key: cord-1040868-s6229sbc authors: Guo, S.; An, R.; McBride, T. D.; Yu, D.; Fu, L.; Yang, Y. title: Social Distancing Interventions in the United States: An Exploratory Investigation of Determinants and Impacts date: 2020-05-30 journal: nan DOI: 10.1101/2020.05.29.20117259 sha: 458924a78c9df5c48223e12a69a73258128c3f99 doc_id: 1040868 cord_uid: s6229sbc Background To combat the Covid-19 pandemic in the United States, many states and Washington DC enacted Stay-at-Home order and nonpharmaceutical mitigation interventions. This study examined the determinants of the timing to implement an intervention. Through an impact analysis, the study explored the effects of the interventions and the potential risks of removing them under the context of reopening the economy. Method A content analysis identified nine types of mitigation interventions and the timing at which states enacted these strategies. A proportional hazard model, a multiple-event survival model, and a random-effect spatial error panel model in conjunction with a robust method analyzing zero-inflated and skewed outcomes were employed in the data analysis. Findings To our knowledge, we provided in this article the first explicit analysis of the timing, determinants, and impacts of mitigation interventions for all states and Washington DC in the United States during the first five weeks of the pandemic. Unlike other studies that evaluate the Stay-at-Home order by using simulated data, the current study employed the real data of various case counts of Covid-19. The study obtained two meritorious findings: (1) states with a higher prevalence of Covid-19 cases per 10,000 population reacted more slowly to the outbreak, suggesting that some states may have missed the optimal timing to prevent the wide spread of the Covid-19 disease; and (2) of nine mitigation measures, three (non-essential business closure, large-gathering bans, and restaurant/bar limitations) showed positive impacts on reducing cumulative cases, new cases, and death rates across states. Interpretation The opposite direction of the prevalence rate on the timing of issuing the mitigation interventions partially explains why the Covid-19 caseload in the U.S. remains high. A swift implementation of social distancing is crucial-the key is not whether such measures should be taken but when. Because there is no preventive vaccine and because there are few potentially effective treatments, recent reductions in new cases and deaths must be due, in large part, to the social interventions delivered by states. The study suggests that the policy of reopening economy needs to be implemented carefully. determinants of the implementation timing of mitigation strategies. Why did some states in the United States respond rapidly to the pandemic while others did not? Of various determinants, we are primarily concerned with the prevalence rate of the confirmed cases vis-à-vis the time when each mitigation strategy was enacted. Delay in taking intervention actions during this period may directly affect the speed and scope of the Covid-19 spread. The second question focuses on the correlations between the timing of enacting mitigation and the serial incidence of Covid-19 cases and deaths. For this preliminary assessment of the impact of mitigation interventions, the observational window was extended to 4/15/2020. The impact study aims to address a key question: Is an initial response observed? Precisely during this period, many states had begun to reopen their economies and public life. By May 11, 32 states chose to do so. 6 In light of these decisions, the purpose of this paper is to explore the initial impact of nonpharmaceutical mitigation interventions in the event that re-opening the economy is associated with a second wave of Covid-19 disease and a second round of mitigation becomes necessary. In this study, we define the enactment of social distancing strategies as the exact hour at which a mitigation intervention was implemented, as stipulated by state government documents. The study window is defined as the period from the time when a National Emergency was declared to the time when the last enacting state issued a Stay-at-Home order. More precisely, the zero hour for this study is 12 PM March 13, U.S. Eastern Time, and the ending hour is 6 PM April 7, which is one hour after the last enacting state (i.e., South Carolina) implemented its Stay-at-Home order (5 PM April 7, U.S. Eastern Time). States declined to issue a mitigation order are considered censored at 6 PM on April 7, 2020. Within this 25.25-day or 606-hour window, we study not only whether a state enacted mitigation interventions, but also the exact times at which these interventions were implemented. The timing of the National Emergency is a natural starting point for the study because, before that date, none of the migration strategies was enacted in the United States. A special caveat should be put on the word "zero": it does not mean that the Covid-19 pandemic started at that point, because as Anderson et al. pointed out: "Most countries are likely to have spread of COVID-19, at least in the early stages, before any mitigation measures have an impact." 6 Qualitative content analysis was used to analyze large quantities of text data from governmental documents. We examined all relevant documents (a total of 1,470 executive orders) pertinent to the social distancing orders from the state-government websites for all 50 states and Washington DC. A content-coding/analysis 7 was performed by two researchers. Review and coding guidelines were developed in advance. The two reviewers separately examined all documents, and resolved discrepancies in coding by consensus in a discussion. Considering variation across states in the frequency of use of alternative social distancing strategies and prior literature on social distancing, we extracted information on 9 types of the mitigation interventions: Stay-at-Home order, strengthened Stay-at-Home order, public school closure, all school closure, large-gathering ban of more than 10 people, any gathering ban, restaurant/bar limit to dining out only, non-essential business closure, and mandatory self-quarantine of travelers. The exact timing to implement each of these 9 nonpharmaceutical interventions (i.e., the hour when a specific strategy was implemented) was recorded. A state that did not enact a particular intervention by the end of the study window (i.e., 6 PM April 7) was considered censored for that strategy and was coded as having the longest time of 606 hours on length of time it took to enact the measure. Censoring under the current setting refers to the fact that the event of interest did not occur by the end of the study window but might occur in future. The theoretical model guiding the selection of potential determinants for inclusion in our models of social distancing orders is the "social determinants of health" (SDOH) framework. 8 Originally developed by WHO, the framework emphasizes the strong relationship between societal factors and health outcomes, especially crucial activities needed to close gaps and inequalities in health care concerning socioeconomic status, gender, tradition/culture, and race/ethnicity. To adapt this framework for the U.S., the National Academies of Science, Engineering, and Medicine organized a special group to further articulate strategies for integrating social care into the delivery of health care to improve the nation's health. Report of the Academies was published on September 25, 2019. Immediately following its publication, a Congressional Briefing was convened to discuss the implementation of SDOH-related strategies. 9 The Academies' report underscores five key health care activities to better address the SDOH and integrate social care with health care: awareness, adjustment, assistance, alignment, and advocacy. These activities help us to identify important variables for the determinant study. 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. Under the guidance of the five key SDOH activities and based on the available census data and other published statistics, we identified the following five blocks of state-level variables for inclusion as potential determinants in modeling social distance orders: demographic characteristics, economic well-being information, public heath infrastructure, information related to politics, and international connectivity. To measure state governmental awareness of the Covid-19, we included in the study a time-varying covariate of cumulative Covid-19 cases per 10,000 population one-day before the time when a mitigation strategy was enacted. This measure is analogous to the prevalence rate of Covid-19 disease. To explore the impacts of mitigation interventions, we conducted statistical analyses that link the timing of each mitigation order to various case counts of Covid-19. This is a correlational rather than causal analysis, due to lacking an accurate or even a proxy estimate of counterfactual. A counterfactual in the current setting refers to "What would have happened had we not enacted a mitigation strategy?" The outcome data come from Johns Hopkins University Coronavirus Data Stream that combines WHO and CDC case data. 10 Based on the original input data, we employed 9 variables in the investigation: cumulative cases, cumulative deaths, new cases, new deaths, cumulative cases per 10,000 population, cumulative deaths per 10,000 population, cumulative new cases per 10,000 population, cumulative new deaths per 10,000 population, and death rate defined as the number of cumulative deaths divided by the number of cumulative cases. These measures are computed as of the time of this study and may underestimate the actual cases and deaths. The study employed daily counts on each of the 9 outcome measures from March 11 (two days before the National Emergency was declared) to April 15 (about one week after the last state enacted the Stay-at-Home order). The analysis of determinants of each mitigation strategy employed a proportional hazards model. 11, 12 The analysis of overall determinants pulled together all nine episodes of mitigation interventions and employed the WLW model. This model is designated by the first initials of the developers' last name. 13, 14 The WLW model is a robust approach developed to control for the autocorrelations of study events. Under the framework of Cox proportional hazards model, the hazard function for the i th state to enact the k th type of mitigation intervention is expressed as = ( 1 , … , )′ denotes the covariate vector for the i th state with respect to the k th type of order, 0 ( ) with ( = 1, … , ) is the unspecified baseline hazard function, K is the maximum number of order types in a state, and = ( 1 , … , )′ is a × 1 vector of unknown regression parameters. Let Tik be the time when the k th type of event occurs in the i th state, and let Cik be the corresponding censoring time. Define Xik = min (Tik, Cik), and Δik = I(Tik