Skip to main content
Top
Published in:

Open Access 05-11-2024 | Special Issue Paper

The uneven geography of the health system and its effect on the individual probability of death by COVID-19

Authors: Grace Carolina Guevara-Rosero, Víctor Hugo Hinojosa, Christian L. Vásconez

Published in: The Annals of Regional Science | Issue 4/2024

Activate our intelligent search to find suitable subject content or patents.

search-config
loading …

Abstract

The COVID-19 pandemic caused many life losses, which were not uniform across territories. Several factors can explain geographical differences, including the health and sanitary infrastructure and the economic performance. The present study seeks to determine the probability of death by COVID-19 of individuals by analyzing their individual and contextual characteristics related to their canton of residence with emphasis on the structure of the health system. Using combined data at the individual and contextual level, a logit multilevel model is estimated. The results show that cantonal differences explain 12.8–22.6% of the total variance of the individual probability to die. Cantons where people were more likely and less likely to die by stage were identified. Level 1 health care centers played an important role in reducing the probability of death by COVID-19 of individuals from the isolation stage to the vaccination stage.
Notes

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1 Introduction

The COVID-19 pandemic was an unprecedented event with tremendous impacts at the social and economic level. Many unvaluable lives including health personnel were lost during this phenomenon. According to the World Health Organization (2023), there have been 7 million deaths reported due to COVID-19 worldwide by July 26th, 2023. This quantity of deaths is unevenly distributed across the globe, with more deaths in the Americas, representing 41% of the total number of deaths (2.96 million). In terms of the population (referred in the literature as mortality: number of deaths/total population), the number of deaths per million inhabitants was 3083 in South America, 2795 in Europe, 1758 in North America, 357 in Asia and 179 in Africa. In Ecuador, the case study of this investigation, the number of COVID-19 deaths per million inhabitants was 952.22 by March 2022. Those differences indicate that the context where people are matters for their health outcomes. Several factors can explain geographical differences, from the population structure of countries to underlying conditions of the economies including the health and sanitary infrastructure and the economic performance. In fact, in Ecuador, the evolution of the pandemics was different across regions due to the differences in terms of population density and economic activities. As shown in Fig. 1, Metropolitan cities, Quito and Guayaquil, exhibit higher number of infected people compared to smaller cities. Regarding the cantonal lethality rate (number of deaths/number of infected people) (shown in Fig. 2), it varies across regions and phases, and it is not necessarily related to the number of infected people. Indeed, although Quito and Guayaquil were the epicenters of the COVID-19 pandemic in Ecuador, they do not head the lethality rate ranking. They are in the 94th and 194th positions out of 221 cantons, respectively. This could be related to a better health infrastructure in these main cities. Conversely, smaller cantons record a high lethality rate.
In this line, several studies have analyzed the effect on lethality of contextual factors. Contextual factors are elements and circumstances of the environment that influence an individual's behavior, decisions and experiences. These factors can be of various kinds and affect the way in which a person relates to the world and makes decisions. For instance, Giancotti et al. (2021) have studied the lethality rate at an aggregated level for European countries and obtained that the percentage of older people is the main predictor of the COVID-19 lethality, followed by the number of intensive care units. Khan et al. (2020) and Rađenović et al. (2022) show a positive and statistically significant correlation between health spending, either as a percentage of GDP or per capita, and the Global Health Security (GHS) index, which shows the efficiency of the health system. Investment in health systems in aspects such as care unit beds and health personnel is a key factor to reach sustainable healthcare that could ensure pandemic preparedness (Silverman et al. 2020). Another aggregate study for Chinese cities conducted by Yu et al. (2021) showed that the case fatality rate increases with confirmed cases and with the air quality index, and reduces with the number of doctors, with the percentage of humidity and with the population density. Using individual-level data, Brandén et al. (2020) studied the residential context of patients and their COVID-19 mortality, focusing on household characteristics such as generation composition in households, care homes and single or multifamily households, population density in neighborhoods and the COVID-19 incidence in boroughs.
While these studies, mostly with aggregate analysis, give important insights, the present study goes beyond aiming to determine the influence of aggregate contextual factors on the probability of death of individuals. In situations where there is a spatial disparity, the context where people live gains relevance since in places with a better/worse endowment of health and sanitary infrastructure and health personnel, people would be less/more exposed to bad outcomes regarding their own health. The emphasis is put on the availability of health system services, which are not well distributed in the country of study, Ecuador, as in other developing countries. In developed countries such as Sweden, studied by Brandén et al. (2020), less concerns on this issue are in play since the distribution of health services might be more uniform across the territory. For this reason, although Brandén et al. (2020) analyze the contextual factors in the individual mortality risk, they do not consider the availability of health services. In this sense, the present study contributes to the understanding the probability of death of individuals, considering the health service geographical disparity, which is a common characteristic across developing countries.
This analysis is relevant since the context where people live matters for their probability to get medical assistance in case of contagion and in turn, their probability of death. People living in laggard regions with many deficiencies will suffer more in a pandemic scenario. As in many countries, Ecuador registers high levels of regional disparity not only in economic terms but also in terms of health infrastructure. The health system composed by different levels (primary-level 1, secondary-level 2 and tertiary-level 3) is unevenly distributed across the territory. Level 1 health centers, intended to prevent and detect diseases early, are present in all cantons of Ecuador, though in not optimal conditions. Regarding level 2 health centers, there are, on average, 0.13 per 100 000 inhabitants. There are 67 cantons above this average and 154 below it. Level 3 health centers are present in only 5 out of 221. And only 32 out of 221 cantons account for intensive care beds. This implies that the population in the remaining 216 cantons does not have immediate access to ICUs to be treated in severe COVID-19 scenarios.
The present study seeks to determine the probability of death by COVID-19 of individuals by analyzing their individual and contextual characteristics related to their canton of residence with emphasis on the structure of the health system. Using combined data at two levels: individual and cantonal, from different information sources, a logit multilevel model is estimated. Our results show that the probability of death by COVID-19 of individuals not only depends on the individual characteristics but also on the contextual characteristics. The health infrastructure and economic features of regions play an important role in determining the probability of death by COVID-19 of individuals. The probability of dying varies according to the pandemic phase analyzed and across cantons. Cantons where people were more likely and less likely to die were identified. Our main conclusion is that the unequal geographical dispersion of COVID-19 lethality in Ecuador is closely related to the existing spatial disparities of medical care facilities, health care and demography (e.g., aging) in regions.
This article is structured as follows. Section 2 describes the probability of death in regions of Ecuador. Section 3 describes data and methodology. Section 4 discusses the results and Sect. 5 concludes.

2 Literature review

A growing literature on the causes and effects of the COVID-19 pandemic has been developed from various fields such as economic, health, psychology, among others. From the health perspective, studies have focused on clinical characteristics of patients and their effect on the evolution of the disease. Simpson-Yap et al. (2022) studied the probability of hospitalization and death, focusing mainly on clinical characteristics such as multiple sclerosis, anti-CD20 medication and disability. Mahmoudabadi, A. (2021) studied the chance of mortality in patients infected with COVID-19, also focusing on medical individual characteristics such as age, symptoms and underlying diseases in general, without distinguishing diseases. Hu et al. (2020) developed a clinical model obtaining that age, lymphocyte count, and other medical characteristics are informative for survival of patients. Other economic studies identify socioeconomic factors that influence the COVID-19 incidence and subsequent hospitalization such as income, type of employment and educational level (Upshaw et al. 2021). Literature about public health indicates that past pandemics have disproportionately affected people whose individual and contextual characteristics are unfavorable (Ahmed et al. 2020). For example, the Black Death in the fourteenth century reduced the world's population by one-third, and the highest number of deaths was observed among the poorest population (Duncan and Scott, 2005). Additionally, people of low socioeconomic status have a disproportionately higher representation in essential work settings such as agricultural establishments, farms, public transportation, factories, commerce, street vending, etc. Working in these settings represents a higher probability of spread to SARS-CoV-2 and mortality from COVID-19, due to factors such as close contact with the public and other workers, the impossibility of working from home, not having sick leave, among others (Cortés and Ponciano 2021). Then, access to private services of high quality and therefore higher cost contributes to a lower risk of progression of COVID-19 (Seligmanid et al. 2021). In this sense, the COVID-19 pandemic had a reinforcing effect on income inequalities within cities (Castells-Quintana and Royuela 2023; Reveiu and Constantin 2023).
Many of these studies, especially in health, focus on individual characteristics of patients, disregarding the contextual aspects. However, those aspects are of high importance. In fact, Abrams and Szefler (2020) show that in the late nineteenth and early twentieth century in Europe, improved housing, less overcrowding, pasteurization, among other factors that improved their living conditions in the community, contributed to a precipitous decline in tuberculosis rates (Butler-Jones and Wong 2016). In the case of the COVID-19 pandemic, an efficient health system was crucial to maintaining low mortality rates. For example, Costa Rica stood for initially having one of the lowest COVID-19 mortality rates in the Americas, which was largely attributed to its strong universal health infrastructure system, rapid response led by key national leaders and strong institutional support from public and private organizations (Lal et al. 2021). By contrast, in US counties with less primary health care, there was a 20% increase in their excess deaths relative to counties with more primary health care during the 2019–2020 period (Stokes et al. 2021).
Contextual aspects influencing in the COVID-19 pandemic that have been identified include the economic situation of regions or countries, health spending, health personnel, education, population density and housing and health infrastructure (Guevara-Rosero 2023; Seligmanid et al. 2021; Wagstaff 2002; Sassen and Kourtit 2021; Arauzo-Carord et al. 2023).
The results of Khan et al. (2020) and Rađenović et al. (2022) show a positive and statistically significant correlation between health spending, either as a percentage of GDP or per capita, and the GHS index, which shows the efficiency of the health system. Higher health spending is also associated with better prevention and better health system capacity. Silverman et al. (2020) emphasize the importance of public government commitment to systematically strengthening health system capacity and parallel progress toward universal health coverage and global health security, ensuring pandemic preparedness and the implementation of sustainable healthcare. Therefore, a key determinant influencing overall health system efficiency as well as prevention and health system capacity is the investment in the health system such as the number of hospital intensive care unit beds, number of nurses and doctors (Eissa 2020; Giancotti et al. 2021). Another aggregate study for Chinese cities conducted by Yu et al. (2021) showed that the case fatality rate increases with confirmed cases and with the air quality index, and reduces with the number of doctors, with the percentage of humidity and with the population density. Giancotti et al. (2021) have studied the lethality rate at an aggregated level for European countries and obtained that the percentage of older people is the main predictor of the COVID-19 lethality, followed by the number of intensive care units. Brandén et al. (2020) studied the residential context of patients and their COVID-19 mortality, focusing on household characteristics such as generation composition in households, care homes and single or multifamily households, population density in neighborhoods and the COVID-19 incidence in boroughs.
Most of these studies that consider contextual factors are conducted at aggregate levels. Although Simpson-Yap et al. (2022) applied a multilevel mixed-effects ordered probit model, and Brandén et al. (2020) considered residential contextual factors, they disregard the effect of the health system and sanitary infrastructure (relevant in the COVID-19 scenario) on the individual health outcome.

3 Methodology

3.1 Data

To analyze the individual probability of death by COVID-19, data on COVID-19 infections and deaths at the individual level is used. These data combine information from three sources, the National Direction of Health Intelligence, the General Coordination of Strategic Development of Health from the Ministry of Public Health of Ecuador (MSP, acronym in Spanish) and the National Emergency Operations Committee (COE, acronym in Spanish). Such database accounts for information at a daily basis of COVID-19 cases with dates of infection and dates of death if it occurred, gender, age and canton of residence of the infected people, from February 27th, 2020, to April 3rd, 2022. Regarding individual-level data, only age and sex of individuals is available in the database provided by the MSP.
In the data pre-processing phase, outlier values and inconsistent values were treated and eliminated (older than 110 years old, observations with a confirmed date of death before the registration of the infection). Both kinds of issues were only present during the first weeks of the pandemic. However, there were limitations in the collection of the information due to lack of infrastructure to register them. There might be under-registration of information since only information reported by clinics that undertook COVID-19 tests was reported. Despite this, this database is considered the most complete and reliable data, due to the fact that the organizations (MSP and COE) that collect these data were those with the assigned responsibilities of managing the COVID-19 pandemic by the national government.
For this analysis, COVID-19 deaths are those associated to confirmed and probable cases if they ended in death. According to the World Health Organization nomenclature, a probable case is when the patient meets the clinical criteria for a suspect and was in contact with a probable or confirmed case. A confirmed case is when the patient records a positive reverse transcription-polymerase chain reaction (RT-PCR) test or rapid antigen detection test for SARS-CoV-2, regardless of signs and symptoms WHO (2020). Suspected and discarded cases are not considered. The resulting database accounts for 1′046,060 observations.
The period of analysis was divided into four stages. The first stage is called "Isolation" and constitutes the period from February 29, 2020, to May 3, 2020. This stage starts with the first confirmed case of COVID-19 in Ecuador. On March 11, 2020, the WHO declared COVID-19 as a pandemic. On March 16, 2020, local authorities declared a state of emergency and quarantine for a 60-day period. During this emergency period, the government decided to relax the containment measures and to start a phase of epidemiological risk assessment on May 4, 2024.
The second stage is called “Social Distancing” stage. It goes from May 4, 2020, to February 26, 2021. It is characterized by the use of cantonal traffic lights and by the progressive reopening of commercial activities, considering the so-called biosecurity measures (as the mandatory use of masks in public spaces and social distancing). Traffic light colors of each canton was stablished depending on the local contagion level and the availability of free spaces in health centers. They were red (high risk), yellow (medium risk) and green (low risk).
The third stage is called the “vaccination” stage, which goes from February 27, 2021, to November 29, 2021. From the first day of this period, the vaccination plan promoted by the national government begun and was completed on November 29, 2021. According to government's official bulletins, about 9 million people were inoculated during this stage.
The fourth post-pandemic stage started in November 30, 2021, and ended in April 3, 2022. This period is marked by the relaxation of biosecurity measures decreed by the national government. During this stage, the use of masks in open spaces became voluntary, and social distancing was not mandatory. This would define the beginning of the progressive return to normality. The end of this period was in April 3, 2022. At this date, only one case of infection and no deaths due to COVID-19 at the national level were reported.
To introduce the case study, Fig. 3 shows the time evolution of infections in the left scale (blue line), and the time evolution of the number of deaths in the right scale (red line). Vertical yellow lines mark the four phases. The level of infections increases between the isolation phase and the social distancing phase, and at the end of the vaccination phase, a pronounced reduction of infections is observed. However, during the post-pandemic phase, a precipitous increase in infections occurred, which could be due to the Christmas and New year’s Eve holidays. The time series of deaths shows four notable peaks. The first peak occurred at the beginning of the pandemic, approximately from April 13th to April 25th, 2020. The second peak took place from July 2nd to August 5th, 2020. A third peak is observed from April 2nd to April 22nd, 2021. Finally, the fourth peak occurred from January 15th to January 30th, 2022. This behavior is coherent with COVID-19 lethality in Colombia and Peru where 4 peaks stand out, among which 2 peaks coincide during months of July/August 2020 and January/February 2022 (Global Change Data Lab 2023). The last peak on January /February 2022 coincides with other countries due to Christmas and New Year festivities in these countries. Therefore, the time series of COVID-19 infections and deaths has a seasonal component.
To determine the effect of contextual factors on the probability of death of individuals, complementary aggregated data at the cantonal and provincial level are used. Table 1 describes the variables1 and their information source. Although variables such as air pollution and walkability align with the concept of contextual variables, they were not included in the model due to lack of information at the regional level.
Table 1
Description of variables
Variable name
Description
Information source
Age
Patient's age at the time of reporting the infection
Ministry of Public Health (MSP, 2021 and COE)
Sex
Binary variable that takes the value of 1 if the person is a man and 0, otherwise
 
People with respiratory diseases per 10,000 inhab
Hospital patients treated for respiratory diseases (classification J) per 10,000 inhabitants in cantons
Statistical Register of Hospital Beds and Discharges (2019). Diseases are classified according to the International Statistical Classification of Diseases and Related Health Problems 10th Revision
People with severe renal diseases per 10,000 inhab
Hospital patients treated for kidney diseases (classification N) per 10,000 inhabitants in cantons
 
People with diabetes per 10,000 inhab
Hospital patients treated for diabetes according (classification E) per 10,000 inhabitants in cantons
 
People with severe heart disease per 10,000 inhab
Hospital patients treated for heart disease (classification I) per 10,000 inhabitants in cantons
 
People with liver diseases per 10,000 inhab
Hospital patients treated for liver diseases (classification K) per 10,000 inhabitants in cantons
 
Level 1 health centers per 10,000 inhab
Level 1 health centers per 10,000 inhabitants in a canton
GEOSALUD system, Ministry of Public Health (2020)
Level 2 health centers per 10,000 inhab
Level 2 health centers per 10,000 inhabitants in a canton, which provide services in four basic specialties: internal medicine, gynecology, general surgery, anesthesiology and hospitalization of patients referred from level 1 health centers
 
Level 3 health centers per 10,000 inhab
Level 2 health centers per 10,000 inhabitants in a canton, which provide all health care services
 
ICU beds per 10,000 inhab
Number of ICU beds in health facilities per 10,000 inhab. in each cantona
Ministry of Public Health (MSP, 2021)
Public health investment
per capita
Public investment in health per capita in thousands of dollars in a province
Annual investment plan (PAI, acronym in Spanish, 2021)
% households with access to drinking water
Percentage of households with access to potable water in a canton
Economic Environmental Information in Decentralized Autonomous Governments (2021)
Economic Environmental Information in Decentralized Autonomous Governments (2021)
% households with access to sewerage
Percentage of households with access to sewage in a canton
 
% cantonal population with complete COVID-19 vaccination schedule
Percentage of the population with complete vaccination schedule (two doses)
Ministry of Health (MPS, 2021)
% growth of daily infections
% growth of daily infections in the canton of residence in the period of study
Ministry of Public Health (MSP, 2021 and COE)
% people with adequate employment
Percentage of people with adequate employment over the total economically active population in a province
Employment, Unemployment and
Underemployment Survey (ENEMDU, 2019)
GVA per capita
Gross value added per capita in thousands of dollars in a canton
Central Bank of Ecuador (BCE, acronym in Spanish, 2020)
aThe number of doctors was also included in the model. However, it was eliminated due to multicollinearity issues

3.2 Method

To analyze the death probability of individuals, considering contextual factors and individual factors, a probit multilevel model is employed as in Mahmoudabadi (2021) and Simpson-Yap et al., (2022). The data structure is hierarchical, i.e., individuals are nested in cantons. In this case, individuals within cantons are very likely to be correlated to each other since that due to their geographic proximity, they share common regional factors such as environmental conditions and local policies. Therefore, the assumption of independence in residuals in one-level models is not fulfilled, and standard errors will be underestimated, leading to higher significance of variables. Correct standard errors are estimated using multilevel models which consider between-group-variance (University of Bristol, 2011). In this manner, the multilevel model allows to integrate individual and geographically aggregate data.
Given that our dependent variable, \({y}_{ij}\), is a binary variable that takes the value of 1 if a person died due to COVID-19 and 0, otherwise, the model in terms of the mean or the expected value of \({y}_{ij}\) is expressed as:
$$E\left( {y_{ij} {|}x_{ij} ,u_{j} } \right) = { }\beta_{0} + \beta_{1} X_{ij} + u_{j}$$
(1)
And since \(E\left( {y_{ij} {|}x_{ij} ,u_{j} } \right) = \pi_{ij} = \Pr \left( {{ }y_{ij} { } = { }1} \right){ }\), the generalized linear model with random intercept is as follows:
$$E\left( {y_{ij} {|}x_{ij} ,u_{j} } \right) = \pi_{ij} = \Pr \left( {{ }y_{ij} { } = { }1} \right){ }$$
(2)
where \({F}^{-1}\) is the inverse cumulative distribution function and residuals \({u}_{j}\) are assumed to be independent and follow a normal distribution with zero mean, \({u}_{j}\sim N\left(0,{\sigma }_{u}^{2}\right)\).
This model is estimated for 4 identified stages of the COVID-19 pandemic: isolation, social distancing, vaccination and post-pandemic. For the models for the isolation and social distancing stages, \({X}_{j}\) corresponds to the vector of all the explanatory variables at the canton level indicated in Table 1 except for the percentage of people vaccinated with a full schedule in a canton. During the vaccination and post-pandemic stages, the vector \({X}_{j}\) includes all variables in Table 1. Additionally, during all stages, the vector \({W}_{ij}\) corresponds to the individual-level variables of each patient \(i\) in canton \(j\). The vector \({u}_{0j}\) corresponds to the random effect intercept that varies according to canton \(j\).
The variance of \({y}_{ij}\) is \(var\left({y}_{ij}\right)={\sigma }_{u}^{2}+{\sigma }_{e}^{2}\), where \({\sigma }_{u}^{2}\) is the between-canton variance and \({\sigma }_{e}^{2}\) is the individual variation. To determine the amount of variation that can be attributed to differences between groups, the variation partition coefficient is used.
$${\text{VPC}} = \frac{{\sigma_{u}^{2} }}{{\sigma_{u}^{2} + \sigma_{e}^{2} }}$$
(3)
where the level 1 variance, \(\sigma_{e}^{2} = { }\frac{{\pi^{2} }}{3} = 3.29\). To determine the suitability of a multilevel model, the likelihood ratio test is performed. Table 2 shows in column (2), the ratio of the log likelihood ratio which corresponds to the difference between the log likelihood ratio of the empty model at one level and the log likelihood ratio of the two-level empty model, in column (3), the test statistic of each model, which is \(799,8{ }\left( { - 2{*}\left( {logLik} \right)} \right)\). Since the test statistic is higher than the Chi-square distribution with 1 degree of freedom (3.84), the null hypothesis that the variance between communities is equal to zero is rejected for all models, which indicates that multilevel models are suitable.
Table 2
LogLikelihood tests for multilevel models
 
Between-canton variance \({\sigma }_{u}^{2}\)
Likelihood ratio (log)
Test statistic
Result
Isolation
0.4865
− 383.2597
613,062.2161
(p < 0.0001)
Social Distancing
0.4554
− 1459.3520
2,334,379.4592
(p < 0.0001)
Vaccination
0.4817
− 1208.4470
1,933,031.8212
(p < 0.0001)
Post-pandemic
0.9610
− 369.0373
590,312.0651
(p < 0.0001)
Table 3 shows the descriptive statistics of our variables. There are 1.78% of people whose final status is death. There are 50.75% of women and the average age is 40 years. At the cantonal level, there are 1.08 level 1 health centers per 10,000 inhabitants in each canton in average, 0.06 level 2 health centers and 0.02 level 3 health centers. Drinking water and sewerage coverage at the cantonal level is, on average, 87.05% and 77.05%, respectively. By March 2021, the average vaccination rate at the cantonal level was 86.76%. The average investment in health infrastructure per capita by provinces $66.34 thousand dollars.
Table 3
Descriptive statistics of variables
Dependent variable
 
Relative Frequency
Absolute Frequency
 
% of alive cases
% of death cases
Alive cases
Death cases
Final status of patients
98.22
1.78
1,027,522
18,638
Quantitative independent variables
 
Media
Standard Deviation
Minimum
Maximum
Age
40.35
17.44
0.002
95
Proxies of individual variables
Incidence of diseases
People with respiratory diseases per 10,000 inhab
56.93
21.19
6.42
237.70
People with kidney diseases per 10,000 inhab
19.47
6.57
0
74.72
People with diabetes diseases per 10,000 inhab
10.15
4.89
0
51.37
People with heart diseases per 10,000 inhab
25.59
8.21
1.62
64.08
People with liver diseases per 10,000 inhab
3.66
1.35
0
16.84
Contextual variables
Health infrastructure variables
Level 1 health care centers per 10,000 inhab
1.08
1.31
0.31
16.53
Level 2 health care centers per 10,000 inhab
0.06
0.12
0.00
1.46
Level 3 health care centers per 10,000 inhab
0.02
0.03
0.00
0.24
ICU beds per 10,000 inhab
0.66
0.44
0.00
1.51
Public health investment per capita
66.34
59.23
8.15
148.17
% households with access to drinking water
87.05
17.50
8.64
100
% households with access to
sewerage
77.02
23.13
0.01
100
% cantonal population with complete COVID-19 vaccination schedule
86.76
9.50
19.27
100
Urban green index
20.59
32.24
0.014
259.03
Economic variables
% people with adequate employment in the province
41.27
11.50
15.68
65.17
Cantonal GVA per capita
6.08
2.97
0.77
45.18
% households living in overcrowded conditions
8.91
5.00
3.61
24.09
Pandemic variables
% growth of daily infections in the canton of residence
160.65
616.81
-99.20
22,500.00
Qualitative independent variable
 
Relative Frequency
Absolute Frequency
 
% of female people
% of male people
0 (Female)
1 (Male)
Male (Binary variable)
50.75
49.25
530,969
515,191
Regarding economic variables, the average gross value added by cantons $6.08 thousand dollars per capita. In average, 41.27% of the provincial population has adequate employment, and 8.91% of population live in overcrowded conditions. The growth rate of daily infections by cantons was 160.65 infections. As for incidence of diseases in each canton, there are 56.93 cases of respiratory diseases per 10,000 inhabitants, 25.59 cases of heart diseases per 10,000 inhabitants, 10.15 cases of diabetes per 10,000 inhabitants and 3.66 cases of liver diseases per 10,000 inhabitants.

4 Results

Figure 4 shows the caterpillar plots for cantonal random effects (\({u}_{j}\)), estimated using the complete model for each stage of the pandemic. These figures show how different the cantonal residuals are from the average (red line) at a 95% confidence interval (vertical lines of each observation). The population of cantons that lie above the line and do not overlap with the red line are significantly more likely to die due to COVID-19 (red in Fig. 5). The population of cantons that lie under the red line are significantly less likely to die due to COVID-19 (blue in Fig. 5). As shown in Table 4, during the social distancing phase and the vaccination phase, more cantons record significant values higher and lower than the mean than during the isolation phase and the post-pandemic phase. Only Huaquillas remains above the average probability of dying in all stages, and there are no cantons that are always below the average (Table 4). Huaquillas recorded lethality rates of 15.67%, 3.4%, 1.97% and 0.30% during the isolation phase, social distancing phase, vaccination and post-pandemic phase, respectively. This canton register a low level of health infrastructure (0 ICU beds, 0.82 level 1 health centers, 0.16 level 2 health centers and 0 level 3 centers) than the national average (0.1025, 2.74, 0.12 and 0.0055, respectively). There were 138 cantons that did not have significant residuals in any stage of the pandemic. Quito, the capital of the country, recorded a higher probability of death than the mean in the isolation and in the post-pandemic (0.0209) stages, whereas it recorded a lower probability of death than the mean in the vaccination phase and in the social distancing phase. This result suggests that the pandemic process was dynamic and not monotonic in all regions. Guayaquil, the epicenter of the pandemic, another main city of Ecuador, records significant residuals below the mean in the stages of isolation, social distancing and vaccination. Cuenca, another important city, was not significantly different from the other cantons in the isolation stage, while it had a lower-than-average probability of death during the last 3 stages of the pandemic. The differences in the probability of dying across regions can be explained by several factors such as the size of a city, the availability of health infrastructure and the pandemics dynamics itself. Cantons where people are more likely to die are mainly small cantons with less health infrastructure. However, in this group, there are also medium-sized cities, big cities and metropolitan cities where population density is high, and congestion of health services is more probable. In the low probability of dying group, the representation of medium-sized, big and metropolitan cities is higher than in the high probability of dying group. These cities have better endowment of health infrastructure and may conducted a better management of the pandemic due to more personnel and organizational structures. In this group of low probability of dying, there are also small cities which could be more isolated and less affected by the pandemic. Nevertheless, there is not a general rule of characteristics explaining the probability of dying of people since as in the case of Quito, cantons change their status of high or low probability of dying across stages, showing that the pandemic dynamics in places could also matter and other factors such as the governmental management of the pandemic.
Table 4
Cantons with significantly different probability of dying by COVID-19 than the mean by pandemic stages
Higher probability of death
Isolation
Social distancing
Vaccination
Post-pandemic
11 cantons:
33 cantons:
 
37 cantons:
 
8 cantons:
Riobamba
Tulcán
El Carmen
Sígsig
El Carmen
Arenillas
Machala
Espejo
Jipijapa
Latacunga
Jipijapa
Huaquillas
El Guabo
Latacunga
Junín
Pujilí
Manta
Esmeraldas
Huaquillas
La Maná
Rocafuerte
Saquisilí
Montecristi
Atacames
Playas
Riobamba
Tosagua
Arenillas
Rocafuerte
Pueblo viejo
Quevedo
Colta
24 de Mayo
El Guabo
Santa Ana
Manta
Jipijapa
Guamote
Ambato
Huaquillas
Sucre
Quito
Manta
Guano
Cevallos
Santa Rosa
24 de Mayo
Santo Domingo
Montecristi
Penipe
Pelileo
Esmeraldas
Jaramijó
 
Jaramijó
Machala
Píllaro
Balzar
San Vicente
 
Quito
El Guabo
Tisaleo
Loja
Archidona
 
 
Huaquillas
Yantzaza
Macará
Mera
 
 
Pasaje
Lago Agrio
Saraguro
Pedro Moncayo
 
 
Santa Rosa
Orellana
Zapotillo
Pelileo
 
 
Playas
Santo Domingo
Portoviejo
Píllaro
 
 
Macará
La Concordia
Bolívar
Zamora
 
 
La Libertad
 
Chone
Yantzaza
 
   
Cascales
La Libertad
 
   
Santo Domingo
  
Lower probability of death
Isolation
Social distancing
Vaccination
Post-pandemic
8 cantons:
16 cantons:
 
22 cantons:
 
2 cantons:
Guayaquil
Cuenca
Samborondón
Cuenca
San Jacinto De Y
Cuenca
Daule
Gualaceo
San Jacinto De Y
Guaranda
Grl. Antonio Elizalde
Ambato
El Empalme
Guaranda
Antonio Ante
Chimbo
Gonzanamá
 
Milagro
Azogues
Loja
San Miguel
Baba
 
Samborondón
Atacames
Catamayo
Riobamba
Vinces
 
San Jacinto De Yaguachi
Guayaquil
Vinces
Colta
Olmedo
 
Portoviejo
Daule
Morona
Pasaje
Morona
 
 
Milagro
Quito
Guayaquil
Quito
 
   
Milagro
Rumiñahui
 
   
Naranjal
Ambato
 
   
Samborondón
San Cristóbal
 
Table 5 presents the results of the logistic multilevel model for 4 stages of the pandemic: i. isolation, ii. social distancing, iii. vaccination and iv. post-pandemic. Two models are presented for each stage, the empty model in columns a, b, c and d), the complete model in columns (a.1, b.1, c.1 and d.1) and marginal effects are shown in columns a.2, b.2, c.2 and d.2.
Table 5
Results logit model
 
Isolation
Social distancing
Vaccination
Post-pandemic
 
(a) Null Model
(a.1) Coefficient
(a.2) Marginal Effects
(b) Null Model
(b.1) Coefficient
(b.2) Marginal Effects
(c) Null Model
(c.1) Coefficient
(c.2) Marginal Effects
(d) Null Model
(d.1) Coefficient
(d.1) Marginal Effects
Intercept
− 1.93477***
− 5.1918***
 
− 3.63975***
− 7.6505***
 
− 3.8905***
− 7.8352***
 
− 5.6932***
− 9.1877***
 
  
(0.42)
  
(0.24)
  
(0.32)
  
(0.59)
 
Man
 
0.4893***
0.0325
 
0.5983***
0.0121
 
0.4536***
0.0074
 
0.4703***
0.0013
  
(0.05)
  
(0.02)
  
(0.03)
  
(0.07)
 
Age
 
0.0819***
0.0054
 
0.0875***
0.0018
 
0.081***
0.0013
 
0.0947***
0.0003
  
(0.00)
  
(0.00)
  
(0.00)
  
(0.00)
 
Health infrastructure variables
            
Level 1 health care centers per 10,000 inhab
 
− 0.1208*
− 0.008
 
− 0.0753**
− 0.0015
 
− 0.1462***
− 0.0024
 
− 0.022
− 0.0001
 
(0.06)
  
(0.02)
  
(0.03)
  
(0.04)
 
Level 2 health care centers per 10,000 inhab.
 
− 0.5603
− 0.0372
 
− 0.3126
− 0.0063
 
0.3989
0.0065
 
0.5056
0.0014
 
(0.66)
  
(0.25)
  
(0.31)
  
(0.50)
 
Level 3 health care centers per 10,000 inhab.
 
− 0.7125
− 0.0473
 
1.0487
0.0212
 
− 0.5948
− 0.0097
 
− 0.4844
− 0.0014
 
(1.86)
  
(1.34)
  
(1.63)
  
(1.94)
 
% population with complete COVID-19 vaccination schedule
       
0.0081*
0.0001
 
0.0017
0.000
       
(0.00)
  
(0.01)
 
Public investment in health per capita
 
− 0.0026
− 0.0002
 
− 0.0009
0.000
 
− 0.0021
0.000
 
− 0.0158***
0.000
 
(0.00)
  
(0.00)
  
(0.00)
  
(0.00)
 
ICU beds per 10,000 inhab
 
− 0.0963
− 0.0064
 
0.3312*
0.0067
 
0.2022
0.0033
 
0.3038
0.0009
 
(0.22)
  
(0.15)
  
(0.18)
  
(0.22)
 
Economic variables and infrastructure
            
% people with adequate employment in the province
 
− 0.0242**
− 0.0016
 
− 0.0173***
− 0.0004
 
− 0.0158**
− 0.0003
 
− 0.0138
0.000
 
(0.01)
  
(0.00)
  
(0.01)
  
(0.01)
 
Cantonal GVA per capita
 
− 0.3008**
− 0.02
 
− 0.0688
− 0.0014
 
0.0834
0.0014
 
0.0201
0.0001
 
(0.11)
  
(0.07)
  
(0.09)
  
(0.12)
 
% households with access to drinking water
 
− 0.0028
− 0.0002
 
− 0.0032
− 0.0001
 
− 0.0024
0.000
 
0.0026
0.000
 
(0.00)
  
(0.00)
  
(0.00)
  
(0.00)
 
% households with access to sewerage in the canton
 
0.001
0.0001
 
0.0007
0.000
 
− 0.0002
0.000
 
− 0.009**
0.000
 
(0.00)
  
(0.00)
  
(0.00)
  
(0.00)
 
Urban Green Index
 
0.0019
0.0001
 
0.0016
0.000
 
− 0.004
− 0.0001
 
− 0.0059
0.000
 
(0.00)
  
(0.00)
  
(0.00)
  
(0.00)
 
Incidence of Diseases
            
% growth in daily COVID-19 infections in the canton of residence
 
− 0.057***
− 0.0038
 
− 0.0165***
− 0.0003
 
− 0.033***
− 0.0005
 
− 0.029*
− 0.0001
 
(0.01)
  
(0.00)
  
(0.00)
  
(0.01)
 
No. of people with respiratory diseases / 10,000 inhab. in each canton
 
− 0.006
− 0.0004
 
− 0.0106***
− 0.0002
 
− 0.0088**
− 0.0001
 
− 0.0181***
− 0.0001
 
(0.00)
  
(0.00)
  
(0.00)
  
(0.00)
 
No. of people with diabetes / 10,000 inhab. in each canton
 
0.0053
0.0003
 
− 0.0028
− 0.0001
 
0.0037
0.0001
 
0.0004
0.000
 
(0.02)
  
(0.01)
  
(0.01)
  
(0.02)
 
No. of people with severe heart disease / 10,000 inhab. in each canton
 
− 0.0158
− 0.0011
 
0.0072
0.0001
 
− 0.0015
0.000
 
− 0.0145
0.000
 
(0.01)
  
(0.01)
  
(0.01)
  
(0.01)
 
No. of people with severe kidney disease / 10,000 inhab. in each canton
 
0.0457**
0.003
 
0.0119
0.0002
 
− 0.0008
0.000
 
0.0115
0.000
 
(0.01)
  
(0.01)
  
(0.01)
  
(0.01)
 
No. of people with liver disease / 10,000 inhab. In each canton
 
0.0366
0.0024
 
− 0.0009
0.000
 
0.0703**
0.0002
 
0.0852*
0.0002
 
(0.05)
  
(0.02)
  
(0.02)
  
(0.04)
 
Observations (N)
26,252
26,252
 
387,349
387,349
 
320,798
320,798
 
311,772
311,772
 
Variance between individuals \(\frac{{\pi }^{2}}{3}\)
3.29
3.29
 
3.29
3.29
 
3.29
3.29
 
3.29
3.29
 
Individual variation (%)
87.12%
93.59%
87.84%
93.37%
87.23%
90.53%
77.39%
91.76%
Between-canton variance
0.4865***
0.2252***
 
0.4554***
0.2335***
 
0.4817***
0.3442***
 
0.961***
0.2955***
 
Standard error of the between-canton variance
0.6975
0.4745
 
0.6748
0.4833
 
0.694
0.5867
 
0.9803
0.536
 
Between-canton variance (%)
12.88%
6.41%
 
12.16%
6.63%
 
12.77%
9.47%
 
22.61%
8.24%
 
Total Variation Model
3.7765
3.5152
 
3.7454
3.5235
 
3.7717
3.6342
 
4.251
3.5855
 
(Percentage)
100%
100%
 
100%
100%
 
100%
100%
 
100%
100%
 
Odds of dying by COVID-19
12.6%
0.55%
 
2.6%
0.48%
 
2%
0.04%
 
0.3%
0.01%
 
Wald test between-canton variance
613,062.2161
  
2,334,379.459
  
1,933,031.821
  
590,312.0651
  
(p-value)
(0.00)
  
(0.00)
  
(0.00)
  
(0.00)
  
The standard deviation is shown in parentheses. Statistical significance *** 0.00 ** 0.05 *0.1
At the bottom of Table 5, the variance between individuals and the variance between cantons are shown for empty and complete models. Overall, in all models, the variance of the probability of death is mainly explained by the variation across individuals (shown at the bottom of columns a, b, c, d), ranging from 77.39% \(\left( {{\text{CPV}} = \frac{{\sigma_{e}^{2} }}{{\sigma_{u}^{2} + \sigma_{{e^{*} }}^{2} }} = \frac{3.29}{{0.961 + 3.29}} = 77.39\% } \right)\) in the post-pandemic phase to 87.84% in the social distancing phase \(\left( {{\text{CPV}} = \frac{{\sigma_{e}^{2} }}{{\sigma_{u}^{2} + \sigma_{{e^{*} }}^{2} }} = \frac{3.29}{{0.4554 + 3.29}} = 87.84\% } \right)\). The remaining variance is explained by the between-canton variance, which is not negligible, ranging between 12.16% \(\left( {{\text{CPV}} = \frac{{\sigma_{u}^{2} }}{{\sigma_{u}^{2} + \sigma_{{e^{*} }}^{2} }} = \frac{0.4554}{{0.4554 + 3.29}} = 12.16\% } \right)\) in the social distancing phase and 22.61% in the post-pandemic phase \(\left( {{\text{CPV}} = \frac{{\sigma_{u}^{2} }}{{\sigma_{u}^{2} + \sigma_{{e^{*} }}^{2} }} = \frac{0.961}{{0.961 + 3.29}} = 22.6\% } \right)\) (shown at the bottom of columns a.1, b.1, c.1 and d.1). According to the Wald test, shown at the bottom of Table 5, the variation between cantons in all stages is significant.
When including the explanatory variables in the models, the cantonal variance explanation is reduced reaching percentages ranging from 6.41 to 9.47% (columns a.1, b.1, c.1 and d.1).
The first row of Table 5 shows the values of the intercepts of models, which are used to calculate the probability of death due to COVID-19 for an average canton. In the isolation phase, this value is -1.93477 and the corresponding probability is 12.6% (computed as exp(− 1.93477)/(1 + exp(− 1.93477)). At the end of the Table 5, columns a, b, c and d show the probability of dying by COVID-19 in each phase, in the social distancing phase, it was 2.6%, in the vaccination phase, it was 2% and in the post-pandemic phase, it was 0.3%. When including explanatory variables (columns a.1, b.1, c.1, d.1), the probability of dying in the isolation phase is estimated at 0.55%, in the social distancing phase at 0.048%, at 0.04% in the vaccination phase, and at 0.01% in the post-pandemic phase. All these estimates are statistically significant at the 99% confidence level.
Under the assumption that \({u}_{j}\) follows a normal distribution, we would expect around 95% of cantons to have a value of \({u}_{j}\) within 2 standard deviations of the mean zero. As shown in Table 6, considering complete models, the probability of death due to COVID-19 in the isolation phase lies between 0.002 and 0.014 in the middle 95% of cantons. Such interval reduces over time so that in the post-pandemic phase, the probability of death due to COVID-19 lies between 0.000034 and 0.0003.
Table 6
Coverage interval of the predicted probability of dying due to COVID-19, using the complete multilevel models
 
\({\beta }_{0}\)
\({\sigma }_{u}^{2}\)
2 standard deviations
\(\widehat{\pi }\) Below
\(\widehat{\pi }\) Above
Isolation
− 5.19
0.2252
0.9491
0.00215
0.01417
Social distancing
− 7.65
0.2335
0.96644
0.00018
0.00125
Vaccination
− 7.84
0.3442
1.17337
0.00012
0.00128
Post-pandemic
− 9.19
0.2955
1.0872
0.000034
0.0003
Regarding the results of individual characteristics, sex and age are important factors that influence the probability of death. In accordance with the existent literature, older people are more likely to die after getting infected with the CoV-Sars-2 virus. This is mainly related to the preexistence of comorbidities associated to the advanced age such as diabetes mellitus and chronic respiratory diseases (Plasencia-Urizarri et al. 2020; Wang et al. 2020). Our results show that men are more likely to die from COVID-19 than women. These results were also obtained in Abate et al. (2020) and Hallak et al. (2022) who explained that men have more risk factors for exposure to SARS-CoV-2, such as smoking, alcohol consumption, exposure during burials, work in primary sectors with higher occupational risks that require physical activity outside their homes and a higher degree of interaction. Moreover, hygienic habits in men such as infrequent hand washing could increase their risk of getting infected. These variables are statistically significant and their marginal effects in both cases progressively decrease between stages. The difference in the probability of death between men and women decreases from 3.25 higher probability of death in men compared to women in the isolation stage to a 0.13% difference in the post-pandemic stage. Likewise, the effect of age on the probability of death from COVID-19 decreases along phases from 0.54 in the isolation phase to 0.03% in the post-pandemic phase. This decrease in the marginal effects of age and sex may be attributable to the vaccination process, which was preferentially applied to vulnerable groups, including the elderly (Russell and Greenwood 2021) and did not discriminate between men and women. Therefore, this vaccination application reduced the probability of death for older people and make men and women equally likely to die.
The number of level 1 health care facilities per 10,000 inhabitants is negative and significant for the stages of isolation, social distancing and vaccination, indicating that more health care facilities are associated with a lower probability of death of individuals during these stages. The marginal effects show that an increase of one unit of health care facility is associated with a lethality reduction of -0.80%, -0.15% and − 0.24% during the isolation phase, social distancing phase and the vaccination phase, respectively. When infected people go to a health care center, they get diagnosed by the health personnel of the center and continue with treatment. Thus, a high availability of health care centers where people can get assistance would reduce the probability of dying. Despite the severity of the initial pandemic situation and in many cases, nonoptimal conditions of level 1 health centers, it seems that these centers played an important role in disease prevention, through established actions for the detection, screening, isolation and surveillance of infected people (Organización Panamericana de la Salud, 2021). However, it is worth noting that even though they are not well equipped in Ecuador, it seems that their role was very important to reduce the probability of death by COVID-19. In the last stage of Post-pandemic, the number of level 1 health care centers is no longer significant to explain the probability of death. This result can be explained by the fact that in average more than 75% of the population in cantons were fully vaccinated, which reduce the symptoms of COVID-19 (Sadarangani et al. 2021) and the contagion rate. Surprisingly, level 2 and level 3 health centers have a nonsignificant effect on the probability of death. This could be explained by the health system saturation.
The individual probability of death is positively associated with the percentage of population with a complete vaccination schedule in cantons. This result was not expected since vaccination could reduce the probability of death (López et al. 2022; Scruzzi et al. 2022, López et al. 2022; Scruzzi et al. 2022). In our case, the correlation between the percentage of vaccinated population and the percentage of dead people over the infected people is 0.470. It might indicate that places with a high proportion of dead people are places with high population density and therefore higher vaccination coverage.
The per capita investment in health in a given province is associated to a lower probability of death of individuals only in the post-pandemic phase when the health system saturation point has passed. During the stages of isolation, social distancing and vaccination, the nonsignificant effect indicates that the preexisting health expenditure was not sufficient to reduce the probability of death. It implies that universal health coverage is important to prevent deaths (Kruk et al. 2018). When health spending is low, disease mortality increases as shown for the Indian case by Balakrishnan et al. (2021). In places with low investment in health, health care centers might lack medical inputs or health personnel to assist people, negatively affecting their probability to recover from illness.
The number of ICU beds per 10,000 inhabitants in a given canton is positively associated to the probability of death of individuals. This result can be explained by the fact that the presence of ICU beds in specific cantons attracts individuals with worse symptoms, where the number of existing beds is exceeded, and patients begin to die. This variable is significant only in the social distancing stage and then loses significance possibly due to the decrease in severe cases of COVID-19 because of vaccination.
In cantons with high rates of adequate employment, the individual probability of death from COVID-19 is lower. This indicates that people with minimum labor conditions in terms of income (minimum salary) and work time (forty hours per week) are more likely to face the pandemic favorably since they can increase their private or public health consumption if they were infected. In the case of Ecuador, the differentiation between adequate employment and inadequate employment is crucial. According to the employment classification established by the Institute of Statistics and Census of Ecuador (INEC, (2015), workers with adequate employment are those who during the reference week earned a labor income equal or higher than the minimum salary and worked forty or more hours per week, regardless of their willingness or availability to work additional hours. People with inadequate employment are those who earn less than the minimum salary and/or work less than forty hours per week. In this case, if workers with inadequate employment get infected by COVID-19, they most likely will not be able to cope with medical charges. For this reason, in this study, the adequate employment is considered rather than the general employment rate. Béland et al. (2002) also show that the unemployment rate influences the perceived individual health in Quebec regions. In addition, when more people have access to sewerage in a canton, the probability of dying due to COVID-19 is lower. Proper management of human waste and proper sewage disposal is critical to reduce the spread of infectious diseases such as COVID-19.
The growth rate of COVID-19 infections2 in the canton is statistically significant during all stages of the pandemic; however, its sign is negative, which is apparently contradictory. However, for the last stages, it could be explained by the effect of herd immunity. This variable was included in the model since lethality is related to infected people, therefore the probability of death is expected to increase when the contagion rate increases. According to Randolph and Barreiro (2020), if a fraction of the population has immunity, indicating that it was previously infected and recovered, the probability of contagion between infected and susceptible patients is reduced, therefore, mortality rate would reduce.
To control comorbidities, the prevalence of diseases at the cantonal level is used. Only the prevalence of liver and kidney diseases is positively associated to higher probabilities of death by COVID-19, which is in line with medical literature indicating that patients with COVID-19 and these preexisting chronic diseases are more likely to have multiorgan failure (Ortiz-Hernández and Pérez-Sastré, 2020). The prevalence of respiratory diseases in a canton is negative for the probability of death of individuals during the last 3 stages of the pandemic. The negative sign might indicate that people with respiratory diseases might have developed defenses that are used to face the COVID-19 illness (Dominguez 2020).
These results emphasize the importance of contextual factors in the spread and impact of COVID-19 in Ecuador. For instance, demographic factors like age and gender, as well as social determinants like education level and access to healthcare, significantly affected individuals' vulnerability to the virus and their responses to public health measures. Moreover, the intention to use COVID-19 vaccines has been shaped by a complex interplay of beliefs, attitudes, perceived barriers and social influences, highlighting the need for tailored communication strategies to address vaccine hesitancy. Understanding these contextual factors is crucial for developing effective interventions and predicting compliance with containment measures, which is essential for managing current and future public health crises.

5 Conclusion

This study measured the individual probability of dying by COVID-19 in different stages of the pandemic using not only individual characteristics but also contextual characteristics where people live, putting special emphasis on health care services. From our descriptive analysis, COVID-19 pandemic affected unevenly to the Ecuadorian cantons. There were cantons with many daily deaths whereas in other cantons, any death was registered, which shows that the context matter for the individual probability of death. Our estimations results show that a non-negligible 12% to 22% of the variance of the individual probability to die is explained by cantonal differences. Cantons where people were more likely and less likely to die by stage were identified. Those differences could be attributable to the city size, the availability of health infrastructure and the pandemics dynamics itself. This result can serve as basis for policy actions for local governments. Regarding factors, only the number of level 1 health care centers has a negative association with the probability of death of individuals during the first three stages of the pandemic. However, level 2 and level 3 health centers, which are more specialized were not significant to explain the probability of dying. To increase the influence of level 2 and level 3 health centers to reduce the probability of dying, these facilities must be strengthened so that they could be used to face negative shocks. Based on the positive association between the ICU beds and the probability of dying, which is explained by a concentration of these facilities in big cities, it is important to redistribute them across the territory so people can get medical assistance timely.
Key elements of resilience of Ecuadorian cities during the COVID-19 pandemic were related to adequate employment, health public investment and sanitary infrastructure. When people have a stable income, they can have savings to face emergencies and reduce fatal outcomes. It is then important to improve the employment level in cantons, by focusing on those with less opportunities. It is imperative to increase the coverage of tapping water to reach the whole population. In addition, the preexistent public investment in the health sector was associated with a lower probability of dying. However, it might not be sufficient, so further increases in health expenditure would contribute to facing future pandemics and health emergencies. As Wagstaff (2002) argues, to effectively address the COVID-19 pandemic, it is important to have an integral system approach with different elements and dimensions. All in all, the spatial disparities in terms of health, sanitary and economic aspects are prominent determinants of the probability of death by COVID-19 of individuals. One limitation of this study is that we could not account for specific individual variables such as comorbidities and socioeconomic characteristics which play a part in determining their probability of death. We include disease incidence in each canton to address this problem. In addition, it is worth noting that at the beginning of the pandemic, recording information was problematic, which could lead to underreport COVID-19 infections and deaths.

Acknowledgements

Christian L. Vásconez acknowledges the partial funding of EPN project PIEX-DFIS-MSP-2022.

Declarations

Conflict of interest

This manuscript has not been published or presented elsewhere in part or in entirety, and is not under consideration by another journal. There are no conflicts of interest to declare.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://​creativecommons.​org/​licenses/​by-nc-nd/​4.​0/​.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Footnotes
1
1 Different variables were tested in the model. To avoid multicollinearity, some of them were not included. Here, an explanation of what variables were tested but not included is given. The number of doctors was included in the model; however, this variable showed high correlation with the variable's No. of health centers 1, 2 and 3. This variable was negative but not significant in the first two stages. In the last 2 stages, it became significant and changed its sign. When removing the variable No. of health centers from the model, the variable of the number of doctors kept the negative sign and gained statistical significance in the second stage of the pandemic. Due to these circumstances, this variable was not considered, and we kept the variable of the number of health centers which showed stability across models. The variable of provincial household overcrowding was also included in the model; however, it showed a high level of multicollinearity with variables such as the number of households with access to drinking water, sewerage and cantonal GVA. The estimate of the average household overcrowding was negative and significant in all stages of the pandemic, indicating a counter-intuitive result. In addition, the variable of growth in daily COVID-19 infections in the canton of residence captures somehow the population density. Therefore, this variable more associated to the COVID-19 context was preferred over the population density, measured by the household overcrowding.
 
2
Other measures such as number of infected people in each period and the number infected people in each period over the cantonal population were also included in the model and the estimates are still negative.
 
Literature
go back to reference Abrams EM, Szefler SJ (2020) COVID-19 and the impact of social determinants of health. Lancet Respiratory 8:659–661CrossRef Abrams EM, Szefler SJ (2020) COVID-19 and the impact of social determinants of health. Lancet Respiratory 8:659–661CrossRef
go back to reference Arauzo-Carod JM, Giménez-Gómez JM, Llop M (2023) Defining an ‘Epidemiological Risk Index’to analyse COVID-19 mortality across European regions. Annals Regional Sci 1–23 Arauzo-Carod JM, Giménez-Gómez JM, Llop M (2023) Defining an ‘Epidemiological Risk Index’to analyse COVID-19 mortality across European regions. Annals Regional Sci 1–23
go back to reference Béland F, Birch S, Stoddart G (2002) Unemployment and health: contextual-level influences on the production of health in populations. Soc Sci Med 55(11):2033–2052CrossRef Béland F, Birch S, Stoddart G (2002) Unemployment and health: contextual-level influences on the production of health in populations. Soc Sci Med 55(11):2033–2052CrossRef
go back to reference Brandén M, Aradhya S, Kolk M, Härkönen J, Drefahl S, Malmberg B, Rostila M, Cederström A, Andersson G, Mussino, E (2020) Residential context and COVID-19 mortality among adults aged 70 years and older in Stockholm: a population-based, observational study using individual-level data. Lancet healthy longevity 1(2):e80–e88 Brandén M, Aradhya S, Kolk M, Härkönen J, Drefahl S, Malmberg B, Rostila M, Cederström A, Andersson G, Mussino, E (2020) Residential context and COVID-19 mortality among adults aged 70 years and older in Stockholm: a population-based, observational study using individual-level data. Lancet healthy longevity 1(2):e80–e88
go back to reference Castells-Quintana D, Royuela V (2023) On the association between income inequality and covid spread: a view into spanish functional urban areas. In: Celbiş M, Kourtit K, Nijkamp P (eds) Pandemic and the City. Springer, Cham, pp 127–138CrossRef Castells-Quintana D, Royuela V (2023) On the association between income inequality and covid spread: a view into spanish functional urban areas. In: Celbiş M, Kourtit K, Nijkamp P (eds) Pandemic and the City. Springer, Cham, pp 127–138CrossRef
go back to reference Hu C, Liu Z, Jiang Y, Shi O, Zhang X, Xu K, Chen X (2020) Early prediction of mortality risk among patients with severe COVID-19, using machine learning. Int J epidemiol 49(6):1918–1929CrossRef Hu C, Liu Z, Jiang Y, Shi O, Zhang X, Xu K, Chen X (2020) Early prediction of mortality risk among patients with severe COVID-19, using machine learning. Int J epidemiol 49(6):1918–1929CrossRef
go back to reference López L, Portugal W, Huamán K, Obregón C (2022) Efectividad de vacunas COVID-19 y riesgo de mortalidad en Perú: un estudio poblacional de cohortes pareadas. Anales de La Facultad de Medicina, 83(2). https://doi.org/10.15381/anales López L, Portugal W, Huamán K, Obregón C (2022) Efectividad de vacunas COVID-19 y riesgo de mortalidad en Perú: un estudio poblacional de cohortes pareadas. Anales de La Facultad de Medicina, 83(2). https://​doi.​org/​10.​15381/​anales
go back to reference Mahmoudabadi A (2021) Developing a probit regression model for estimating the chance of mortality for coronavirus disease-2019 patients. Public Health Open J 6(2):62–67CrossRef Mahmoudabadi A (2021) Developing a probit regression model for estimating the chance of mortality for coronavirus disease-2019 patients. Public Health Open J 6(2):62–67CrossRef
go back to reference Plasencia-Urizarri TM, Aguilera-Rodríguez R, Mederos LEA (2020) Comorbilidades y gravedad clínica de la COVID-19: revisión sistemática y meta-análisis. Revista Habanera de Ciencias Médicas 19:3389 Plasencia-Urizarri TM, Aguilera-Rodríguez R, Mederos LEA (2020) Comorbilidades y gravedad clínica de la COVID-19: revisión sistemática y meta-análisis. Revista Habanera de Ciencias Médicas 19:3389
go back to reference Rađenović T, Radivojević V, Krstić B, Stanišić T, Živković S (2022) The efficiency of health systems in response to the COVID-19 Pandemic: evidence from the EU Countries. Problemy Ekorozwoju 17(1):7–15CrossRef Rađenović T, Radivojević V, Krstić B, Stanišić T, Živković S (2022) The efficiency of health systems in response to the COVID-19 Pandemic: evidence from the EU Countries. Problemy Ekorozwoju 17(1):7–15CrossRef
go back to reference Reveiu A, Constantin DL (2023) The impact of the COVID-19 pandemic on regional inequalities in Romania. Spotlight on unemployment and health conditions. Reg Sci Policy Pract 15(3):644–658CrossRef Reveiu A, Constantin DL (2023) The impact of the COVID-19 pandemic on regional inequalities in Romania. Spotlight on unemployment and health conditions. Reg Sci Policy Pract 15(3):644–658CrossRef
go back to reference Sassen S, Kourtit K (2021) A post-corona perspective for smart cities: ‘Should I stay or should I go?’ Sustainability 13(17):9988CrossRef Sassen S, Kourtit K (2021) A post-corona perspective for smart cities: ‘Should I stay or should I go?’ Sustainability 13(17):9988CrossRef
go back to reference Scruzzi GF, Aballay LR, Carreño P, Anabel G, Rousseau D, Franchini CG, Cecchetto E, Willington AP, Barbás MG, López L (2022) Investigación original Vacunación contra SARS-CoV-2 y su relación con enfermedad y muerte por COVID-19 en Argentina. https://doi.org/10.26633/RPSP.2022.39 Scruzzi GF, Aballay LR, Carreño P, Anabel G, Rousseau D, Franchini CG, Cecchetto E, Willington AP, Barbás MG, López L (2022) Investigación original Vacunación contra SARS-CoV-2 y su relación con enfermedad y muerte por COVID-19 en Argentina. https://​doi.​org/​10.​26633/​RPSP.​2022.​39
go back to reference Silverman R, Krubiner C, Chalkidou K, Towse A (2020) Financing and scaling innovation for the Covid fight financing and scaling innovation for the Covid fight: a closer look at demand-side incentives for a vaccine. Center Global Develop 1(1):1–7 Silverman R, Krubiner C, Chalkidou K, Towse A (2020) Financing and scaling innovation for the Covid fight financing and scaling innovation for the Covid fight: a closer look at demand-side incentives for a vaccine. Center Global Develop 1(1):1–7
go back to reference Simpson-Yap S, Pirmani A, Kalincik T, De Brouwer E, Geys L, Parciak T, Helme A, Rijke N, Hillert JA, Moreau Y, Edan G, Sharmin S, Spelman T, McBurney R, Schmidt H, Bergmann AB, Braune S, Stahmann A, Middleton RM, Peeters LM (2022) Updated results of the COVID-19 in MS global data sharing initiative. Neurology—Neuroimmunology Neuroinflammation. https://doi.org/10.1212/NXI.0000000000200021CrossRef Simpson-Yap S, Pirmani A, Kalincik T, De Brouwer E, Geys L, Parciak T, Helme A, Rijke N, Hillert JA, Moreau Y, Edan G, Sharmin S, Spelman T, McBurney R, Schmidt H, Bergmann AB, Braune S, Stahmann A, Middleton RM, Peeters LM (2022) Updated results of the COVID-19 in MS global data sharing initiative. Neurology—Neuroimmunology Neuroinflammation. https://​doi.​org/​10.​1212/​NXI.​0000000000200021​CrossRef
go back to reference University of Bristol (2011). LEMMA (Learning environment for multilevel methodology and applications). Natl Centre Res Methods. University of Bristol (2011). LEMMA (Learning environment for multilevel methodology and applications). Natl Centre Res Methods.
go back to reference Wagstaff A (2002) Poverty and health sector inequalities. Policy and Practice 80(2):97–105 Wagstaff A (2002) Poverty and health sector inequalities. Policy and Practice 80(2):97–105
go back to reference Yu H, Lao X, Gu H, Zhao Z, He H (2021) Determinants on COVID-19 case fatality rates of cities in china: a logit-nb hurdle model analysis. Research Square Yu H, Lao X, Gu H, Zhao Z, He H (2021) Determinants on COVID-19 case fatality rates of cities in china: a logit-nb hurdle model analysis. Research Square
Metadata
Title
The uneven geography of the health system and its effect on the individual probability of death by COVID-19
Authors
Grace Carolina Guevara-Rosero
Víctor Hugo Hinojosa
Christian L. Vásconez
Publication date
05-11-2024
Publisher
Springer Berlin Heidelberg
Published in
The Annals of Regional Science / Issue 4/2024
Print ISSN: 0570-1864
Electronic ISSN: 1432-0592
DOI
https://doi.org/10.1007/s00168-024-01325-7