key: cord-1035734-a6l97lm6
authors: Brueningk, S. C.; Klatt, J.; Stange, M.; Mari, A.; Brunner, M.; Roloff, T.-C.; Seth-Smith, H.; Schweitzer, M.; Leuzinger, K.; Kobberoee Soegaard, K.; Albertos Torres, D.; Gensch, A.; Schlotterbeck, A.-K.; Nickel, C.; Ritz, N.; Heininger, U.; Bielicki, J.; Rentsch, K.; Fuchs, S.; Bingisser, R.; Siegemund, M.; Pargger, H.; Ciardo, D.; Dubuis, O.; Buser, A.; Tschudin-Sutter, S.; Battegay, M.; Schneider-Sliwa, R.; Borgwardt, K.; Hirsch, H. H.; Egli, A.
title: Determinants of SARS-CoV-2 transmission to guide vaccination strategy in a city
date: 2020-12-17
journal: nan
DOI: 10.1101/2020.12.15.20248130
sha: 1292df4219b81d807ddf95f4d31d2259f997dcb5
doc_id: 1035734
cord_uid: a6l97lm6

Transmission chains within cities provide an important contribution to case burden and economic impact during the ongoing COVID-19 pandemic, and should be a major focus for preventive measures to achieve containment. Here, at very high spatio-temporal resolution, we analysed determinants of SARS-CoV-2 transmission in a medium-sized European city. We combined detailed epidemiological, mobility, and socioeconomic data-sets with whole genome sequencing during the first SARS-CoV-2 wave. Both phylogenetic clustering and compartmental modelling analysis were performed based on the dominating viral variant (B.1-C15324T; 60% of all cases). Here we show that transmissions on the city population level are driven by the socioeconomically weaker and highly mobile groups. Simulated vaccination scenarios showed that vaccination of a third of the population at 90% efficacy prioritising the latter groups would induce a stronger preventive effect compared to vaccinating exclusively senior population groups first. Our analysis accounts for both social interaction and mobility on the basis of molecularly related cases, thereby providing high confidence estimates of the underlying epidemic dynamics that may readily be translatable to other municipal areas.

). For living space per person or percentage of senior citizens, mobility was comparable 145 between tertiles with a trend towards higher mobility within the younger population groups. 146 Dynamic changes in mobility were assessed by agglomerating normalized traffic counts for public 147 and private transport modalities ( Figure 2C ). There was a clear drop in mobility for both public 148 and private transport modes around the onset of the national lockdown date (12th March, 2020) . 149 The decrease was more pronounced for public transport, resulting in a weighted average mobility 150 drop of approximately 50% ( Figure 2D ). Figure 2D also shows the dynamic change in social in-151 teraction contribution to B.1-C15324T case numbers obtained from the estimation of the effective 152 reproductive number using a Kalman filter 26, 27 . Despite noticeable fluctuation, social interaction 153 contribution decreased on average over time. This data also reflects variation in case reporting 154 which affected the estimated effective reproductive number. Importantly, since the B.1-C15324T 155 was eventually eradicated, a final social interaction contribution of zero was expected. These re-156 sults, in terms of spatial and temporal variation, serve as input to the following SEIR-model. the transmission, as 1.9% of 2,019 residents showed detectable antibody levels, corresponding 165 to 77.2% unreported cases overall, and 87.5% for the sequenced B.1-C15324T strain. Figures interaction. Importantly, there was a significant difference (2% achieved significance) in R ef f 171 between statistical blocks of the highest and lowest median income. The significance level was here 172 scored based on a comparison to 99 random partitioning of the statistical blocks (see Table S1 ).

For all socioeconomic partitions these differences are summarized as histograms in Figures 3G-J. 174 We found that blocks with a higher median income (2% achieved significance), or higher living 175 space per person (1% achieved significance), or lower share of 1-person households (2% achieved 176 significance) had a significantly lower R ef f (< 1.7) relative to the maximum R ef f observed in the 177 relevant partition. A partitioning based on the share of senior residents did not result in significant 178 differences in R ef f (45% achieved significance). Differences in R ef f are due to two factors: the 179 effective mobility contribution ( Figure 2B ), and the modelled reproductive number (R, eq.(8)). 180 In particular, the tertile with the lowest share of 1-person households (T1) showed less mobility 181 compared to T2 and T3, leading to significant differences in R ef f . In contrast, mobility in the T1 182 and T3 tertiles of living space per person were similar ( Figure 2B ), yet differences in R ef f were 183 significant, indicating that the transmission was not dominated by mobility alone. 184 9 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 developments of the first wave of the epidemic under the assumption of different mobility sce-186 narios and modelled two future vaccination strategies. All estimated parameters (including R ef f , 187 see methods for details) were employed for this scenario building. Figure 4A (Figures S4-S6G ) 188 displays the results for mobility scenarios as observed with up to 50% mobility reduction (scenario 189 MO), 100% mobility (scenario M1), and no mobility (scenario M2). Peak case numbers (April 190 12 th ) would have been approximately three times higher in the case of no reduction in mobility 191 (M1). However, the decrease in peak case numbers assuming zero mobility (M2) as modelled, was 192 not as pronounced. Mobility reduction hence played a vital role for the containment of CoV-2 during the first wave.

To study the optimized delivery of vaccines we used data-driven estimates for number of vaccines 195 and their efficacy (70% 28 and 90% 29, 30 ) in the early 2021 period. Scenarios were compared based 196 on absolute case numbers and on the time to reach 50% of intensive care unit (ICU) capacity as 197 quantifiable endpoint to judge the degree of burden on the healthcare system. We model the sce-198 nario of a single exposed individual being introduced into the Basel-City population. Figure 4B 199 (Figures S4-S6H) shows the results for an outbreak scenario (denoted as V1) in which a specific 200 fraction of the population (33% or 66%) will be vaccinated at either 70% or 90% efficacy. As 201 expected, we show that higher vaccine efficacy or higher population fraction vaccinated reduces 202 the slope and plateau (i.e. overall reduction of case burden) of the epidemic curve. Based on these 203 results, vaccination of a third of the city's population at 90% efficacy would cause a decrease in the 204 slope of the epidemic curve resulting in approximately 36 day delay to reach 50% ICU capacity.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 It should be noted, that vaccination of the population at random is an artificial scenario, applied 206 here only to demonstrate the impact of vaccination efficacy relative to the population fraction vac-207 cinated. This scenario will serve in the following as a baseline comparison for two more realistic 208 vaccination strategies given in Figure 4C , and D ( Figures S4-S6I) . Figure 4C shows a scenario 209 where selected demographics with lowest income, that had less options to socially distance and 210 hence were more likely to be exposed to and/or transmit the virus (reflected by a higher R ef f , 211 scenario V2), were prioritized for vaccination. With this strategy, the slope of the epidemic curve 212 would be reduced compared to randomly vaccinating the same number of subjects from the whole 213 population leading to a substantial further delay of approximately 30 days to reach the 50% ICU 214 capacity mark (see S9 for more detailed illustration regarding the development of ICU occupancy 215 under the different vaccination scenarios), and an overall reduction of the case burden for the whole 216 city of up to 6.9 folds relative to V1. 217 Figure 4D shows a scenario where priority was given to the population group with the highest 218 share of senior residents (aged > 64), which had lower mobility than the rest of the population 219 ( Figure 2B ) but constitutes 60% of ICU cases (scenario V3). We observe that scenario V3 resulted 220 in a steeper epidemic curve and would yield 50% ICU capacity at a similar time as a random vac-221 cination strategy. However, the total number of cases at this time would be approximately double 222 in scenario V3 compared to V1.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 sized (<500k residents) city, including local transmission analysed by phylogenetic tree inference 226 and clustering, and the overall spread described by a compartmental SEIR-model. We harness 227 these rich and detailed data sets for the optimization of vaccination strategies within a city. The 228 main strength of this study lies within the high degree of diverse and detailed data available, which 229 included on a per case basis, whole genome sequencing of the virus, the residential address, so-230 cioeconomic data, symptom onset (if any), estimated place and time of infection as well as contact 231 tracing information. This was complemented by high resolution, both spatially and temporally, 232 mobility data as well as serology data for the estimation of the fraction of unreported cases, in one 233 comprehensive study. Despite the small size of Basel-City compared to previous city-models that 234 focused on large metropolises 20 , we benefit from high data density and quality with respect to the 235 number of residents, and address here a representative example of a European, medium-sized city 236 for which studies are currently lacking.

By basing our modelling efforts only on sequenced data, despite reducing the number of included 238 cases, transmission dynamic analysis did not need to account for potential new introductions. This 239 is a strong advantage of our approach over others 15, 31-33 since a continuum model, such as a SEIR- CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10. 1101 /2020 previous reports within Europe 8, 34 .

The use of mobility and socioeconomic data in our models is also unique among studies published 246 to date. Our analysis is based on information regularly collected and analysed by the statistical 247 office of Basel city, providing a high spatial and temporal resolution network of the inner city mo-248 bility patterns. In contrast to the evaluation of mobile phone data which has frequently been used 249 by others 19, 20, 35, 36 our mobility information is not restricted to groups with a device but represents 250 long-term analysis of different traffic streams. Such data is not subjected to privacy legislation and 251 is hence expected to be more readily available for other cities, making our analysis transferable. 252 We do not hold information on the duration and specific location of individuals, but a continuum 253 estimate of population mixing that aligns well with the concept of a compartmental ODE model. 254 We enrich this information with time-variable data on the degree of capacity utilization for the 255 specific study period.

Based on these data, we evaluated the effective reproductive number for different socioeconomic 257 groups of the population, rather than at a purely geographical level. We found that socioeconomic 258 brackets characterized by low median income and a smaller living space per person, were associ-259 ated with a significantly larger effective reproductive number than the socioeconomically stronger 260 groups. This observation is in line with previous results 35 , suggesting that population groups from 261 a weaker socioeconomic background (here low median income) are more mobile and at higher 262 risk for SARS-CoV-2 infection/transmission. This SEIR-model analysis complements our find-263 ings based on the analysis of phylogenetic clusters. We observed clusters predominantly within 264 higher socioeconomic groups, implying that these are spread within the same social network. It 265 13 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10. 1101 /2020 is likely that those individuals are retired, or have had the ability to work from home during the 266 first wave, a pattern that has been observed also in other cities 37 . In contrast, infections within 267 lower socioeconomic groups may result from multiple sources. This observation aligns with the 268 possibility that low socioeconomic status may relate to jobs requiring higher personal contact, and 269 compulsory mobility 38 , which has been shown to increase the risk of infection by 76% 39 .

Mobility and the reduction thereof has been suggested as a proxy to evaluate the reduction of the corresponding to the WHO estimate) is infeasible, it is essential to strategically optimize vac-279 cine distribution. According to our model the largest reduction in case numbers would occur in 280 a scenario where priority is given to those population groups that cannot effectively reduce their 281 contacts or that are highly mobile corresponding to low median income groups. The overall reduc-282 tion in case burden by choosing a targeted distribution is estimated reach up to 6.9 folds compared 283 to a random vaccine distribution strategy. Also, the significant delay of transmission dynamics 284 provides additional time to increase overall vaccination coverage throughout the population. By CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101/2020.12.15.20248130 doi: medRxiv preprint be achieved. We showed that the senior citizen population was not driving the transmission of 287 SARS-CoV-2 during the first wave. Hence, although vaccinating high-risk groups would reduce 288 the number of hospitalized and ICU patients in the short term, the spread of the pandemic would 289 be more effectively contained by vaccinating the transmission drivers. A vaccine would be most 290 effective if applied in areas where it is not possible to trace transmission chains, as is the case for 291 mobile and socioeconomically weaker groups. In contrast, we observed that cases within the more 292 senior population groups clustered, indicating that contact tracing strategies would be efficient for 293 this group where transmission chains were detected and contained. By restricting vaccination to 294 risk groups only, a larger fraction of the general population will be exposed to SARS-CoV-2 imply-295 ing that contact and travel restrictions remain vital to contain transmission. Such measures come 296 at great economic cost 41 . Also, reliable estimations on the efficacy of vaccines in different age 297 groups is yet to be established and it remains uncertain if the vaccines will reach the reported 42, 43 298 high efficacy also in the senior population. We did not account for this effect in our SEIR-model 299 and performed vaccination scenario building accounting only for parameters based on data-driven 300 estimates. However, in the case of a less effective vaccine in senior population groups this would 301 intensify the differences between the simulated scenarios in favour of vaccination of the trans-302 mission drivers. Our vaccine scenario simulations are based on a simple, yet data-driven model 303 that is currently specific to Basel-City and we further did not account for vaccination ramp-up, 304 groups with co-morbidity, or differences in vaccine efficacy directly as has been suggested by oth- CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint 

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint 

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101/2020.12.15.20248130 doi: medRxiv preprint 

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this this version posted December 17, 2020. ; 

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this this version posted December 17, 2020. ; Figure 4 : Scenario Simulations for a partition based on median income. A) Influence of the mobility pattern on the total number of infected cases during the first wave (sum of reported and unreported cases) modelling either no change in mobility (no lockdown scenario, M1), or full shut-down of all inner city mobility (M2). For comparison the observed scenario (MO) is shown. B) Simulation of future vaccination effects if a specific percentage of all citizens was randomly selected for vaccination at given efficacy (V1). We compare this to the scenario of no vaccine (V0). C) Simulation of future vaccination effects based on a partition according to median income. Scenario V2 models vaccination of 33% of all citizens selected from the tertile with the lowest median income (T1). A maximum 6.9 fold difference in cases was observed at day 99. D) Simulation of future vaccination effects based on a partition according to the share of senior residents. Scenario V3 models vaccination of 33% of all citizens selected from the tertile with the highest share of senior residents (T3). We model 90% vaccine efficacy and compare with scenarios V0 and V1. In C) and D) we model 90% vaccine efficacy and compare with scenarios V0 and V1. Dots indicate the time of reaching a 50% ICU occupancy.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. quencend/unreported cases. We accounted for susceptibles (S), exposed (E, incubation time T inc ), 586 and pre-symptomatic yet infectious cases independent of reported/unreported status (P , repro-587 ductive number R). After a presymptomatic time T inf P , cases were separated according to the 588 estimated proportion of reported and sequenced cases, p sq , into either reported infectious (I), or 589 unreported infectious (U i , reproductive number R). Since our data did not include information on 590 recovered patients, a 'recovered' compartment was not included following compartment I. It was 591 assumed that reported cases remained isolated. The unreported compartment on the other hand 592 transitions to recovery (U r ) after an infectious time T inf U . To allow for connectivity and exchange 593 34 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 between the defined partitions, cross contamination was included through the mobility matrix M jk 594 described above. Over the course of the studied period the Swiss government imposed a partial 595 lockdown over the country resulting in notable temporal variation of mobility and social interaction 596 patterns of the population of Basel-City which we account for with two time dependent weighting 597 factors α mob (t) and α soc (t) introduced above. In summary this resulted in the following system of 598 ODEs for the spread within each socioeconomic tertile j:

For all compartments, the initial number of susceptibles was fixed to the relevant population. All 607 other compartments were initialized as zeros, with the exception of a seed in the exposed compart-608 ment corresponding to the first reported cases. In summary, our model is based only on six free 609 model parameters, including the respective reproductive number per tertile R j (three parameters, 610 range [0, 20]), the initial number of exposed in a single tertile (range [0, 5]), and the infectious times 611 T inf U and T inf P (range [1.5, 12] days). We assumed a latency period T inc and infectious time prior 612 to symptom onset T inf P of two days each 54 .

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10. 1101 /2020 window average was taken to account for reporting bias on weekends, and cumulative numbers of 616 infected cases (compartment I) were calculated. Due to the loss of single sequencing plate, miss-617 ing numbers on the 29th, 30th and 31st of March were imputed by assuming a constant ratio of the 618 B.1-C15324T strain amongst the sequenced samples. Simulations were initialized on the 22nd of 619 February, the estimated date of the occurrence of the initial exposed cases 23 . 620

The ODE system was implemented in python (version 3.8.) using the scipy functions odeint to 621 iteratively solve the system of equations and minimize (with L-BFGS algorithm, cut-off tolerance 622 of 10 −7 ) for parameter fitting based on the average sum of squared differences between the loga-623 rithm of estimated and recorded cases. Points with cumulative case numbers below 15 were not 624 scored in the cost-function since the continuum assumption of the model may not be satisfied for 625 small case numbers. The fit was performed simultaneously for all four socioeconomic partitions 626 to account for the shared parameters T inf U (obtained 1.8 days) and T inf P (obtained 2.1 days). We compare effective reproductive numbers corresponding to the normalization of R by the the 633 36 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101/2020.12.15.20248130 doi: medRxiv preprint

Scenario simulation. The impact of mobility relative to social interaction was analysed by re-636 calculating the predicted epidemic trajectory under the constraint of constant intra-urban mobility . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 We thank all authors, who have shared their genomic data on GISAID, especially the Stadler Lab from CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 AG. KL provided virological expertise. AB provided serology samples from the blood transfusion service. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 Figure S1: The Canton of Basel City and its delineation with respect to statistical blocks colored according to the partition into tertiles T1, T2, and T3 of increasing fraction of 1-person households per block as provided by the canton's office for statistics. Inset: resulting mobility-graph, with nodes representing tertiles and edges representing effective connectedness through mobility by means of various modes of transport (thicker/thinner edges indicating weaker/stronger connectedness), as computed from the traffic-model provided by the traffic department of the Canton of Basel City.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101 https://doi.org/10. /2020 Figure S2 : The Canton of Basel City and its delineation with respect to statistical blocks colored according to the partition into tertiles T1, T2, and T3 of increasing fraction of residents aged older than 64 per block as provided by the canton's office for statistics. Inset: resulting mobility-graph, with nodes representing tertiles and edges representing effective connectedness through mobility by means of various modes of transport (thicker/thinner edges indicating weaker/stronger connectedness), as computed from the traffic-model provided by the traffic department of the Canton of Basel City.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101/2020.12.15.20248130 doi: medRxiv preprint Figure S3 : The Canton of Basel City and its delineation with respect to statistical blocks colored according to the partition into tertiles T1, T2, and T3 of increasing living space per person as provided by the canton's office for statistics. Inset: resulting mobility-graph, with nodes representing tertiles and edges representing effective connectedness through mobility by means of various modes of transport (thicker/thinner edges indicating weaker/stronger connectedness), as computed from the traffic-model provided by the traffic department of the Canton of Basel City.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101/2020.12.15.20248130 doi: medRxiv preprint Figure S4 : Data fit, reproductive number, and simulation of mobility and vaccination scenarios for a partition according to living space per person. A-C) Model fit to the case number time-series. Data points are shown together with model predictions based on undisturbed data (solid lines), and fifty bootstraps from disturbed data (bands) for the different tertitles T1(A), T2 (B) and T3 (C). D-F) The dynamic variation of the effective reproductive number for each of the tertiles shown in A-C). G) Influence of the mobility pattern on the total number of infected cases (sum of reported and unreported cases) assuming either no change in mobility (M1, 100% mobility), or full shutdown of all inner city mobility (M2, zero mobility). For comparison the observed scenario (M0) is shown. H) Prediction of vaccination effects if a specific percentage of all citizens was randomly selected for vaccination at given efficacy (scenario V1) compared to a simulation in the absence of any vaccination (V0). I) Prediction of vaccination effects if a specific percentage of all citizens was selected for vaccination from tertile T1 according to living space per person (scenario V2) together with the relevant scenarios V0 and V1. Dots indicate the time of reaching 50% ICU occupancy.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101/2020.12.15.20248130 doi: medRxiv preprint Figure S5 : Data fit, reproductive number, and simulation of mobility and vaccination scenarios for a partition according to the share of 1-person households. A-C) Model fit to the case number timeseries. Data points are shown together with model predictions based on undisturbed data (solid lines), and fifty bootstraps from disturbed data (bands) for the different tertitles T1(A), T2 (B) and T3 (C). D-F) The dynamic variation of the effective reproductive number for each of the tertiles shown in A-C). G) Influence of the mobility pattern on the total number of infected cases (sum of reported and unreported cases) assuming either no change in mobility (M1, 100% mobility), or full shut-down of all inner city mobility (M2, zero mobility). For comparison the observed scenario (M0) is shown. H) Prediction of vaccination effects if a specific percentage of all citizens was randomly selected for vaccination at given efficacy (scenario V1) compared to a simulation in the absence of any vaccination (V0). I) Prediction of vaccination effects if a specific percentage of all citizens was selected for vaccination from tertile T2 according to the share of 1-person households (scenario V2) together with the relevant scenarios V0 and V1. Dots indicate the time of reaching 50% ICU occupancy.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; Figure S6 : Data fit, reproductive number, and simulation of mobility and vaccination scenarios for a partition according to the share of senior residents. A-C) Model fit to the case number timeseries. Data points are shown together with model predictions based on undisturbed data (solid lines), and fifty bootstraps from disturbed data (bands) for the different tertiles T1(A), T2 (B) and T3 (C). D-F) The dynamic variation of the effective reproductive number for each of the tertiles shown in A-C). G) Influence of the mobility pattern on the total number of infected cases (sum of reported and unreported cases) assuming either no change in mobility (M1, 100% mobility), or full shut-down of all inner city mobility (M2, zero mobility). For comparison the observed scenario (M0) is shown. H) Prediction of vaccination effects if a specific percentage of all citizens was randomly selected for vaccination at given efficacy (scenario V1) compared to a simulation in the absence of any vaccination (V0). I) Prediction of vaccination effects if a specific percentage of all citizens was selected for vaccination from tertile T1 according to the share of senior residents (scenario V2) together with the relevant scenarios V0 and V1. Dots indicate the time of reaching 50% ICU occupancy.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101/2020.12.15.20248130 doi: medRxiv preprint Figure S7 : Urban quarters in Basel-City.

socioeconomic indicator achieved significance level living space per person 1% median income 2% fraction of 1-person households 2% fraction of residents aged above 64 45% Table S1 : Achieved significance level (ALS) of maximum differences in R eff associated with a partition of housing blocks according to various socioeconomic indicators. ALSs have been obtained by comparing these differences in R eff with those obtained from 99 bootstrapping random partitions.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10.1101/2020.12.15.20248130 doi: medRxiv preprint Figure S8 : Lineage identity (pangolin) of PCR-confirmed COVID-19 cases from 26th of February until 22nd of April, 2020, in Basel-City with B.1-C15324T as dominant variant highlighted.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; Figure S9 : Modelling of ICU capacity for the three vaccination scenarios V0 (no vaccine), V1 (random vaccination of 33% of the population) and V2 or V3 (vaccination of 33% of the population prioritizing the specified tertile). Results are shown for all socioeconomic partitions including median income (A), share of senior residents (B), living space per person (C), and share of 1person households (D). Dots indicate the time of reaching 50% ICU occupancy.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 17, 2020. ; https://doi.org/10. 1101 /2020 

SARS-CoV-2 patients with ICU stay, 40% were younger than 64 years resulting in a probability of 657 an under 64 year old

Acknowledgements We greatly appreciate the input and data received from Construction-and Traffic 660 department Canton

Statistical Office of the Canton of Basel-City, and want to specifically thank Björn Lietzke (Statistical Office 662 of the Canton of Basel-City), Madeleine Imhof (Statistical Office of the Canton of

Kathrin Grotrian (Construction-and Traffic 664 department Canton Basel-City), Matthias Hofmann (Basler Verkehrs-Betrieb)

Baselland Transport AG) for their support

University Hospital Basel) for excellent technical assistance with next generation sequencing