Structure and work process in primary care and hospitalizations for sensitive conditions

ABSTRACT OBJECTIVE The objective of this study is to investigate whether the characteristics of the structure of primary health units and the work process of primary care teams are associated with the number of hospitalizations for primary care sensitive conditions. METHODS In this ecological study, we have analyzed data of Brazilian municipalities related to sociodemographic characteristics, coverage of care programs, structure of primary health units, and work process of primary care teams. We have obtained the data from the first cycle of the Brazilian Program for Improving Access and Quality of the Primary Care, of the Department of Information Technology of the Brazilian Unified Health System, the Brazilian Institute of Geography and Statistics, and the United Nations Development Programme. The associations have been estimated using negative binomial regression coefficients (β) and respective 95% confidence intervals, with a hierarchical approach in three levels (alpha = 5%). RESULTS In the adjusted analysis for the outcome in 2013, in the distal level, the coverage of the Bolsa Família Program (β = -0.001) and private insurance (β = -0.01) had a negative association, and the human development index (β = 1.13), the proportion of older adults (β = 0.05) and children under the age of five (β = 0.05), and the coverage of the Community Health Agent Strategy (β = 0.002) showed positive association with hospitalizations for primary care sensitive conditions. In the intermediate level, minimum hours (β = -0.14) and availability of vaccines (β = -0.16) showed a negative association, and availability of medications showed a positive association (β = 0.16). In the proximal level, only the variable of matrix support (β = 0.10) showed a positive association. The variables in the adjusted analysis of the number of hospitalizations for primary care sensitive conditions in 2014 presented the same association as in 2013. CONCLUSIONS The characteristics of the structure of primary health units and the work process of the primary care teams impact the number of hospitalizations for primary care sensitive conditions in Brazilian municipalities.


INTRODUCTION
The rate of hospitalizations for primary care sensitive conditions (ICSAP) represents an indicator of access and quality of the primary health care (PHC) 17,21,23 . International studies have shown that factors associated with higher rates of ICSAP are: male gender, senility, living in rural areas, low income, and socioeconomic disadvantage 1,2 .
In Brazil, a cross-sectional study has identified a higher probability of ICSAP in females, children under five years of age, persons with less schooling, history of emergency medical consultation in the previous month, and hospitalization in the previous year 15 . In a cohort of adults, higher rates of ICSAP have been observed in persons over 60 years of age and non-white; however, no gender differences have been identified 8 .
There is evidence that the expansion of the Family Health Strategy (FHS) is associated with a tendency to reduce this indicator 4,7,10,12 . Therefore, although the rates of ICSAP are still high when compared to those of other countries 24 , there was a decrease of more than 5% per year between 2008 and 2014 in Brazil 12 . It is possible that only the increase in coverage of the FHS will not be sufficient for the consistent reduction of this indicator. However, few studies have analyzed the characteristics of the quality of the care involving elements of structure and work process in PHC.
The association of these characteristics with the indicators of ICSAP is not well established. Gonçalves et al. 8 have found no association between quality of services in PHC and rates of ICSAP. A Spanish ecological study has found no significant association with the number of physicians per inhabitant, but it has shown that the physician workload has a significant direct association with sensitive hospitalizations, and it suggests that it is a result of the impairment of the quality of care offered to the patient 13 . Therefore, given the need to better exam the relation between the characteristics of PHC and ICSAP, the hypothesis of this work is that the better the structure and the work process in PHC, the lower the number of ICSAP, even after adjusting for the sociodemographic variables and variables of coverage of care programs.

METHODS
This study is ecological, with population data available on different databases. The study population consisted of 5,565 Brazilian municipalities, 38,812 Basic Health Units (BHU), and a sample of 17,202 primary care teams (PCT) existing in 2012. From these data, aggregate indicators were constructed for the municipal level.
All variables used in this study are described in the Figure The structure data of the BHU came from module I of the first cycle of the External Evaluation of the National Program for Improving Access and Quality of the Primary Care (PMAQ-AB), from on-site observation. The work process in PHC was obtained by interviews with the PCT that were part of the PMAQ-AB. The work process variables were referral to specialist, home visit, and matrix support. The PMAQ-AB data were collected by electronic questionnaire using tablets c . After data collection, the data were validated and sent to the database Hours of operation were considered minimum when the BHU operated at least two shifts a day, five days a week. An indicator was constructed for the municipality calculating the average percentage of BHU with minimum hours. % of BHU in the municipality with availability of at least 75% of the 12 vaccines of the basic calendar The availability of vaccines was estimated by the existence and availability of immunobiological agents advocated by the National Immunization Policy.
The indicator corresponded to the average percentage of BHU that always had more than 75% of the 12 vaccines of the basic calendar. The cutoff point refers to the proportion of adequate coverage a .
% of BHU in the municipality with 80% or more of the medications of the basic component of the National Essential Medicine List (RENAME) The medications of the basic component of the RENAME were verified, in sufficient quantity (antiparasitics, multivitamins and minerals, antiasthmatics, hormones, antihypertensives and of cardiovascular action, antidiabetics, antibacterials, analgesics/antipyretics, treatment and prevention of osteoporosis, antacids/antiemetics/antisecretors). The indicator refers to the sum of medications available in sufficient quantity, divided by the total number of medications in the list, multiplied by 100. The cutoff point refers to the goal of 80% availability of medications b .
% of PCT in the municipality responsible for the user's referral to specialized care Percentage of PCT that, when a user needs to be referred to a specialist, schedules and informs the date of appointment with the specialist to the user.
Module II of cycle 1 of the PMAQ-AB (2012) % of PCT in the municipality with home visit The indicator of home visit for the municipality represents the average percentage of teams that reported visits, taking into account the CHA and other team members.

% of PCT in the municipality with matrix support
The matrix support is a methodology of care management and is a technical and pedagogical support to teams 6  centralized by the Ministry of Health, which later made them available on the website of the Department of Primary Care c .
For the statistical analysis of the data, we used the software Stata 12.0 (Stata Corp., College Station, TX, USA). Initially, three databases were tabulated. The first one corresponded to the data of the municipalities, the second one, to the questions of Module I of the PMAQ-AB, and the third one, to the work process variables of the PCT. Then, the structure and work process variables were aggregated for the municipal level, estimating the average percentage of variables per municipality. The variable used to link the data was the IBGE code.
The descriptive analysis estimated: averages (X) ± standard deviations (SD), as well as medians (Md) and interquartile deviations (Q1 and Q3), in addition to minimum (min) and maximum (max) values.
Univariate and multivariate analyses were performed using negative binomial regression considering municipal population. The regression coefficients (β) and respective 95% confidence intervals (IC95%) were calculated.
For the modeling, we used a hierarchical approach in three levels. The distal level corresponded to the sociodemographic variables and variables of coverage of care programs. The intermediate level presented the structural variables of the BHU in the municipalities.
The proximal level consisted of the characteristics of the work process of PCT (Figure).
In the adjusted analysis, all variables of the distal level were initially included in the model, remaining the variables with p < 0.10. Then, the variables of the intermediate level were included in the model, and we removed the variables of this level with p > 0.10, adjusted for the variables that came from the previous level. The same procedure was repeated for the proximal level, adjusted for the variables of the previous levels.
Alpha of 5% was used as a criterion of statistical significance. To this end, the regression estimates and 95%CI were verified in their corresponding level. All analyses considering  Number of ICSAP the variables of the proximal level (work process) were weighted for unequal selection probabilities, as we used the entire population in the previous levels (municipalities and BHU), but only one sample in the proximal level (by auto-inclusion). As the proportion of adherence of the PCT to PMAQ-AB was different among the Brazilian municipalities, the estimates were weighted by the inverse of the probability of adherence. This variable was constructed as the ratio of total existing PCT by the number of teams that were part of the program in the municipality.
In compliance with the Resolution of the National Health Council, the study was approved by the Ethics Committee of the Universidade Federal de Pelotas (Official Letter 38/12, in May 10, 2012). All persons interviewed signed the informed consent.  Table 1).  Table 1).

Among
The absolute number, rate, and proportion of ICSAP decreased, as shown in Table 2.
The unadjusted analysis for outcome in 2013 and 2014 showed a similar association. There was a significant negative association between the proportion of children under five years of age, coverage of private health insurance, minimum hours, and availability of vaccines. There was a significant positive association between the proportion of persons aged 60 years or over, coverage of BFP, EACS, and FHS, availability of medications, and matrix support (Table 3).
In the adjusted analysis of 2013, in the distal level, municipalities with higher coverage of BFP (β = -0.001) and higher coverage of private insurance (β = -0.01) had a lower number of ICSAP. Municipalities with higher MHDI (β = 1.13), higher proportion of older persons (β = 0.05), higher proportion of children under the age of five years (β = 0.05), and higher coverage of EACS (β = 0.002) showed a higher number of hospitalizations.
In the intermediate level, municipalities with BHU operating at least with minimum hours (β = -0.14) and with at least 75% of available vaccines (β = -0.16) presented a lower number of ICSAP. Municipalities with at least 80% of available medications showed a higher number of hospitalizations (β = 0.16).
In the proximal level, only the variable of matrix support (β = 0.10) showed a positive and significant association, which indicates that municipalities with more teams that receive matrix support have a higher number of hospitalizations (Table 3).
In the adjusted analysis of 2014, the variables of the distal and intermediate level presented the same association as in the adjusted analysis of 2013. The matrix support lost statistical significance only in the proximal level (β = 0.10, p = 0.092) ( Table 3).

DISCUSSION
After adjusting for the sociodemographic variables and variables of coverage of care programs, the variables of structure of the BHU (minimum hours, and availability of medications and vaccines) and the work process of the PCT (matrix support) were associated with the number of ICSAP.
The indicators of ICSAP decreased as it has been found in other studies, which have related this reduction to a better coverage of health services in recent years 7,11,17 . A large dispersion of this indicator was observed in the municipalities. The number of ICSAP presents a significant geographic variation 3 , what makes our study appropriate because of the uncovering of the characteristics of the primary health services associated with the number of ICSAP.
Regarding the care programs analyzed, although the coverage of FHS was not associated with the lowest indicators of ICSAP and the coverage of EACS showed an association with a higher number of hospitalizations, other articles show that the implementation of FHS seems to be associated with reduced ICSAP 11,12 . Dourado et al. 7 show that socioeconomic variables reduce the impact of FHS in reducing hospitalizations.
These strategies are the main sources of access to primary services d ; therefore, we expected to find a lower number of ICSAP in municipalities with higher coverage. Two possible explanations are suggested: first, the coverage of the FHS in the country may be overestimated, and second, the result found may reflect the higher frequency of health problems in the population of municipalities covered by the EACS, since it is aimed at more vulnerable locations with high rates of exclusion in the access to health services and within a disarticulated care network with poor structure 24 , which would justify more hospitalizations. For the moment, it is important to recognize the articulating role of community health agents in primary health care and reinforce the continuing education and supervision activities of their actions to ensure the effectiveness of their practices 5 .
The significant inverse association with BFP suggests that municipalities with a higher percentage of families that are part of income transfer programs have fewer hospitalizations. The descriptive analysis shows that, on average, 80% of families of municipalities with criteria for the monitoring of the BFP are covered by the program, ensuring a better income condition for the poor and extremely poor families benefited. Given the characteristics of the BFP, we highlight that its coverage is an important indicator of social vulnerability, with the population exposed to a diversity of situations of poverty and social inequalities 24 . This population may reside in municipalities with a specialized health services network of poor structure, so that services are not accessed because of a lack of supply, and not because of a lack of need of individuals 24 . Thus, a study suggests that the resolution of socioeconomic problems in poorer areas can impact the number of ICSAP 2 .
As the hospitalizations computed refer only to public hospitalizations, we already expected that municipalities with higher average coverage of private health insurance would have fewer hospitalizations as they are diverted to the private network. This result has been verified. The higher number of hospitalizations in municipalities with a higher proportion of older adults and children under five years of age was also found in another study, which shows higher rates of ICSAP at these extreme ages 15 .
Regarding the indicators of structure of the BHU, we did not identify an association between the ratio of physicians and ICSAP. The association with this indicator is inconclusive in the literature and should be analyzed with care, as it does not reflect the potential number of users per physician 16 . When assessing structural aspects, a systematic review has found an association between longer physician workloads and higher rates of ICSAP 23 . The authors have also identified that the availability of nurses decreases the rate of hospitalization for some diagnoses, such as asthma, and increases the rate for diabetes, although this risk decreases when care is accompanied by a community health agent 23 . There was an association between the minimum hours of operation of the BHU and lower number of hospitalizations. A systematic review shows that most studies find an inverse association between access indicators and ICSAP 15,21 .
Among the hospitalizations that can be avoided, we can mention those resulting from vaccine-preventable diseases. The increase in vaccine coverage has reduced the infant morbidity and mortality of target diseases 9 . Therefore, we can explain the lower number of ICSAP in municipalities with at least 75% of immunobiological agents available, because of the better availability of vaccines.
The availability of medications was the worst indicator of structure. The unavailability of medications impairs the access to this resource and supply is still insufficient and unequal in Brazil, despite the National Essential Medicine List (RENAME) having increased free access 18 . The fragility in the selection and use of RENAME in many Brazilian municipalities can justify this finding and alert managers to invest in pharmaceutical care management 14 .
The adjusted regression analysis showed that the number of ICSAP was higher in municipalities with at least 80% of medications available, although access to them and their rational use are essential to avoid hospitalizations and deaths e . This study shows an increase in the number of hospitalizations from drug side effects and intoxications and shows an increasing tendency of problems related to the use of medications by older adults 21 , but this evidence is not enough to explain the higher number of hospitalizations. We suggest that the indicator of availability of medications is not the most adequate to explain the number of ICSAP, because it does not measure access or use of medications, besides the possibility of reverse causality -first there is the hospitalization and then the need for medication use. In addition, better structured municipalities would have a greater supply of health services and supplies, including medical appointments, medications, and hospital beds. The association observed may result from the possibility of access to hospital beds in municipalities with better structure of health services. It should also be noted that the FHS has great difficulty in asserting itself as a priority model, because of the high complexity supply and hospital coverage offered in these municipalities 24 .
From the characteristics of the work process, the matrix support that the FHS team received was the only one that showed direct association with outcome. This management tool provides specialized support to PCT, but the scarcity of specialized services induces its misuse in a substitute way, which can distort its function 6 .
An important activity of the care team is to guarantee to the users the care of other care points when necessary 22 . As almost all teams (99.6%) reported home visits, the indicator may not have been able to identify associations. However, this result, coupled with the non-association of ICSAP with the referral of the user to a specialist and the opposite association with matrix support, leads us to question the quality of the work process of the teams.
A systematic review 23 provides strong evidence that the continuity of the care decreases the risk of ICSAP at all ages and in different countries. However, it points out the lack of evidence of the positive effect of other organizational aspects of the PHC in ICSAP 23 .
A cohort study in Southern Brazil has concluded that primary services with less than ideal quality appear to have no impact on ICSAP. The authors suggest that, in order to reduce ICSAP, investments need to be made on the quality of the PHC and the reorganization of the work process, effectively turning the PHC into the coordinator of the network 8 .
This study had some limitations. The regression analysis was not stratified by age group for ICSAP or by type of hospitalization. However, it was adjusted for the proportion of extreme ages. However, we suggest future stratified analyses to better understand the relations in question. Data from the Health Information Systems have weaknesses both in relation to the hospitalization diagnosis and the coding of diagnosis using the ICD. However, because they present a good proportion of valid information, they are considered as important databases 20 .
The data were collected by previously trained evaluators with validation and checking strategies. In spite of this, the collection was made from the observation of BHU in module I and it may present problems of standardization. The data of module II were collected from interviews with a professional of the PCT, a situation that can lead to bias as it was an evaluation process.
Questions related to work process may be overestimated, since adherence to the PMAQ-AB may be conditioned by a better situation of the health teams. In an attempt to balance the difference in adherence among municipalities, we weighted the variables of the proximal level using the variable weight in the regression analysis. For the municipal and BHU variables, there was no selection bias, since data were collected for the entire study population.
The study had a large sample, which ensures greater precision of the estimates. Negative binomial regression allows the adjustment of the number of hospitalizations by the size of the population of the municipality, and the hierarchical regression analysis in three levels allowed the reduction of confounding bias.
An ecological study does not allow inferences at the individual level. The interest of the ecological effect in this study is to verify if ICSAP result from interactions between the sociodemographic factors and the factors of coverage of care programs, structure, and work process of the PHC.
There are apparently contradictory results, such as the association of the lower number of ICSAP with the indicators of "municipalities with at least 80% of available medications" and "municipalities with higher population with BFP". However, the indicators indicate municipalities with infrastructure problems in the supply of services and materials (medications). If municipalities do not offer these resources and have few hospital beds, consequently, there will be fewer hospitalizations, because persons will have more barriers to overcome and to obtain access to these resources (medications and hospitalizations) in places farther away from where they live. Thus, the findings are plausible.
We conclude that the characteristics of the structure of BHU and the work process of PCT impact the number of ICSAP in Brazilian municipalities. Investments aimed at the qualification of actions in PHC will potentially reduce ICSAP.