Study design and data sources

This is an ecological cross-sectional study on hospitalizations for ACSC occurring in adult populations in Brazil and Portugal in 2015. The unit of analysis in this study is the municipality: 5,570 for Brazil and 278 for mainland Portugal. The average size of the municipal units in Brazil is 1,526 km², and the average population was 36,706 (minimum: 813; maximum: 11,967,824; SD: 215,590). The average size of the municipal units in Portugal is 320 km², and the average population was 35,393 (minimum: 1,717; maximum: 504,471; SD: 56,807).

This study used the hospitalization databases provided by the Brazilian Hospital Admissions Information System and the Portuguese Central Administration of the Health System for the year 2015. A total of 11,522,004 and 1,000,670 hospitalizations were registered for Brazil and continental Portugal in 2015, respectively. Both databases are produced to reimburse hospitals and, therefore only cover public hospitals. In both countries, the physicians evaluate the patients and determine the principal and secondary diagnosis code, according to the International Classification of Diseases (ICD) (the 9^th revision for Portugal and the 10^th revision for Brazil). In addition, external auditors frequently check the hospital data bases, to ensure quality and identify potential errors. The data is anonymized and was analyzed according to the municipality of residence of the patient.

Data on PHC supply and the socioeconomic characteristics of municipalities were selected according to the literature and data availability, and the sources were the Brazilian Institute of Geography and Statistics (IBGE), the Brazilian Health System Informatics Department (DATASUS), the Statistics Portugal database (SP), and the Portuguese Central Administration of the Health System (ACSS). Table 2 details the variables used and data sources. The ecological variables were: proportion of people aged 65 years or older in the population, population density, proportion of people living in rural areas, economic level (mean of household income in Brazilian reais for Brazil; relative purchase power with the national purchase power used as reference (= 100) for Portugal), proportion of people with low education, physician supply in FHUs and in PHC centers in general, and population coverage of FHUs (for Brazil, this was the number of family health teams * 3,450/population and, for Portugal, this was the number of users registered at FHUs/population). Primary Health Care data for Portugal was retrieved from the periodic publication on number of patients registered on PHC services [35].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Variables information.

https://doi.org/10.1371/journal.pone.0219262.t002

Definition of hospitalizations for ACSC

The definition of which hospitalizations were avoidable was determined according to the methodology of the US Agency for Healthcare Research and Quality (AHRQ), which identifies prevention quality indicators (PQIs) according to the codes of the principal and secondary diagnoses (AHRQ). This methodology was applied for all admissions of patients aged 18 years and older; it excluded obstetric admissions and transfers from other health care facilities. Cases with missing values for the variables age, sex, diagnosis, and municipality of residency were also excluded. This list has a solid theoretical basis, is periodically revised for inclusion and exclusion of cases, and can be applied for both ICD-9 and ICD-10. The use of a single list allows for comparison between both countries.

Analysis was performed separately for the composite indicators PQI 91 (acute conditions) and PQI 92 (chronic conditions). The acute conditions analyzed by this methodology were bacterial pneumonia, urinary tract infection, and dehydration. The chronic conditions were hypertension, congestive heart failure, chronic obstructive pulmonary disease (COPD) or asthma in older adults, asthma in younger adults, short-term and long-term complications of diabetes, uncontrolled diabetes, and lower-extremity amputation among diabetics. Details on disease codes used and methods of calculation can be found in the AHRQ guidelines [3].

Spatial statistical analysis

Rates of hospitalizations for ACSC were presented as number of hospital admissions per 100,000 people over 18 years, as defined by the AHRQ methodology. Descriptive statistics, percentiles, coefficient of variation, and ratio of variation were used to visualize rates and geographic variation of ACSC rates across Brazil and Portugal for each category of ACSC. Spearman’s correlation was used to assess the relationship between rates of acute and chronic ACSC in both countries.

A spatial scan statistic was employed to identify clusters with higher risk of hospitalizations for acute and chronic ACSC in each country separately. The spatial scan statistic employed is a methodology proposed by Kulldorff [36] to test if the number of cases were randomly distributed across different circular windows or if significant spatial clusters exists, according to the corresponding relative risk (RR). The Poisson model was employed as it deals with a discrete variable (number of hospitalizations). The spatial scan statistic is based on a maximum likelihood ratio for each potential cluster, to test the hypothesis of clustering against the hypothesis of uniformity. One important assumption was the scan through circular window shapes, as there is no evidence of the presence of other specific shapes (default). The maximum spatial cluster size was defined as 20% of the population at risk; this parameter identifies clusters in useful sizes for the development of local strategies. The likelihood p-value for the hypothesis test was estimated using Monte Carlo simulations (999 simulations), as the exact distribution of the test statistic cannot be defined. Kulldorff [36] provides more details on the statistical procedure.

A chi-square test was used to analyze if there was a relationship between clusters of acute and chronic ACSC in each country.

The non-parametric Mann-Whitney U-test was performed to compare if significant differences for the socioeconomic variables and regional PHC quantitative measures existed between areas identified as clusters with high-risk of hospitalization for ACSC and non-cluster areas, for each category of ACSC. The spatial scan analysis was performed using SatScan 9.4 and statistical analysis was carried out using IBM SPSS 21.0.

Key findings

The results of this study show that: (i) there are high variations in rates of hospitalizations for ACSC within and between Brazil and Portugal, with higher variations found in Brazil; (ii) there is a more evident pattern of rates in Portugal (with the northern half of the country presenting higher rates); there is no clear pattern in Brazil, only that the northern region had fewer municipalities identified as high risk clusters; (iii) the differences in PHC supply and socioeconomic characteristics between areas identified as high risk clusters and the rest of each country varied between category of ACSC and between Brazil and Portugal; and (iv) rates and cluster distribution of acute and chronic ACSC had a significant agreement between them for both countries. The results presented here agree with previous studies that indicate that hospitalizations for ACSC vary across geographic units and have different associated factors [8,12–14,19].

Regional variations in distribution of hospitalizations for ACSC, both within and between Brazil and Portugal, indicate a possible difference in the underlying factors associated with avoidable hospitalizations and, consequently, which interventions could be more successful for reducing such admissions. Given the use of hospitalizations for ACSC as a performance indicator, it is expected that the variations between and within countries indicate differences in the accessibility and quality of PHC service delivery. Despite a similar approach to providing PHC and similar ACSC hospitalization composition, Brazil and Portugal have very distinct dynamics with regard to mean values of PHC supply and coverage between critical and non-critical areas. In both countries, areas identified as clusters at high risk of acute ACSC had a lower supply of physicians in FHUs, but for chronic conditions these areas had a higher supply in Brazil and no difference for Portugal.

Some studies in Brazil have found an association between the expansion of FHUs and lower ACSC hospitalization rates (even when controlled for socioeconomic factors) [37–39]. Conflicting results on the association of the impact of FHUs on ACSC were found for different regions of Brazil [40], corroborating the idea of variability of ACSC and associated factors across the country. It is important to emphasize that the choice of methodology used to select ACSC codes leads to differences in the results [41,42]. Previous studies in Brazil used the country-specific list developed in 2009, which includes conditions not considered in this study, such as vaccine-preventable conditions, angina, gastroenteritis, nutritional deficiencies, and cellulitis, among several others [43].

In Portugal, high-risk clusters for acute ACSC had lower coverage of FHUs and lower physician supply compared with non-cluster areas, indicating that the FHUs might be associated with lower rates of avoidable hospitalizations for acute conditions. This difference however, could be due to other unobserved factors that are associated with where the FHUs were implemented. Although the supply of primary care physicians is a notable component of access [44], similarities and differences in other dimensions of PHC between countries and for smaller geographic regions should be explored in future studies.

Previous studies have found that the geographic variation in avoidable hospital admission rates were more associated with socioeconomic and health characteristics of the population than with quantitative measures of PHC supply [10,17]. For Brazil and Portugal, there were significant differences in the mean values of both PHC supply and socioeconomic characteristic variables between critical and non-critical areas. These differences indicate the existence of complex dynamics leading to the variation in rates and existence of critical areas. This complexity makes the cross-country learning more difficult and it impacts the interpretation of ACSC as an indicator for performance assessment.

In Brazil, municipalities belonging to high-risk clusters of acute avoidable hospitalizations presented higher economic levels and education levels than non-cluster municipalities. At first glance, such direction of association seems contrary to what is expected and discussed in the literature [45–47]. However, some studies have found that higher economic and education levels in Brazil are associated with higher rates of hospitalizations in general [48,49]. These studies suggest that, in Brazil, people with higher economic levels have better access to health services, including hospitalizations, either because of their understanding of the health system or their financial situation. Therefore, hospitals are used as the preferential access point to the health system for this socioeconomic group.

In Portugal, municipalities in critical areas for acute avoidable hospitalizations presented lower education levels than non-critical municipalities. Low education may lead to decreased quality of life (due to difficulties in obtaining well-paid employment and accessing goods and services), and can hinder the capacity to manage one’s own health and adopt healthy lifestyle and behaviors [50]. While the effect of education on ACSC hospitalizations in Portugal was the same as found in previous studies [46,51], in Brazil the inverse was found for acute conditions. Whether this is a reflection on PHC and hospital use or associated with other health determinants or health behavior should be explored further.

Municipalities in critical areas had a higher proportion of elderly and lower population density mean in both countries. The former reflects a concerning situation given the ageing of the population globally, especially for Portugal which has one of the largest proportion of elderly in the world [29]. As for the latter, most of the clusters were located in the interior of the countries, while the majority of the Brazilian and Portuguese populations live near the coastal regions. In Portugal, the existing FHUs are also concentrated along this region [21]. The reduced geographical proximity between primary health centers and patients can help explain the inequality between rural and urban areas. Not having a close provider of health services can be considered a barrier to access, because people can postpone seeking help until the condition requires hospitalization [52]. The remoteness of such areas can also be an obstacle to attracting and retaining health professionals [53]. For Portugal, it is important to note that the city of Lisbon (the most populated city in the country, with the fourth highest population density) presented high rates of both types of ACSC hospitalizations and was a high risk cluster on its own. The causes and possible associations of this finding should be studied further.

In Brazil, critical areas for acute conditions had a lower proportion of people living in rural areas. Previous studies have pointed out that the highest percentage of families registered at FHUs was in the rural areas of the country [54], and that accessibility and consolidation of PHC is a challenge in large urban centers [23]. Both Brazil and Portugal have FHUs coverage differential across their territories; therefore, the implementation and development of the PHC reforms were not uniform across each country. Results indicate that the PHC reforms, with similar organizational characteristics in different contexts, did not produce similar results between or within countries.

As for the stratification of ACSC between acute and chronic, the Spearman correlation between rates and the chi-square for the municipalities which belong or do not belong to clusters indicated a significant level of agreement between both categories. Nonetheless, the Mann-Whitney test indicated that the mean values of the ecological variables had contrasting differences between both categories for both Brazil and Portugal. Mostly, studies on hospitalizations for ACSC use this indicator as an aggregate of all the conditions deemed avoidable [17,55]. Results indicate that, although the identification of critical areas may be done using ACSC as an aggregated indicator, it is important to analyze the characteristics of these areas more deeply and separately when designing interventions, because the heterogeneity of mean values of the ecological variables could indicate that factors associated with each category of ACSC can be different. The findings of this study suggest the importance of using hospitalizations for ACSC to assess performance on a national level, while taking further actions to reduce them locally, given the context of each smaller region.

Strengths and limitations

This study used large national databases covering all hospitalizations registered in public hospitals in Brazil and Portugal in 2015, as well as ecological data on different dimensions that can be associated with avoidable hospitalizations. A further strength of the study is represented by the well validated spatial scan approach used, which allows for local health authorities of both countries to identify critical regions to focus on, with important implications for health policy. This methodology can be expanded to other contexts as necessary. The comparison of hospitalizations for ACSC between Brazil and Portugal is a valuable opportunity to analyze variations between two settings with similar PHC organizations and important differences in country areas, demographics, epidemiologic characteristics, and levels of economic development.

One important limitation of this study is that, because of its ecological approach, it is not possible to establish causal relationships between variables. Nonetheless, this approach seems appropriate to analyze ACSC hospitalizations, because some studies recommend that this analysis should be performed at a group level [18,56]. We did not standardize the rates of ACSC hospitalizations, so the composition of populations had impact on results. Although to standardize rates is common practice to compare distinct contexts, we wanted to identify what are the real geographic areas that should receive more detailed attention. We wanted to identify these critical areas in real populations and analyze if they are related to similar characteristics, including ageing. For example, if prevalence of elderly was the only different characteristic between cluster and no cluster areas, it would mean that it was mostly important to improve older people health care. Our study showed that this is not the case. In addition, the ecologic variables were not standardized either.

The use of routinely collected administrative data is another limitation of this study, because the validity of diagnosis can vary according to ICD coding, across diseases, hospitals, and countries. Furthermore, the analysis performed only covers part of the complicated framework of factors associated with ACSC, because other important unobserved variables were not considered in this study.