Introduction

COVID-19^1,2 was reported for the first time in China, on December 31^st 2019. It spread rapidly and soon afterwards was declared to be a pandemic by the World Health Organization (WHO).³ Globally, over 11 million people contracted COVID-19, resulting in 500,000 deaths due to the disease by August 28^th 2020. Brazil is in second place among the most affected countries,⁴ accounting for 3,761,391 cases and 118,649 deaths also as at 28/08/2020.⁵ COVID-19 has already affected all the Brazilian Federal Units and 99% of Brazilian municipalities had notified at least one case by August 2020.⁶ It is estimated that between 13% and 24% of cases progress to the more severe form of the disease, with acute respiratory failure.⁷ Other estimates indicate that 19% (95% confidence interval: 95%CI 13%;26%) of individuals with COVID-19 need to be hospitalized in intensive care units (ICU), and that 9% (95%CI 6%;13%) need lung ventilators.⁸

In order to structure a timely response to COVID-19, a critical element to consider is understanding where cases and deaths are concentrated. The emergence of clusters of COVID-19 cases and the dynamics of increases and decreases in these cases must be monitored so that timely interventions can be conducted. Analyses of spatio-temporal clusters are useful⁹ for identifying significant clusters of cases and deaths and for performing trend analyses.^10-12

In turn, analysis of spatio-temporal clusters using the emerging hot spot approach enables identification of trends to be shared by neighbors in time and space, creating conditions for the generation of useful insights for the evidence-based decisions.¹³ Despite being used in several different areas, such as Traffic Engineering¹⁴ and Veterinary Medicine,¹⁵ this tool is still little explored by Epidemiology.¹⁶

The objective of this study was to explore possibilities of reorganizing the health system with regard to tertiary care for COVID-19, by means of an approach based on spatio-temporal clusters, considering the epidemiological situation of the disease in Brazil and the possible benefits obtained by using emerging hot spot analysis.

Methods

This is a longitudinal ecological study based on a spatio-temporal approach using secondary data.¹⁷

Three secondary data sources taken from public administration databases were used for the analysis of the urgency and emergency care network regarding COVID-19. As such, it was not necessary to submit the study project to a Research Ethics Committee for approval.

The first database used was the COVID-19 case record database released by the Brazilian Ministry of Health.⁵ This database contained the historical record of confirmed COVID-19 cases, as well as reported deaths, right from the confirmation of the first case in Brazil on February 26, 2020 until August 27, 2020, this being the period selected by the researchers. For the purposes of analysis, the number of confirmed new cases per day was used.

The second database contained the municipal population projections for the year 2019. Information was retrieved from this source on the total number of inhabitants in each of the country’s 5,570 municipalities.

The third and final database used was the National Health Establishment Registry. The information retrieved from the Registry allowed us to characterize ICU beds in Brazil, both regular ICU beds and those dedicated exclusively to people with COVID-19. Moreover, this database enabled geolocalization of all registered health care facilities. The data on beds related to April 2020.

The study’s unit of analysis consisted of each of the 5,570 Brazilian municipalities. To evaluate the adequacy of the efforts to reorganize the Brazilian National Health System (SUS) in relation to COVID-19, three analytical steps were structured. First, we investigated spatial-temporal clustering patterns, by epidemiological week, of COVID-19 incidence in Brazil, using emerging hot spot analysis. In parallel, we assessed the degree of accessibility to hospital ICU beds in Brazil in the period before the pandemic, as well as the access patterns to new ICU beds introduced to respond to COVID-19. The data from the clustering process and the characterization of ICU bed accessibility were compared in order to identify care gaps in regions with COVID-19 incidence and mortality above the national medians. In this way, it was possible to identify regions with care gaps and high volumes of cases and deaths, weighted by population. Once these regions had been identified, the accessibility index for new ICU beds for COVID-19 was overlaid spatially, with the purpose of identifying priority regions for efforts to reorganize the urgency and emergency network, regarding emergency service coverage of health care gaps.

Results

As of August 27, 2020, Brazil had confirmed 3,761,391 COVID-19 cases and 118,649 COVID-19 deaths (Table 1). The Southeast (35%) and Northeast (30%) regions had the highest percentage of cases and deaths: 45% and 29%, respectively. Figure 1 represents COVID-19 incidence and mortality in Brazil. In all, data from 8,023 hospitals were analyzed, totaling 59,695 existing ICU beds, distributed over 3,543 cities. Data on a further 14,671 ICU beds exclusively for COVID-19 in 646 hospitals distributed over 340 cities in April 2020 were added to this total.

Source: COVID.SAUDE/Ministry of Health.

Figure 1 COVID-19 case incidence rate (per 100,000 inhab.) and COVID-19 case mortality rate (per 100,000 inhab.), Brazil, as at Epidemiological Week 35/2020 

When analyzing the COVID-19 incidence clustering in space and time, 54% of the 5,570 Brazilian municipalities were categorized as having an increasing pattern over time, concomitant with statistically significant spatial clustering (Figure 2; Table 1). The oscillating hot spot category was present in 81% of the 3,695 municipalities that had a detected pattern. This category marks a transition of the pattern between cold and hot spots. If in the past these municipalities had a small number of cases, as the epidemiological weeks progressed, they transitioned to a statistically significant spatio-temporal pattern of increasing cases. Eighty-four percent (84%) of the municipalities in the Midwest region fell into the oscillating hot spot category, followed by the Southern region with 63% and the Northern region with 62% of its municipalities categorized in this way (Figure 2; Table 1). In the state of Santa Catarina, 100% of its municipalities fell into this category, and in the states of Rondônia, Sergipe, Tocantins, Goiás and Espírito Santo, more than 90% of the municipalities followed the same pattern.

It can be seen that 67% of the municipalities in the state of Roraima were categorized as consecutive hot spots. This category indicates that the location was uninterruptedly categorized as a statistically significant hot spot in the final time intervals analyzed (Figure 2; Table 1).

Figure 2 Spatio-temporal cluster analysis of COVID-19 cases (A) and spatio-temporal cluster analysis of COVID-19 deaths (B), Brazil, Epidemiological Week 35/2020 

There are variations in the categorizations found in the Southeast region of Brazil. The state of Minas Gerais, for example, has 13% of municipalities classified as diminishing cold spots, that is: clusters that in the past were cold spots but for which statistical association has been weakening (Figure 2; Table 1). This means that in the near future those municipalities may be taken as proxies for regions with the potential to become hot spots.

Spatio-temporal clustering of mortality indicated the same increasing pattern in 31% of municipalities over the epidemiological weeks studied. Of the municipalities for which a mortality rate pattern was determined, 29% were classified in the oscillating hot spot category (Figure 2; Table 1). The Northeast region had the highest percentage (16%) of municipalities in transition from cold spots to hot spots (oscillating hot spots). More than 80% of the municipalities in the states of Alagoas, Ceará, Espírito Santo and Sergipe were categorized as oscillating hot spots (Figure 2; Table 1)

It can be seen that the states of Acre (45%) and Roraima (40%) had a high percentage of municipalities considered to be consecutive hot spots. Forty-seven percent (47%) of the municipalities in Roraima were categorized as sporadic hot spots.

Figure 3 shows the accessibility indices for all ICU beds (A) and for ICU beds exclusively for COVID-19 (B). We found that 25% (1,380) of Brazilian municipalities had the lowest accessibility index (≤0.000112/inhab.) for regular ICU beds, with 47% (213 of 450) of the Northern region municipalities falling into this category. Only 250 cities achieved the highest accessibility index (≤0.001075/inhab), mostly in the Midwest region (16.1%) (Figure 3; Table 1)

a) ICU: unit care intensive.

Figure 3 Accessibility index for ICU^a beds (A), accessibility index for ICU^a beds exclusively for COVID-19 (B), municipalities with a low accessibility index for ICU^a beds (C) and municipalities with a high accessibility index for ICU^a beds exclusively for COVID-19 (D), Brazil, May 1, 2020 

Regarding exclusive beds for COVID-19, 29% (1,614) of Brazilian municipalities had an accessibility index ≤0.000036/inhabitant and only 5.3% (293) reached the highest index on the scale, that is, ≤0.000219/inhabitant. In the Northern region, which is strongly affected by lack of access, 76.2% (343) of its municipalities had the lowest accessibility indices (Figure 3; Table 1).

With regard to municipalities with accessibility below the national median (0.000204/inhab.), low access to ICU beds was attributed to municipalities in the interior regions areas of the country. This pattern of spatial distribution of access draws attention to the challenge of responding to COVID-19 as the disease spreads to interior regions over time. Taking hot spots selected based on the spatio-temporal clustering of COVID-19 incidence and overlaying accessibility of regular ICU beds revealed that the Midwest and Northern regions both had 44% of their municipalities in hot spots with a care gap for regular ICU beds. Regarding mortality, the regions with the highest number of municipalities in hot spots with a care gap were the Northeast (25%) and the North (21%) (Figure 4).

Figure 4 Hot spots with low access to regular ICU^a beds (A, incidence; B, mortality) and ICU^a beds exclusively for COVID-19 in areas with low access to ICU^a beds (C, incidence; D, mortality), Brazil, May 1, 2020 

With the aim of responding to COVID-19 with due agility, 14,671 ICU beds specifically for COVID-19 were made available throughout Brazil by May 1^st (Table 1). When observing the distribution of these beds and the municipalities in areas with care gaps, it is evident that 28% of these municipalities were benefited. In the Northeast region, 41% of municipalities had specific COVID-19 ICU beds in areas with care gaps, while in the Southeast region 10% of municipalities were in this situation (Figure 4).

Regarding mortality, only 14% of the municipalities in areas with care gaps were covered by new ICU beds exclusively for COVID-19. The Southern region had the highest coverage of this measure: 50% of its municipalities in hot spots areas were covered. In the Northeast region, 19% of its municipalities received ICU beds intended for this specialty, as well as 8% of the municipalities in the Midwest region, 5% in the Southeast, and 4% in the North of Brazil.

Discussion

This study sought to explore the reorganization of the health system regarding tertiary care in response to the COVID-19 pandemic. We were able to identify regions with care gaps not covered by the provision of new ICU beds for COVID-19.

Adequate spatial overlaying of the population’s need for services in relation to service supply structures is an important proxy for access.²⁰ Despite the importance of geospatial models, these approaches are underutilized in analytical efforts to understand patterns of disease spread. A PubMed search in July 2020 found 361 papers using GIS to analyze COVID-19 data and only one paper with a spatio-temporal clustering approach using emerging hot spot analysis.

Although examining spatial clustering patterns, without the use of longitudinal approaches, is useful for the instrumentalization of specific actions, it lacks a broader perspective capable of providing an overview of trends. In this study, the use of the detailed approach sought precisely to overcome this limitation. Space-time cluster analysis allowed us to identify groups of Brazilian municipalities that have been reporting an increase in incidence and mortality.

The methodological steps defined in this study make it possible to qualify the efforts to increase access to ICU beds in response to COVID-19 and, furthermore, to identify regions that have higher needs than others, in order to prioritize, as well as to motivate possible reallocations, based on the principle of solidarity of public entities and the federal pact.

Gravity models (2SFCA), like the one used in this study, enable identification of regions with health assistance gaps. Regions indicated as having low ICU bed availability are marked by peculiarities and additional challenges to access, characterized as geographical barriers.²¹ Unavailability of ICU beds within two hours of traveling time is an additional COVID-19 risk parameter, since in order for an individual with an acute form of the disease to have respiratory support they will have to travel long distances. The existence of care gaps in Brazil is a chronic problem, which relates to the organizational principle of regionalization and hierarchization of the Brazilian National Health System (SUS). With the exception of the South, all other regions of the country have a high number of municipalities located in regions with care gaps. Overlaying regions categorized as COVID-19 incidence and mortality hot spots and regions with low access to regular ICU beds, pointed to fragile situations in the North, Northeast, and isolated portions of the Southeast.

Despite the challenges inherent to a rapid response to COVID-19, a very expressive expansion of ICU beds in Brazil can be seen, beginning in mid-March 2020. In all, more than 14,000 ICU beds were made available exclusively for COVID-19. In order to optimize these scarce resources, actions should be directed to those places characterized as having ICU care gaps. The spatial overlay analyses were able to identify that the new ICU beds made available were not always directed to regions with care gaps. According to the National Health Establishment Registry database, access to new ICU beds in response to COVID-19 was optimized mainly in the Northeast, Southeast and Southern regions. It can be seen that in large parts of the North and Midwest regions, expansion of access to ICU occurred in places that are not hot spots or do not have care gaps. Thus, it can be seen that part of the response efforts to COVID-19 could be better directed, aiming to supply regions with greater challenges resulting from geographic barriers to ICU access.

The world as a whole has faced a shortage of medical equipment, human health care resources, personal protective equipment and respirators.²² Providing evidence of these problems is necessary in order to instrumentalize management decisions with the objective of optimizing scarce resources. In Brazil, although new ICU beds were made available they were unable to provide coverage broad enough to assist municipalities with care gaps and where COVID-19 incidence and mortality was worsening. It is important to discuss the possibility of incorporating new beds for COVID-19 in the SUS on a permanent basis, which could reduce the care gaps historically identified in these regions. The findings of this study demonstrate how spatio-temporal clustering can be a useful tool to help decision making with regard to resource allocation.

This study has limitations. Among them, it is worth mentioning case underreporting, which could modify the grouping patterns of municipalities categorized as incidence and/or mortality hot spots.²³ Another point to be highlighted is the fact that we did not analyze data on human health resources or equipment. Surveillance systems are known to have limitations with regard to the time taken to report cases. Delays in case notification can cause temporal association patterns to change. Finally, the option of using Euclidean distance for estimating the accessibility index is another point worthy of discussion. The diversity in land use patterns in Brazil causes other 2SFCA approaches to present inaccurate results when trying to estimate accessibility in densely populated regions, rural regions, or in the Amazon which is a huge and hard-to-access territory. Therefore, Euclidean distance proves to be the approach which presents fewest inaccuracies in very divergent contexts, as is the case of Brazil.²⁴

Looking to the future, it is important to discuss the need to incorporate, in the routine monitoring and evaluation of the response to COVID-19, dynamic analysis techniques, capable of generating evidence. The versatility of the approach used and defended in this study, means that it can be applied to other scopes and diseases, such as severe acute respiratory syndrome and vaccine-preventable diseases (for redistribution of vaccines and specialized services), during emergencies due to natural disasters.