guatemala children

11
Tropical Medicine and International Health volume 5 no 2 pp 145–155 february 2000 Health-seeking behaviour for child illness in Guatemala Noreen Goldman 1 and Patrick Heuveline 2 1 Office of Population Research, Princeton University, Princeton, USA 2 Population Research Center, NORC & The University of Chicago, Chicago, USA Summary Relying on data from the 1995 Guatemalan Survey of Family Health (EGSF), we analyse the relationship between child illness and health-seeking behaviour. Information on illness was collected for 3193 children. This analysis is based on 870 of these who became ill with diarrhoeal or respiratory disease during a 13-day period prior to interview. Estimates are derived from logistic models of the probability of seeing any or a specific type of health care provider as a function of characteristics of the illness on a given day and the child. The results indicate that modern medical care plays a major role in the treatment of infectious illness among children in rural Guatemala, with visits to pharmacists, doctors and the staff at government health facilities occurring much more frequently than visits to curers and other traditional practitioners. In general, families are much more likely to seek out a health care provider when a child experiences fever and gastro- intestinal symptoms than when suffering from respiratory and other symptoms, and when a mother perceives the illness to be serious. The results also indicate that infants, low parity children, and children assessed as having generally been in good health are more likely to visit health care providers than other children. However, the particular associations often vary by type of health care provider. keywords diarrhoea, ARI, Guatemala, health care behaviour correspondence Dr Noreen Goldman, Office of Population Research, Princeton University, 21 Prospect Avenue, Princeton, NJ 08544, USA. E-mail: [email protected] Introduction In recent years, epidemiologists and social scientists have devoted increasing attention to studying health-seeking behaviour associated with the two leading causes of child mortality, namely diarrhoeal illness and acute respiratory infection (ARI). Among childhood deaths in developing countries in 1993, about 27% have resulted from ARI and another 23% from diarrhoea (Unicef 1995). Yet, our knowl- edge about how and when families in developing countries seek treatment for these prevalent illnesses remains seriously incomplete, for two principal reasons: limitations of the samples used in existing studies and lack of sufficiently detailed information about the nature and timing of illness and treatment behaviour. Many studies of morbidity in developing countries are based on samples of patients in hospitals, clinics, or other facilities. Epidemiologists are well aware that these studies are seriously compromised by selection biases. While medical records for these samples may provide useful information for certain populations in industrialized countries, they are grossly inadequate in most developing countries primarily because most illnesses in these countries receive little formal treatment (Kalter 1992). Although hospital or clinic-based samples have been used to study the aetiology of illness in developing countries – for example Bale (1990) for studies of ARI – these samples are not appropriate for examining health care behaviour, even in the case of severe illness. Many studies rely instead on samples drawn from one or several communities (e.g. Van der Stuyft et al. (1996) for a study of child illness in two Guatemalan villages). Un- fortunately, these studies almost always suffer from relatively small sample sizes (rarely more than 500 and often less than 100), thereby restricting the types of statistical analyses that can be done. In addition, few are based on probability samples, which limits their generalizability. Health interview surveys appear to offer the best vehicle for analysing treatment behaviour on a large representative sample of children. These surveys typically involve a single cross-sectional interview, based on a random sample of a defined population, in which respondents (or mothers, in studies of child illness) are asked to report about the illnesses TMIH527 © 2000 Blackwell Science Ltd 145

Upload: karlpoor

Post on 13-Sep-2015

214 views

Category:

Documents


1 download

DESCRIPTION

research

TRANSCRIPT

  • Tropical Medicine and International Health

    volume 5 no 2 pp 145155 february 2000

    Health-seeking behaviour for child illness in Guatemala

    Noreen Goldman1 and Patrick Heuveline2

    1 Office of Population Research, Princeton University, Princeton, USA2 Population Research Center, NORC & The University of Chicago, Chicago, USA

    Summary Relying on data from the 1995 Guatemalan Survey of Family Health (EGSF), we analyse the relationshipbetween child illness and health-seeking behaviour. Information on illness was collected for 3193 children.This analysis is based on 870 of these who became ill with diarrhoeal or respiratory disease during a 13-dayperiod prior to interview. Estimates are derived from logistic models of the probability of seeing any or aspecific type of health care provider as a function of characteristics of the illness on a given day and thechild. The results indicate that modern medical care plays a major role in the treatment of infectious illnessamong children in rural Guatemala, with visits to pharmacists, doctors and the staff at government healthfacilities occurring much more frequently than visits to curers and other traditional practitioners. In general,families are much more likely to seek out a health care provider when a child experiences fever and gastro-intestinal symptoms than when suffering from respiratory and other symptoms, and when a mother perceivesthe illness to be serious. The results also indicate that infants, low parity children, and children assessed ashaving generally been in good health are more likely to visit health care providers than other children.However, the particular associations often vary by type of health care provider.

    keywords diarrhoea, ARI, Guatemala, health care behaviour

    correspondence Dr Noreen Goldman, Office of Population Research, Princeton University, 21 ProspectAvenue, Princeton, NJ 08544, USA. E-mail: [email protected]

    Introduction

    In recent years, epidemiologists and social scientists havedevoted increasing attention to studying health-seekingbehaviour associated with the two leading causes of childmortality, namely diarrhoeal illness and acute respiratoryinfection (ARI). Among childhood deaths in developingcountries in 1993, about 27% have resulted from ARI andanother 23% from diarrhoea (Unicef 1995). Yet, our knowl-edge about how and when families in developing countriesseek treatment for these prevalent illnesses remains seriouslyincomplete, for two principal reasons: limitations of thesamples used in existing studies and lack of sufficientlydetailed information about the nature and timing of illnessand treatment behaviour.

    Many studies of morbidity in developing countries arebased on samples of patients in hospitals, clinics, or otherfacilities. Epidemiologists are well aware that these studies areseriously compromised by selection biases. While medicalrecords for these samples may provide useful information forcertain populations in industrialized countries, they are

    grossly inadequate in most developing countries primarilybecause most illnesses in these countries receive little formaltreatment (Kalter 1992). Although hospital or clinic-basedsamples have been used to study the aetiology of illness indeveloping countries for example Bale (1990) for studies ofARI these samples are not appropriate for examining healthcare behaviour, even in the case of severe illness.

    Many studies rely instead on samples drawn from one orseveral communities (e.g. Van der Stuyft et al. (1996) for astudy of child illness in two Guatemalan villages). Un-fortunately, these studies almost always suffer from relativelysmall sample sizes (rarely more than 500 and often less than100), thereby restricting the types of statistical analyses thatcan be done. In addition, few are based on probabilitysamples, which limits their generalizability.

    Health interview surveys appear to offer the best vehiclefor analysing treatment behaviour on a large representativesample of children. These surveys typically involve a singlecross-sectional interview, based on a random sample of adefined population, in which respondents (or mothers, instudies of child illness) are asked to report about the illnesses

    TMIH527

    2000 Blackwell Science Ltd 145

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    experienced and health services or treatment used within aspecified period prior to interview. While the samplingdesigns of these surveys are generally adequate, most do notcollect sufficiently detailed data to permit meaningful analy-sis of the relationship between illness and treatment behav-iour. Specifically, few studies have obtained information onthe full range of symptoms experienced by the child, all treat-ments sought (including home remedies, biomedical prac-titioners, and traditional health care providers), and thetiming of both symptoms and treatment. For example, in thesecond and third rounds of the Demographic Health Surveys(DHS) programme, respondents were asked about the pres-ence of several symptoms (typically diarrhoea, cough withrapid breathing, and fever) during the most recent-two weekperiod. However, they were not asked to provide any infor-mation about additional symptoms that may have occurredduring this time or about the duration or timing of any of thereported symptoms (Ryland & Raggers 1998).

    The lack of adequately detailed information on illness andtreatment seriously compromises the resulting analysis ofexisting studies in several respects. First, illnesses cannot beproperly described or classified without a fairly completereport of symptoms. For example, as demonstrated byGoldman et al. (1998a) and shown below, most children ex-periencing some symptom of diarrhoea or ARI suffer frommore than a single symptom and many experience both gas-trointestinal and respiratory symptoms during a given illness.Second, the failure of many health interview surveys to obtaincomplete reports of nonbiomedical sources of care necessarilyleads to biased estimates of the frequency and determinants ofhealth-seeking behaviour. And third, the absence of data ontiming makes it impossible to interpret estimates of the preva-lence of illness or of health-seeking behaviour. For example,some illnesses or sick children may be less likely to receivetreatment than others simply because the illnesses last for ashorter duration. The need to account for duration is particu-larly important when the information from health interviewsurveys pertains to a very recent period prior to interview: Inthese cases, many recent illnesses are still in progress at thetime of the survey and may eventually receive treatment even ifthey have not done so as of the interview date. More generally,the absence of data on the timing of illness prevents the ana-lyst from selecting an appropriate sampling frame of illnessepisodes to include in the analysis (e.g. those which begin in aspecified period; Ross & Vaughan 1986) and inevitably leads tobiased estimates of illness frequency and associated measuresof treatment.

    In this paper, we rely on survey data from the GuatemalanSurvey of Family Health (EGSF 1995) to analyse the relation-ship between child illness and health-seeking behaviour. Asdescribed in more detail below, through both its samplingdesign and questionnaire construction, the EGSF has avoided

    many of the limitations that affect earlier studies of treatmentbehaviour. Of particular importance for our study is the inclu-sion of a two-week daily calendar of morbidity and treatmentbehaviour for children age five and younger. Based on thesecalendar data, we use statistical models to examine how char-acteristics of diarrhoeal and respiratory illnesses amongchildren affect their use of biomedical and nonbiomedicalhealth care providers.

    Subject and methods

    Guatemala is the largest country in Central America with apopulation of approximately 10.5 million in 1995 (CELADE1997). Although only 108 889 km2 in size (Instituto Nacionalde Estadstica 1988), many rural areas remain relatively iso-lated from urban Guatemalan life, which is centred primarilyin the capital, Guatemala City. Guatemala remains a highlystratified society with large income inequality and the vastmajority of the population living below the poverty line (Steele1994). Most of the rural population does not have adequateaccess to such public services as water, sanitation and electric-ity (Steele 1994). Roughly half of the population is indigenous i.e. descendants of Maya and other preconquest groups while the other half, referred to as ladinos, speak Spanish, wearEuropean clothing, identify with the national Guatemalan cul-ture, and are of both indigenous and European origins.

    As elsewhere in Latin America, Guatemalan mortality rateshave fallen since the 1950s, although they remain among thehighest in the region. The most recent estimates of infant mor-tality indicate a national level of about 50 per 1000 for theearly 1990s (Instituto Nacional de Estadstica et al. 1996;World Bank 1995). Treatment for illness is available from bio-medical health care providers (both through the publiclyfinanced health care system and through private doctors), fromtraditional practitioners (such as midwives and curers), andfrom popular practitioners who often dispense modern drugsand give injections without biomedical training (Cosminsky &Scrimshaw 1980; Pebley et al. 1996).

    The Guatemalan Survey of Family Health

    The data for this analysis come from the Guatemalan Surveyof Family Health (known in Spanish as the EncuestaGuatemalteca de Salud Familiar or the EGSF), conducted byPrinceton University, RAND, and the Instituto de Nutricin deCentro Amrica y Panam in 1995. The EGSF is based on asample of households in rural communities (i.e. communitieswith 20010 000 inhabitants) within four departments ofGuatemala (Chimaltenango, Totonicapn, Suchitepequez andJalapa). The four departments were selected on the basis ofsocial, economic, and environmental diversity as well as ethniccomposition.

    2000 Blackwell Science Ltd146

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    Sixty communities were included in the survey, 15 in eachof the selected departments. Communities were selected withprobability proportional to population size to yield self-weighting samples within departments. Approximately 50women aged 1835 received detailed individual question-naires in each of selected communities, for a total of 2872women. Interviews with community informants and healthcare providers were also conducted in each of the sampledcommunities.

    The analysis presented below is derived entirely from theindividual questionnaires administered to women aged 1835and based largely on the section devoted to childrens ill-nesses. In this section of the questionnaire, mothers wereasked questions related to diarrhoeal and respiratory illnessfor a maximum of two children born since 1990. They werefirst asked whether each of eight specific symptoms related toacute respiratory infection or diarrhoea occurred during thepreceding two weeks. The eight symptoms are: constantcough; boiling of the chest; panting, wheezing, or difficultybreathing; high fever; weakness, apathy, or lethargy; diar-rhoea more than three times a day; blood in stools; andvomiting. Some of these symptoms have been shown in otherstudies to have high sensitivity and specificity (Kroeger 1983;Kalter et al. 1991; Boerma & Van Ginneken 1992). The symp-toms were adapted to the rural Guatemalan setting on thebasis of medical anthropological research and our own pilotstudy. For example, Guatemalan mothers frequently men-tioned boiling of the chest (hervor de pecho in Spanish) torefer to the noise made by congestion; this symptom wasfound to be associated with cough, bronchitis and broncho-pneumonia (INCAP 1994).

    If a child experienced any of the eight specified symptoms,mothers were asked when the symptom began and on whichdays during the past two weeks the symptom had beenpresent. They were also asked about any other symptomsexperienced during this time, whether the symptoms wereperceived as serious, whether the mother asked others(relatives, neighbours or friends) for advice or visited healthcare providers regarding their childs illness, and whetherthey or anyone else administered any treatment. Informationregarding the presence of symptoms, seriousness and treat-ment was recorded in the appropriate days of the calendar,indexed from 14 (14 days or two weeks before interview) tozero (the day of interview). Additional information in-cluding the nature of the advice and treatment as well as thecost and perceived effectiveness was subsequently obtained(in tabular format) about each of the persons, health careproviders and treatments recorded in the calendar. Forfamilies with more than one living child born since 1990,the entire section of the questionnaire was asked for boththe youngest and the penultimate child (Peterson et al.1997).

    Statistical analysis

    Illness information was obtained from 3193 children in theEGSF. Overall, 45% (1446) experienced at least one symptomduring the two-week calendar period. Of these, 870 childrenexperienced their first symptom subsequent to the start of thecalendar period. This latter group forms the basis for allresults pertaining to treatment behaviour presented in thisstudy. Children whose illnesses began two or more weeksprior to the interview were excluded for two reasons. First,information regarding very recent visits to health careproviders is more likely to be accurate than data on earliervisits. And second, data on the timing of treatment were notobtained for days prior to the start of the calendar period.

    Results pertaining to the relationship between health-seeking behaviour and characteristics of illnesses and childrenare derived from logistic models in which the day of illness isthe unit of analysis. These models are far better suited to ananalysis of treatment behaviour than conventional regressionmodels in which children or illnesses are the units of analysis,since the latter can neither account for varying exposures tothe risk of seeking treatment across children with illnesses ofdifferent durations nor for right censoring of observations,namely that many children are still experiencing symptoms atthe time of interview and may subsequently seek out a healthcare provider.

    Both (binomial) logistic and multinomial (polytomous)logistic models are estimated (Hosmer & Lemeshow 1989).The former are used when the outcome variable denotes visits(on a given day of illness) to any health care provider,whereas the latter are used when the outcome represents visitsto specific types of health care providers. All visits to healthcare providers (i.e. not only the first for a given illness) areincluded in the analysis.

    The logistic models are based on a sample of days pertain-ing to the 870 children whose illnesses began in the calendarperiod. The day of interview is excluded from the samplebecause it represents an incomplete day of exposure to seek-ing treatment. In addition, days in which children did nothave any symptoms are excluded because no respondentreported seeing a health care provider on these days. The finalsample for analysis i.e. all days with symptoms between 1and 13 days before interview includes 4344 days, yielding anaverage of 5 days of illness for each child in the sample.

    The logistic models consider the probability of seeingany health care provider (or a specific type of health careprovider) on a day of illness as a function of characteristics ofthe illness on that day and of characteristics of the child. Theformer set of variables includes the nature of symptoms,duration of symptoms, assessed severity, and beliefs related tothe cause of the symptoms.

    A series of exploratory logistic models were estimated in

    2000 Blackwell Science Ltd 147

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    order to identify a parsimonious representation of the differ-ent symptoms (which included the 8 solicited symptoms aswell as others volunteered by respondents) and combinationsof symptoms experienced by children in the sample. Theresults indicated that there was little difference in the prob-ability of seeking a health care provider for various symptomswithin a general category (e.g. cough, difficulty breathing, orother respiratory symptoms), and that the presence of fever,especially in combination with gastrointestinal symptomssuch as vomiting and diarrhoea, was strongly associated withvisits to a health care provider. These findings led to a four-category classification of symptoms:

    only respiratory symptoms; fever with gastrointestinal symptoms; fever without any gastrointestinal symptoms (but pos-

    sibly with other symptoms);

    all other symptoms alone or in combination.Duration of illness is measured as the number of con-

    secutive days with symptoms. Of the 870 children in theanalysis, 57 experienced two episodes of illness i.e. one ormore consecutive days without any symptom was recordedbetween two days with (potentially different) symptoms.Duration is represented by four categories: day 1, day 2, days35, and days 6 and higher. Severity of illness is modelled as adummy variable indicating whether or not the childs motherperceived the symptoms to be serious on a given day. The twodummy variables denoting health beliefs are derived from amothers (potentially multiple) responses about the causes ofher childs symptoms. The first variable identifies beliefsrelated to hygiene or contamination and includes suchresponses as children putting dirty food or other items intheir mouths, mothers not washing hands, and the presenceof microbes or an infection (Goldman et al. 1998b). Thisvariable is intended to identify families holding relativelymodern beliefs about illness causation, although it is likelythat many of the responses in the hygiene/contaminationcategory do not actually reflect knowledge about germ theory(McKee 1987; Pebley et al. 1999). The second variable ident-ifies beliefs related to traditional folk illnesses such as the evileye and empacho a folk illness with gastrointestinalsymptoms.

    Variables pertaining to the child include the childs age,parity, and an overall assessment of the childs health status.Age is represented by a dummy variable indicating whetherthe child is an infant (i.e. < 12 months of age), whereas parityis modelled as a continuous variable. A mothers perceptionof the general health status of her child since the time ofbirth is represented by two dummy variables, denoting verypoor and very good health, respectively; the referencecategory contains children whose health status was assessedas poor, fair or good.

    In addition to these variables describing the illness and thechild, the logistic models also incorporate a set of dummyvariables denoting the four departments in which the surveytook place. The inclusion of these variables compensates forthe fact that the EGSF sample is not self-weighting acrossdepartments.

    Results from the binomial and multinomial logistic modelsare presented in terms of exponentiated coefficients i.e.odds ratios in the former models and relative risk ratios in thelatter. Relative risk ratios are considerably more difficult tointerpret than odds ratios. For example, in a multinomialmodel, a relative risk ratio greater than unity for a dichot-omous variable does not always imply that the probability ofthe particular outcome is greater when the dichotomous vari-able is unity as compared with zero. For ease of interpret-ation, we present predicted (or simulated) percentages basedon the multinomial model in the text (Table 5) and relativerisk ratios in the appendix. The predicted values wereobtained by (1) retaining all variables except those under con-sideration at their observed values for each observation in thesample (i.e. a day of illness); and (2) setting the variablesunder consideration at a preselected value (e.g. 1 or 0 in thecase of categorical variables).

    Results and discussion

    Symptoms of illness

    In Table 1, we present estimates of the two-week prevalenceand median duration of each symptom and the frequencywith which mothers considered each symptom to be serious.The results reveal that the frequency of diarrhoea and ARI-related symptoms is high in the rural Guatemalan popu-lation. During the most recent two-week period, nearly half(45.3%) of children age five and under experienced at leastone of the eight symptoms solicited. There is substantial vari-ability in the length of the different symptoms, ranging fromabout two days for vomiting, blood in the stools and highfever to 11 days for constant cough. Overall, just under one-quarter of days with symptoms were considered to beserious, with the prevalence of severity varying by the type ofsymptom and being especially high for vomiting.

    Table 2 presents the distribution of children in the sampleby the number of symptoms reported during the calendarperiod (including nonsolicited symptoms), along with theaverage number of symptoms experienced on a given day.Among the 1446 children with at least one symptom, abouttwo-thirds had more than one. On average, 1.8 symptomswere reported on a day containing at least one symptom. Theaverage number of symptoms reported on a given day is con-siderably less than the total number of symptoms recorded inthe calendar because few children experienced all of the

    2000 Blackwell Science Ltd148

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    reported symptoms simultaneously throughout the period ofillness. Rather, individual symptoms typically began andended on different days, thereby changing the nature of theillness over even a relatively short duration. One consequenceof this complexity is that classification of illness becomesproblematic. For example, results not presented here revealthat among children with at least one respiratory symptom,42% were reported to also have at least one gastrointestinalsymptom during the calendar period. These findings high-light the need to go beyond conventional categories of illnessand consider patterns of symptoms in models of treatmentbehaviour.

    A description of treatment behaviour

    The EGSF collected information on treatments administeredto the child regardless of whether the treatments were rec-ommended or administered by the mother, a relative orfriend, or a health care provider as well as visits to healthcare providers. Almost 90% of the 870 children in our samplereceived some form of treatment. The vast majority weregiven medicines, whereas a relatively small proportion weregiven herbs, herbal teas or other home remedies (Heuveline& Goldman 2000); most of these treatments were recom-mended by the mother or another member of the familyrather than a health care provider. As shown in the firstcolumn of Table 3, only about one-third of sick children (i.e.children with any symptom) visited a health care providerduring the calendar period. Pharmacists, who frequently dis-pense advice as well as medication but generally do notreceive professional training (Van der Stuyft et al. 1996), werethe health care providers most likely to have been consulted,and doctors and the staff of government-sponsored healthposts or centres were seen more frequently than curers (i.e.persons who at least prior to recent years relied primarilyon folk remedies to treat illness) and other types of healthcare providers.

    The results in the right-hand panel of Table 3 reflect thedistribution of health care providers seen by the number ofconsecutive days that the child experienced any symptom.The data suggest that, overall, health care providers arevisited more often during the first few days (most notably thesecond day) than later during the course of longer illness. Forexample, just over 10% of second days of illness are charac-terized by a visit to a health care provider in contrast to only

    2000 Blackwell Science Ltd 149

    Two-week period Median duration Percent of days Symptom prevalence (%)* (days) serious

    Constant cough 19.8 10.8 22.6Boiling of the chest 12.0 08.7 25.7Panting 05.6 07.2 30.8High fever 24.4 02.3 30.4Weakness 12.6 07.7 28.5Diarrhoea 21.8 04.5 26.8Blood in the stools 01.4 02.5 27.2Vomiting 04.8 01.7 42.5Any of the eight solicited symptoms 45.3 05.8 22.8

    *Estimates of prevalence are based on 3193 children; Estimates of median duration are basedon life tables and refer to the number of consecutive days with the symptom, for symptomsbeginning subsequent to day 14 of the calendar. A given child may contribute more than oneepisode to the estimate if the symptom stopped and resumed on a later day during the two-week period; Estimates of percent of days when symptoms are considered serious are basedon the total sample of 10 742 days with symptoms.

    Table 1 Prevalence of symptoms, medianduration, and percent of days with a givensymptom that was considered serious, bytype of symptom

    Table 2 Mean number of symptoms per day, among children with agiven number of symptoms in the calendar

    Number of symptoms Number of Percent of Mean number ofin the calendar children children symptoms per day*

    None 1747 054.71 0507 015.9 1.02 0392 012.3 1.43 0254 008.0 1.94 0140 004.4 2.45 0088 002.8 2.86 0049 001.5 3.27 0013 000.4 3.98 0002 000.0 4.39 0001 000.0 5.9Total 3193 100.0 1.8

    *Based on all days between the first and last occurrence of asymptom. Based on days with at least one symptom.

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    5% or 6% from day six onward. However, the patterns varyby the type of health care provider. We will return to the issueof the timing of health care provider visits in the statisticalanalysis presented below.

    Modelling treatment behaviour

    The results presented in Table 4 are derived from a logisticmodel that examines the probability that a day of illnessresults in a health care provider visit as a function of charac-teristics of the child and his or her symptoms. The table pre-sents both estimated and observed odds ratios, along withP-values and confidence intervals associated with the former.The estimates are based on a total of 4323 days (rather thanthe full sample of 4344 days shown in Table 3) because ofmissing values for the variables denoting health beliefs. Thevast majority (94%) of visits to health care providers in thissample are first visits.

    The odds ratios are consistent with the duration effectsnoted earlier in Table 3: health care providers are most fre-quently sought on the second day of illness and are leastlikely to be visited after the fifth day, but the effects are smalland not statistically significant. The odds ratios for symp-toms are large (and significant) and indicate that childrenwith respiratory symptoms are least likely to seek health careproviders, while those with fever and especially fever togetherwith gastrointestinal symptoms are especially likely to visithealth care providers. Moreover, days of illness in whichmothers perceive the symptoms to be serious are much morelikely to result in health care provider visits than other days.

    Results from a logistic model (not shown) of the prob-ability that mothers perceive symptoms to be serious reveal

    that mothers are far more likely to perceive the combinationof fever and gastrointestinal symptoms as serious and lesslikely to assess respiratory symptoms as serious comparedwith other patterns of illness. However, the logistic modelshown in Table 4 which controls for both the nature ofsymptoms and their perceived severity indicates that factorsbeyond perceived severity are responsible for the higher fre-quency of health care provider visits associated with certainsymptoms. In fact, estimates not presented here indicatedthat the coefficients associated with the symptom variableschange little when seriousness is added to the logistic model.

    The odds ratios associated with characteristics of the childindicate that infants and low parity children are more likelyto be seen by health care providers than other children. Themothers perception of her childs health status is also signifi-cantly associated with treatment behaviour. Children per-ceived as generally having very good health are much morelikely to visit health care providers while children viewed asgenerally being in very poor health are least likely (althoughnot significantly so because of the small number of childrenin this category) to do so. The inclusion of a variable denot-ing gender (not shown) revealed no significant sex differencein health-seeking behaviour, a finding that is consistent withresults from other studies in Latin America.

    The results related to health beliefs indicate that familiesholding modern beliefs are more likely to see a health careprovider than families holding other beliefs. (The estimatedodds ratio remains virtually unchanged when the sample isrestricted to only those days with gastrointestinal symptoms i.e. the type of symptoms most closely associated with lackof hygiene.) By contrast, families who believe that traditionalfolk illnesses are the cause of their childrens symptoms are

    2000 Blackwell Science Ltd150

    Table 3 Distribution of type of health care provider seen among children and by duration of symptoms

    Distribution of type of health care provider seen by duration of symptoms(% of days)

    % children seeing Number of consecutive days with any symptomHealth care health care provider provider within calendar period 1 2 3 4 5 6 7 81

    Pharmacist 0009.7 003.0 003.4 002.3 000.8 001.6 001.3 001.1 000.8Doctor 0006.9 001.6 002.1 001.8 001.9 001.8 001.6 000.4 000.8Someone in the health post/centre* 0007.1 001.2 002.1 001.7 001.9 001.8 000.9 000.7 001.1Curer 0004.4 000.7 001.2 001.2 000.8 002.3 001.3 001.5 002.1Other 0005.4 001.7 001.6 001.4 001.5 001.0 000.6 000.7 001.1No health care provider 0067.9 091.8 089.8 091.6 093.1 091.5 094.4 095.6 094.1Number of children 0870Number of days 4344 927 829 654 481 386 320 271 476

    *Also includes clinics and health technicians. Other health care providers primarily include promotors (volunteers associated with the Ministryof Health who receive minimal training in basic health issues), midwives and nurses. The number of first days of illness exceeds the total num-ber of children in the analysis because 57 children experienced two episodes of illnessi.e. one or more consecutive days without any symptomwas recorded between two days with (potentially different) symptoms.

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    not significantly different (in terms of the probability of see-ing a health care provider) from other families.

    The results in Table 4 suggest that numerous characteristicsof the illness and of the child are strongly associated with thelikelihood of visiting a health care provider. In order to assessthe extent to which these covariates have disparate effectsacross health care providers, we estimated a multinomialmodel with six outcomes (pharmacist, health post, doctor,curer, other and the reference category of no health careprovider). This model has almost the same covariate structureas the logistic model in Table 4; the only difference is theexclusion of the variable that denotes very poor health,because of inadequate sample size.

    The results from the multinomial model are presented interms of predicted percentages (together with observed per-centages) in Table 5, and as relative risk ratios in the appen-dix. For the sake of parsimony, the percentages for parity arepresented for one relatively low (2) and one moderately high(6) value, rather than for each observed value (parity rangesfrom one to 12 in the sample). When interpreting the result-

    ing values, it is important to note that these percentages referto a single day of illness rather than the entire duration andhence are relatively small in comparison to the likelihood thata health care provider is sought sometime during the courseof an illness.

    The predicted values in Table 5 indicate that the type ofsymptoms experienced and the perceived severity of thesymptoms have large impacts on the likelihood of seeing ahealth care provider. Children experiencing fever and es-pecially the combination of fever and gastrointestinal symp-toms are much more likely to be taken to a biomedical healthcare provider than children with respiratory or other symp-toms, whereas the differentials by type of symptom are muchmore modest and follow a different pattern for those seek-ing curers. Interestingly, days characterized by serious symp-toms are significantly more likely than other days to result inhealth care provider visits, with the exception of visits tohealth centres and posts (as shown in the appendix, this dif-ference is not statistically significant). This finding is prob-ably due to the minimal fees associated with visits to these

    2000 Blackwell Science Ltd 151

    Odds ratioNumber 95% confidence of days Observed Estimated P-value interval

    Illness characteristicsDay 1 0925 1.00 1.00Day 2 0827 1.28 1.23 , 0.23 0.88 1.72Days 35 1515 0.96 0.99 , 0.95 0.73 1.35Days 6 1 1056 0.65 0.78 , 0.18 0.54 1.13Other symptoms 1778 1.00 1.00Respiratory only 1165 0.58 0.66 , 0.03 0.46 0.95Fever, no gi 1046 1.86 1.87 , 0.01 1.41 2.47Fever & gi 0334 3.17 2.82 , 0.01 1.99 4.01Symptoms not serious 3437 1.00 1.00Symptoms serious 0886 2.58 2.29 , 0.01 1.78 2.94No hygiene beliefs 4034 1.00 1.00Hygiene beliefs 0289 1.94 2.08 , 0.01 1.42 3.06No folk illness beliefs 4011 1.00 1.00Folk illness beliefs 0312 1.60 1.17 , 0.43 0.79 1.75

    Child characteristicsAges 15 3167 1.00 1.00Under age 1 1156 1.41 1.52 , 0.01 1.18 1.95Parity 0.91 0.89 , 0.01 0.83 0.94Poor-good health 4175 1.00 1.00Very poor health 0046 0.27 0.27 , 0.20 0.04 1.98Very good health 0102 3.36 3.54 , 0.01 2.10 5.98Pseudo R2 0.083

    Total number of days 4323

    *All models include a set of dummy variables to represent the four departments in which theEGSF took place. Because parity is modelled as a continuous variable and, thus, has noobserved odds ratio, we have presented the estimated value from a univariate logistic model(i.e. a logistic model in which parity is the only explanatory variable).

    Table 4 Observed and estimated odds ratiosfor the logistic model of the probability ofseeing any health care provider on a day ofillness, by characteristics of the illness andthe child

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    government-sponsored health facilities. Days of illness per-ceived as serious are most likely to be treated by doctors orpharmacists. The percentages by duration indicate that,whereas families are most likely to visit biomedical healthcare providers on the first or second day of illness, they areprogressively more likely to seek curers the longer the illnesslasts.

    The results in Table 5 also reveal that mothers who holdmodern health beliefs are more likely to visit health post/centre staff, doctors, and those classified as other health careproviders, but are less likely to seek traditional curers(although not all of the results are statistically significant). Bycontrast, mothers holding beliefs related to traditional folkillnesses are significantly more likely to see curers than othermothers. These findings suggest that health beliefs play animportant role in determining whether women seek tradi-tional or biomedical health care providers for their children.

    The remaining values in Table 5 indicate variation acrosshealth care providers in the association with a childs age,parity, and overall health assessment. The reversal of the agedifferential for pharmacists is not surprising since families areapt to be reluctant to administer medicines to infants, eventhough they are apparently more likely to take infants to doc-tors, curers or other health care providers. Interestingly, chil-dren perceived to have been in very good health are morelikely to be taken to doctors and to curers than other chil-dren, but are no more likely to visit health posts and centres.

    Although some of the differentials in Table 5 appear to bemodest, recall that the simulated percentages reflect only oneday of illness and capture changes in a single variable at atime. The absolute differences in the probability of seeing ahealth care provider (e.g. between nonserious and serioussymptoms) over longer time spans such as a week are muchlarger than those presented here. Moreover, the differentials

    2000 Blackwell Science Ltd152

    Table 5 Observed and predicted percentages seeking different types of health care providers on a day of illness, by selected characteristics ofthe illness and the child, estimated from the multinomial model in the appendix

    Pharmacist Health post Doctor Curer Other Observed Predicted Observed Predicted Observed Predicted Observed Predicted Observed Predicted

    Illness characteristicsDay 1 0003.0 2.9 1.2 1.2 1.6 1.6 00.7 0.6 1.7 1.6Day 2 0003.4 3.1 2.1 2.0 2.1 1.9 01.2 1.1 1.6 1.4Day 35 0001.7 1.6 1.8 1.8 1.8 1.7 01.4 1.4 1.3 1.3Day 6 1 0001.0 1.2 1.0 1.0 1.0 1.1 01.7 2.0 0.9 1.0Other symptoms 0001.6 1.6 1.0 1.0 1.4 1.3 01.8 1.6 0.7 0.7Respiratory only 0001.6 1.7 0.9 0.9 0.8 0.9 00.3 0.3 0.4 0.5Fever, no gi 0003.3 3.0 2.4 2.3 2.1 2.0 01.0 1.0 2.7 2.6Fever & gi 0003.6 3.1 3.6 3.4 4.2 3.4 03.0 2.9 3.6 2.8Symptoms not serious 0001.6 1.7 1.4 1.4 1.1 1.1 01.1 1.1 1.1 1.1Symptoms serious 0004.2 3.6 2.0 1.8 3.6 3.4 02.1 1.9 2.5 2.3No hygiene beliefs 0002.1 2.1 1.5 1.4 1.4 1.4 01.3 1.3 1.1 1.1Hygiene beliefs 0002.4 2.1 2.1 2.3 3.8 3.8 00.4 0.4 4.8 6.2No folk illness beliefs 0002.1 2.1 1.6 1.5 1.6 1.7 01.0 1.0 1.4 1.4Folk illness beliefs 0002.6 2.0 1.0 1.0 1.3 1.0 05.5 2.8 1.3 1.2

    Child characteristicsAges 15 0002.5 2.4 1.3 1.3 1.6 1.5 00.7 0.8 1.1 1.0Under age 1 0001.2 1.3 2.1 2.1 1.6 1.8 02.8 2.3 2.1 2.5Parity 2 0002.5 2.3 1.2 1.7 1.8 1.7 02.8 1.6 2.1 1.7Parity 6 0002.0 1.9 2.3 1.3 0.5 1.4 00.3 0.7 0.3 0.7Very poor-good health 0002.2 2.2 1.5 1.5 1.5 1.5 01.0 1.0 1.3 1.3Very good health 0000.0 * 1.0 1.1 3.9 4.4 13.7 7.9 2.9 3.7

    Number of days 4323

    *Parameter cannot be estimated because there are no children in very good health that visited a pharmacist. For example, the predicted percent-age seeing a pharmacist for days with only respiratory symptoms (1.7) was obtained from the sample of days of illness as follows: all variablesexcept the three dummy variables denoting symptoms were retained at their observed values, whereas the three dichotomous variables denoting(1) respiratory symptoms only (2) fever but no gastrointestinal symptoms, and (3) fever plus gastrointestinal symptoms were set to 1, 0 and 0,respectively. Also includes health centres, clinics and health technicians. Other health care providers primarily include promotors (volunteersassociated with the Ministry of Health who receive minimal training in basic health issues), midwives and nurses. Parity is modelled as a con-tinuous variable; predicted values are shown for two selected values of parity (2 and 6).

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    are larger when several variables are considered simul-taneously. For example, results from the multinomial model(not shown here) reveal that on a day of illness characterizedby fever and gastrointestinal symptoms that were assessed asserious, an infant faces a 31% chance of seeing some type ofhealth care provider. By contrast, on a day with only respir-atory symptoms not considered serious, an older child facesthe corresponding daily probability of only 3%.

    Summary and conclusions

    The findings from this analysis confirm and quantify previousresults, derived largely from small community-based studies,that modern medical care plays a major role in the treatmentof infectious illnesses among children in rural Guatemala.Although only one-third of illnesses result in a visit to ahealth care provider, pharmacists, doctors, and personnel atgovernment-sponsored health facilities are much more likelyto be seen than are curers and other traditional practitioners.Visits to biomedical health care providers are far more com-mon than explicit biomedical health beliefs about the causesof illness among womens own children.

    The results from statistical models indicate that the likeli-hood of a health care provider visit depends considerably onthe characteristics of the child and his or her illness. In gen-eral, families are much more likely to seek treatment from ahealth care provider when a child experiences fever andgastrointestinal symptoms such as vomiting or diarrhoeathan when showing respiratory and other symptoms, andwhen mothers perceive the symptoms to be serious. Whilegastrointestinal illnesses are likely to benefit from medicalattention, particularly given the high risks of malnutritionand mortality associated with dehydration from these ill-nesses, it is likely that many of the respiratory symptomsexperienced by these children also need attention. The respir-atory symptoms selected for explicit mention in the EGSFquestionnaire (e.g. panting/wheezing/difficulty breathing) arefrequently associated with lower respiratory infections, suchas bronchitis and pneumonia, which often warrant medicalintervention.

    Our findings from the EGSF agree with those of Yoder andHornik (1996), who use sample survey data from six sites inAsia and Africa to demonstrate that both the types of symp-toms experienced and the perceived severity have separateeffects on treatment choice. The types of symptoms most fre-quently identified as serious by Yoder and Hornik vomiting,fever, and lassitude are generally similar to those identifiedin this study. Both sets of findings suggest that familiesperceptions about the efficacy of treatments in curing orameliorating specific symptoms, along with their perceptionsof severity, are likely to affect their use of health careproviders.

    Results from the multinomial model indicate that factorsaffecting the likelihood of seeking treatment often vary by thetype of health care provider. In particular, the illness charac-teristics that are associated with the highest frequency ofvisits to curers, namely beliefs related to traditional folk ill-nesses and higher durations of illness, are contrary to thoseassociated with the highest frequency of visits to biomedicalhealth care providers. This result suggests that families seekdifferent types of health care providers for contrastingreasons and at varying stages of illness. Moreover, some vari-ables which are important determinants of visits to doctors,such as perceived severity of illness and the general healthstatus of the child, show little association with visits to thevirtually free government health facilities. The apparentfavouritism toward very healthy children with regard to doc-tor (and curer) visits, but not with respect to governmentfacilities, suggests that families may be more willing to investtheir scarce resources in those children who are most likely tolead long productive lives.

    This study highlights the advantages of the calendarapproach for studying treatment behaviour associated withchild morbidity. The collection of detailed data on the timingand nature of symptoms and treatments in the EGSF permitsstatistically unbiased estimation of the association between abroad range of characteristics of illnesses and children andthe likelihood of visits to different types of health careproviders. To the best of our knowledge, such estimates oftreatment, derived from a large probability sample, have notpreviously been obtained for developing countries.

    In addition, the collection of extensive information in theEGSF about family, household and community characteristicspermits considerable extension of the model of treatmentbehaviour considered in this analysis. For example, as sug-gested by previous research in Guatemala and elsewhere inLatin America (Annis 1981; Young 1981; Van der Stuyft et al.1996), a familys social network and support systems, a fam-ilys economic resources, and the availability and accessibilityof health care providers and facilities in the community arelikely to affect the frequency with which families seek healthcare providers for their childrens illnesses, as well as the typeof health care providers sought. Inclusion of a broad range offamily, household, and community-level variables in thestatistical analysis is currently underway.

    Acknowledgements

    The Guatemalan Survey of Family Health was carried out incollaboration with the Instituto de Nutricin de CentroAmrica y Panam and RAND. We acknowledge supportfor this project from NICHD grants R01 HD31327,P30HD32030, and R01 HD27361, and thank GermnRodrguez, Anne Pebley, Michele Gragnolati, Barbara

    2000 Blackwell Science Ltd 153

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    Vaughan, Dana Glei and Michelle Bellessa for their adviceand assistance.

    References

    Annis S (1981) Physical access and utilization of health services inrural Guatemala. Social Science and Medicine 15D, 515523.

    Bale JR (1990) Etiology and epidemiology of acute respiratory tractinfection in children in developing countries. Reviews of InfectiousDiseases 12, S861S1083.

    Boerma JT & Van Ginneken JKV (1992) Comparison of substantiveresults from demographic and epidemiological survey methods. In:Measurement of Maternal and Child Mortality, Morbidity andHealth Care: Interdisciplinary Approaches (ed. JT Boerma),Editions Derouaux Ordina, Lige, pp. 2762.

    CELADE (1997) The demographic situation of Central America. In:Demographic Diversity and Change in the Central AmericanIsthmus (eds AR Pebley & L Rosero-Bixby), RAND, SantaMonica, CA.

    Cosminsky S & Scrimshaw M (1980) Medical pluralism on aGuatemalan plantation. Social Science and Medicine 14B, 267278.

    Goldman N, Vaughan B & Pebley A (1998a) The use of calendars tomeasure child illness in health interview surveys. InternationalJournal of Epidemiology 27, 505512.

    Goldman N, Pebley AR & Beckett M (1998b) Diffusion of ideasabout personal hygiene and contamination in poor countries:evidence from Guatemala. Paper prepared for the Workshop onSocial Processes Underlying Fertility Change in DevelopingCountries, Committee on Population, National Academy ofSciences, January, 1998.

    Heuveline P & Goldman N (2000) A description of child illness andtreatment behaviour in Guatemala. Social Science and Medicine50, 345364.

    Hosmer DW & Lemeshow S (1989). Applied Logistic Regression.John Wiley & Sons. New York.

    INCAP (1994) Household Management of ARI, Multi-Centre Studyin Central America. Progress Report, February 1994.

    Instituto Nacional de Estadstica (1988). Anuario Estadstica.Instituto Nacional de Estadstica, Guatemala.

    Instituto Nacional de Estadstica, Ministerio de Salud Pblica yAsistencia Social, Agencia para el Desarrollo Internacional, Fondode las Naciones Unidas para la Infancia & Macro International

    Inc (1996) Encuesta Nacional de Salud Materno Infantl 1995.Macro Internationl Inc. Maryland.

    Kalter HD (1992) The validation of interviews for estimating mor-bidity. Health Policy and Planning 7, 3039.

    Kalter HD, Gray RH, Black RE & Gultiano SA (1991) Validation ofthe diagnosis of childhood morbidity using maternal health inter-views. International Journal of Epidemiology 20, 193198.

    Kroeger A (1983) Health interview surveys in developing countries: areview of the methods and results. International Journal ofEpidemiology 12, 465481.

    McKee L (1987) Ethnomedical treatment of childrens diarrhoeal ill-nesses in the highlands of Ecuador. Social Science and Medicine25, 11471155.

    Pebley AR, Goldman N & Rodrguez G (1996) Prenatal and deliverycare and childhood immunization in Guatemala: do family andcommunity matter? Demography 33, 231247.

    Pebley AR, Hurtado E & Goldman N (1999) Beliefs about childrensillness. Journal of Biosocial Science 31, 195219.

    Peterson C, Goldman N & Pebley A (1997) The 1995 GuatemalanSurvey of Family Health (EGSF): Overview and Codebook.RAND, Los Angeles.

    Ross DA & Vaughan JP (1986) Health interview surveys in develop-ing countries: a methodological review. Studies in Family Planning17, 7894.

    Ryland S & Raggers H (1998) Demographic and Health Surveys.(Comparative Studies 27): Childhood Morbidity and TreatmentPatterns. Macro International, Calverton, MD.

    Steele D (1994) Guatemala. Indigenous People and Poverty in LatinAmerica: an Empirical Analysis (eds G Psacharopoulos & HAPatrinos), World Bank, Washington, DC, pp. 105140.

    UNICEF (1995). The State of the Worlds Children. OxfordUniversity Press, Oxford.

    Van der Stuyft P, Sorensen SC, Delgado E & Bocaletti E (1996)Health-seeking behaviour for child illness in rural Guatemala.Tropical Medicine and International Health 1, 161170.

    World Bank (1995). World Development Report (1995) OxfordUniversity Press, Oxford.

    Yoder SP & Hornik RC (1996) Symptoms and perceived severity ofillness as predictive of treatment for diarrhoea in six Asian andAfrican sites. Social Science and Medicine 43, 429439.

    Young JC (1981) Medical Choice in a Mexican Village. RutgersUniversity Press, New Brunswick, NJ.

    2000 Blackwell Science Ltd154

  • Tropical Medicine and International Health volume 5 no 2 pp 145155 february 2000

    N. Goldman and P. Heuveline Health-seeking behaviour for child illness in Guatemala

    2000 Blackwell Science Ltd 155

    Appendix Estimated relative risk ratios for a multinomial model of the probability of seeing specific types of health care providers on a day ofillness, by characteristics of the illness and the child

    Pharmacist Health post Doctor Curer Other RRR P-value RRR P-value RRR P-value RRR P-value RRR P-value

    Illness characteristicsDay 1 1.00 1.00 1.00 1.00 1.00Day 2 1.09 , 0.75 1.72 , 0.17 1.21 , 0.60 1.79 , 0.28 0.90 , 0.78Day 35 0.55 , 0.03 1.51 , 0.26 1.08 , 0.82 2.31 , 0.09 0.80 , 0.53Day 6 1 0.40 , 0.01 0.82 , 0.67 0.66 , 0.32 3.45 , 0.02 0.60 , 0.24Other symptoms 1.00 1.00 1.00 1.00 1.00Respiratory only 1.03 , 0.93 0.84 , 0.67 0.69 , 0.35 0.15 , 0.01 0.75 , 0.60Fever, no gi 1.95 , 0.01 2.39 , 0.01 1.65 , 0.10 0.66 , 0.28 4.21 , 0.01Fever & gi 2.19 , 0.03 3.70 , 0.01 3.02 , 0.01 2.14 , 0.06 4.92 , 0.01Symptoms not serious 1.00 1.00 1.00 1.00 1.00Symptoms serious 2.37 , 0.01 1.40 , 0.25 3.45 , 0.01 2.12 , 0.02 2.46 , 0.01No hygiene beliefs 1.00 1.00 1.00 1.00 1.00Hygiene beliefs 1.11 , 0.80 1.87 , 0.16 3.10 , 0.01 0.33 , 0.28 7.11 , 0.01No folk illness beliefs 1.00 1.00 1.00 1.00 1.00Folk illness beliefs 0.93 , 0.85 0.65 , 0.49 0.59 , 0.33 2.96 , 0.01 0.84 , 0.76

    Child characteristicsAges 15 1.00 1.00 1.00 1.00 1.00Under age 1 0.54 , 0.04 1.77 , 0.03 1.24 , 0.44 3.09 , 0.01 2.66 , 0.01Parity 0.94 , 0.26 0.92 , 0.21 0.94 , 0.36 0.79 , 0.01 0.79 , 0.01Very poor-good health 1.00 1.00 1.00 1.00 1.00Very good health * ,* 0.83 , 0.86 3.45 , 0.03 10.94 , 0.01 3.63 , 0.05

    Pseudo R2 0.112Number of days 4323

    Reference category refers to no health care provider seen. *Parameter cannot be estimated because there are no children in very good health thatvisited a pharmacist. All models include a set of dummy variables to represent the four departments in which the EGSF took place. Alsoincludes clinics and health technicians. Other health care providers primarily include promotors (volunteers associated with the Ministry ofHealth who receive minimal training in basic health issues), midwives and nurses.