key: cord-0744341-dmk1itsc authors: Baker, R. E.; Park, S. W.; Yang, W.; Vecchi, G. A.; Metcalf, C. J. E.; Grenfell, B. T. title: The impact of COVID-19 non-pharmaceutical interventions on the future dynamics of endemic infections date: 2020-06-23 journal: nan DOI: 10.1101/2020.06.22.20137588 sha: 5ae07b5fc8a925ea3e722691f3763aa0379b0aff doc_id: 744341 cord_uid: dmk1itsc Non-pharmaceutical interventions (NPIs) have been employed to reduce the transmission of SARS-CoV-2, yet these measures are already having similar effects on other directly-transmitted, endemic diseases. Disruptions to the seasonal transmission patterns of these diseases may have consequences for the timing and severity of future outbreaks. Here we consider the implications of SARS-CoV-2 NPIs for two endemic infections circulating in the United States of America (USA): respiratory syncytial virus (RSV) and seasonal influenza. Using laboratory surveillance data from 2020, we estimate that RSV transmission declined by at least 20% in the USA at the start of the NPI period. We simulate future trajectories of both RSV and influenza, using an epidemic model. As susceptibility increases over the NPI period, we find that substantial outbreaks of RSV may occur in future years, with peak outbreaks likely occurring in the winter of 2021-2022. Results for influenza broadly echo this picture, but are more uncertain; future outbreaks are likely dependent on the transmissibility and evolutionary dynamics of circulating strains. : Reduction in RSV and Influenza cases since March 2020. The percent positive laboratory tests for a) RSV and b) Influenza across four US states. Data from 2020 are highlighted in red (RSV) and light blue (influenza). Data from previous seasons (2016-2019) are highlighted in grey. c) 2020 change relative to seasonal mean for influenza for all available US states (RSV surveillance data is only available for select states). Dashed lines show timing of the declaration of national emergency these two respiratory infections. We focus primarily on RSV, with the simpler limit cycle dynamics presenting an opportunity to probe interactions with NPIs. We first evaluate the influence of control measures targeting SARS-CoV-2 using influenza and RSV surveillance data. Since changes to physician visits for both viruses could be driven by behavioral responses to control measures, we look at the percent positive tests for both viruses as reported from laboratory surveillance data. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint COVID-19 pandemic was declared on March 13th 2020 in the US, shown with the dashed line. Following the declaration, many states put in place control measures to limit the spread of SARS-CoV-2. Despite the declaration occurring after the typical seasonal peak in cases, a decline in prevalence is observed beyond mean seasonal levels. In Florida, where RSV cases tend to persist throughout the year [11, 25] , observed RSV prevalence is reduced to near zero in March 2020. A similar pattern is visible in Hawaii for influenza, where cases are normally persistent. In Fig 1c we show the 2019-2020 change in percentage positive influenza tests relative to weekly mean over the previous four seasons. The 2019-2020 influenza season appears to have been more severe than average, with a relative increase in prevalence prior to March 2020 possibly driven by increased circulation of influenza subtype B (Fig. S2 ). However, following the declaration of emergency, declines to below average levels can be observed across almost all reporting states. To explore the possible implications of control (i.e. NPI) periods for the future dynamics of influenza and RSV we use epidemiological models and consider a range of possible scenarios for the length and intensity of control measures. Given the current uncertainty in the future course of the COVID-19 pandemic, and how responses might change over time, we cannot make precise predictions of future outcomes. For RSV, we use the time-series SIR model [26, 27] , fitted to historic US case data described in previous work [11] . Specifically, we evaluate how NPI perturbations impact the epidemic limit cycles of RSV. We first consider a range of control period lengths and percent reduction in transmission based on Florida and Texas seasonality. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. We use RSV laboratory surveillance data for Florida and Texas to parameterize the actual reduction in transmission caused by SARS-CoV-2 NPIs. We find that a reduction in transmission of 20% is able to conservatively capture the decline in prevalence recently observed in the surveillance data (Fig. 2c, Fig. S3 ). Using this model parameterization, we run simulations 6 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint county-level case data (Fig 3) [11] . Counties with short time series (less than fives years of data) and sparse numbers of cases (under ten at peak) are removed. We compare the impact of two periods of control: lasting six months (Fig 3a,c) and lasting one year (Fig. 3b,d) . Although the six month control period occurs outside the peak season of the virus, substantial RSV outbreaks are still projected as a lagged response to the SARS-CoV-2 NPIs. In general, the longer, one-year control period results in larger RSV outbreaks, yet complex interactions with seasonality arise. For New York county, the shorter control period results in a large outbreak in the following winter (2022), but the longer control period results in a more persistent but less intense outbreak. In contrast, a large RSV outbreak is observed in Miami after a year of control measures. In most cases, simulated dynamics eventually return to the pre-NPI attractor. For Boulder county, by contrast, control periods have complex interactions with the seasonal biennial epidemics of the disease. In these deterministic simulations, a longer control period in Boulder county causes the epidemic trajectory to shift to a different attractor (Fig 3e, Fig. S7 ). In general, the timing and size of future outbreaks will depend on the interaction between the dynamics of susceptibility and the seasonality of transmission. Compared to RSV, influenza epidemics exhibit a less uniform seasonal pattern. Gradual evolution of the influenza virus' antigenic sites (antigenic drift), means population susceptibility changes over time [21] and different subtypes may circulate each year with different levels of transmission [28] . In our preliminary analysis we therefore focus on the overall dynamics of susceptibility, ignoring year-to-year differences in circulating strains. We simulate influenza using a Susceptible-Infected-Recovered-Susceptible (SIRS) model, developed in previous analyses to explore influenza seasonality in the USA, where R 0 varies between a maximum and minimum value driven by changes in absolute humidity [12, 13] . To capture the variability in transmission rates, we consider two scenarios: R 0max = 3 and R 0max = 2.2, based on the range of prior estimates [13, 28] . In both scenarios, R 0min = 1.2. We simulate the model using the 8 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. after the NPI period are slightly elevated. In contrast, longer control periods provide more time for the susceptibility to build, resulting in an earlier influenza epidemic starting in the summer months. In the low transmission scenario, this is followed by a large outbreak in 2021. While these results suggest a more uncertain impact of NPI periods on future influenza outbreaks, 9 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint dynamics will likely differ in locations with more persistent influenza cycles, such as the tropics [9] . Non-pharmaceutical interventions put in place to limit the spread of SARS-CoV-2 are already beginning to affect the transmission of other directly-transmitted, endemic diseases. Our results suggest that a build up of susceptibility during these control periods may result in large outbreaks in the coming years. Results for RSV in the USA suggest that these outbreaks may reach their peak in the winter of 2022. This finding appears robust even when we account for possible imported cases (results not shown). Following perturbation, RSV generally returns to the endemic attractor, but more complex behavior is possible (Fig. 3e) . Preliminary results for influenza suggest outbreaks may occur outside of the typical season, coinciding with the end of the control period. However, we do not address complex features of the influenza virus such as circulating subtypes or the implications of global NPIs for antigenic drift. The latter may prove significant, for example, if the evolution rate first declines with NPIs, then rebounds [29] . More broadly, our results suggest that healthcare systems may need to prepare for future outbreaks of non-COVID-19 infections, as NPIs are relaxed. These outbreaks may occur several years after initial NPIs were put into place. There are several caveats to these results. First, we are at the early stages of understanding the implications of SARS-CoV-2 NPIs for endemic infections. In our model we used a fixed reduction in transmission however, this may not capture heterogeneities in NPIs across locations and over time. As more surveillance data becomes available, tracking further changes to endemic disease prevalence will be important. Serological surveys, currently used to measure exposure and potential immunity to SARS-CoV-2, could similarly be employed to monitor these polymicrobial responses [30, 31, 32] . Second, an influx of COVID-19 cases could artificially lower the percent positive test data we use to calibrate reductions in RSV and influenza transmission. For RSV this seems unlikely as the mean age of infection is much lower than 10 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint sharp decrease observed across states right after the national emergency declaration suggests that it is unlikely to be primarily driven by this factor. Finally, interactions between the SARS-CoV-2 virus and endemic viruses may be more complex than described here. Immunological relationships between viruses, both competitive and cooperative, may have broad scale implications for future infection dynamics [33] . The impact of NPIs on strain structure of RSV [34] is an important area for future work. Finally, although we have primarily focused on the USA, outcomes may be more severe in southern hemisphere locations where NPI timing aligns with the peak season for seasonal wintertime diseases. Our results also illustrate the potential for COVID-19 NPI to impact the dynamics and persistence of a much wider range of infections. Increased surveillance, serological surveys and local modeling efforts will help determine the future dynamics and risk from these circulating infections. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. RSV epidemics exhibit distinct dynamic patterns driven by local climate [11, 25] . Locations in the USA with a large seasonal variation in specific humidity tend to experience biennial RSV dynamics where a large outbreak is followed by smaller outbreak. Locations with a moderate seasonal variation in specific humidity tend to experience annual outbreaks and locations with less seasonal variation in specific humidity tend to experience more persistent RSV outbreaks. In these latter locations, such as in Florida and southern Mexico, outbreaks tend to coincide with periods of elevated rainfall. RSV epidemic dynamics exhibit a stable limit cycle structure. For a few locations, co-existing attractors are possible, with perturbations to the system driving a shift to an alternate stable limit cycle (Fig. S7) [35] . Although there are two circulating RSV strains, strain-level data were not available for the USA, so our models add across strain structure [34] . We first calculate location-specific seasonal transmission rates using the time series Susceptible-Infected-Recovered model (TSIR), a discrete time adaptation of the SIR model [36, 26] . Countylevel transmission rates were calculated for a previous study [11] . The TSIR model describes the number infected and susceptible individuals as a set of difference equations. The number of 12 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint susceptible individuals is given by: where S t is the number of susceptible individuals, I t is the number of infected individuals, B t is births and u t is additive noise, with E[u t ] = 0. The time period t is the generation time for RSV, set at 1 week. The susceptible population can be rewritten as S t =S + Z t whereS is the mean number of susceptible individuals in the population and Z t is the unknown deviation from the mean number of susceptible individuals at each time step. Equation (1) is rewritten in terms of these deviations and iterated starting at Z 0 : where ρ is the reporting rate which accounts for both under-reporting of RSV hospitalizations as well as infections that did not result in hospitalization and Ir k is the reported incidence. Using this formulation, it is shown that a linear regression of cumulative births on cumulative cases, gives Z t as the residuals, assuming u t is small. The inverse of the slope of the regression line provides an estimate of the reporting rate ρ.S is calculated by defining the expected number of infected cases at each time step, E[I t+1 ], as: which is log-linearized as: where β t are biweekly factors that capture the seasonal trend in transmission rate and α is a constant that captures heterogeneities in mixing and the discretization of a continuous time process. We fix α at 0.97 to be consistent with prior studies [37, 11] . Equation 4 is fit using a Poisson regression with log link. The mean number of susceptible individuals,S, can then 13 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint be estimated using marginal profile likelihoods from estimating equation (4), for a range of candidate values. Following [11] , we add one to zero observations in the infected time series which represents continual low-level background transmission resulting in the lack of epidemic extinction we observe in the data. For fitting the TSIR we use the tsiR package [38] . When fitting to state-level data for Texas and Florida, we use a locally-varying spline regression for equation (2) , which accounts for macro-scale changes in reporting over time. We generate forward simulations using county seasonal transmission rates, βt, assuming a constant population and birth rate (based on average population and average birth rates from the historic time series). Model results are shown in terms of incidence per capita. The simulations are initially run for 40 years to remove transient dynamics. The control period is introduced to the model by lowering the seasonal transmission rates by a fixed proportion, starting on week 11 of 2020 (the week when a National Emergency was declared). For all simulations we lower the transmission by 20% unless otherwise specified. The percentage reduction in transmission is estimated by comparing model simulations to laboratory RSV data from 2020 for Texas and Florida. State-level data from Minnesota are not used because the laboratory data does not capture the biennial cycles of incidence that exist in this state [11] . Other states do not have multiple years of available data to compare present reductions in prevalence. Lab test data are scaled to the model projection using the 2016-2020 mean i.e. T scaled = (T /T ) * P where T is the laboratory test data and P is the model projections. Simulations are run using reductions in transmission ranging from 0% to 90% in 10% intervals. For Florida, a 20% reduction in transmission, starting in week 11 (when the national emergency was declared), is found to be the best fit (Mean Absolute Error) reduction based on available data (Fig. S3 ). For Texas, both 10% reduction and 20% reduction give similarly good fits. Joint error is minimized using the 20% reduction rate. For influenza we use a climate-driven Susceptible-Infected-Recovered-Susceptible model 14 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint [13, 28, 14] . Antigenic drift of the influenza virus results in a seasonal return to susceptibility meaning TSIR methods are not appropriate for this infection. The model is described by a series of differential equations: As before, S is the susceptible population, I is the number of infected individuals and N is the total population. D, the mean infectious period, is set a 4 days. L, the duration of immunity, is fixed at 40 weeks, allowing the influenza epidemic to recur each season. β(t) is the contact rate at time t and is related to the basic reproductive number by R 0 (t) = β(t)D. R 0 is related to specific humidity q(t) using the equation: R 0 (t) = exp(a * q(t) + log(R 0max − R 0min )) + R 0min (7) where a = −180, based on earlier findings [12, 13, 28] . R 0min is minimum reproductive number, fixed at 1.2, following [28] . R 0max is the maximum reproductive number. In Fig. 4 , we use values of R 0max = 2.2 and R 0max = 3, based on plausible ranges observed in [13, 28] . 15 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137588 doi: medRxiv preprint Differential effects of intervention timing on covid-19 spread in the united states An investigation of transmission control measures during the first 50 days of the covid-19 epidemic in china Effect of non-pharmaceutical interventions to contain covid-19 in china The effect of travel restrictions on the spread of the 2019 novel coronavirus (covid-19) outbreak The effect of large-scale anti-contagion policies on the coronavirus (covid-19) pandemic Respiratory virus shedding in exhaled breath and efficacy of face masks Impact assessment of non-pharmaceutical interventions against coronavirus disease 2019 and influenza in hong kong: an observational study The calendar of epidemics: Seasonal cycles of infectious diseases Influenza in tropical regions Synchrony, waves, and spatial hierarchies in the spread of influenza Epidemic dynamics of respiratory syncytial virus in current and future climates Absolute humidity modulates influenza survival, transmission, and seasonality Absolute humidity and the seasonal onset of influenza in the continental united states Susceptible supply limits the role of climate in the early sars-cov-2 pandemic Travelling waves and spatial hierarchies in measles epidemics The dynamics of measles in sub-saharan africa Impact of vaccination on the spatial correlation and persistence of measles dynamics Demographic variability, vaccination, and the spatiotemporal dynamics of rotavirus epidemics Quantifying seasonal population fluxes driving rubella transmission dynamics using mobile phone data Long-term dynamics of measles in london: Titrating the impact of wars, the 1918 pandemic, and vaccination Molecular mechanisms of variation in influenza viruses The evolution of seasonal influenza viruses Mapping the antigenic and genetic evolution of influenza virus Global, regional, and national life expectancy, all-cause mortality, and cause-specific mortality for 249 causes of death, 1980-2015: a systematic analysis for the global burden of disease study Environmental drivers of the spatiotemporal dynamics of respiratory syncytial virus in the united states Time series modelling of childhood diseases: a dynamical systems approach Dynamics of measles epidemics: scaling noise, determinism, and predictability with the tsir model Inference of seasonal and pandemic influenza transmission dynamics Impact of cross-protective vaccines on epidemiological and evolutionary dynamics of influenza Revealing measles outbreak risk with a nested immunoglobulin g serosurvey in madagascar Measles virus infection diminishes preexisting antibodies that offer protection from other pathogens A global immunological observatory to meet a time of pandemics Virus-virus interactions impact the population dynamics of influenza and the common cold Understanding the transmission dynamics of respiratory syncytial virus using multiple time series and nested models A simple model for complex dynamical transitions in epidemics Dynamics of measles epidemics: estimating scaling of transmission rates using a time series sir model. Ecological monographs 72 Interpreting time-series analyses for continuous-time biological models-measles as a case study tsir: An r package for time-series susceptible-infectedrecovered models of epidemics