key: cord-0699361-hdbhzk39 authors: Cevallos-Valdiviezo, Holger; Vergara-Montesdeoca, Allan; Zambrano-Zambrano, Gema title: Measuring the impact of the COVID-19 outbreak in Ecuador using preliminary estimates of excess mortality, March 17–October 22, 2020 date: 2020-12-19 journal: Int J Infect Dis DOI: 10.1016/j.ijid.2020.12.045 sha: c679156e5a16420149cbcfc1f316fffc955af432 doc_id: 699361 cord_uid: hdbhzk39 OBJECTIVES: Ecuador is among the worst-hit countries in the world by the COVID-19 pandemic. In terms of confirmed deaths per million inhabitants, as of October 22, Ecuador ranks fourth in the Americas and ninth worldwide according to data from the World Health Organization. In this report we estimate excess deaths due to any cause in Ecuador since the start of the lockdown measures on March 17, 2020 until October 22, 2020. METHODS: Estimates of excess deaths were calculated as the difference between the number of observed deaths from all causes and estimates of expected deaths from all causes. Expected deaths were estimated for the period March 17 to October 22, 2020 from forecasts of an ARIMA model of order (3,0,1) with drift which was applied to daily mortality data for the period January 1, 2014 to March 16, 2020. RESULTS: The number of all-cause excess deaths in Ecuador was estimated to be 36,922 (95% bootstrap confidence interval: 32,314–42,696) during the study period. The peak in all-cause excess mortality in Ecuador may have occurred on April 4, 2020, with 909 excess deaths. CONCLUSIONS: Our results suggest that the real impact of the pandemic in Ecuador was much worse than indicated by reports from national institutions. Estimates of excess mortality might provide a better approximation of the true COVID-19 death toll. These estimates might capture not only deaths directly attributable to the COVID-19 pandemic but also deaths from other diseases that resulted from indirect effects of the pandemic. capture not only deaths directly attributable to the COVID-19 pandemic but also deaths from other diseases that resulted from indirect effects of the pandemic. (Olson et al., 2020) . For this analysis, daily mortality data from the National Institute of Statistics and Census (Instituto Nacional de Estadística y Censos, 2020), the National Civil Registry (Dirección General del Registro Civil, Identificación y Cedulación, 2020) and the National Risk and Emergency Management Service (Servicio Nacional de Gestión de Riesgos y Emergencias, 2020) of Ecuador were used. All-cause excess deaths for any day were estimated as the difference between the number of observed deaths from all causes and the estimate of expected deaths from all causes obtained for that day from historical data. To obtain estimates of expected deaths from all causes for the period of March 17 to October 22, 2020 we used forecasts from a statistical model built on daily mortality data from the period January 1, 2014 to March 16, 2020. This is a similar approach to that used by Blanton et al. (2007) and Olson et al. (2020) . Different models of exponential smoothing, cubic smoothing splines, simple moving average and ARIMA were considered for this study. These methods have been used in similar studies (see e.g., Adair et al., 2020; Rossen et al., 2020; Vieira et al., 2020) . To compare the forecast performance of these models we used out-of-sample (OOS) testing with rollingwindows (Tashman, 2000) on the model-building dataset. In particular, the fit period was successively set to start on March 17 and end on March 16 of the following year to produce forecasts for the subsequent test period of March 17 to October 22. Because mortality data were available from January 1, 2014 onwards, the first fit period was set from January 1, 2014 to March 16, 2015. Five fit-test periods were generated in total in the model-building dataset. Note that this evaluation procedure emulates the forecast problem of this study. We used as measure of performance the root mean squared error J o u r n a l P r e -p r o o f (RMSE) computed from the observed and forecasted values in the five testing periods considered. For each method, only the model with the lowest RMSE estimate in the OOS evaluation is displayed in Table 1 . As shown in Table 1 , the ARIMA model of order (3,0,1) with drift shows the best forecast performance and was therefore selected for the analysis. Figure 1 shows the estimated curve of excess mortality based on this model and compares it with the daily number of confirmed and probable COVID-19 deaths in the period January 1 to October 22, 2020. Note that the sequence of bars for the daily numbers of confirmed and probable COVID-19 deaths shows a pattern similar to the curve of excess mortality, except for being shifted to the right, which may reflect reporting delays by national institutions during the pandemic. Note also that the levels of estimated excess mortality in general exceed those of the number of confirmed and probable COVID-19 deaths. This difference may represent deaths directly attributable to COVID-19 that were not officially reported but also deaths from other diseases that resulted from indirect effects of the pandemic, such as the fear of being treated in hospitals overwhelmed with COVID-19 patients or a shortage of medication, physicians, intensive care unit beds, ventilators or a general lack of access to healthcare services. These results suggest that the real impact of the pandemic in Ecuador was much worse than indicated by reports from national institutions. Estimates of all-cause excess mortality might provide a better approximation of the true COVID-19 death toll. From March 17 to October 22, 2020, a total of 80,108 deaths from all causes were registered in Ecuador, of which 36,922 (95% bootstrap confidence interval: 32,314-42,696) were estimated to be in excess of expected levels. As shown in Figure 1 -Levels of all-cause mortality from December 24, 2019 to December 31, 2019 were lower compared to all-cause deaths registered on previous days. Deaths not registered on these vacation days were probably only registered on January 1, 2020, and as result excess mortality was estimated to be much higher than 0 for that day. Levels of deaths in the period March 13, 2020 to March 19, 2020 were also unusually low and as a result negative estimates of excess deaths were obtained in a systematic way for those days. These results were obtained in a first analysis with the ARIMA (3,0,1) model with drift, which showed the best performance with the OOS J o u r n a l P r e -p r o o f evaluation. To improve the fit of the model, we fitted a simple exponential smoothing model with multiplicative errors (ETS(M,N,N) ) on the time series data preceding each of these periods and used the corresponding forecasts to replace the actual values of these days. ETS(M,N,N) produced the lowest RMSE in a last block evaluation for each of these 2 periods when compared to other models of exponential smoothing, cubic smoothing splines, simple moving average and ARIMA. Reported results in Figure 1 and Table 1 are based on these modified data. -Data on confirmed and probable COVID-19 deaths from the National Risk and Emergency Management Service of Ecuador were negative for some days. We set those values to 0 in the plot shown in Figure 1 . -A number of 362 probable COVID-19 deaths on October 8, 2020 appears to be an outlier with respect to levels of COVID-19 deaths around that period of time. Many of them may represent deaths that possibly occurred at an earlier point in time and that were only registered on October 8. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. None. Bloomberg Philanthropies Data for Health Initiative, Civil Registration and Vital Statistics Improvement Update: Influenza activity--United States and worldwide, 2006-07 season, and composition of the 2007-08 influenza vaccine Nacimientos y Defunciones -Información Histórica Preliminary Estimate of Excess Mortality During the COVID-19 Outbreak -New York City Excess Deaths Associated with COVID-19, by Age and Race and Ethnicity -United States Informes de Situación e Infografías -COVID 19 -desde el 29 de Febrero Out-of-sample tests of forecasting accuracy: an analysis and review Rapid Estimation of Excess Mortality during the COVID-19 Pandemic in Portugal -Beyond Reported Deaths World Health Organization, WHO Director-General's opening remarks at the media briefing on COVID-19 -11 World Health Organization, WHO Coronavirus Disease (COVID-19) Dashboard We declare no conflicts of interest. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. 14 12 0 15-04-2020 455 19 146 0 16-04-2020 398 15 50 0 17-04-2020 375 18 43 0 18-04-2020 283 35 56 0 19-04-2020 365 18 86 0 05-2020 189 49 292 0 07-05-2020 142 36 77 0 08-05-2020 135 50 72 0 09-05-2020 144 13 135 0 10-05-2020 131 410 0 0 11-05-2020 162 18 0 0 12-05-2020 140 182 58 0 13-05-2020 147 7 0 0 14-05-2020 68 4 17 0 15-05-2020 119 256 52 0 16-05-2020 109 94 19 0 17-05-2020 108 48 22 0 18-05-2020 95 63 13 0 19-05-2020 113 40 25 0 20-05-2020 105 49 113 0 21-05-2020 105 51 75 0 22-05-2020 96 117 12 0 23-05-2020 85 40 94 0 24-05-2020 93 12 0 0 Table 1 . RMSE estimates of the best models for each of the methods considered in this study based on the OOS evaluation.