Impact of public health policy and mobility change on transmission potential of severe acute respiratory syndrome coronavirus 2 in Rhode Island, March 2020 – November 2021

ABSTRACT To study the SARS-CoV-2 transmission potential in Rhode Island (RI) and its association with policy changes and mobility changes, the time-varying reproduction number, Rt, was estimated. The daily incident case counts (16 March 2020, through 30 November 2021) were bootstrapped within a 15-day sliding window and multiplied by Poisson-distributed multipliers (λ = 4, sensitivity analysis: 11) to generate 1000 estimated infection counts, to which EpiEstim was applied to generate Rt time series. The median Rt percentage change when policies changed was estimated. The time lag correlations were assessed between the 7-day moving average of the relative changes in Google mobility data in the first 90 days, and Rt and estimated infection count, respectively. There were three major pandemic waves in RI in 2020–2021: spring 2020, winter 2020–2021 and fall-winter 2021. The median Rt fluctuated within the range of 0.5–2 from April 2020 to November 2021. Mask mandate (18 April 2020) was associated with a decrease in Rt (−25.99%, 95% CrI: −37.42%, −14.30%). Termination of mask mandates on 6 July 2021 was associated with an increase in Rt (36.74%, 95% CrI: 27.20%, 49.13%). Positive correlations were found between changes in grocery and pharmacy, Rt retail and recreation, transit, and workplace visits, for both Rt and estimated infection count, respectively. Negative correlations were found between changes in residential area visits for both Rt and estimated infection count, respectively. Public health policies enacted in RI were associated with changes in the pandemic trajectory. This ecological study provides further evidence of how non-pharmaceutical interventions and vaccination slowed COVID-19 transmission in RI.


Introduction
The coronavirus disease 2019 (COVID-19) pandemic is in its fourth year despite the best efforts of public health experts and clinical practitioners.The Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) spreads as infected individuals release droplets of microscopic particles through breathing, sneezing, or coughing [1].
The COVID-19 pandemic impacted physical health and economic stability across the United States (US) [2].A rise in stress, anxiety, and depression among the general population and survivors of the SARS-CoV-2 infection is well documented [3].The far-reaching consequences of the pandemic are factors considered by policymakers as they make federal, statewide, and countywide policies for the general population [4].Data surrounding cases, deaths, and risk of infection are an integral part of risk communication to the general population, depicting the dire situation of the pandemic [5].
RI is unique in that it highlights the trajectory of COVID-19 in a progressive state.The factors associated with high COVID-19 vaccination rates include older age [9], a desire to protect family members, and politics.The median age in RI was 40.0 years compared to 38.2 years for the US as a whole in 2020 [10].For an eastern state, the percentage living in multi-generational households, which represent 5.1% of all households, is high.Democrats are more likely to accept COVID-19 vaccines than Republicans [11].Recent elections have tilted toward the Democratic Party.In the 2020 presidential election, 38.6% of the population voted Republican [12].
Across RI on 31 January 2022, 829,885 residents were reported to have completed primary vaccine efforts (i.e. two doses of the Pfizer or Moderna vaccine or one dose of the Johnson & Johnson vaccine), whereas 987,640 residents were at least partially vaccinated.A total of 382,692 residents, i.e. 36.2% of the total population had received a booster dose [8].
Time-varying reproduction number (R t ) is the mean number of secondary cases generated by an infectious individual in the presence of public health interventions, behavioral changes, and increase in community immunity level.Hence, R t changes over time throughout an epidemic.Timely estimates of R t inform policymakers about how implemented policies and behavioral changes at the state and county level are likely to impact COVID-19 transmission.When R t >1, transmission is sustained, whereas when R t is maintained below 1, the epidemic will eventually die out [13,14].The purpose of this study is to retrospectively assess the impact of various policies on the SARS-CoV-2 transmission potential across RI in 2020-2021 and to examine the cross-correlations between changes in mobility patterns in RI with estimated infection counts and R t , respectively, in the first 90 days of the pandemic in RI.

Study design and data acquisition
This is an ecological study.The cumulative case count data for each county across RI were obtained from the New York Times GitHub data repository up till 16 December 2021 [15].The first COVID-19 case in RI was reported on 1 March 2020.From 1 March through 24 March 2020, county names were listed as 'unknown' in the RI COVID-19 confirmed cases dataset.Since 25 March 2020, the RI had started categorizing the cases into one of the five counties [16].We decided to drop the unknown county-level data and then merged the known county-level data to obtain an aggregate for the statewide data.Therefore, the time series of the data analyzed in this study began with cases reported on 25 March 2020.Probable cases identified through antigen testing were included in the case counts.For subsequent analysis, incident case count data (Figure S1) was derived from the cumulative case count data.

R t estimation
R t was estimated using the instantaneous reproduction number method [13].The EpiEstim package version 2.2-4 in R version 4.1.0was utilized for the data analysis [17].The serial interval distribution was parametrically defined as mean = 4.60 days and standard deviation = 5.55 days [18].(Details of the sensitivity analysis are in the next paragraph.)The time series of daily incidence were shifted 9 days backward to account for the mean incubation period of 6 days and a median testing delay of 3 days [19,20].The cutoff point was moved forward 1 week early to allow for a 15-day sliding window.Therefore, the time frame for this study was 16 March 2020 -30 November 2021 by the assumed date of infection.The case count data was then bootstrapped 100 times and then multiplied by 10 multipliers drawn from a Poisson distribution with a mean of 4 (or 11 in the sensitivity analysis) to obtain 1000 time series of estimated number of infections.These infection count time series were used to generate R t estimates.Furthermore, we analyzed and presented R t in two ways: first, as a moving average R t in 7-day sliding window, and second, an average R t over a non-overlapping time period between two changes in public health policy.The latter (shorthanded as 'policy change R t ') was used to quantify the change in R t at particular time points when the policy changed.The median and the 95% credible intervals (CrI) of the 7-day sliding window R t and the policy change R t reported were generated from 10,000 random values drawn from the 1000 R t posterior distributions (10 per estimated infection count time series) for each time window.Percentage change was determined for the non-overlapping time window R t using the formula: R t2 À R t1 R t1 � 100, where R t2 represents the R t estimate of the time window after a new policy was implemented and R t1 refers to the previous window.The percentage change in R t and its 95% CrI were obtained through bootstrapping 1,000 times.For county-level data, we only applied the analysis to the main scenario (Poisson-distributed multiplier of 4 and serial interval distribution with a mean of 4.60 days and a standard deviation of 5.55 days).Further technical details are provided in the following paragraphs.

Sensitivity analysis: serial interval
Owing to the variability of the mean serial interval based on prior studies [21], we conducted a sensitivity analysis to estimate R t using the instantaneous reproduction number method [13], with an alternative serial interval distribution with a mean of 4.55 and a standard deviation of 3.30 [21,22].The sensitivity analysis was conducted for state-level data only.

Addressing the uncertainty in data reporting
We used two separate approaches to address the two sources of uncertainty in data reporting: the uncertainty in reporting delay and the uncertainty in underreporting.

Bootstrapping of incident case count time series.
Accounting for the uncertainty in reporting delay, we carried out bootstrapping of incident case count data (that were already shifted backward by 9 days to approximate the date of infection).Using a sliding window of 15 days (7 days prior to the time t through 7 days after time t), we sampled with replacement 100 numbers for the time t from the incident case count data in the time window.This was repeated for each day in the incidence curve, and we generated 100 bootstrapped time series of incident case counts.The 95% uncertainty bound represents the 2.5th quantile and 97.5th quantile of these bootstrapped case count time series at each date.

Estimation of infection counts utilizing a multiplier.
Next, to account for underreporting and its uncertainty, a total of 1000 time series of estimated infection counts would then be estimated from the 100 bootstrapped incident case count time series (10 infection count time series per bootstrapped incident case count time series).For each day of each incidence curve, 10 multipliers would be randomly drawn, with replacement, from a Poisson distribution of a mean of 4 and 11, respectively.The bootstrapped case count data would be multiplied by the multipliers to generate the infection count time series.The 95% uncertainty bound represents the 2.5th quantile and 97.5th quantile of these infection count time series at each date.
The multipliers of 4 and 11, respectively, were chosen as illustrations for our method as two different CDC web pages suggested that among those infected with SARS-CoV-2 in the US, only 1 in 4 cases or 1 in 11 cases are being reported and notified [23,24].The multiplier of 4 was used in this study as the main case scenario that is applied to both state-and county-level data.The multiplier of 11 was applied to state-level data only.

Choice of policy as cutoff points for policy change R t
For the policy change R t estimation, the implementation, and relaxation of specific non-pharmaceutical interventions (NPIs) were chosen to study how policy change might be associated with changes in R t (see Supplementary Table S1).Meanwhile, the dates of the rollout of vaccination expansion across RI were chosen arbitrarily to earmark periods representing phases of the pandemic in RI.They included the first vaccination rollout on 14 December 2020 (label 'D'), the expansion of vaccination efforts to 16 years and older on 3 March 2021 (label 'E'), the expansion of vaccination efforts to 12 years and older on 12 May 2021 (label 'F'), and the expansion of vaccination efforts to 5 years and older on 3 November 2021 (label 'H').We note that these dates should be interpreted as arbitrary cutoff points as the impact on the change in R t would be gradual as vaccines took time to be administered, and it took time to develop immunity upon vaccination.

Google mobility: Correlation analysis
Data on mobility were downloaded from the Google database to assess changes in the number of visits to the following locations grouped into distinct categories: (a) grocery and pharmacy, (b) retail and recreation, (c) parks, (d) transit, (e) residential, and (f) workplaces [25].The data provided insight into how visits and duration of stay across multiple locations changed compared to the median values before the pandemic began, from the corresponding day of the week.The data were representative of Google users who provided consent to utilizing their location histories.
The time-lagged cross-correlation between the 7day rolling mean of relative mobility (percentage change from baseline) and the median R t estimates from the 7-day sliding window, as well as the estimated daily number of SARS-CoV-2 infection were assessed, respectively.The analysis was performed for the first 90 days of the pandemic (i.e. between 16 March 2020, through 16 May 2020); the time lag used was between −3 days and +3 days.The correlation with the highest magnitude was reported in the text.The R t used in the main analysis was estimated with a serial interval with a mean of 4.60 days and a standard deviation of 5.55 days; in the sensitivity analysis, the R t used was estimated with the alternative serial interval distribution (mean, 4.55; standard deviation, 3.30).

Ethics
The Georgia Southern University Institutional Review Board made a non-human subject determination for this research project (H20364) under the G8 exemption category according to the Code of Federal Regulations Title 45 Part 46.

Results
There were three major waves of COVID-19 cases in RI over the study period from March 2020 through November 2021.The first wave was in Spring 2020.The second wave started in winter 2020-2021 and ended prior to summer 2021.The third wave began in late summer 2021, and there was a major surge by November of 2021 (Figure 1).Daily incident case count and cumulative case count of the five counties in RI by four specific dates in the study time frame were presented in Supplementary Figures S2  and S3.County-level bootstrapped case counts, estimated infection counts, and R t results were summarized in the Supplementary Materials and in Figures S4-S8.

7-day sliding window R t estimates at the state level
At the state level, the 7-day sliding window R t in RI state was around 2 in late March 2020, R t dropped from around 2 in March 2020 to below 1 in late April 2020.In fall 2020 (R t >1), sustained transmission preceded the major wave in winter 2020.A decline in R t saw the case count dropped and leveled off in Spring 2021 before dropping to a low level in Summer 2021.A surge in R t to 1.5 in Summer 2021 preceded a surge in cases in Fall 2021 and a subsequent wave in November 2021 (Figure 1).
It is important to note that the 95% CrI of the 7-day sliding window R t includes 1 for multiplier of 4 through a significant portion of the time frame as represented in Figure 1.We observed that if we made the alternative assumption regarding the infection-to-case ratio and changed the multiplier from 4 to 11, the 95% CrI of the 7-day sliding window R t narrows (Figure 2).Changing the assumption of the serial interval distribution might change the 95% CrI a little bit, but it did not change the trend nor the conclusion (Figures 3 and 4).

Policy change R t
Supplementary Table S1 presents a list of COVID-19 related policies in RI.The R t estimated using nonoverlapping windows suggested the possible effects of NPIs on the increase and decrease of SARS-CoV-2 transmission.We also presented how R t changed over different pandemic phases since vaccination was rolled out.All the following numbers refer to the main scenario with an infection-to-case ratio of 4 and a mean serial interval of 4.60 (standard deviation, 5.55) days.
The stay-at-home order was issued in 43 out of 50 states across the US when the pandemic first hit.The main aim was to decrease the interpersonal contacts and thus the transmission potential of SARS-CoV-2 [26,27].The Eleventh Emergency Declaration on 28 March 2020, enacted the stay-at-home order, and limited public gathering to 5 or fewer people (label 'A').It was associated with a statistically significant decrease in R t at the county-level in Washington County (−29.70%,95% CrI: −56.97%, −8.33%)only (Supplementary Tables S2-S7).
Termination of mask mandates on 6 July 2021 (label 'G'), coincided with a wave of cases of the Delta variant, was associated with an increase in R t at the state and county levels: 36 5 and Supplementary Tables S2-S7).
The following time period of 12 May -6 July 2021 (Label 'F') was observed to be associated with a significant decrease in R t across the state and all counties except Washington County: a change by − 21 S7).

Correlation analysis
changes in mobility to retail and recreation establishments, grocery stores and pharmacies, workplaces, parks, transit locations, and residential locations in RI are presented in Figures S9-S16.Using the 7-day moving average of the relative changes in Google mobility data for the first 90 days, its respective correlation with Rt and estimated infection count were explored (Figure 7).
Relative changes in the volume of visits to transit locations, workplaces, retail and recreation centers centers, and grocery and pharmacies were found to have a significant positive correlation with R t , respectively, in decreasing order of magnitude (r = 0.865, p < 0.001; r = 0.822, p < 0.001; r = 0.815, p < 0.001; r = 0.434, p < 0.001).However, relative changes in the volume of visits to residential buildings had a negative correlation with R t (r = −0.789,p < 0.001) (Table 1).
The analysis suggests relative changes in the volume of visits to workplaces, transit, retail and recreation centers, and grocery and pharmacies were positively correlated with the estimated infection count respectively, in decreasing order of magnitude (r = 0.813, p < 0.001; r = 0.809, p < 0.001; r = 0.739, p < 0.001; r = 0.603, p < 0.001).In addition, mobility changes to residential areas were negatively correlated with the estimated infection count (r = −0.760,p < 0.001) (Table 1).
Sensitivity analysis performed using the R t generated with the alternative serial interval (mean, 4.55; standard deviation, 3.30) produced very similar results with the same direction of correlation in similar magnitude (Table S8, Figure S17).

Discussion
Like various states across the US, RI began implementing NPI mandates and orders in March 2020 to prevent SARS-CoV-2 transmission.Meanwhile, COVID-19 vaccines became available, and their coverage has expanded since December 2020 [28].There were statistically significant decreases in R t statewide and at the county level in the period immediately following 14 December 2020 (the date of the first vaccination rollout).The date following 12 May 2021 (the expansion of vaccination to 12 years and older) was associated with a statistically significant decrease in R t across the state of RI, Bristol County, Kent County, Newport County, and Providence County.However, the period that began on 3 March 2021 (the expansion of vaccination to 16 years and older) was significantly associated with an increase in R t for the state and in Kent County, and Providence County.The increase was insignificant for Bristol County and Newport County.In Washington County, there was a statistically significant decrease (Figure S8).Also, the Omicron wave coincided with the period immediately following 3 November 2021 (the expansion of vaccination to 5 years and older) that was associated with a significant increase in R t in Kent County, Newport County, and Providence County, while the increase was insignificant across the state, and in Bristol County, and Washington County (Figure 6 and Figures S4-S8).Its apparent association with an increase in R t should be interpreted with caution as association does not imply causality.Moreover, the choice of using the dates of expansion of vaccination as cutoff points is arbitrary since individuals within the population do not get vaccinated simultaneously; surely, variations in the timing of vaccination exist.In addition, it takes 2 weeks for individuals to become fully vaccinated after the final dose in primary series, and it takes about 3 or 4 weeks apart from each dose of a two-dose regimen of the mRNA vaccines [29].With pharmaceutical interventions like vaccination, concomitant events of different nature are occurring simultaneously that contribute to fluctuations in R t .Whereas for NPIs like face mask mandates, an immediate change is observed following the implementation of the policy.To further illustrate this, mask mandate termination was statistically significantly associated with an increase in R t both at the state and county levels.Likewise, we observed a statistically significant association between the mask mandate and a decrease in R t at the state and county levels except Bristol County and Newport County, where the decrease was statistically insignificant (Figure 5).
A study on the COVID-19 pandemic in Arkansas and Kentucky in 2020 presented similar findings, such as a decrease in R t following the implementation of the mask mandates, and an increase in R t when the states began to reopen after the stay-at-home order and the termination of mask mandates [30].All five counties across RI displayed similar trends in R t , which may be attributable to adherence to state-wide mandates at the beginning of the pandemic.By June 2020, RI implemented a culturally tailored SARS-CoV-2 testing service that provided convenient testing services with access to translators, multiple modalities for accessing test results and did not require documentation or insurance [31].The sustained transmission of COVID-19 across the US in summer 2020 could be attributed to sustained transmission among young adults, a group that overwhelmingly resumed normal level of activities [32].Meanwhile, the surge of cases in summer and fall 2021 was likely associated with the Delta variant [33].
The variations in R t with policy level suggest that the levels of compliance to NPIs for SARS-CoV-2 across the state differed at the county level.Compliance with NPI policies may decrease over time due to pandemic fatigue [34].Health information campaigns that address misinformation and educate the public on how they can protect themselves via vaccination and NPIs remain crucial [34,35].
Moreover, in November 2021, the Omicron variant emerged as the dominant variant of concern in the US that led to a new wave of cases by November 2021.The new variant coincided with the expansion of vaccination to children aged 5 years and older across the country [36,37].The elevated R t was likely a consequence of the increased transmissibility of the Omicron variant [38,39].Thus, the apparent increase in R t after the expansion of vaccination to children should be appropriately attributed to the spread of the Omicron variant and interpreted as such.

Limitations
Our study is subject to limitations.First, incident case counts reported by county across RI had irregular intervals, and thus required data interpolation.Second, the number of reported cases was likely a fraction of the number of infections.The CDC estimated that from February 2020 to September 2021, only one-quarter of COVID-19 infections in the US have been reported [40].A number of factors contributed to this phenomenon.Accessibility issues related to testing and healthcare provision could have contributed to the undertesting and underreporting of cases [41].Asymptomatic cases might not get tested and thus not be confirmed cases; yet they might contribute to the SARS-CoV-2 transmission that the EpiEstim package did not account for.In this study, we attempted to estimate the infection count from incident case count using a Poisson-distributed multiplier.It is noteworthy that starting 19 January 2022, the US federal government made available at-home SARS-CoV-2 antigen tests, which Americans can order online at COVIDTests.gov and will be mailed directly to their household addresses [42].However, this new development was beyond the time frame of this study.Third, this is an ecological study.The pitfalls of ecological fallacy should be cautioned: what we found at the aggregate level cannot be inferred at the individual level.Individuals' compliance to certain policies was not measured and might vary between individuals.Furthermore, as are other observational studies, other factors might confound any association that we might observe.For example, the emergence of a new highly transmissible variant that coincided with the time period in which there was an expansion of vaccination efforts to a younger age group would have confounded our interpretation of the R t change.Any suggestion that an expansion of vaccination to younger age groups would have led to an increase in R t would be misguided.That said, the policy change R t estimates could provide some indications of the potential effect of policies as they were implemented in real life if they were interpreted cautiously.Fourth, while we accounted for the incubation period and the delay to testing by shifting the case count data by 9 days to approximate the dates of infection, these remained estimates.Fifth, while we used the assumed dates of infection in our analysis, the actual dates of infection are unknown.This had the potential to introduce potential confounders in the analysis which are unaccounted for.We acknowledge that we did not use the deconvolution method, which is computationally intensive and conceptually challenging, to estimate the dates of infection.Nonetheless, our simple method of shifting backward by the incubation period and the reporting delay was deemed 'tolerable' by Gostic et al. in their comparison between different methods to account for the delay from infection to case reporting [43].Sixth, we dropped the data for RI that were reported from 1 March through 24 March 2020.However, we chose to do so to maintain the consistency of our data given that we chose to reconstruct the state-level data from county-level data.Seventh, we used bootstrapping and a 15-day sliding window to account for uncertainty in the reporting delay.However, the choice of a 15-day sliding window is arbitrary.Eighth, the multiplier for underreporting of infections as cases was assumed to be Poisson-distributed.This implied the expected value (the mean) equaled the variance.This was an arbitrary assumption.If underreporting was highly variable, a negative binomial distribution could be an alternative.

Public health implications
Heterogeneous data reporting at the county level might distort our view of the epidemic at the state level.RI as a state and each county within it reported its data heterogeneously.Counties did not provide daily case count updates; instead, data were updated weekly.If the reporting date was not uniform across counties, the raw data on the daily count of new cases at the state level could have been contributed by one county yesterday and by another county today.If we took the state-level data at face value, our understanding of the epidemic would have been confounded by the underlying issue of heterogeneous data reporting.Here, our potential contribution was the reconstruction of the state-level incident case count data after interpolating the county-level data, thus attaining a better retrospective view of the COVID-19 pandemic in RI (Figure S1).In turn, this may provide policymakers with a solid ground upon which the relative contributions of the policy mandates introduced during the pandemic could be discussed and revisited retrospectively.

Conclusions
This study analyzed the impact of public health policy and mobility change on the transmission potential of SARS-CoV-2 using incident case count data from 1 March 2020, through 16 December 2021 (by date of report), within RI and its five counties.Sustained transmission in fall 2020 (R t >1) preceded the major wave in winter 2020.A decline in R t saw the case count dropped and plateaued in Spring 2021 before dropping to a low level in Summer 2021.A surge in R t to 1.5 in Summer 2021 preceded a surge in cases in Fall 2021 and a subsequent wave in November 2021.Mask mandate (18 April 2020) was associated with a decrease in R t by 25.99% in RI.Vaccination roll-outs appeared to be associated with a decrease in R t with some exceptions; however, such associations were confounded by subsequent introduction of variants of concern, such as the Omicron variant.In the first 90 days of the pandemic in RI, positive correlations were found between changes in grocery and pharmacy, retail and recreation, transit, and workplace visits, for both R t and estimated infection count, respectively; and negative correlations were found between changes in residential area visits for both R t and estimated infection count, respectively.

Disclosure statement
ICHF has invested in equity in Alphabet, Inc. (GOOGL).ACS reports grants to her institution from Gilead Sciences.Other coauthors have no conflicts of interest to declare.

Figure 1 .
Figure 1.COVID-19 epidemic curves and R t curves of Rhode Island.Scenario: Multiplier of 4; serial interval: mean, 4.60; standard deviation, 5.55.Panel 1: Bootstrapped (n = 100) case count data (with a sliding window of 15 days); Panel 2: Estimated daily number of new infections through multipliers drawn from a Poisson distribution with a mean of 4; Panel 3: 7-day sliding window R t using the estimated daily number of new infections as inputs; Panel 4: Non-overlapping window R t , representing policy change in Rhode Island using the estimated daily number of new infections as inputs.The government policies represented by the alphabets in the figure are: A: Eleventh Supplemental Emergency Declaration-Staying at Home, Reducing Gatherings, Certain Retail Business Closures and Further Quarantine Provisions, B: Mask Mandate, C: Phase 1 Re-opening and restrictions lessen, D: First Vaccination roll-out, E: Expansion of Rhode Island's Vaccination efforts [16+], F: Expansion of Vaccination Efforts to Adolescent [12+], G: Mask Mandate Terminated, H: CDC Recommends Pediatric COVID-19 Vaccine for Children [5-11 years].

Figure 2 .
Figure 2. COVID-19 epidemic curves and R t curves of Rhode Island.Scenario: Multiplier of 11; serial interval: mean, 4.60; standard deviation, 5.55.Panel 1: Bootstrapped (n = 100) case count data (with a sliding window of 15 days); Panel 2: Estimated daily number of new infections through multipliers drawn from a Poisson distribution with a mean of 11; Panel 3: 7-day sliding window R t using the estimated daily number of new infections as inputs; Panel 4: Non-overlapping window R t , representing policy change in Rhode Island using the estimated daily number of new infections as inputs.The government policies represented by the alphabets in the figure are: A: Eleventh Supplemental Emergency Declaration-Staying at Home, Reducing Gatherings, Certain Retail Business Closures and Further Quarantine Provisions, B: Mask Mandate, C: Phase 1 Re-opening and restrictions lessen, D: First Vaccination roll-out, E: Expansion of Rhode Island's Vaccination efforts [16+], F: Expansion of Vaccination Efforts to Adolescent [12+], G: Mask Mandate Terminated, H: CDC Recommends Pediatric COVID-19 Vaccine for Children [5-11 years].

Figure 3 .
Figure 3. COVID-19 epidemic curves and R t curves of Rhode Island.Scenario: Multiplier of 4; serial interval: mean, 4.55; standard deviation, Panel 1: Bootstrapped (n = 100) case count data (with a sliding window of 15 days); Panel 2: Estimated daily number of new infections through multipliers drawn from a Poisson distribution with a mean of 4; Panel 3: 7-day sliding window R t using the estimated daily number of new infections as inputs; Panel 4: Non-overlapping window R t , representing policy change in Rhode Island using the estimated daily number of new infections as inputs.The government policies represented by the alphabets in the figure are: A: Eleventh Supplemental Emergency Declaration-Staying at Home, Reducing Gatherings, Certain Retail Business Closures and Further Quarantine Provisions, B: Mask Mandate, C: Phase 1 Re-opening and restrictions lessen, D: First Vaccination roll-out, E: Expansion of Rhode Island's Vaccination efforts [16+], F: Expansion of Vaccination Efforts to Adolescent [12+], G: Mask Mandate Terminated, H: CDC Recommends Pediatric COVID-19 Vaccine for Children [5-11 years].

Figure 4 .
Figure 4. COVID-19 epidemic curves and R t curves of Rhode Island.Scenario: Multiplier of 11; serial interval: mean, 4.55; standard deviation, 3.30.Panel 1: Bootstrapped (n = 100) case data (with a sliding window of 15 days); Panel 2: Estimated daily number of new infections through multipliers drawn from a Poisson distribution with a mean of 11; Panel 3: 7-day sliding window R t using the estimated daily number of new infections as inputs; Panel 4: Non-overlapping window R t , representing policy change in Rhode Island using the estimated daily number of new infections as inputs.The government policies represented by the alphabets in the figure are: A: Eleventh Supplemental Emergency Declaration-Staying at Home, Reducing Gatherings, Certain Retail Business Closures and Further Quarantine Provisions, B: Mask Mandate, C: Phase 1 Re-opening and restrictions lessen, D: First Vaccination roll-out, E: Expansion of Rhode Island's Vaccination efforts [16+], F: Expansion of Vaccination Efforts to Adolescent [12+], G: Mask Mandate Terminated, H: CDC Recommends Pediatric COVID-19 Vaccine for Children [5-11 years].

Figure 5 .
Figure 5. Median percentage change (95% credible intervals, CrI) of policy change R t estimates for Rhode Island before and after social and non-pharmaceutical public health interventions.

Figure 6 .
Figure 6.Median percentage change (95% credible intervals, CrI) of policy change R t estimates for Rhode Island before and after the dates of various phases of vaccination roll-out representing different phases of the pandemic from December 2020 to November 2021.The observed increase in R t after November 3, 2021 happened to coincide with the Omicron wave.

Figure 7 .
Figure 7. Time-lag correlation coefficients between relative mobility (percentage change from baseline) for trips to grocery and pharmacy, parks, residential, retail and recreation, transit, workplace categories respectively and the median 7-day sliding window R t of SARS-CoV-2 (upper panel) and daily number of new SARS-CoV-2 cases (lower panel) over the first 90 days (i.e. between March 16, 2020, through May 16, 2020) of the pandemic in Rhode Island.The estimated number of infections was estimated with a Poisson-distributed multiplier of 4 and R t estimated with a serial interval with a mean of 4.60 and a standard deviation of 5.55.For examining the null hypothesis of zero cross-correlation, the gray lines represent the lower and upper limit of the confidence bands based on the standard test statistics; the blue dashed lines represent the lower and upper limit of the confidence bands based on the robust test statistics.