Drivers of respiratory health care demand in Acre state, Brazilian Amazon: a cross-sectional study

Respiratory illnesses are among the main sources of the global burden of disease according to [1], and the same is true in Brazil, where they are the third main cause of mortality [2]. A recent large international survey conducted by [3] with 5196 individuals from seven developed and seven developing countries, including Brazil, estimated that only 30% of patients with sore throat, a symptom associated with influenza-like illnesses and acute respiratory illnesses [4], seek a general practitioner. Similarly, high rates of refusal and delay in seeking care were found in national-level surveys [4‐7]. With this, the impairment of one’s own health and labour productivity may be extended [8], and mortality and local contagion may be increased [5, 6].

Such negative social effects are more probable in less developed regions such as the Brazilian Amazon. There, the per capita number of families receiving poverty alleviation cash transfers is 1.5 times that of the whole country [9, 10],¹ and large-scale exposure to pollution from agricultural fires is seasonal and comparable in its intensity to exposure to urban pollution in megacities such as London and Mexico City [11, 12]. Additionally, the prevalence of both tuberculosis (TB) and influenza-like illnesses is 1.3 times larger than the country’s average (of six cases per 10,000 inhabitants [10, 13];).² At the same time, in the Amazon, the number of physicians per capita is thirtyfold below the world average, and 0.09 hospital beds per 1000 inhabitants are available, compared with 3.2 in the world [14, 15].

In addition to the supply-side gap, previous studies in Brazil also have detected demand-side barriers to health care, such as a low capacity to recognize, both in oneself and in children cared for, diseases, their severity and contagiousness [16, 17], a limited capacity to obtain and use health-relevant information, mainly by the low-educated [18, 19], household income incompatible with the transport costs required for visiting a health facility, which are more likely for rural residents [20], and reliance on self-medication [21].

Nevertheless, none of these studies is based on patient surveys uncovering factors enabling and disabling care seeking for respiratory illnesses. Such surveys are rare in Brazil and, as far as can be assessed, are not available for the Amazonian region. Nevertheless, to stimulate health care seeking, there is a need to detect patients’ characteristics predicting whether such a course of action would be pursued spontaneously or not. This is required for knowing which social groups should be prioritized by specific interventions [22], such as behavioural nudges [23] and educational campaigns [7], and how these must be designed. Seeking to fill such gaps, this study detects predictors of health care seeking among respiratory ill individuals in Acre state, western Amazon.

Twelve recently published papers were consulted to identify potential factors influencing the demand for health care, as a basis for selecting covariates for quantitative analysis and for discussing and complementing the results. Summaries of all papers are provided in the Additional Information, and here, a short synthesis and taxonomy are presented.

The revised papers highlighted the relevance of multiple factors, including the characteristics of patients, social groups (households and communities), social contexts and macroeconomic features such as levels of development and economic growth (Table 1). These factors operate through multiple mechanisms at the individual (e.g., identification of illness from symptoms), household (e.g., women autonomy to seek care for children), community (e.g., support for seeking care) and macroeconomic levels (e.g., economic growth expanding national health systems). Therefore, the demand for health care is both a multidimensional and multilevel process that can hardly be apprehended with only one research method. This is clear from the larger diversity of factors in the literature compared with the datasets used in this paper (Table 1). Nevertheless, it also is clear that a significant fraction of factors is captured, with the uncaptured (macroeconomic) factors being left to be accounted for while discussing the results (Conclusions). Additionally, the focus on demand-side variables is coherent with the consulted literature, especially with the finding by [5, 6] that delays in receiving TB treatment were mostly demand-driven.

Acre state, located in the western Brazilian Amazon, is among the least developed and health care-endowed states of Brazil, ranking 21^st of the countries’ 27 states in human development [30] and 23^rd in physicians per capita [10, 15]. Despite limited resources, respiratory illnesses are targeted by multiple actions, including annual vaccination campaigns [31] and the Family Health primary care programme, including home visits by community health agents (CHAs), who provide key advice on, for instance, home-based preventive measures, health care seeking and the administration of prescribed drugs.³ Medication for asthma and rhinitis also is subsidized [32].

The prevalence of respiratory illnesses in Acre state is summarized in Table 2, based on the most updated set of ambulatory visit records available at the time of the cough survey (2019). The four most frequent respiratory disease classes were, in decreasing order, (i) acute pharyngitis and tonsillitis, (ii) other acute infections of the upper airways, (iii) influenza-like illness or flu and (iv) pneumonia. Their joint frequency amounted to 78%. In complement, the TB incidence rate in Acre state was of 5.5 confirmed cases per 10 thousand inhabitants in the average across 2013 to 2019 [53, 54], a prevalence rate that was 13-fold smaller than that of pneumonia, one of the top four respiratory illnesses in Table 2; such comparison is nevertheless limited by TB frequency being based not only on ambulatory visits but also on visits to emergency rooms and hospitals. Regarding approximate aetiology, diseases caused by viral or bacterial pathogens (at least one of those) were most frequent both in Table 2 and based on respiratory mortality data for Acre state from 2013 to 2018 [55], with 56% of ambulatory visits and 47% of deaths (approximately 90% of the latter owed to pneumonia and approximately 9% to TB). Mixed (asthma), environmental and behavioural (smoking) sources accounted for 10% of morbidity and 42% of mortality, and this last figure was nearly entirely due to chronic obstructive diseases.

A structured quantitative survey was conducted in Rio Branco (~ 400 k inhabitants), the most populous town of Acre state, from March to July 2017, following a pilot wave in October–November 2016 with 146 participants. The goal was to elicit the willingness to pay (WTP) to avoid the recurrence of the most recent respiratory illness episode. A hypothetical vaccine was offered as a product capable of yielding full prevention. Since WTP estimates are not directly related to the goal of this paper,⁴ only the survey’s first module, on the description of the illness episode, and the third module, on the socioeconomic and health attributes, are detailed here.

The sample was selected based on a mix of conglomerate-quota randomization with the help of local CHAs (details are found in [56], SI, Section 4). In the first stage, ten neighbourhoods of Rio Branco’s urban perimeter were selected, initially randomly, and, afterwards, depending upon whether CHA support proved insufficient,⁵ new neighbourhoods were introduced as substitutes, being selected to ensure the representativeness of income and respiratory illnesses’ prevalence at the whole perimeter level. The shares of eight combinations of two genders and four age intervals (0 to 4, 5 to 17, 18 to 65, above 65) in the study region’s population were moderately altered to end with considerable numbers of sampled children and elderly. Such percentage quotas, after multiplying by 18, which was the number of interviews a CHA could support, were informed to such professionals, who were instructed to write down a patient’s list with a size two to fourfold that of each absolute quota. Research assistants then drew randomly, for each age–gender group, a number of individuals matching the quota. The interviews were then conducted in interviewees’ houses, and whether selected individuals were not found or refused to participate, the next listed subject of the age–gender group was approached (this was repeated in case of a new failure). Enumerators were accompanied by CHAs in interviews, which both minimized rejection and helped, whenever needed, with rephrasing questions in ways more understandable to respondents.⁶

A total of 519 valid interviews were conducted for the purposes of this paper. In 278 interviews, adults responded about illnesses acquired by them, and in the remaining 241, they responded about illnesses acquired by their children. Caregivers, generally mothers or fathers, were interviewed on behalf of children.

Now turning to the questionnaire, its first module retrieved descriptive data on the most recent illness episode, including (i) a sequence of yes/no/can’t remember questions on whether 21 symptoms were felt,⁷ (ii) a three-point Likert scale on the subjective pain experienced, (iii) the symptoms beginning date and duration, (iv) whether medical care was sought and (v) diagnosed illness.⁸ Children aged above 5 years were asked to confirm caregivers’ responses. The third module asked about age, household size, years of schooling (according to the Brazilian educational system), and household income. The latter was asked in a comprehensive way comprising six main income sources. This was complemented by the inquiry on the ownership of durable goods that were highly correlated with income according to the National Household Survey [57]. Additional questions covered smoking, health insurance, distance to the health unit where treatment for the first module’s illness could be found, and seven serious lung diseases (asthma, bronchitis, lung cancer, chronic obstructive pulmonary disease, emphysema, pneumonia and TB).

From December 2019 to the middle of March 2020,⁹ a new survey designed as an improvement of the one described in the previous section also was run in Acre state, but this time it also included its second most populous municipality, Cruzeiro do Sul.¹⁰ The survey consisted of a discrete choice experiment estimating the WTP for a hypothetical remedy¹¹ able to reduce the number of days with dry cough. Thus, its main particularity was the focus on one single respiratory symptom, which was selected from medical literature due to its high recurrence across respiratory illnesses [58, 59].¹² Dry cough was considered because its suppression is more commonly targeted by treatment than wet (with phlegm) cough [62‐64], which made the selection of the attributes of the remedy easier (with antitussive drugs being the focus).

The same sampling and children inclusion approaches of the respiratory illness survey were adopted. The modules used in this paper also were near-equivalent to those of the previous survey, except for focussing on the latest dry cough episode and inquiring about income in the form of eight intervals,¹³ whose midpoints were used in the analysis, instead of the open-ended format adopted in the previous survey. Before the final stage of the survey, a prepilot comprising 33 interviews was conducted in July 2019, and a pilot with 54 interviews was conducted in December 2020. The latter is included here as part of the analysed data due to minor differences with the final-stage questionnaire.

In the respiratory illness survey, two research assistants were trained by the first author and monitored in the field during the pilot wave. These last-year undergraduate students also were qualified to train enumerators. All questionnaires were revised after they were applied, and if needed, the enumerator returned to respondents’ houses, or the questionnaire was discarded. In both cases, the decision was based both on the missing or ill-filled fields and the last questionnaire module, the latter filled out by the enumerator alone, as an assessment of the quality of the responses received. In the cough survey, similar procedures were adopted, with the exception that the whole team was trained by the first author himself (except for one enumerator who conducted 18 interviews).

Standard central tendency, position and dispersion statistics were applied to describe the data. The proportion difference test, based on the standard normal distribution, was applied to compare the adult and child subsamples. Econometric analysis was pursued with standard index multivariate regressions for discrete dependent variables (whether health care was sought), whose disturbances followed either the Gaussian cumulative distribution function (“probit”) or the logistic function (“logit”). Estimators’ standard errors were robust for heteroscedasticity ([65], Chap. 14, [66], Chap. 15). Only models that were globally significant, as informed by a chi-square test of nullity for all covariates’ coefficients ([66], Section 15.5.1), are reported.

In the respiratory illnesses survey, the average age and income were 30 years and 2.5-fold the minimum wage, respectively, and the frequencies of respondents receiving poverty alleviation cash transfers and with a lower secondary educational level were 30 and 62%, respectively (Table 3). The latest respiratory illnesses reported by participants caused 10 symptoms, lasted 10 days and inflicted moderate to extreme pain in 95% of cases. Only 14% of the episodes occurred in the dry season (i.e., from June to September, in the case of Acre state). Only 38% of respondents sought health care, 32% when the illness was acquired by an adult and 46% when a child was the patient, a statistically significant difference (p value < 5%). A specific disease was diagnosed in 79% of cases, and influenza-like illness was the most recurrent diagnosis (64%), followed by pneumonia (12%) and then asthma (6%), with these three collectively accounting for 82% of diagnosed diseases (Table B.1 in the Additional Information document). Viral or bacterial aetiology was dominant (79% of cases). All symptoms with high frequency (> 50%) either attacked the nose or throat or caused cough (dry and wet), fever, headache, body aches, shortness of breath or malaise.

In the cough survey, income was smaller, twice the minimum wage, in line with a higher rate of poverty of 47%,¹⁴ and age and education were near-equivalent to the data from the respiratory illness survey (Table 4). On average, the most recent dry cough episode lasted 10 days, caused moderate to extreme pain in 78% of cases and occurred in the dry season in 38% of cases. The overall health care seeking rate was 33%: 26% for adults and 39% for children. A significant difference was found at the 10% significance level, and the rate of (pharmacy-purchased) medication was 66%: 57% for adults and 75% for children, with a statistically significant difference at the 5% level. A specific disease was diagnosed in 66% of cases, with influenza-like illness again being the most recurrent diagnosis (67%), followed by pneumonia (8%) and asthma (8%). These three diseases amounted to 83% of diagnosed diseases; two cases of TB were reported (Table B.2 in the Additional Information document). Viral or bacterial aetiology was dominant (83% of cases).

In summary, most respondents were of low to medium income, reported a moderately painful influenza-like illness episode over 10 days and were more likely to seek care and purchase medication for children than for adults.

Three models were estimated for each survey. First, there was a baseline model with all factors suggested by the literature review that could be retrieved in the surveys (as in Table 1 above). Second, there was an extended version of the former, including a female gender vs. household size interaction in order to capture that the more members there are in the household, the smaller may be the time female members allocate to domestic tasks and thus the larger the time available to seek health care (this captures intrahousehold effects as in Table 1). Third, there was a child-only model to test whether the kinship tie between child and respondent, supposedly the caregiver, i.e., whether the latter was the mother, father or other relative, influenced health care seeking (which again captures the influence of intrahousehold phenomena).

The results in Tables 5 and 6 show that health care seeking was mainly driven, considering the two surveys, by health condition characteristics, first of all, subjective pain but also severity as measured by the number of symptoms (significant only in the respiratory illness survey) or duration. Factors that were significant only in one survey corresponded to respondents’ health, specifically whether a lung disease was ever experienced (cough survey), and to individual characteristics, specifically age (respiratory illness survey), and, at last, household characteristics, particularly household size (cough survey). In addition, in the cough survey, whether the symptoms occurred in the dry season increased care seeking, as expected, due to the better condition of roads and thus the easier access to facilities and because in the colder dry season, cough could happen together with more or more severe symptoms [67]. There were three factors whose insignificance was less expected. The first was education, revealing that schooling may not capture health knowledge or that the latter was not significantly related to care seeking in the particular context. The second, income, was due to respondents resorting to free care offered by the public system (which also explains why poverty was irrelevant). The last, distance to health unit, could have been caused by most respondents living close to units (75% within 2 km in the illness survey and 94% within 30 minutes in the cough survey), a consequence of the considerable geographical diffusion of primary care facilities and, probably, of CHAs’ priority towards patients located nearer to facilities in routine home visits.

Among the remaining nonsignificant factors, waiting time inside the health unit (cough survey) was irrelevant, probably due to insufficient variation (58% waited up to 1 h). Resort to nonprofessional health care (cough survey), which mostly captured household-made medication, was nonsignificant, probably because it was not a question posed to all respondents but, instead, a spontaneous statement made as part of the choice experiment whenever medication was not chosen. This probably biased the variable’s significance downwards. Household-level phenomena, captured by the interaction of female gender and household size and by the relative who answered on behalf of the child (supposedly the caregiver), did not significantly influence the health care-seeking decision.

In this section, the main results are concisely linked with the literature classified in Table 1 for each class of factors influencing health care seeking. The macroeconomic factors class is omitted for not being at a scale compatible with a survey on individuals, but they are addressed in the next subsection (on other factors and policy).

Some words are merited on the macroeconomic factors that were not directly controlled for. The linking of results with the macroeconomic evidence provided by [8, 27] suggests that subjective factors preventing respiratory ill individuals from seeking health care (such as pain being sensed as tolerable) have macroeconomic implications. The duration of the temporary human capital reduction is extended at the individual level and, possibly, also at the social level in the case of viral and bacterial diseases [5]. Economic growth may be diminished through the long-run relationship that connects it with health care demand and health level, according to the mentioned articles.

Any consideration of implications for policy-making should be made in light of two key pieces of evidence. First, the low level of care seeking, with at most 40% of respondents seeking care, in contradiction with the severity of the respiratory conditions respondents experienced, which lasted 10 days, involving ten symptoms on average, or at least dry cough, and mainly causing moderate or extreme pain. What should be noted is not at odds with previous studies ([3, 4, 7]).¹⁸

The second relevant information is the most frequent classes of diseases in the target population. In this sense, even with the disease most frequently reported by participants (influenza-like illness) belonging to the acute viral category, thus requiring mild medical intervention, the second most reported was pneumonia, a contagious disease for which treatment is needed, thus avoiding complications that could lead to hospitalization and even death [50], with the latter especially likely in the case of small children [70]. Contagion inside the household, nursing homes and local communities also is a worry [50]. A considerable frequency of diagnosed pneumonia (7%) also was observed in Acre state’s ambulatory records (Table 2),¹⁹ together with a 30% frequency of acute pharyngitis and tonsillitis, which also are transmissible diseases.

Therefore, synthesizing the two pieces of evidence, the health care-seeking gap is both sizeable and a cause of avoidable human and economic costs. Pneumonia transmission is a clear demonstration of such costs, but influenza-like illnesses, pharyngitis and tonsillitis, even requiring milder medical intervention, may lead to work absenteeism, lower productivity and failure to undertake household tasks, which compress human capital and GDP [71]. Whether it pays off to foster increased health care seeking for these nonsevere diseases depends on the intensity of the trade-off between the cost imposed on the health system and on households and the remaining economy. Even being out of the scope of this paper to measure such intensity, by highlighting the need for assessing it as part of the process of improving Acre state health policy, an already useful policy contribution is made.

The results also support the conjecture that supply-side policy recommendations, such as performance-based increases in government expenditures on health care (suggested by [27]), may bring less-than-expected results due to demand-side barriers. One clear barrier is the subjective underestimation of respiratory illnesses’ consequences, both for own health and for the broader society, as suggested both by a rate of care seeking not above 40% and by the dominant role of subjective predictors. In this regard, a two-pronged policy strategy may be successful. First, regarding respiratory illnesses that are both severe and transmissible, implementation of microeconomic behavioural policies seeking to change subjective perceptions of symptoms and illnesses, via nudges [23] but also through educational health interventions (as proposed by [6, 7]), is advisable. One possibility, suggested by the lower likelihood of adults seeking care compared to children, is CHAs’ educational actions stressing that adults’ illnesses, if contagious, could be acquired by their children. Influenza vaccination campaigns that achieve over 80% coverage in Acre state [31] also could better inform the population about the symptoms and contagiousness (as recommended for TB by [6]) of severe diseases. A specific study on alternatives to increase health care usage during the wet season is needed, which requires assessing the possibility of temporarily expanding health workers’ teams in rural areas when rain is intense enough to prevent rural-urban displacements.

Regarding nonsevere diseases, such as influenza-like illnesses, campaigns informing about household-based management would probably be more cost-effective, avoiding the overload of an already deficient health system (according to Section 2.3). In particular, primary care home visits could be intensified by hiring more CHAs. However, again, assessing the trade-off between the health system and economic costs is needed to know whether some of the recommendations in the previous paragraph also are reasonable for nonsevere diseases.

Two key limitations are worth stressing as gaps to be filled by future research. First is the circumscription to urban centres, which, albeit concentrating most of the population, may have diminished accuracy in the estimation of the significance of potential predictors, with travel cost being a clear example. Relatedly, Brazilian Amazon-wide valid estimates of predictors’ influences would require expanding the survey to the other eight states integrating the region. Second, health knowledge and reliance on traditional medicine were proven to require specific and detailed sets of questions, the same also being true for intrahousehold interactions and decision-making. Finally, the results achieved here are generalizable only to other developing countries that also have a public health system targeting universal coverage and offering free health care (as Malawi [26]).