key: cord-0279430-a6mjdcql
authors: Zheng, Z.; Warren, J. L.; Artin, I.; Pitzer, V. E.; Weinberger, D. M.
title: Relative timing of RSV epidemics in summer 2021 across the US was similar to a typical RSV season
date: 2021-09-23
journal: nan
DOI: 10.1101/2021.09.22.21256632
sha: 354ff31fe760946971bcec7ef49c9f008fd81e1c
doc_id: 279430
cord_uid: a6mjdcql

We evaluated the timing of respiratory syncytial virus (RSV) epidemics in 2021 using Google search data. Despite the unusual out-of-season timing, the relative timing of RSV epidemics between states in 2021 shared a similar spatial pattern with typical RSV seasons. These insights can inform prophylaxis administration and planning of clinical trials.

Respiratory syncytial virus (RSV) infection is the most common cause of respiratory hospitalizations in young children [1] . Anticipating the timing of RSV epidemics is important.

Administration of available prophylaxis against RSV for high-risk infants needs to be timed with the onset of RSV epidemics. Moreover, the resumption of clinical trials for various RSV prevention strategies needs to be timed to the initiation of the epidemics to ensure sufficient power.

Anticipating epidemic timing for RSV is currently a challenge because the typical seasonal dynamics of the virus were disrupted by restrictions related to the COVID-19 pandemic. In the United States (US), the timing of typical RSV epidemics follows a notable spatial pattern, with epidemics that start earliest in Florida and proceed towards the north and the west [2] . During normal RSV seasons, the epidemics generally begin in the fall and peak in the winter. However, many states in the US observed an unusual increase in RSV cases during the spring and summer of 2021 after the absence of RSV epidemics in the winter of 2020-2021 [3] . As climate conditions differ from typical RSV seasons, these out-of-season RSV epidemics offer an opportunity to probe the drivers of RSV spatial transmission and improve predictions for future seasons.

Google search data have been validated as a good indicator of RSV epidemic patterns [4] [5] [6] . It is especially valuable for its timeliness and extensive geographical coverage compared with traditional clinical and laboratory data. Thus, we employed search engine data to estimate the . CC-BY-ND 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 (which was not certified by peer review) preprint onset of RSV epidemics in each of the 47 continental US states relative to Florida. We set out to evaluate the spatial variation in epidemic timing in summer 2021 compared to a typical season across the US. Our findings can inform the timing of administration of RSV prophylaxis and the planning of clinical trials for RSV vaccines during future epidemics in the US.

We used Google searches as indicators of RSV activity. Google searches for the keyword "RSV" were extracted from Google Trends for the 48 continental states in the US from July 2016 to August 2021 using the R package gTrendsR [7] . Google Trends provides normalized search data by week, which range from 0 to 100, to reflect the relative interests in relevant topics. The timing of search interest for RSV during a typical season and in 2021 is highly correlated with the timing of RSV estimated from the CDC's National Respiratory and Enteric Virus Surveillance System [4, 6] , RSV hospitalizations [4, 6] , and pediatric RSV encounters [5] .

We estimated the onset of RSV epidemics for each state in each epidemiological year. The epidemiological year was defined as July to the following June for typical RSV seasons and September 2020 to August 2021 for the out-of-season RSV epidemics. To estimate onset, we fit smooth curves to the search data using generalized additive models with penalized basis splines.

We defined the onset of typical RSV seasons as the time when the second derivative of the smooth functions reached its maximum, implying maximum increase in growth rates of RSV . CC-BY-ND 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 second derivative method is less preferable in 2021 as many states have not reached the turning point of the epidemic cycles. It is possible that the increase will continue and the epidemics have not reached the maximum increase in growth rates. Therefore, we determined a threshold based on the average search volume at the onset time in each previous season, with a separate threshold for each state. We then estimated when the search volume crossed these thresholds to determine the epidemic onset time for 2021.

We calculated the relative timing of RSV onset by subtracting the week of RSV onset in Florida (the earliest state) from the week of RSV onset in other states for each season.

We used a hierarchical Bayesian regression model to assess the similarity between timing of RSV epidemic onset in a typical year and the timing observed in 2021 across all states (Supplementary Methods). Onset time in a typical year was the predictor, and onset in 2021 was the outcome. We anticipated that if there were no major differences in timing between 2021 and previous years, onset time in a typical year would provide an unbiased estimate of the onset time in 2021(i.e., the intercept would be near zero and the slope would be near one). We included three state-level covariates that could potentially explain differences in the relative timing of RSV epidemics between 2021 and previous years [8] : population density, average household size, and stringency index of non-pharmaceutical interventions against COVID-19 [9] . Additionally, . CC-BY-ND 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 we also included spatially correlated random effects to account for geographic similarity in timing across states. The model is described in detail in the Supplementary Methods. Model parameters and their interpretations are described in Table S1 . Data and code to reproduce this study are available from Github: https://github.com/weinbergerlab/RSVtiming.git

In spring and summer 2021, RSV epidemics started in Florida and proceeded towards the north and west ( Figure 1A) . States in the northeastern corner (e.g. Maine and Vermont) and the central mountain states had the latest RSV epidemic onsets ( Figure 1A ).

In general, the relative timing of RSV epidemics in 2021 was 5.4 weeks later relative to Florida compared with that of typical RSV seasons ( Figure 1B and Table S1), while the ordering of epidemic onset and speed of spread were similar. Neither the stringency of non-pharmaceutical interventions, population density, nor household size were associated with the shift in the relative timing of the out-of-season RSV epidemics (Table S1) . Most states had a delayed timing compared with typical RSV seasons (95% credible intervals for intercepts greater than zero) except for Georgia and Alabama, the two southeastern states that border Florida (Table S1) .

These states had similar timing relative to Florida for typical RSV seasons and also formed a spatial cluster that was notably different from other parts of the US (Table S1 ).

Understanding the timing of RSV epidemics is crucial for clinical practice, including administration of prophylactic antibodies against RSV and for planning clinical trials. We sought . CC-BY-ND 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 to better understand the spatiotemporal pattern of RSV epidemics and the underlying factors associated with these patterns. We found that the spatiotemporal pattern of out-of-season RSV epidemics in 2021 was similar to that of typical seasons, but the relative timing of out-of-season RSV epidemics was delayed in most of states compared to three states in the southeastern US.

Stringency of non-pharmaceutical interventions, household size, and population density did not explain differences in the relative timing of RSV epidemics in 2021.

Other than two southeastern states (Georgia and Alabama), most states had a delayed start compared to Florida, but similar relative timing compared to previous RSV seasons. Thus, Florida and other southeastern states could serve as a bellwether in future years to adjust timing of the initialization of seasonal prophylaxis administration. Those conducting clinical studies could time recruitment of participants based on the relative timing in other states.

The reason for the delayed timing of RSV onset in other states compared to Florida, Georgia, and Alabama is unclear. However, the onset of the out-of-season RSV epidemics in other states coincided with increased volume of domestic air travel [10] . Future genomic research on virus circulation may help to explain whether these dynamics result from regional and national patterns of movement and transmission or whether other local factors are responsible.

Previous studies have suggested that climate factors play an important role in RSV seasonality and spatiotemporal patterns [11] . However, given that the spatiotemporal pattern of RSV epidemics in spring-summer 2021 was similar to that of previous fall-winter epidemics, it is unlikely that climate factors alone can explain variations in epidemic timing between locations.

. CC-BY-ND 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. Our study is subject to several limitations. First, we used Google search data on RSV as an indicator of RSV activity, which have lower specificity and sensitivity compared with inpatient and laboratory surveillance. While several studies validated the correlations between Google search data and clinical data sources [4, 5] , these correlations could be disrupted by increased attention to RSV in the news media, which can influence search behavior. Second, our analysis was performed at the state level, which neglected potential local variations. The relatively small number of data points may also lead to wide confidence intervals, masking the significance of potential predictors of relative timing. Future studies using finer spatial units may help improve the accuracy of projections.

In conclusion, the unusual out-of-season RSV epidemics in 2021 followed a similar spatiotemporal pattern compared to typical RSV seasons. The onset of RSV epidemics in Florida can serve as a baseline to adjust the initiation of prophylaxis administration and clinical trials in other states.

. CC-BY-ND 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-ND 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-ND 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 September 23, 2021. ; https://doi.org/10.1101/2021.09.22.21256632 doi: medRxiv preprint

The burden of respiratory syncytial virus infection in young children

Reduced-Dose Schedule of Prophylaxis Based on Local Data Provides Near-Optimal Protection Against Respiratory Syncytial Virus. Clinical infectious diseases : an official publication of the Infectious Diseases Society of

Respiratory syncytial virus tracking using internet search engine data

Using Google Trends to Predict Pediatric Respiratory Syncytial Virus Encounters at a Major Health Care System

Google searches accurately forecast RSV hospitalizations

Community factors associated with local epidemic timing of respiratory syncytial virus: A spatiotemporal modeling study

COVID-19 GOVERNMENT RESPONSE TRACKER

Department of Homeland Security Transportation Security Administration. TSA checkpoint travel numbers (current year versus prior year(s)/same weekday)

Environmental Drivers of the Spatiotemporal Dynamics of Respiratory Syncytial Virus in the United States