key: cord-0897678-bokvvu6t
authors: Quijano-Angarita, A.; Espinosa, O.; Mercado-Reyes, M. M.; Walteros, D.; Malo, D. C.
title: Analysis of the interventions adopted due to the COVID-19 on ARI morbility for Colombia
date: 2020-09-14
journal: nan
DOI: 10.1101/2020.09.12.20193334
sha: 98f1b22b5ad5ef4233cda316caf700f581a1e8e5
doc_id: 897678
cord_uid: bokvvu6t

Acute Respiratory Infections are among the leading causes of death globally, particularly in developing countries, and are highly correlated with the quality of health and surveillance systems and effective early interventions in high-risk age groups. According to the World Health Organization, about four million people die each year from mostly preventable respiratory tract infections, making it a public health concern. The official declaration of a pandemic in March 2020 due to the Sars-CoV-2 virus coincided with the influenza season in Colombia and with environmental alerts about low air quality that increase its incidence. The objective of this document is the application of a flexible model for the identification of the pattern and monitoring of ARI morbility for Colombia by age group that shows atypical patterns in the reported series for 5 departments and that coincide with the decisions implemented to contain the COVID-19

Acute Respiratory Infection (ARI), defined as a set of infections caused by viral and bacterial 26 microorganisms to the respiratory system of high dissemination, presents the highest morbidity in 27 the world and is among the leading causes of medical care and death in children under 5 years in 28 Colombia, with an accentuated pattern in certain seasons, in the form of epidemic outbreaks, which 29 vary according to the climatic and epidemiological characteristics of the regions. Cases of acute 30 respiratory infection include clinical diagnoses that involve everything from common colds to 31 more serious cases that require, as in the case of pneumonia, specialized hospital care that includes 32 managing patients in hospitalization or intensive care areas because they are a threat to life. 33

According to the WHO (1), lower respiratory tract infections cause more than four million deaths 34

per year at high rates in low-and middle-income countries and in children under 5 years of age. In 35 fact, in 2015 for the 0-5 age group, respiratory infections caused 920.136 deaths, representing 15% 36 of deaths among all causes worldwide. In Colombia, according to data calculated from vital 37 statistics report (Table 1 from National Statistics Department of Colombia -DANE data), in 2017 38 and 2018 deaths from diseases in the respiratory system in children under five years of age 39 represented 8.3% and 8.5% respectively of the total number of all causes of death in this age group, 40 where between 5.6% and 5.8% of deaths were caused by influenza. As the cases due to ARI 41 represent a high risk of death and, in addition, a high demand for medical assistance, they represent 42 an event of interest in public health also because attendance at health services leads to treatment 43 failure and high risk of death. 44

The WHO declaration of a global pandemic (March 11, 2020) due to the novel coronavirus 45 COVID-19 and the appearance of the first case imported into Colombia (March 6) from Italy of 46 this easily transmitted infection due to the Sars-CoV-2 virus, coincides with the beginning of the 47 influenza season in the country, the air pollution alerts in the main cities of the country such as 48

Bogotá and Medellín and the humanitarian crisis due to the massive migration of Venezuelan 49 citizens that, is expected to have an impact on the number of cases reported of Acute Respiratory 50 Infection, and therefore, that there is concern about the hospital capacity and the pressure this could 51 cause in the Colombian health system (leading to a collapse in the hospital network causing an 52 increase in deaths from preventable causes). On the contrary, that because of the epidemic and the 53 measures taken the peaks of IRA morbility fall and capacity can be released. 54

The lockdown decreed by the Colombian government on March 24, and extended until April 27 55 throughout the country, was defined after the measure to close schools and universities and cancel 56 massive events (March 11 and 12, 2020) and was intended to reduce the speed of contagion from 57 Sars-CoV-2 and avoid a possible collapse of health systems that in the Latin American region are 58 quite fragile (2) . Therefore, the assessment and monitoring of health events, especially those 59 related to respiratory diseases, are important, especially for different age groups and for outpatient 60 and serious cases involving hospitalization. 61 Therefore, this paper first presents an analysis of the number of cases adjusted by the population 62 and a Bayesian model of binomial negative response, to determine the departmental pattern of ARI 63 morbility in outpatient (including medical emergencies) and inpatient (including medical intensive 64 care) by age groups (0-4, 5-19 , 20-59, 60+ years) to enable the evaluation of interventions 65 implemented at the national level to contain COVID-19 and decision-making in prevention and 66 control. 67 68 69 70 . CC-BY-NC 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 14, 2020. . https://doi.org/10.1101/2020.09.12.20193334 doi: medRxiv preprint

Data 72

For the analysis, we used the database of Sivigila (Sistema Nacional de Vigilancia en Salud 73 Pública) of event 995 -Morbility due to ARI of the Instituto Nacional de Salud -INS (National 74

Institute of Health), which contained information on counts of Acute Respiratory Infection 75 outpatient and emergency, hospitalization and critical care by departments and large cities and for 76 the 52 epidemiological weeks of the year, in addition to information on other causes of disease 77 outpatient, emergency and hospitalization (including critical care). The database was processed 78 from the first week of 2017 to week 13 of the year 2020 for risk calculation at the departmental 79 level and for Bayesian model (time interval with the best quality records). In addition, the 80 estimated and projected population presented by the National Administrative Department of 81 Statistics (DANE) of Colombia was obtained to calculate the population at risk by using an 82 exponential interpolation per weeks, using official population data between 2018 and 2021. The 83 data was processed by department, merging large cities data with the department to which the 84 belong to produce an aggregate analysis. 85

First, for an exploratory analysis the risk is calculated at a departmental level per week, defined 87 as, 88

(1) 89 where risk (1) is expressed as the quotient between the new ARI cases ( !,# $,% ) per week t and 90 department k observed for outpatients (O) and inpatients (I) and the population at risk ( !,# ), 91

. CC-BY-NC 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 14, 2020. . https://doi.org/10. 1101 /2020 which was obtained at a weekly level by using exponential interpolation defined as: & exp( ) 92 where r is the weekly growth rate, and & is the initial population. 93

Second, since we have cases of new reported cases of ARI morbility, we use a negative binomial 94 model as simple as possible that considers the variability and dependence between epidemic weeks 95 and one that allows for dealing with the apparent over-dispersion present in the counts of some 96 departments of Colombia due to the quality of the reported data. Let ' ~( ' , ) be the 97 number of cases of ARI morbility by weekly (from week 1 of 2017 to week 13 of 2020), age group 98 and for inpatients and outpatients, with = 1, . . . ,169, we use the parameterization presented by 99

Ntzoufras3 whose density function is defined as, 100

. Then, we model the interest variable 102

' from a Bayesian model defining parameters from (2) as follows, 103

with, 105 ' = / ! , 106 and the predictor lineal ' , considering year and week variables as covariates, 107

(3) 108

In (3) we considered an intercept that varies for each epidemiological week (1-52) and considers 109 the dependence between them and a linear combination of dummy variables for each year (2017-110 2020) with, 111

,+

. CC-BY-NC 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 model was fitted using JAGS, simulating 3 chains of 30,000 draws each with a burning period 117 of 15,000. To evaluate convergence, the Gelman-Rubin convergence diagnostics is followed (4) 118 and is considered a threshold of 1.1 as suggested by Gelman and Rubin (5) . 119

The estimation accuracy is measured by the symmetric Mean Absolute Percentage Error (sMAPE), 120 which has the advantage of avoiding the asymmetry of MAPE and is less sensitive to outliers and 121 satisfy 0 < sMAPE < 1 (by dividing by 2 the original sMAPE which considers an arithmetic 122 mean in the denominator). Thus, the sMAPE used is defined as follows 1 , 123

.

(4) 124 10, when the first case was reported. The trend observed is an exaggerated increase between week 131 10 and 11 of outpatient cases in the 20-59 age groups for the 5 departments, cases that decrease by 132 1 Makridakis (6) shows the benefits of this index, which is one of the most used in the international literature on forecast calculations. 2 When descriptively analyzing the risk of the 33 departments of Colombia, these 5 were the ones that exposed the temporal trends of greatest alarm (abrupt drop in health care).

. CC-BY-NC 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 copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101/2020.09.12.20193334 doi: medRxiv preprint

week 11 when it is decided to close schools, universities and ban massive events in the country. 133 A striking behavior in the series is that in the department of Valle del Cauca for the 5 to 19 year 134 age group where the number of cases of patients in hospitalization increased with respect to 135 previous years. Between week 11 and 13, when the lockdown is decreed and children over 70 are 136 prohibited from commute, the series falls below the level of previous years, an effect that is 137 perceived in all departments. By week 13 in the department of Huila, there is an increase in the 138 number of cases of hospitalization in the 5-19 age group, and their evolution must be monitored. 139

The adjusted Bayesian model presents a good quality of the estimates obtained (see table 2 ). From 140 the analysis of the symmetric Mean Absolute Percentage Error (sMAPE), it is found that the age 141 groups of higher risk, under 5 years and over 60 years show a percentage deviation of less than 142 10%, except for the model for Huila of hospital patients that rises to 26%. In general, the models 143 perform well, and the best fit is for inpatient and outpatient cases for children under five which is 144 a risk group of interest. 145

The outcome of the Bayesian negative binomial response model in 2020 for the mentioned 146 departments is shown in Figures 6 to 10 , in which the 95% credibility interval for the follow-up of 147 these ARI cases that seem striking in the descriptive analysis is presented as red dots in the graphs. 148

In fact, by weeks 11 and 12 in the 5 departments in the 20-59 age group, the observed values of 149 outpatients were outside the range of values expected by the warning at week 10 of the first case 150 of COVID-19 in Colombia, which could have generated a generalized panic that led people to seek 151 medical attention for related symptoms. However, this effect diminished when mobility 152 restrictions were enacted as outpatient cases dropped for all departments in almost all age groups. 153

In fact, by week 13 (lockdown) the number of outpatient cases and hospitalizations of children 154 under five years old for Bogotá and Valle is lower than expected. For Huila, there is an increasing 155 . CC-BY-NC 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 14, 2020. . https://doi.org/10.1101/2020.09.12.20193334 doi: medRxiv preprint trend of hospitalization cases for the 5-19 year old age group and they are shown as outliers in 156 Figure 9 But in general, there is a drop in outpatient consultations by week 13. 157 158

Although the descriptive analysis is adequate to find atypical patterns in the reported cases of 160 Acute Respiratory Infection in Colombia, the Bayesian model presented allows for the timely 161 detection of those cases that present unexpected behavior since they fall outside the 95% credibility 162 interval estimated from information from epidemiological weeks of previous years, which allows 163 for the evaluation of causes and decision making. Considering the negative binomial distribution 164 that deals with the problem of overdispersion and that is used for discrete variables that represent 165 counts makes the inference reliable and that in places with deficit in the report, it is possible to 166 have a tool to evaluate atypical patterns in the series, without considering the assumption of normal 167 distribution that is used in some surveillance methods and that can lead to biases. As a method of 168 surveillance during the pandemic, the result presents sufficient elements for the evaluation of those 169 rare cases that require assessment and allow intervention and control in health systems. 

193 . CC-BY-NC 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 14, 2020 . . https://doi.org/10.1101 /2020 CC-BY-NC 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 14, 2020. . https://doi.org/10.1101/2020.09.12.20193334 doi: medRxiv preprint 

. CC-BY-NC 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-NC 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 14, 2020. . https://doi.org/10.1101/2020.09.12.20193334 doi: medRxiv preprint CC-BY-NC 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 14, 2020. . https://doi.org/10. 1101 /2020 

The Global Impact of Respiratory Disease 172 -second edition. Sheffield; 2017. 173 2. The Lancet. The unfolding migrant crisis in Latin America

Bayesian modeling using WinBUGS

Inference from iterative simulation using multiple sequences

Bayesian data analysis

Accuracy measures: theoretical and practical concerns