Patterns of healthcare seeking among people reporting chronic conditions in rural sub‐Saharan Africa: findings from a population‐based study in Burkina Faso

Non‐communicable diseases are rapidly becoming one of the leading causes of morbidity and mortality in sub‐Saharan Africa. Yet, little is known about patterns of healthcare seeking among people with chronic conditions in these settings. We aimed to explore determinants of healthcare seeking among people who reported at least one chronic condition in rural Burkina Faso.


Introduction
Non-communicable diseases (NCDs) are one of the major challenges of the 21st century. Each year, they are estimated to account for around 15 million premature deaths between the ages of 30 and 69 [1]. Facing a particularly dramatic epidemiological transition due to increased longevity, poor dietary behaviours and limited physical activity, today already more than three quarters of global NCD deaths occur in low and middle income countries (LMICs) Data collection was possible thanks to funding by the World Bank Health Results Innovation Trust Fund as part of the Impact evaluation of the Performance Based Financing programme rolled out in the country. We are grateful to the funding agency and the Ministry of Health in Burkina Faso for the opportunity to collect additional data on community health and use them for independent analyses. 1542 [2,3]. In sub-Saharan Africa (SSA), NCDs are projected to exceed communicable, maternal, perinatal and nutritional conditions as the most common causes of death by 2030 [4]. In 2017, the Global Burden of Disease Study estimated the age-standardised disease burden in SSA to be almost equal for NCDs and communicable diseases, with more than 20 000 DALYs lost per 100 000 population [5].
In light of this emerging disease burden, healthcare systems in SSA are often unable to meet the needs of people living with NCDs, due to shortages in supplies and equipment and constrained capacity to distribute resources evenly among population groups [6]. In spite of the efforts made to foster universal health coverage (UHC) and to increase social health protection [7], as stipulated in the 2030 Agenda for Sustainable Development [8], NCD care is still largely subject to user charges in many SSA countries [9][10][11][12]. Payments at point of use have repeatedly been shown to limit access, especially for the most vulnerable and socio-economically disadvantaged segments of the population [13], ultimately aggravating their disease burden [14].
Care for NCDs (hereafter referred to as chronic conditions, excluding human immunodeficiency virus and tuberculosis) requires continuity in form of regular access to affordable healthcare services. To date, only a handful of studies have addressed healthcare seeking for people affected by chronic conditions in SSA, documenting the role played by socio-demographic [11,15], economic factors [11,[16][17][18][19], perceived severity of illness [17,18] and health system factors [18]. Specific to Burkina Faso, chronic conditions have been estimated to cause almost one third of all deaths in 2017 [20], yet only one study has so far explored related healthcare seeking patterns, but focused exclusively on women with chronic breast-related symptoms [21]. This study explored determinants of healthcare seeking among people with chronic conditions in rural Burkina Faso.

Study setting
In Burkina Faso, around 70% out of a population of approximately 18 million live in rural areas [22]. Most healthcare provision occurs within a three-tiered system of over 2000 public health facilities distributed across 63 health districts [23]. Medical pluralism persists, with traditional healers operating in parallel to formal healthcare providers. In 2015, the government introduced a 5-year National NCD Strategy to reduce NCD mortality [24]. This strategic plan indicates that NCD care, including laboratory-based work-ups and glycaemia management, should be offered at the second level of the primary healthcare system, specifically at district hospitals [25]. While user fees for all services for maternal and child healthcare services have been abolished in June 2016, NCD care remains subject to user payments at point of use. In 2014, the government of Burkina Faso launched a Performance Based Financing (PBF) pilot targeting maternal and child healthcare services in 12 districts distributed across six regions [26]. NCD care was neither part of the PBF incentivised service package nor of any other targeted intervention having taken place in the country over the past decade.

Study design and data sources
Our cross-sectional study used data from the endline household survey conducted between March and June 2017 in 24 districts (the above-mentioned 12 PBF intervention districts plus an additional 12 districts selected as controls) within the framework of the PBF impact evaluation [26]. The sample was derived in a three-stage cluster sampling procedure, described in detail elsewhere [26]. In brief, clusters were defined in relation to primary healthcare facility (Centre de Sant e et Promotion (CSPS)) catchment areas. Within each cluster, one village was randomly selected, and within each village, 15 households were randomly selected among households with at least one woman currently pregnant or having completed a pregnancy in the prior two years. As such, the overall sample is biased to include households with a higher proportion of women of reproductive age than what would be observed in the population at large.
The survey collected information on household socio-demographic and economic characteristics from household heads or representatives. A distinct module asked all household members to report up to three chronic conditions and related health seeking behaviour. Primary caregivers responded on behalf of children below the age of 14.

Variables and measurement
In line with prior studies conducted using self-reported NCD morbidity [11,16], we operationalised our outcome, healthcare seeking for individuals with chronic health conditions as an individual's self-reported illness status and related health service use in the four weeks prior to the survey date. We distinguished three use categories: formal provider, informal provider and no care. The choice of explanatory variables was based on the Andersen's behavioural model of health service use which defines an individual's healthcare seeking behaviour to be modulated by barriers and facilitators classified as (i) predisposing, (ii) enabling, (iii) need and (iv) health system factors [27]. Table 1 provides an overview of all variables included in our study and their measurement.

Analytical approach
First, we descriptively analysed the data, using chi-square (v 2 ) tests to investigate potential selection bias in our study sample compared to the overall survey population. Second, we used three regression models (Model A, B and C) to capture associations between explanatory variables and health service use categories. Model A and Model B were logistic regression models comparing 'no care' to 'formal provider' (Model A) and 'no care' to 'informal provider' (Model B). For models A and B, bivariate regressions were conducted by including only one of the explanatory variables. Model C was a multiple multinomial logistic (MNL) regression analysis including all three service use categories. 'No care' was defined as the reference category in all models. We relied on robust standard errors to adjust for clustering at the household level. Significance level a was set at 5%. Complete case analysis was conducted using Stata version 15 IC, and residual plots were performed using R version 3.5.2.

Results
Of 52 562 surveyed individuals, 1173 reported at least one chronic condition. Among those, we had complete information on 1124 (95.8%) individuals, who, compared to the population sample, tended to be: older, Christian, household heads and living in households led by a literate head (see Appendix, Table S1). Most frequently reported chronic conditions were related to gastric ulcers (n = 348), joint pain (n = 323) as well as cardiovascular problems operationalised as heart problems (n = 184) and hypertension (n = 157); the least frequently reported conditions were diabetes (n = 12) and cancer (n = 5) ( Table 2). Table 3 shows the sample distribution according to our outcome variable classification. Of 1124 individuals, 749 (66.6%) did not seek any care for their chronic condition in the four weeks prior to the survey, while 256 (22.8%) and 119 (10.6%) sought care at formal and informal providers, respectively. Table 4 displays results from the three multiple regression models (A, B, C). Relevant results from the bivariate analyses are shown in Table S2. For the MNL regression analysis (Model C), both the suest-based Hausman and the Small-Hsiao test confirmed that the model did not violate the independence of irrelevant alternatives (IIA) assumption. Model C shows that households with the *Based on assumption of intra-household allocation [28]. †Constructed through an asset-based approach using multiple correspondence analysis [29]. ‡According to average numbers from most recent population statistics [30]. §Based on WHO classification of non-communicable diseases [31]. ¶Derived from Global Positioning System coordinates measured in km and dichotomised in close and far using the median distance, that is 5 km to CSPSs and CMs and 25 km to CMAs and CHRs, respectively. #Has been described in more detail elsewhere [26].

Discussion
As one of the very first studies assessing healthcare seeking practices for chronic conditions in rural SSA, and to our knowledge the first population-based one in rural Burkina Faso, our study makes an important contribution to the scarce literature on the topic. Although the survey was not initially designed to capture chronic condition reporting and associated healthcare seeking, its geographical scope offered an unprecedented opportunity to collect relevant information on a sample covering over one third of all districts in the country. This wide coverage is a key feature of our study and sets it apart from other studies on healthcare seeking for chronic conditions in rural SSA whose geographical focus is much more restricted [11,[16][17][18][19]21]. The first finding deserving particular attention pertains to the small proportion of individuals, approximately 2% of the total sample, who reported a chronic condition. At first glance, this low proportion may appear startling given that epidemiological studies indicate the prevalence of hypertension and of diabetes to be, respectively, at 14.8% and 4.4% among adults aged 25-64 years in rural communities [32]. This discrepancy between our findings and published epidemiological estimates, however, can be explained in light of the fact that unlike the STEPS surveys [33], our study included all population strata and not exclusively adults. Moreover, our population-based survey did not attempt to measure the actual prevalence of NCDs, but purposely relied on individual self-reporting of all chronic conditions as an indication of illness awareness and as such, as the essential starting point in the healthcare seeking process [27]. Nevertheless, this discrepancy between self-reported and epidemiological data remains worrisome since it suggests that many people in rural Burkina Faso may not be aware of asymptomatic early stage conditions and of their related early healthcare needs.
To this regard, findings from our population-based survey resonate with the wider literature on care cascades for NCDs [34] and suggest that screening services are probably not adequate to create sufficient awareness of one's illness status. For instance, the fact that in a sample of over 50,000 individuals, only 15 reported being diabetic and only 5 reported being affected by cancer denotes fundamental shortcomings in respect to early detection and routine screening for common chronic conditions [21,24], a challenge previously highlighted by facility-based studies in East Africa [35,36]. In a context characterised by poor geographical accessibility [11], the fact that district hospitals, rather than CSPS, serve as entry care points for chronic conditions likely introduces a substantial barrier to early diagnosis and proper disease management [24]. Individuals in need of early therapeutic interventions may thus go undiagnosed for years until their health status worsens and more likely result in premature deaths [37].
The second relevant finding pertains to the fact that even among those who reported a chronic condition, less than one fourth sought care at a formal health facility in the month prior to the survey date. While not all chronically ill individuals may need care on a monthly basis,    the utilisation rate detected in our study is worrisome and below the one reported in Kenya (33.1%) [16], and Malawi (42.5%) [17], but aligned with the one reported in Nigeria (20.1%) [19]. Low utilisation rates suggest untreated chronic conditions, which contribute to a rising burden and premature death [37]. These findings point at the urgent need to review existing service provision models and call for the integration of screening and early management of chronic conditions in the CSPS service package to increase the uptake of treatment seeking at formal care providers. Such a shift would entail closing implementation gaps to translate into everyday practice what is already outlined as an objective in the national NCD plan, that is implementing the WHO Package of essential non-communicable disease interventions (PEN) at primary level [38]. Third and last, our results from the MNL shed some light on existing barriers to accessing care. Similar to findings from Kenya [16], Ethiopia [18], and Nigeria [11], individuals from wealthier households were found to be more likely to seek care at formal providers. This underlines the decisive role that households' ability to pay plays in shaping access to services, suggesting the persistence of pervasive inequities in fee-based systems [39]. Albeit beyond the scope of our study, reliance on direct user payments is also likely to result in extreme financial distress for those households which decide not to forego care, as suggested by prior literature [17]. In line with prior evidence from Malawi [17], the fact that individuals from households with multiple chronically ill members were less likely to seek formal care provides additional evidence that considerations on affordability and intra-households resource allocation limit proper health seeking [28]. Our findings suggest that policies aimed at reducing morbidity and mortality due to chronic conditions must not only aim at reforming healthcare provision structures, bringing quality services closer to communities, but also aim at removing direct payments at point of use, possibly expanding the scope of the recent gratuit e policy [38].
Echoing results from studies conducted in Malawi [17]Ȩ thiopia [18], and Uganda [15], perceived illness severity was observed to increase the probability of seeking care at both formal and informal providers. This finding suggests that in the absence of functioning early screening and detection options, people act upon their illness only when major discomfort arises at an advanced disease stage. Considering the potential negative public health consequences derived from such behaviour and in line with what already mentioned earlier, this finding calls for the immediate expansion of early detection and routine screening, accompanied by proper information campaigns to increase awareness on the need to seek care even during asymptomatic stages [15].
Contrary to what was reported in prior literature for Nigeria [11] and Uganda [15], our findings do not highlight the existence of a gender bias in health seeking for chronic conditions. This discrepancy between our findings and prior published literature is interesting and surely deserving of further qualitative inquiry. It is curious that confirming findings from a very early study conducted in Burkina Faso [28], our work suggests that while resource scarcity appears to shape decisions on intra-household resource allocation, gender does not do so, at least not sufficiently strongly to be captured by our quantitative models. Burkina appears to stand aside from its continental counterparts on such matters, and understanding why this may be the case is surely worth investigating to inform policy developments adequately.
In line with the early onset of symptoms and premature health decline noted by Miszkurka et al. (2012) [40], our findings suggest that younger individuals are more likely to seek treatment at informal providers. It is likely that due to resource constraints, younger individuals, not yet affected severely, stay away from a healthcare system that may impose a considerable financial burden upon seeking care. As such, most likely, informal providers are used to overcome potential barriers in accessing formal services. Again, in line with what stated earlier, given the central role that early access to treatment plays in curbing the chronic disease epidemic, this calls for action in improving accessibility and affordability of adequate healthcare services, following the example set by LMICs, which introduced health insurance coverage as a an important policy tool [41].

Methodological considerations
Besides its originality and the strength derived by a large population-based sample covering almost a third of the country, several limitations inevitably affect this study calling for cautious interpretation of results. First, we assessed healthcare seeking behaviour within a 4-week recall period, potentially introducing recall biases that lead to incomplete data and a subsequent underestimation of the actual number of individuals seeking care after reporting a chronic condition. Second, measurement bias could have affected our results since we relied on self-reported health status information from individuals rather than objective disease measures. This leaves open the possibility that people could not distinguish between acute and chronic conditions. Yet, such limitation does not threaten the validity of our results per se since we explicitly aimed at exploring healthcare seeking behaviour based on one's own reported health. This goal was further enriched by applying a prominently used theoretical framework allowing the inclusion of variables that holistically reflect important determinants in health service use. Third, the households' sampling strategy (privileging households with a recent history of pregnancy) might have resulted in a sample not truly representative of the Burkinab e population. Last, we cannot exclude illness reporting bias, meaning that some individuals, in our case household heads and individuals who were of older age, Christians and living with a literate household head, were more likely to report a chronic condition. It is unlikely that these features characterise the underlying distribution of chronic conditions at the population level since various authors identified variables such as socio-economic status [40,[42][43][44], age [40,42,44], gender [40,44], household food insecurity [42] or urban residency [42] as characterising features for chronic conditions and their risk factors. In our case, it is rather likely that they are an indication of illness awareness, possibly introducing some bias in the truncated sample used for analysis [45].

Conclusion
Findings from this population-based study suggest the urgent need to invest in interventions aimed at improving service provision and enhancing affordable access for early detection and management in peripheral settings in order to reduce the burden of chronic health conditions. Our study calls for further qualitative inquiry into the reasons inducing people to forgo formal care seeking as means of understanding more comprehensively their implications and inform policy design adequately.

Supporting Information
Additional Supporting Information may be found in the online version of this article: