Journal of Epidemiology and Global Health (2013) 3, 175–185

http : / / www.elsev ier .com/ locate / jegh

Impact of the Family Health Program ongastroenteritis in children in Bahia, NortheastBrazil: An analysis of primary care-sensitiveconditions

Laura J. Monahan a,b, Gregory S. Calip b,c,d,*, Patricia M. Novo b,e,Mark Sherstinsky b,f, Mildred Casiano b, Eduardo Mota g, Ines Dourado g

a Division of Pediatric Critical Care, New York University School of Medicine, 462 1st Avenue, Suite 8S7-8,New York, NY 10016, United Statesb New York University, Global Institute of Public Health, 240 Greene Street, 2nd Floor, New York,NY 10003, United Statesc Department of Epidemiology, University of Washington School of Public Health,University of Washington, Box 357263, Seattle, WA 98195, United Statesd Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, 1100 Fairview AvenueNorth, M4-B402, Seattle, WA 98109, United Statese Department of Psychiatry, New York University School of Medicine, 423 East 23rd Street, Suite 17015W,New York, NY 10010, United Statesf State University of New York College of Optometry, 33 West 42nd Street, New York, NY 10036,United Statesg Institute of Collective Health (Instituto de Saude Coletiva), Federal University of Bahia (UFBA),Rua Basılio da Gama, Campus Universitario Canela, Cep: 40.110-040, Salvador, BA, Brazil

Received 16 January 2013; received in revised form 4 March 2013; accepted 7 March 2013Available online 13 April 2013

22ht

No

nom

KEYWORDSFamily Health Program(Programa Saude daFamılia; (PSF); NationalUnified Health System(Sistema Unico de Saude;(SUS); Pediatric gastroen-teritis; Primary care

10-6006/$ - see front matp://dx.doi.org/10.1016/

* Corresponding author. Arth, M4-B402, Seattle, WE-mail addresses: laura.vo@nyumc.org (P.M. Novaines@ufba.br (I. Dourado

tter ª 201j.jegh.201

ddress: CA 98109-10monahan@o), sherstin).

Abstract In seeking to provide universal health care through its primary care-ori-ented Family Health Program, Brazil has attempted to reduce hospitalization ratesfor preventable illnesses such as childhood gastroenteritis. We measured rates ofPrimary Care-sensitive Hospitalizations and evaluated the impact of the FamilyHealth Program on pediatric gastroenteritis trends in high-poverty Northeast Brazil.We analyzed aggregated municipal-level data in time-series between years 1999–2007 from the Brazilian health system payer database and performed qualitative,in-depth key informant interviews with public health experts in municipalities in

3 Ministry of Health, Saudi Arabia. Published by Elsevier Ltd.3.03.002

ancer Prevention Program, Fred Hutchinson Cancer Research Center, 1100 Fairview Avenue24, United States. Tel.: +1 206 667 3209; fax: +1 206 667 7850.nyumc.org (L.J. Monahan), gcalip@fhcrc.org, calip@u.washington.edu (G.S. Calip), patricia.sky@gmail.com (M. Sherstinsky), mc3327@gmail.com (M. Casiano), emota@ufba.br (E. Mota),

Open access under CC BY-NC-ND license.

176 L.J. Monahan et al.

Bahia. Data were sampled for Bahia�s Salvador microregion, a population of approx-imately 14 million. Gastroenteritis hospitalization rates among children aged lessthan 5 years were evaluated. Declining hospitalization rates were associated withincreasing coverage by the PSF (P = 0.02). After multivariate adjustment for garbagecollection, sanitation, and water supply, evidence of this association was no longersignificant (P = 0.28). Qualitative analysis confirmed these findings with a frameworkof health determinants, proximal causes, and health system effects. The PSF, withother public health efforts, was associated with decreasing gastroenteritis hospital-izations in children. Incentives for providers and more patient-centered healthdelivery may contribute to strengthening the PSF�s role in improving primary healthcare outcomes in Brazil.

ª 2013 Ministry of Health, Saudi Arabia. Published by Elsevier Ltd. All rightsreserved.

1. Introduction

The Brazilian health care system has become oneof the world�s most ambitious and celebrated mod-els of primary health care (PHC), constitutionallyreforming its health system to attempt to provideuniversal health coverage along the lines of thePHC elements defined by the World Health Organi-zation (WHO) emphasizing equity, community par-ticipation, integration, shared financing among thedifferent levels of government, participation bythe private sector, and decentralization of healthgovernance from the federal to the municipal level[1,2].

Through its National UnifiedHealth System (Siste-ma Unico de Saude, or SUS), Brazil launched theFamily Health Program (Programa Saude da Famılia,or PSF) in 1994. The PSF provides a broad range ofprimary care services through local family healthteams, which include at least one physician, onenurse, one nurse assistant, and four communityhealth agents (agentes comunitario de saude, orACS) [3]. Each team is assigned to a designated geo-graphical area and is responsible for enrolling andmonitoring the health status of up to 3500 people liv-ing in its area, providingprimary care, health promo-tion and education services, and making referrals toother levels of care. As of 2009, 27,324 PSF teamscovered 98 million people (50% of Brazil�s popula-tion) [4]. The program�s participatory health promo-tion focus attempts to reduce the progressiontoward expensive hospital-based care.

The coverage goal of the SUS faces seriousobstacles from the volume of care needed toachieve universal access as well as the cost to Bra-zil�s health financing system. Given considerableinvestments in the PSF program to date, therehas been little research into the extent to whichthe program is associated with changes in healthstatus while controlling for other variables knownto affect health [5–9].

With the revitalized interest in PHC, there hasbeen a growing amount of research internationallyin the last two decades on indicators measuringPHC effectiveness. One such indicator, AmbulatoryCare Sensitive Conditions (ACSC), developed in1990, refers to a category of diagnoses for whichtimely and effective outpatient care can help re-duce the risks of hospitalization by either prevent-ing the onset of an illness or condition, controllingan acute episodic illness or condition, or managinga chronic disease or condition [10]. ACSC as indica-tors of access and quality of primary care were firststudied in the United States [10] and later in Can-ada [11] and Spain [12]. Several studies have shownthat high rates of hospitalization for ACSC are asso-ciated with gaps in coverage of health servicesand/or deficient PHC [13–15]. The Brazilian Minis-try of Health developed and validated a list of con-ditions sensitive to primary health care and isreferred to as ‘‘primary care sensitive conditions’’or internacoes por condicoes sensıveis a atencaoprimaria (ICSAP). The final ICSAP list was basedon systematic literature review, workshops withresearchers, physicians, and health managers, peerreview by the Brazilian Society for Family and Com-munity Medicine and an official public commentperiod via Internet [16].

A common ICSAP in children in Brazil is gastroen-teritis, which can lead to severe dehydration anddeath. The mortality rate for children aged lessthan 5 years in Brazil was 34 deaths per 1000 in2004; approximately 12% of those deaths wereattributable to diarrheal diseases (not includingneonatal deaths) [17]. Mortality and prevalenceof diarrheal diseases are directly related to socio-economic status [18,19]. Several studies in Brazilhave indicated that good hygiene, improvementsin sanitation, and disposal of garbage have had sig-nificant impacts on the rate of diarrheal illnesses[20,21]. Given the well-known, measureable fac-tors associated with this ICSAP, we examined the

Impact of the Family Health Program on gastroenteritis in children 177

possible association between PSF coverage and de-clines in gastroenteritis hospitalization rates inchildren.

The purpose of this study was to examine the im-pact of the PSF on hospitalization rates for gastro-enteritis in children aged less than 5 years.Rotavirus vaccination programs for infants lessthan 16 weeks of age beginning in 2006 have beenassociated with declines in childhood mortalityand hospitalization rates in Brazil [22]. Therefore,we examine years 1999–2007 before rotavirus vac-cination would influence gastroenteritis hospital-ization rates appreciably in this population. ThisICSAP was evaluated through mixed quantitativeand qualitative measures as a surrogate for the ef-fect of primary care provision in Salvador, Bahiaand its surrounding municipalities.

2. Materials and methods

Our mixed methods approach was utilized in ascer-taining the outcomes of interest to (1) corroborateor test the consistency of findings across differentmethods; (2) elaborate or interpret findings col-lected using the other method; (3) design our ana-lytic plan; and (4) explore theories or hypothesesbased on differing or inconsistent findings usingone or both methods. An adapted conceptualframework guided our approach (Fig. 1).

The geographic area of interest for this studywas the state of Bahia�s Salvador microregion

Figure 1 Adapted conceptual framework: det

including the municipal capital, Salvador, Laurode Freitas, Simoes Filho, Candeias, Camacari, Diasd�Avila, Sao Francisco do Conde Madre de Deus,Itaparica, Vera Cruz, and surrounding municipali-ties of Sao Sebastiao do Passe and Mata de SaoJoao. These 12 municipalities (of 417 in the stateof Bahia) represent a population of approximately14 million. The institutional review board and eth-ics committee of NYU and the ISC-UFBA approvedthis research study.

2.1. Quantitative analysis

We conducted multivariate longitudinal analysesusing panel data for years 1999–2007 to measurecoverage by the PSF and corresponding rates ofpediatric hospitalizations for gastroenteritis. Dataon individual hospitalizations with or without PSFteam exposure would be preferred, although nosuch dataset exists containing the necessary vari-ables per region and year. Hospitalization ratesper 10,000 for gastroenteritis (ICD-10 codes:A02.0, A08, A09, J10.8, J11.8, K52.) in childrenaged less than 5 years were characterized by placeof family residence.

The main data source was the SUS HospitalAdmissions Information System (SIH–SUS) [23].These data are produced for the purpose of reim-bursing hospitals through the SUS and are madepublicly available by the Ministry of Health. TheSIH–SUS files were imported for each state

erminants of hospitalization for ICSAP [32].

178 L.J. Monahan et al.

monthly for each year of the study period [24].These files contain detailed information regardingthe cause of hospital admission (ICD 10 codes),the age, sex and place of residence of the patientfor all SUS hospital admissions. The year of admis-sion was defined by the date of hospital discharge.Population data for calculation of hospital admis-sion rates came from the 2000 census and includedprojections for 1999 and from 2001 to 2007 [25].We used the Brazilian government definition ofPSF enrollment, number of health teams in eachstate by average number of enrollees per team, di-vided by the total yearly state population and ex-pressed as a percentage, used as quintiles [26,27].

Other independent variables known to influenceICSAP include socioeconomic conditions measuredby indicators such as proportion of population withadequate water or sanitation [28]; developmentconditions represented by the human developmentindex (HDI) [29] and proportion of population overage 15 years who are illiterate [23]; and health ser-vice indicators including number of physicians per10,000 [23,28,29].

Fixed effect specification was indicated to cor-rect for serial correlation of repeated measures,accounting for known variability within and be-tween municipalities over time. Regression model-ing was used to control for unobservable municipal-level characteristics [30]. Gastroenteritis hospital-ization rates were calculated over time for individ-ual municipalities and the aggregate. Modeling wasbased on stepwise inclusion of exposure variablesfor regression analyses as well as a priori hypothe-ses. Analyses were performed using Stata Software(StataCorp. 2007. Stata Statistical Software: Re-lease 10. College Station, TX: StataCorp LP.).

2.2. Qualitative analysis

Key informant interviews were performed in orderto further describe and examine findings analyzedfrom quantitative data sources. Key informantswere identified with the assistance of ISC-UFBA inBahia. Thus, subjects for these interviews were se-lected by purposive, convenience sampling toachieve descriptive qualitative data on the healthsystem. Interview formats for all subjects werein-depth and open-ended, guided by questions ina semi-structured individual interview guide thatwas linguistically relevant, and approved by insti-tutional review boards at NYU and ISC-UFBA. A to-tal of 15 key informant interviews were held inprivate or semi-private designated meeting loca-tions, working with translators as necessary. Noincentives were offered to key informants, and norisks were posed to participants.

Interviews were voice recorded with the consentof all individuals present. All voice recordings andinterviews were transcribed and translated fromPortuguese and/or Spanish to English when applica-ble. Coding of responses to questions was exploredindependently, and elucidated descriptive data tobe analyzed using an adapted matrix tool [31]. Fil-tering of data in grounded theory was iterative,occurring first in transcription, dual translation bybilingual research personnel if necessary, and final-ly coding independently for thematic analysis.

Strategies to address the limitations and subjec-tivity of qualitative research were employed. Toheighten the fidelity and credibility of this study,multiple strategies for rigor in the qualitativemethodology were used [31]. Partnership betweeninstitutional partners was collaborative and long-standing in an attempt to ameliorate reactivityand respondent bias. Also, prolonged engagementamong investigators nurtured an environment ofpeer debriefing and member checking to ensureongoing agreement of the study implementation.Lastly, use of a conceptual framework preserveda clear auditing and decision trail.

3. Results

Descriptive statistics (Table 1) show hospitaliza-tions from 1999 to 2007 for gastroenteritis in chil-dren aged less than 5 years declined by 39%overall for the region, although rates varied consid-erably within individual municipalities. Expansionin provision of primary care was described by aninefold increase in PSF coverage. Where availabil-ity of hospital beds increased, the number of doc-tors per capita decreased during this period.Changes in socioeconomic indicators were mixed.Improvements were made in the HDI, proportionof poor population, and decreasing illiteracy rate.Other factors, including urbanization and incomedisparity (GINI index), were shown to have aslightly increasing trend but were mostly stableover the period. Notably, considerable increasesin the availability of clean water, sanitation, andgarbage collection services were observed.

Rates of hospitalization for gastroenteritis acrossmunicipalities were variable in distribution acrossthe region (Fig. 2). Trends in hospitalizations per ca-pita by municipality were variable, but in generalshowed decreasing hospitalization rates among chil-dren aged less than 5 years for gastroenteritis (Ta-ble 1). When not assuming equal variancesbetween the individual municipalities, the meanhospitalization rates for gastroenteritis among chil-dren aged less than five years were not significantly

Table 1 Descriptive characteristics. Hospitalizations for gastroenteritis, children aged less than 5 years (per 10,000population).

Mean hospitalization rates 1999 2007 Change in mean 1999–2007 (%)

Mean SD Mean SD

1999 2007 Change in mean 1999–2007 (%)

Bahia (12 municipalities) 599.31 524.05 362.35 356.14 �236.96 �39.54

Hospitalization rates by municipality Change in rate 1999–2007Camacari 181.70 193.72 12.03 6.62Candeias 163.86 394.09 230.23 140.50Dias d�Avila 40.31 169.38 129.06 320.14Itaparica 1240.04 1271.57 31.54 2.54Lauro de Freitas 536.10 69.72 �466.38 �86.99Madre de Deus 404.04 380.23 �23.81 �5.89Mata de Sao Joao 1010.10 308.55 �701.55 �69.45Salvador 1146.32 169.46 �976.86 �85.22Sao Francisco do Conde 65.62 204.20 138.58 211.20Sao Sebastiao do Passe 595.39 359.80 �235.59 �39.57Simoes Filho 213.81 201.73 �12.08 �5.65Vera Cruz 1594.46 829.99 �764.48 �47.95

1999 2007 Change in mean 1999–2007 (%)

Mean SD Mean SD

Independent and control variablesPSF coverage (%) 7.30 16.21 66.21 33.28 58.90 806.41Hospital beds (per 1,000) 663.25 2074.89 844.67 2597.93 181.42 27.35Illiteracy rate (per 10,000) 14.12 4.30 9.14 3.02 �4.98 �35.27Population with access to permanentclean water supply (%)

78.66 13.03 93.96 9.77 15.29 19.44

Population with access to permanentsewage and sanitation services (%)

54.32 19.11 75.87 30.54 21.55 39.68

Population with access to permanentgarbage collection services (%)

71.84 14.34 95.52 6.85 23.68 32.97

Human development index (HDI) 0.54 0.33 0.60 0.36 0.06 11.34Urbanization (%) 89.43 9.08 92.60 8.75 3.16 3.54GINI index 0.45 0.28 0.46 0.28 0.01 2.58Population poor (%) 47.63 8.46 42.00 7.74 �5.63 �11.83Number of doctors (per capita) 0.26 0.67 0.24 0.70 �0.01 �5.21

Impact of the Family Health Program on gastroenteritis in children 179

different (P = 0.766). This apparent variation be-tweenmunicipalities was further illustrated by bothhigh PSF coverage and greatest hospitalizationsbeing linked together (rather than being inverselyproportional) in municipalities such as Vera Cruz,Mata de Sao Joao and Sao Sebastiao do Passe. Othermunicipalities of Bahia tended to follow the hypoth-esis that as PSF coverage increases gastroenteritishospitalizations will decrease.

PSF coverage is described in regression models(Table 2) that control for within municipal varia-tion of socioeconomic factors with fixed effectsspecification. Coverage by PSF was significantlyassociated with decreases in hospitalization rateswith adjustment for increases in the HDI. However,

after accounting for increasing clean water avail-ability, permanent sanitation, and garbage collec-tion services (Model 3) this association was nolonger statistically significant. Thus, this measureof primary care provision was not shown to be a sig-nificant predictor independent of the effects ofwater availability, sanitation, and garbagecollection.

Qualitative results do suggest some effect of PSFon health outcomes in the region. Results aregrouped using domains identified in the adaptedconceptual framework (Fig. 1). Health determi-nants affecting gastroenteritis included factors inthe physical and political environments, demo-graphics, and socioeconomic conditions (Fig. 3).

Figure 2 Rates of gastroenteritis hospitalizations in children ages <5 years per capita, 1999–2007.

180 L.J. Monahan et al.

3.1. Environment

A number of key factors surfaced in both physicaland political environments. One major themeemerging from key informant interviews in termsof physical environment was that decreases in hos-pitalizations for gastroenteritis were due toimprovements in sanitation, not health systems.‘‘This reduction [of gastroenteritis] is attributableto the sanitation program not the PSF. In Salvadorthe cooperation with the PSF is not high.’’ Anxietywas expressed over urbanization and its effect onhealth. When asked about living conditions, an-other informant remarked, ‘‘poor families haveterrible condition of health and poor living condi-tions as we have a lot of slums.’’ An unexpectedtheme emerging from several interviews was theeffect of the political environment on PSF cover-age. Management of the PSF is decentralized anddirected by each municipality. Thus, local politi-cians� priorities affect implementation of the PSF.In Salvador, there have been delays and difficultiesin increasing PSF coverage partly due to changes inthe political arena and prioritization of PSFimprovements.

3.2. Socioeconomic conditions

Salvador is regarded as a resort town in NortheastBrazil but contains a considerable amount of pov-

erty. Compelling contrasts between luxury hotelsand widespread slums were observed. Some keyinformants remarked that poor home conditionsmake health less of a priority. Parents delay carebecause the PSF is unavailable after working hoursor they cannot afford a private physician. Regard-ing literacy and education, some informants com-mented that patients do not always understandinstructions given by the PSF agents and may notreturn to clinics for follow-up. One informant ob-served, ‘‘some patients need non-verbal instruc-tions (e.g. pictures) to understand PSFdirections.’’

3.3. Proximal causes women�s health andparental health literacy

In Bahia, pregnant women, particularly first-timemothers, are provided with a vast amount of infor-mation. Key informants described this situation asa ‘‘key entry point’’ for accessing health servicesthrough the PSF. While this point seems clear,observations and discussion with researchers andhealth professionals in Salvador divulged the com-plexity of navigating the health system by parents.Unavoidable in conversations of parental health lit-eracy, key informants described how socioeco-nomic factors affected health-seeking behaviorsfor ICSAP among parents. One key informant sta-ted: ‘‘They only bring the children to the doctor

Table 2 Family Health Program (PSF) Coverage. Fixed effect regression modelsc for Bahia municipalities, 1999 to 2007.

Model 1a Model 2c Model 3Coefficient Coefficient Coefficient

Coverage by PSF (SE)b �3.07 �3.12 �1.83(1.14) (1.15) (1.60)

Constant �1792.86 �1211.50 �2565.02(2503.20) (2430.06) (3210.28)

Observations 12 12 12Municipalities 12 12 12R-squared (within) 0.11 0.12 0.30P-value 0.021 0.020 0.278bRobust standard errors are in parentheses. All models are adjusted for number of hospital beds per 1000 population, illiteracyrate, urbanization, GINI index, proportion of poor, and number of doctors per capita.cModel 1: Adjusted specification for PSF coverage.Model 2: Adjusted specification for PSF coverage, controlling for HDI.Model 3: Adjusted specification for PSF coverage, controlling for HDI, permanent sanitation structure, clean water availability, andgarbage collection service.a Significant at P < 0.05.

Impact of the Family Health Program on gastroenteritis in children 181

when the condition worsens... to avoid not skip-ping work or being late.’’ Another key informantdescribed the problematic factors related to theobserved, apparent increase in hospitalizationsfor some ICSAP as failures to address access tocare, stating: ‘‘Good health for the population isbeyond no symptom of illness... the strategy isto improve access to care... and this change is stillin its infancy in this city.’’

3.4. Health system

Most key informants had opinions about the healthsystem overall, as well as the PSF specifically. Atstate and municipal levels, shortage of doctorswilling to work in the PSF was cited as a major rea-son why implementation was not more pervasive.Upstream issues, including medical school culture,result in fewer doctors engaged early on in becom-ing primary care professionals. One PSF doctornoted, ‘‘In my class at medical school, only 1 per-son [of 200] wanted to be a primary care doctor.’’Even those who are attracted initially tend to leaveearly on. Informants noted that about one-half ofPSF teams changed doctors within 1 year and fewremained unchanged after 2 years. Reasons givenfor the high turnover included poor benefits, lackof job security, longer working hours (versus pri-vate or specialty practice), and lower perceivedprofessional status.

3.5. Primary health care

Informants described gastroenteritis among chil-dren as ‘‘the focus of the PSF’’, given ‘‘it is the

easiest to act on.’’ The ACS has the primary promo-tion and prevention role. They are required to livein the community in which they work and these areseen as good positions. ‘‘The ACS is a key figurewho has shown to be having much effect in Bahia.’’Other programs were also described to have an ef-fect. Disease-specific public information does notappear to be widespread, however. One informantnoted, ‘‘[for] immunization, family planning andsexuality there is information, but about diseaseI don�t think so.’’

4. Discussion

The overall goal for this project was to assess theeffect of the PSF on hospitalizations in childrenaged less than 5 years in the context of the ICSAP,gastroenteritis. According to the analysis, gastro-enteritis hospitalizations trended downward be-tween 1999 and 2007 in the 12 municipalitiesexamined in Bahia. Using the conceptual frame-work (Fig. 1), there were a number of factors inaddition to PSF contributing to hospitalizationrates. From a quantitative perspective, factors re-lated to permanent sewage systems and garbagecollection were significantly associated withdecreasing hospitalizations for gastroenteritis,whereas after controlling for these factors PSF cov-erage was not. These negative correlations werefound to remain significant when controlling forfactors including HDI, urbanization, and the pro-portion of poor population in a given municipalunit. Private/permanent water supply was not sig-nificantly associated with gastroenteritis hospital-ization rates in regression models, although other

W

Figure 3 Matrix of topics, themes, and sub-themes. Qualitative analysis from key informant interviews ongastroenteritis.

182 L.J. Monahan et al.

Impact of the Family Health Program on gastroenteritis in children 183

studies describe water sources� greater impact onmitigating burden of gastroenteritis among chil-dren [32].

Qualitative analysis and the literature reviewwere both strongly supportive of sanitation systemimprovements as the primary reason for the de-crease in gastroenteritis hospitalizations. ThePSF, however, has had an impact by providing fam-ilies with education on recognition of symptoms,oral rehydration therapy, and use of proper hand-washing techniques in preventing the spread ofinfection. Health care prevention activities havebeen disseminated through radio, communityhealth fairs, in clinic waiting rooms and by agentsin families� homes.

The paradigm shift described by officials in thePSF has increased the level of health awareness,but not necessarily changed conditions predispos-ing poorer families to delay care, failing to see aprimary care provider early, and resulting in childhospitalizations. Anecdotal evidence from qualita-tive interviews described poor individuals and fam-ilies in Bahia spending an entire month�s salary topurchase health services for their sick child be-cause of both an aversion to being seen as ‘‘lower’’socioeconomic status and concerns for having theirchildren�s needs met, such as being seen in a timelymanner and when parents are available during latehours. Despite such concerns, the PSF also providesa place for acute medical care outside of the hos-pital setting, offering more preventive care. Inmunicipalities with limited access to care and lowPSF coverage, such as in Salvador, the PSF mostlikely has less of an impact than in those with bet-ter coverage. Further, larger studies using time-series analysis of the Northeast region of Brazil orall Brazilian municipalities covered by the PSF havedescribed a robust effect on childhood gastroenter-itis and other ICSAPs. For instance, in an analysis ofaggregated data from Brazil�s 558 microregionsimportant variation was present in hospitalizationrates for ICSAPs stratified by municipalities withpredominantly private or nonprofit hospital beds[33].

Alternatively, recent studies of gastroenteritishospitalization rates associated with national pro-grams for rotavirus vaccination in infants. Follow-ing the initiation of routine vaccination in 2006,significant declines were observed in rates of agesless than 5 years diarrhea-related mortality andhospital admissions in Brazil, with most pro-nounced decreases among children less than age2 years where vaccine exposure was most preva-lent [22]. The stabilization of gastroenteritis fol-lowing control of sanitation and hygiene-relatedfactors along with evidence of subsequent declines

after initiation of routine rotavirus vaccinationlends support to global and national recommenda-tions for vaccination of infants as an importantcomponent to control this primary care-sensitivehospitalization.

SIH–SUS is a limited dataset; only informationfrom the public payer system is included. Less than20% of the population in Bahia has private healthinsurance; however this does not preclude themfrom coverage by the PSF. This makes it challengingto look broadly at public health in the region andmake comparisons between outcomes in the privateversus the public systems. The data are aggregatedat the municipal level, making it difficult to inferindividuals� risk (ecological fallacy). The datasetfor our particular investigation did not include somevariables of interest, such asmortality due to gastro-enteritis, all relevant disease codes, and data weremissing or incomplete for some included variables(e.g. number of nurses per capita). Also, greater var-iation in rates in some municipalities can be de-scribed qualitatively as a function of changes inaccess to care or clinic infrastructure and controlledin regression analyses; validation of reporting to SUSprompted us to analyze allmunicipalities versus onlythose with stable rates, using fixed-effect specifica-tion to account for this variability. Despite this spec-ification, it is possible that residual confounding byimportant covariates that are unknown or unmea-sured exists in our model.

Qualitative data were limited to only a smallpurposive sample. Although information obtainedwas rich in content, such sampling may have con-tributed to a selection bias since all of those inter-viewed were or had been at an administrative levelin the health system with the exception of two phy-sicians. Language and cultural barriers may havecontributed to misunderstanding or misinterpreta-tion of key informant comments. Exposure to moreclinics would have benefitted the study in terms ofadditional direct observations and increasing thenumber of interviews. The study also has limitedgeneralizability due to the small geographic areaexamined relative to all of Brazil.

With the use of these ecological data and ourqualitative findings, this analysis is limited toanswering scientific questions specific to the Salva-dor microregion only. Thus, our study is intendedto be hypothesis-generating rather than providinga basis for causal inference. Our investigation addsto the current literature describing associations be-tween PSF coverage and ICSAPs, including child-hood gastroenteritis hospitalizations, by providinga qualitative context (conceptual framework) tocovariates typically only measureable at an eco-logic level.

184 L.J. Monahan et al.

5. Conclusions

We conclude that public health efforts in the Salva-dor microregion, particularly those pertaining toinfrastructure for sanitation and hygiene, ratherthan coverage by the PSF explain greater variationand improvement in hospitalization rates for gas-troenteritis in children. Direct inputs into thePSF, such as incentives for provider entry and morepatient-centered, convenient working hours, couldcontribute to strengthening the PSF�s role inimproving primary health care outcomes in Brazil.

Competing interests

None.

Ethics approval

This study was conducted with the approval of theinstitutional review board at New York Universityand the ethics committee at Universidade Federalda Bahia.

Contributorship

All authors (L.J.M., G.S.C., P.N., M.S., M.C., E.M.and I.D.) were involved in the study design.G.S.C. and M.S. developed epidemiologic causalmodels. G.S.C. performed the statistical analysisand is the corresponding author for the paper.L.J.M. wrote the initial draft, and G.S.C., P.N.and M.S. assisted with the formatting and editingof the manuscript. All authors participated in theinterpretation of the results, revising of the manu-script for important intellectual content and ap-proval of the final draft.

Financial support

Funding for research in the New York University,Global Institute of Public Health was granted bythe Macy Foundation. G.S.C. is supported by theNational Institutes of Health Cancer PreventionTraining Grant in Nutrition, Exercise & Geneticsat the University of Washington and Fred Hutchin-son Cancer Research Center (R25CA094880).

Acknowledgements

We are grateful to the following individuals and for theirassistance and support: The Macy Foundation; at Institu-to de Saude Coletiva-UFBA, Barbara Laisa Alves Mouraand Renata Castro da Cunha; New York University Global

Public Health faculty, Kristin Bright and James Macinko,and coordinator, Amy Joyce; Data Services and Support,Frank Lopresti, Mashfiqur Khan, Melissa Reese, andHimanshu J. Mistry; Language interpreters and transla-tion, Jose Tadeu Tramontini Filho, Rita Soto, and SolangeFigueiredo; Our colleagues at New York University, espe-cially Ana Krieger. We are thankful to the editor and tworeferees whose constructive comments have helped us toimprove the presentation of the paper.

References

[1] Title VIII-social order; chapter II-social security; section II-health. In: Brazil FRo, editor. Title VIII. Federal Senate,Brasilia, Brazil; 1988.

[2] de Andrade L, Pontes R, Martins Junior T. A descentral-izacao no marco da Reforma Sanitaria no Brasil. Rev PanamSalud Publica 2000;8:85–91.

[3] Programa saude da famılia. Revista Brasileira de Saude daFamılia, Brazilian Ministry of Health 2002;2:7–17D.

[4] Departamento de Atencao Basica, Sistema de Informacaode Saude da Familia. Available from: <http://dtr2004.sau-de.gov.br/dab/abnumeros.php> - numeros; 2007 [accessed24.11.09.]

[5] Escorel S, Giovanella L, Mendonca MH, et al. Avaliacao daImplementacao do Programa Saude da Famılia em DezGrande Centros Urbanos. Sıntese dos Principais Resultados.Brasilia: Ministerio da Saude, Secretaria de Polıticas deSaude; 2002.

[6] Viana A, Pierantoni C. Indicadores de Monitoramento daImplementacao do PSF em Grandes Centros Urbanos.Brasilia: Brazilian Ministry of Health, Department ofPrimary Care; 2002.

[7] Avaliacao normativa do Programa Saude da Famılia noBrasil. Sectaria de Assistencia a Saude. Brasılia: Departa-mento de Atencao Basica, Ministerio da Saude; 2004.

[8] Conill E. Primary care policies and health reforms: anevaluative approach based on an analysis of the FamilyHealth Program in Florianopolis, Santa Catarina, Brazil,1994–2000. Cad Saude Publica 2002;18(Suppl.):191–202.

[9] Serra R. Uma avaliacao empırica do impacto saude dafamılia sobre a saude infantil no estado de Sao Paulo.Trabalho submetido ao Premio Economia da Saude. Brası-lia, DF: IPEA; 2005.

[10] Billings J, Zeitel L, Lukomnik J, Carey T, Blank A, NewmanL. Impact of socioeconomic status on hospital use in NewYork City. Health Affairs 1993. Spring 1993;12:162–73.

[11] Brown A, Goldacre M, Hicks N, et al. Hospitalization forambulatory care-sensitive conditions: a method for com-parative access and quality studies using routinely col-lected statistics. Can J Public Health 2001;92:155–9.

[12] Caminal J, Starfield B, Sanchez E, Casanova C, Morales M.The role of primary care in preventing ambulatory caresensitive conditions. Eur J Public Health 2004;14:246–51.

[13] Ansari Z, Laditka J, Laditka S. Access to health care andhospitalization for ambulatory care sensitive conditions.Med Care Res Rev 2006;63:719–41.

[14] Bermudez D, Baker L. The relationship between SCHIPenrollment and hospitalizations for ambulatory care sensi-tive conditions in California. J Health Care Poor Under-served 2005;16:96–110.

[15] Nedel F, Facchini L, Martin-Mateo M, Vieira L, Thume E.Family Health Program and ambulatory care-sensitiveconditions in Southern Brazil. Rev Saude Publica2008;42(6):1041–52.

Impact of the Family Health Program on gastroenteritis in children 185

[16] Alfradique M, de Fatima Bonolo P, Dourado I, Lima-CostaMF, Macinko J, Mendonca CS, Oliveira VB, Sampaio LF,DeSimoni C, Turci MA. Ambulatory Care Sensitive Condi-tions: elaboration of Brazilian list as tool for measuringhealth system performance (Project ICSAP-Brazil). CadSaude Publica 2009;23:105–17.

[17] WHO. Morality Country Fact Sheet 2006. <http://www.who.int/whosis/mort/profiles/mort_amro_bra_bra-zil.pdf>; 2006 [accessed 31.03.10].

[18] Sergio J, deLeon ACPD. Analysis of mortality from diarrheicdiseases in under-five children in Brazilian cities with morethan 150,000 inhabitants. Cad Saude Publica2009;25(5):1093–102.

[19] Melo M, Taddei J, Diniz-Santos D, et al. Incidence ofdiarrhea in children living in urban slums in Salvador, Brazil.Braz J Infect Dis 2008;12(1):89–93.

[20] Santos CA, Strina A, Amorim LD, Genser B, Assis AM, PradoMS, Barreto ML. Individual and contextual determinants ofthe duration of diarrhoeal episodes in preschool children: alongitudinal study in an urban setting. Epidemiol Infect2011:1–8.

[21] Barreto ML, Genser B, Strina A, Teixeira MG, Assis AM, RegoRF, Teles CA, Prado MS, Matos S, Alcantara-Neves NM,Cairncross S. Impact of a citywide sanitation program inNortheast Brazil on intestinal parasites infection in youngchildren. Environ Health Perspect 2010;118(11):1637–42.

[22] do Carmo GM, Yen C, Cortes J, et al. Decline in diarrheamortality and admissions after routine childhood rotavirusimmunization in Brazil: a time-series analysis. PLoS Med2011;8(4):e1001024.

[23] Ministry of Health of Brazil–Departamento de Atenca~oBasica. Saude da Famılia no Brasil: uma analise de indica-dores selecionados. Brasılia, DF, Brazil: Minsterio da Saudedo, Brasil; 2006.

[24] Brasil. Internacoes Hospitalares do SUS por local deinternacao. Available from: http://tabnetdatasusgovbr/cgi/sih/rxdescrhtm - descvar; 2010 [accessed 10.11.10].

[25] Estatıstica IBdGe. Projecao da populacao do Brasil por sexoe idade 1980–2050. Available from: <http://www.ibge.-gov.br/home/estatistica/populacao/projecao_da_popula-cao/2008/default.shtm>; 2008 [accessed 10.11.10].

[26] Brazil. Family Health Care in Brazil: analysis of selectedindicators. Available from: <http://dtr2004.saude.gov.br/dab>; 2006 [accessed 10.11.10].

[27] Brasil. Teto, credenciamento e implantacao das estrategiasde Agentes Comunitarios de Saude, Saude da Famılia eSaude. Available from: <http://dtr2004saudegovbr/dab/historico_cobertura_sfphp>; 2004 [accessed 10.11.10].

[28] Instituto de Pesquisa Economica Aplicada. Dados macro-economicos e regionais. Available from: <http://www.ipeadata.gov.br>; 2009 [accessed 15.12.09].

[29] Brazilian Institute of Geography and Statistics. Bancomultidimensional de estatısticas. Available from: <http://www.ibge.gov.br>; 2009 [accessed 30.11.09].

[30] He X, Tong X, Sun J, Cook RJ. Regression analysis ofmultivariate panel count data. Biostatistics2008;9(2):234–48.

[31] Mays N, Pope C. Rigour and qualitative research. BMJ1995;11:109–12.

[32] Macinko J, de Souza M, de Fatima M. Evaluation of theimpact of the Family Health Program on infant mortality inBrazil, 1990–2002. J Epidemiol Community Health2006;60(1):13–9.

[33] Macinko J, de Oliveira VB, Turci MA, Guanais FC, Bonolo PF,Lima-Costa MF. The influence of primary care and hospitalsupply on ambulatory care-sensitive hospitalizations amongadults in Brazil, 1999–2007. Am J Public Health2011;101(10):1963–70.