevalence estimates are in all scenarios crude estimates without weighting and adjustment for sensitivity and specificity. The 95% confidence interval for the weekly seroprevalence was computed assuming a Poisson distribution to estimate the rate of positive cases.


16

Figure S7

Sampling probabilities for Munich constituencies. Given the rejection sampling approach for the selection of the constituencies (see Text S2), we estimated the real probability for each constituency in Munich to be included in the set of 100 starting constituencies for the random walks using a Monte Carlo simulation. Within the samples of 100 constituencies that fulfilled the representativeness requirements, we repeated 5000 times the selection of one sample and computed the sampling probabilities as the percentages of how often a constituency was included in the 5000 Monte Carlo samples. The figure shows that all probabilities are estimated to be in a rather narrow range (between 0.12 and 0.15) and thus approximately equal. The rejection sampling procedure does not induce considerable bias.


17

Figure S8

Fraction of individuals with positive Ro-N-Ig result for different household sizes. The total number of tested household members (cyan) and the total number of household members (red) was taken as basis.


18

Table S1 Individual characteristics of the KoCo19-study participants comparing telephone interviewees to those answering the online questionnaire. Characteristics KoCo‐19

Online questionnaire KoCo‐19

Telephone interview Test for

significancea

Individual characteristics

N 4902 411

nmissing n % nmissing n % p‐value

Sex 0 0 0∙006b

Female 2525 51∙5 241 58∙6

Age 0 0 0∙002c

0‐19 yrs 261 5∙3 6 1∙5

20‐34 yrs 1322 27∙0 24 5∙8

35‐49 yrs 1505 30∙7 37 9∙0

50‐64 yrs 1233 25∙2 73 17∙8

65‐79 yrs 504 10∙3 172 41∙8

80+ yrs 77 1∙6 99 24∙1

Country of birth 465 0 0∙035b

Outside Germany 761 17∙2 88 21∙4

Level of education 685 16

Still in school 97 2∙3 3 0∙8 0∙002c

<12 yrs 1103 26∙2 283 71∙6

≥12 yrs 3017 71∙5 109 27∙6

Employment status 575 1 0∙002c

Employed 2825 65∙3 86 21∙0

Self‐Employed 445 10∙3 26 6∙3

Not workingd 970 22∙4 288 70∙2

Otherse 87 2∙0 10 2∙4

Risk employmentf 470 0 0∙001b

Yes 3757 84∙8 178 43∙3

General health 466 0 0∙002c

Fair or poor 150 3∙4 56 13∙6

Good 1485 33∙5 232 56∙4

Very good 2042 46∙0 84 20∙4

Excellent 759 17∙1 39 9∙5

Loss of the sense of smell or taste

Yes 498 0 0∙002c

40 0∙9 13 3∙2

Respiratory allergies 537 0∙012b

Yes 1283 29∙4 96 23∙5

Smoking status 575 1 0∙002c

Never smoker 2363 53∙5 177 43∙1

Ex‐smoker 1258 28∙5 153 37∙2

Current smoker 794 18∙0 81 19∙7

Face mask use 469 0 <0∙001b

Almost always 3080 69∙5 374 91∙0 a Significance test comparing fractions among telephone interviewees to participants who answered the online survey. b Fisher’s exact test for count data. c Pearson’s chi‐squared test with simulated p‐value (based on 500 replicates) d Not working includes unemployed, retired, parental leave, sabbatical, students e Others includes voluntary social year, military service, part‐time jobber, reduced working hours f Considered as risk employment for SARS‐CoV‐2 seropositivity were employees in the: healthcare sector, emergency

service, senior homes, airport, public transport, education, sales, social work and other risk jobs. These also include individuals who have been working for the past year at least


19

Table S2 Prevalence of Roche N pan-Ig seropositivity and risk factor analysis adjusted for age and sex using three different methods by individual, household and health characteristics of the KoCo19 study population. The count n represents the number of positives in the specific covariate group and the % symbol denotes the crude prevalence percentage in that group.

Characteristics Prevalence Crude

(GLM)

Adjusted for clustering

(GLMM)

Bayesian model adjusted for clustering

with multiple imputation (GLMM)

Individual Characteristics

n % ORf 95% CIf ORf 95% CIf ORf 95% CIf

Sex

Male 45 1.8 1 1 1

Female 48 1∙7 0∙98 0∙65 1∙48 0∙99 0∙45 2∙20 0∙92 0∙58 1∙50

Age

0‐19 yrs 4 1∙5 0∙72 0∙21 1∙84 1∙61 0∙17 12∙50 0∙59 0∙14 1∙90

20‐34 yrs 28 2∙1 1 1 1

35‐49 yrs 29 1∙9 0∙90 0∙53 1∙53 2∙75 0∙64 13∙60 0∙84 0∙43 1∙49

50‐64 yrs 19 1∙5 0∙70 0∙38 1∙24 0∙84 0∙17 3∙97 0∙64 0∙33 1∙21

65‐79 yrs 12 1∙8 0∙85 0∙41 1∙64 3∙25 0∙51 20∙60 0∙79 0∙36 1∙79

80+ yrs 1 0∙6 0∙27 0∙02 1∙27 0∙01 0∙00 3∙20 0∙13 0∙01 1∙15

Adjusted for Age (continuous) and Sex

Country of Birth

Germany 67 1∙7 1 1

Outside Germany

14 1∙6 0∙99 0∙53 1∙72 0∙54 0∙09 2∙33 0∙87 0∙41 1∙70

Level of Education

Still in school 2 2∙0 0∙74 0∙11 2∙83 0∙30 0∙01 4∙08 0∙33 0∙04 2∙12

<12 yrs 25 1∙8 1 1 1

≥12 yrs 53 1∙7 0∙84 0∙52 1∙40 0∙62 0∙17 2∙39 0∙72 0∙36 1∙33

Employment Status

Employed 47 1∙6 1 1

Self‐Employed 13 2∙8 1∙89 0∙97 3∙48 1∙20 0∙23 5∙30 1∙80 0∙87 3∙70

Not workinga 19 1∙5 1∙03 0∙58 1∙75 0∙92 0∙26 3∙02 0∙91 0∙47 1∙70

Othersb 2 2∙1 1∙21 0∙20 3∙99 1∙64 0∙04 19∙50 1∙06 0∙13 5∙75

Risk Employmentc

No 61 1∙5 1 1

Yes 20 2∙4 1∙53 0∙89 2∙50 2∙68 0∙81 8∙71 1∙53 0∙80 2∙79

Household Characteristics

Housing Type : Building with

1‐2 aps 27 1∙9 1 1 1

3‐4 aps 3 0∙8 0∙44 0∙10 1∙24 0∙50 0∙00 11∙60 0∙35 0∙05 1∙38

≥5 aps 63 1∙8 0∙92 0∙59 1∙48 1∙10 0∙23 7∙68 0∙88 0∙45 1∙64

Othersd 0 0∙0

Household Type

Single 12 1∙8 1∙26 0∙61 2∙50 1∙88 0∙16 18∙20 1∙40 0∙33 3∙77

Couple 24 1∙4 1 1 1

Family 33 1∙7 1∙11 0∙65 1∙94 1∙04 0∙13 8∙92 1∙19 0∙42 3∙19

Otherse 14 2∙9 1∙90 0∙93 3∙75 1∙92 0∙09 22∙40 3∙03 0∙93 9∙63

Number of Household Members

1 16 2∙0 1 1 1

2 33 1∙6 0∙75 0∙42 1∙40 0∙50 0∙07 4∙06 0∙67 0∙21 1∙81

3‐4 30 1∙5 0∙68 0∙37 1∙31 0∙46 0∙06 3∙99 0∙59 0∙18 1∙69


20

5+ 14 3∙1 1∙35 0∙63 2∙86 0∙83 0∙03 10∙50 1∙96 0∙51 8∙18

Living area per Inhabitant (Quartiles)

≤ 30 m2 27 1∙6 1 1 1

30‐40 m2 28 2∙3 1∙57 0∙92 2∙71 1∙57 0∙18 14∙30 2∙19 0∙94 5∙11

40‐55 m2 9 0∙9 0∙65 0∙28 1∙36 0∙67 0∙02 8∙90 0∙69 0∙27 1∙88

>55 m2 19 2∙1 1∙67 0∙87 3∙16 2∙13 0∙23 20∙70 2∙65 0∙95 7∙27

Net Family Income (Quartiles)

≤ 2500 € 10 1∙7 2∙00 0∙77 5∙57 2∙13 0∙04 312 2∙49 0∙18 11∙5

2500‐4000 € 7 0∙9 1 1 1

4000‐6000 € 19 1∙6 1∙87 0∙82 4∙82 2∙11 0∙10 272 3∙27 0∙27 11∙1

6000+ € 20 1∙8 2∙13 0∙94 5∙45 1∙99 0∙08 262 3∙28 0∙33 14∙1

Health Status

General Health

Fair or poor 4 1∙9 1∙58 0∙42 4∙82 1∙45 0∙08 18∙90 0∙97 0∙22 3∙66

Good 26 1∙5 1∙14 0∙59 2∙41 0∙97 0∙18 5∙85 0∙79 0∙31 1∙79

Very good 39 1∙8 1∙27 0∙68 2∙56 0∙76 0∙16 4∙19 0∙94 0∙37 1∙86

Excellent 12 1∙5 1 1 1

Respiratory Allergies

No 53 1∙6 1 1

Yes 26 1∙9 1∙20 0∙73 1∙90 3∙25 1∙10 10∙30 1∙21 0∙67 2∙16

Skin Allergies

No 67 1∙6 1

Yes 13 2∙0 1∙20 0∙63 2∙13 3∙17 0∙81 12∙00 1∙28 0∙62 2∙64

COPD

No 70 1∙6 1 1

Yes 9 2∙3 1∙48 0∙68 2∙84 2∙81 0∙57 11∙80 1∙58 0∙64 3∙72

Autoimmune Disease

No 75 1∙7 1 1 1

Yes 6 1∙8 1∙10 0∙42 2∙35 1∙15 0∙14 6∙43 0∙98 0∙30 2∙75

Diabetes

No 75 1∙6 1 1

Yes 3 1∙4 1∙09 0∙26 3∙06 0∙39 0∙01 6∙04 0∙78 0∙15 3∙17

CVD

No 65 1∙7 1 1 1

Yes 15 1∙7 1∙24 0∙64 2∙31 1∙01 0∙21 4∙20 1∙16 0∙53 2∙28

Obesity

No 75 1∙7 1 1 1

Yes 6 2∙2 1∙40 0∙54 3∙02 0∙51 0∙03 4∙67 1∙12 0∙35 3∙14

Cancer

No 78 1∙7 1 1 1

Yes 3 1∙2 0∙81 0∙20 2∙27 0∙56 0∙04 4∙80 0∙64 0∙14 2∙34

Loss of the sense of smell or taste

No 70 1∙5 1 1 1

Yes 10 18∙9 16∙3 7∙40 32∙60 41∙3 6∙70 231 28∙2 9∙25 90∙22

Health Behaviour

Smoking status

Never 46 1∙8 1

Ex 21 1∙5 0∙87 0∙50 1∙47 0∙68 0∙18 2∙32 0∙71 0∙37 1∙29

Current 14 1∙6 0∙89 0∙47 1∙59 1∙12 0∙25 4∙41 0∙74 0∙34 1∙49

Use of facemask

<Almost surely 27 1∙9 1∙20 0∙74 1∙92 1∙68 0∙53 5∙27 1∙25 0∙71 2∙21

Almost surely 53 1∙5 1 1 1

a Not working includes unemployed, retired, parental leave, sabbatical, students b Others includes voluntary social year, military service, part‐time jobber, reduced working hours


21

c Considered as risk employment for SARS‐CoV‐2 seropositivity were employees in the: healthcare sector, emergency service, senior homes, airport, public transport, education, sales, social work and other risk jobs. These also include individuals who have been working for the past year at least

d Other types of housing include tents, caravans, or the like e Other household types include shared apartments by e.g., students, subleasing, and assisted accommodation f OR: odds ratio; 95% CI: 95% credible intervals (Bayesian analyses) / 95% confidence intervals (frequentist GLMM)


22

Table S3 Multivariate risk factor analysis for SARS-CoV-2 seropositivity mutually adjusted for all variables in the table. Characteristics Adjusted for clustering

(GLMM)

Bayesian model adjusted for clustering with multiple

imputation (GLMM)

N=4460 N=5313

OR 95% CI ORc 95% CIc

Sex

Male 1 1

Female 1∙00 0∙39 2∙59 0∙93 0∙56 1∙54

Age (continuous) 1∙01 0∙97 1∙04 0∙99 0∙97 1∙01

Risk Employmenta

No 1

Yes 1∙95 0∙52 6∙72 1∙83 0∙90 3∙73

Household Type

Single 1∙54 0∙09 21∙10 1∙07 0∙35 3∙27

Couple 1 1

Family 1∙10 0∙11 12∙20 1∙29 0∙55 3∙08

Othersb 1∙81 0∙05 27∙10 2∙87 0∙91 9∙27

Living Area per Inhabitant (Quartiles)

≤ 30 m2 1 1

30‐40 m2 1∙40 0∙12 16∙2 2∙18 0∙91 5∙32

40‐55 m2 0∙68 0∙01 11∙9 0∙74 0∙24 2∙22

>55 m2 1∙77 0∙10 28∙0 2∙84 0∙92 8∙98

Respiratory Allergies

No 1 1

Yes 2∙70 0∙90 8∙55 1∙30 0∙70 2∙45 a Considered as risk employment for Covid‐19 infections were employees in the: healthcare sector, emergency service,

senior homes, airport, public transport, education, sales, social work and other risk jobs. b Other household types include shared apartments by e.g., students, subleasing, and assisted accommodation c OR: odds ratio; 95% CI: 95% credible intervals (Bayesian analyses) / 95% confidence intervals (frequentist GLMM)


23

Table S4 Classifier’s specificity and sensitivity depending on the manufacturer’s or the optimised cut-offs.

Classifier Cut‐off Specificity

Manufacturer

Specificity evaluated on the data

Sensitivity Manufacturer (low / high)

Sensitivity evaluated on the data

EI‐S1‐IgA (Manufacturer’s cut‐off)

1.100 0.924 0.933 0.917 / 1.000 0.648

EI‐S1‐IgA (Optimised cut‐off)

1.080 0.926 0.648

EI‐S1‐IgG (Manufacturer’s cut‐off)

1.100 0.993 0.980 0.875 / 1.000 0.772

EI‐S1‐IgG (Optimised cut‐off)

1.101 0.978 0.798

Ro‐N‐Ig (Manufacturer’s cut‐off)

1.000 0.998 0.998 0.853 / 0.995 0.855

Ro‐N‐Ig (Optimised cut‐off)

0.422 0.997 0.886

Random Forest 0.998 0.886

Support Vector Machine 0.998 0.845


24

Table S5

Overview on the household size distribution, the number of tested individuals per household and the number of positive tests. Seropositivity is with respect to Ro-N-Ig with optimised threshold.

Number of inhabitants of the household

Number of tested

individuals in household

Number of households

in which no test was

positive

in which 1 test was positive

in which 2 tests were

positive

in total

1 1 768 16 0 784

2 1 223 3 0 226

2 918 24 3 945

3 1 60 0 0 60

2 287 8 1 296

3 120 2 1 123

4

1 44 0 0 44

2 230 5 2 237

3 63 2 0 65

4 51 3 1 55

5

1 12 0 0 12

2 52 1 1 54

3 23 0 0 23

4 19 1 1 21

5 14 3 0 17

6

1 4 0 0 4

2 6 0 0 6

3 6 0 1 7

4 2 0 0 2

5 0 1 0 1

6 2 1 0 3

7

2 2 0 0 2

4 1 0 0 1

5 1 0 0 1

7 1 1 0 2 8 4 1 0 0 1 9 1 1 0 0 1


25

References

1. Hajek J. Asymptotic Theory of Rejective Sampling with Varying Probabilities from a Finite Population. Ann Math Statist. 1964 Dec;35(4):1491–523.

2. Fuller WA. Some design properties of a rejective sampling procedure. Biometrika. 2009;96(4):933–44. 3. Tillé Y. Sampling and Estimation from Finite Populations [Internet]. 1st ed. Wiley; 2020 [cited 2020

Dec 2]. Available from: https://onlinelibrary.wiley.com/doi/book/10.1002/9781119071259 4. Deville C, Lavallée P. Indirect Sampling: The Foundations of the Generalized Weight Share Method.

Survey Methodology. 2006; 32(2):165-176. 5. Deville J-C, Särndal C-E. Calibration Estimators in Survey Sampling. Journal of the American

Statistical Association. 1992 Jun;87(418):376–82. 6. Deville JC. Variance estimation for complex statistics and estimators: linearization and residual

techniques. Survey Methodology. 1999; 25(2): 193-204. 7. Chauvet G, Vallée AA. Consistency of estimators and variance estimators in two-stage sampling. J

Royal Stat Soc. 2020; 82(3): 797-815. 8. Sempos CT, Tian L. Adjusting Coronavirus Prevalence Estimates for Laboratory Test Kit Error.

American Journal of Epidemiology. 2020 Aug 17;kwaa174. 9. Moreira JP de L, Almeida RMVR de, Rocha NC dos S, Luiz RR. Correção da prevalência autorreferida

em estudos epidemiológicos com grandes amostras. Cad Saúde Pública [Internet]. 2016 [cited 2020 Dec 2];32(12). Available from: http://www.scielo.br/scielo.php?script=sci_arttext&pid=S0102-311X2016001204001&lng=pt&tlng=pt

10. Robert Koch Institut. 10. Täglicher Lagebericht des RKI zur Coronavirus-Krankheit-2019 (COVID-19) – Aktualisierter Stand für Deutschland [Internet]. 2020 [cited 2020 Dec 2]. Available from: https://www.rki.de/DE/Content/InfAZ/N/Neuartiges_Coronavirus/Situationsberichte/2020-06-13-de.pdf;jsessionid=2A2033699409CB8BA030EB0F22EA0BD4.internet062?__blob=publicationFile

11. Olbrich et al. A Serology Strategy for Epidemiological Studies Based on the Comparison of the Performance of Seven Different Test Systems - The Representative COVID-19 Cohort Munich. Adjacent manuscript.

12. Molenberghs G, Verbeke G. Models for discrete longitudinal data. New York ; London: Springer; 2005. 683 p. (Springer series in statistics).

13. Brooks ME, Kristensen K, Van Benthem KJ, Magnusson A, Berg CW, Nielsen A, et al. glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. 2017 [cited 2020 Dec 2]; Available from: https://www.zora.uzh.ch/id/eprint/148918

14. Erler NS, Rizopoulos D, Lesaffre EMEH. JointAI: Joint Analysis and Imputation of Incomplete Data in R. arXiv:190710867 [stat] [Internet]. 2020 Sep 1 [cited 2020 Dec 2]; Available from: http://arxiv.org/abs/1907.10867

15. Erler NS, Rizopoulos D, Rosmalen J van, Jaddoe VWV, Franco OH, Lesaffre EMEH. Dealing with missing covariates in epidemiologic studies: a comparison between multiple imputation and a full Bayesian approach. Statist Med. 2016 Jul 30;35(17):2955–74.

16. Verbeke G, Molenberghs G. The Use of Score Tests for Inference on Variance Components. Biometrics. 2003 Jun;59(2):254–62.

17. Molenberghs G, Verbeke G. Likelihood Ratio, Score, and Wald Tests in a Constrained Parameter Space. The American Statistician. 2007 Feb;61(1):22–7.

Documents

Disclaimer: This manuscript has not been peer‐reviewed.€¦ · Disclaimer: This manuscript has not been peer‐reviewed. Research in context Evidence before this study To estimate