key: cord-1010912-lagx4fza authors: Braithwaite, Jeffrey; Tran, Yvonne; Ellis, Louise A; Westbrook, Johanna title: The 40 health systems, COVID-19 (40HS, C-19) study date: 2020-09-30 journal: Int J Qual Health Care DOI: 10.1093/intqhc/mzaa113 sha: a260a9217895d9ff5d7b34d2f37338e6d2c6468e doc_id: 1010912 cord_uid: lagx4fza BACKGROUND: The health, social and economic consequences of the COVID-19 pandemic have loomed large as every national government made decisions about how to respond. The 40 Health Systems, Covid-19 (40HS.C-19) Study aimed to investigate relationships between governments’ capacity to respond (CTR), their response stringency, scope of COVID-19 testing, and COVID-19 outcomes. METHODS: Data to April 2020 were extracted for 40 national health systems on pre-pandemic government capacity to respond (CTR) (Global Competitiveness Index), stringency measures (Oxford COVID-19 Government Response Tracker Stringency Index), approach to COVID-19 testing and COVID-19 cases and deaths (Our-World-in-Data). Multidimensional scaling (MDS) and cluster analysis were applied to examine latent dimensions and visualise country similarities and dissimilarities. Outcomes were tested using multivariate and one-way analyses of variances and Kruskal-Wallis H tests. RESULTS: The MDS model found three dimensions explaining 91% of the variance and cluster analysis identified five national groupings. There was no association between national governments’ pre-pandemic CTR and the adoption of early stringent public health measures or approach to COVID-19 testing. Two national clusters applied early stringency measures and reported significantly lower cumulative deaths. The best performing national cluster (comprising Australia, South Korea, Iceland and Taiwan) adopted relatively early stringency measures but broader testing earlier than others which was associated with a change in disease trajectory and the lowest COVID-19 death rates. Two clusters (one with high CTR and one low) both adopted late stringency measures and narrow testing and performed least well in COVID-19 outcomes. CONCLUSION: Early stringency measures and intrinsic national capacities to deal with a pandemic are insufficient. Extended stringency measures, important in the short-term, are not economically sustainable. Broad-based testing is key to managing COVID-19. https://ourworldindata.org/coronavirus-data [7] Measures: National governments' inherent, pre-pandemic CTR was measured using the Global Competitiveness Index (GCI) 4.0 [4] which assesses factors that drive a state's or territory's productivity, growth, and human development including its resilience, agility, innovation and human-centric approach. The data in this index show that enhanced competitiveness is related to high living standards, income and life satisfaction. We included each jurisdiction's GCI index, as well as their scores for 'policy stability', 'responsiveness to change', and 'government long-term vision'. National pandemic response was measured in terms of early stringency and approach to testing. Index. [1, 5] Until April 29, 2020 the Stringency Index was developed from seven policy decisions relating to 'containment and closure', such as the closing of schools and workplaces, cancellation of public events, and restrictions on internal and international movement. Latvia and Lithuania were not included in OxCGRT-SI so we applied the same methods to estimate their stringency scores. We used the OxCGRT-SI at three time points (at the time of the 1 st , 5 th and 10 th recorded COVID-19 attributed to deaths in each country). Approach to COVID-19 testing was assessed in terms of the adoption of broad or narrow testing criteria, as measured by the index number of tests per case identified from April 13, 2020. Broader testing was indicated by a larger index number, meaning many tests were performed before identifying a COVID-19 case. The April 13 date was when testing data were available from all the states in the sample. The consequences of each national response were measured by COVID-19 deaths per million population; COVID-19 new daily deaths per million population; COVID-19 total cases per million population; and COVID-19 new daily cases per million population. Analysis: Statistical analyses were conducted using SPSS version 25, [8] multidimensional scaling (MDS) analysis, and k-means clustering were conducted using functions from R version 3.5.2. [9, 10] To characterise and group the 40 states based on their national CTR and their actual response to the pandemic, eight variables (four GCI measures, three OxCGRT-SI scores and the tests per COVID-19 case identified) were included in an MDS model. MDS was used to examine the underlying latent dimensions from multivariate data and to visualise the level of similarity or dissimilarity of the nations to each other based on these dimensions. Since the eight variables are not exact metrics, all variables were first standardised by transforming them to z-scores. As MDS examines similarity indices, the variables were converted into a Euclidean distance matrix and the MDS model was visualised on a 3-dimensional plot. Cluster analysis was then conducted on the MDS result using k-means, to determine the clustering structure. Nations within the same cluster are more similar to each other than those in other clusters. Statistical comparisons between cluster groups in terms of government CTR and response variables were conducted using one-way multivariate analysis of variances (MANOVA). Univariate tests were conducted with one-way ANOVAs. For outcome measures consisting of COVID-19 total deaths, new daily deaths, total cases and new daily cases, data were non-parametric. Thus, to test for differences between the clusters and outcome measures, Kruskal-Wallis H tests were conducted for the average of the last five days in March (day 27-day 31) and the last five days in April (day 57-day 61). For all tests, a significance level of .05 was used. Table 1 reports the descriptive statistics for each of the five national clusters for the measures investigated. *p-values are from one-way ANOVAs to examine differences between the 5 cluster groups. Using goodness-of-fit statistics, the number of dimensions with best fit for the MDS model was found at three dimensions, with 91% variance explained. From the MDS plots, we identified each of the dimensions as: (1) government CTR; (2) the level of early stringency response; and (3) approach to testing. Cluster analysis identified five national clusters from the MDS plots as shown in Figure 1 . Figures 1a to 1c each illustrate cluster performance on the three dimensions. Figure 1d is the threedimensional representation of the national health systems in the MDS model. Figure 1a shows three clusters (above the horizonal zero axis) with a higher government CTR compared to two clusters performing with lower government CTR (below the horizontal zero axis). The two clusters (orange and red) on the right of the vertical zero axis applied early stringency measures relative to those on the left. Figure 1b shows the middle green cluster from Figure 1a adopted a much broader testing strategy than all the remaining clusters. There was little relationship between government CTR, adoption of early stringency measures or testing approach (Figures 1a and 1c) . Figure 1a shows that clusters with high CTR and clusters with low CTR both demonstrated early high and low stringency adoption. Figure 1b shows similar testing approaches for 4 of the 5 clusters despite differences in stringency response and Figure 1c shows similar testing approach despite differences in government CTR. There were no clear linear relationships between these factors. Figure 1d b. c. d. A one-way MANOVA found an overall significant difference in the five cluster groups with CTR and response measures, (F(32, 104.9)= 9.4, p<0.001; Wilk's λ= 0.007, partial η 2 = 0.71). The univariate ANOVAs showed that the differences were in all the eight variables (p<0.001) between the five clusters. and Cluster 5 (blue) were both characterised as low early stringency adopters with narrow scope of testing. However, Cluster 1 had low whereas Cluster 5 had high government CTR scores. Cluster 2 (red), like Cluster 1, was below the horizontal zero axis (in Figure 1a ) with low government CTR scores, but it adopted early stringency measures and had medium testing levels. Cluster 3 (green) and Cluster 4 (orange) were similar in their government CTR scores, but Cluster 3 had broader testing, and Cluster 4 had adopted early high stringency measures. For cumulative COVID-19 deaths and daily new COVID-19 deaths, applying Kruskal-Wallis H tests, we found significant differences in daily new deaths between the five clusters (χ 2 =12.0, p=0.02) but not for cumulative deaths (χ 2 = 8.8, p= 0.06) for the last five days in March. By the end of April, there were significant differences between the five clusters for both cumulative COVID-19 deaths (χ 2 =12.0, p=0.02) and daily new COVID-19 deaths (χ 2 =12.0, p=0.02). Figure 4b shows cluster 3 (green) and cluster 4 (orange) substantially lower on new COVID-19 death rates compared to the other clusters. Variation in Government Responses to COVID-19 Blavatnik School of Government Working Paper August, date last accessed) World Health Organisation. WHO Director-General's opening remarks at the media briefing on COVID-19 Responding to global systemic shocks: applying lessons from previous crises to Covid-19 Oxford COVID-19 Government Response Tracker Stringency Index (OxCGRT-SI) Coronavirus Pandemic (COVID-19) -the data IBM SPSS Statistics for Windows RCore Team. R: A language and environment for statistical computing. 2013. R Foundation for Statistical Computing RStudio: Integrated development for R August, date last accessed) How should policy responses to the COVID-19 pandemic differ in the developing world? Democracy, capacity, and coercion in pandemic response -COVID-19 in comparative political perspective. J Health Polit Policy Law 2020 Policy response, social media and science journalism for the sustainability of the public health system amid the COVID-19 outbreak: The Vietnam lessons The Lancet Infectious Diseases. Political casualties of the COVID-19 pandemic Comparison of estimated rates of coronavirus disease 2019 (COVID-19) in border counties in Iowa without a stay-at-home order and border counties in Illinois with a stayat-home order Oxford COVID-19 Government Response Tracker Blavatnik School of Government A conceptual framework for analyzing the economic impact of COVID-19 and its policy implications. UNDP LAC COVID-19 Policy Documents Series Monitoring the level of government trust, risk perception and intention of the general public to adopt protective measures during the influenza A (H1N1) pandemic in the Netherlands Exploring communication, trust in government, and vaccination intention later in the 2009 H1N1 pandemic: results of a national survey Polarization and public health: Partisan differences in social distancing during the Coronavirus pandemic. NBER Working Paper Are high-performing health systems resilient against the COVID-19 epidemic?