Objective To identify determinants of intended versus actual care-seeking behaviours in a pluralistic healthcare system that is reliant on both conventional and non-conventional providers and discover opportunities to catalyse improved healthcare access.
Design Cross-sectional study.
Setting and participants In Haiti 568 households (incorporating 2900 members) with children less than 5 years of age were randomly sampled geographically with stratifications for population density. These households identified the healthcare providers they frequented. Among 140 providers, 65 were located and enrolled.
Outcome measures Household questionnaires with standardised cases (intentions) were compared with self-recall of health events (behaviours). The connectedness of households and their providers was determined by network analysis.
Results Households reported 636 health events in the prior month. Households sought care for 35% (n=220) and treated with home remedies for 44% (n=277). The odds of seeking care increased 217% for severe events (adjusted OR (aOR)=3.17; 95% CI 1.99 to 5.05; p<0.001). The odds of seeking care from a conventional provider increased by 37% with increasing distance (aOR=1.37; 95% CI 1.06 to 1.79; p=0.016). Despite stating an intention to seek care from conventional providers, there was a lack of congruence in practice that favoured non-conventional providers (McNemar’s χ2 test p<0.001). Care was sought from primary providers for 68% (n=150) of cases within a three-tiered network; 25% (n=38/150) were non-conventional.
Conclusion Addressing geographic barriers, possibly with technology solutions, should be prioritised to meet healthcare seeking intentions while developing approaches to connect non-conventional providers into healthcare networks when geographic barriers cannot be overcome.
- health policy
- international health services
- community child health
- public health
Data availability statement
All data relevant to the study are included in the article, uploaded as supplementary information and uploaded to the Dryad data depository. Extra data can be accessed via the Dryad data repository at http://datadryad.org/ with the doi: 10.5061/dryad.wpzgmsbkm.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Strengths and limitations of this study
Inclusive of both conventional and non-conventional healthcare providers, reflecting pluralistic healthcare systems in resource-limited settings.
Uses randomised geospatial sampling methods to minimise the risk of sampling bias that results from logistical constraints to contact isolated and remote households.
Employs a unique application of network-analysis visualisations to quantify relationships between families and their healthcare providers.
A limitation was that comparisons between healthcare seeking intentions (‘would do’) versus healthcare seeking behaviours (‘did do’) were not drawn from exactly equivalent illness scenarios.
A limitation was that non-conventional healthcare providers were more difficult to locate and enrol than conventional, and therefore generated missingness in the dataset. In addition, provider missingness (conventional and non-conventional) may have resulted in an underestimation of households referred from the periphery of the healthcare network.
Improving access to healthcare is one of the highest global health priorities set by the Sustainable Development Goals (SDG) (2015).1 SDG 3.8 seeks to ‘achieve universal health coverage (UHC)’, however, the current rate of progress is insufficient to reach this target by 2030. Low/middle-income countries (LMICs) are the furthest off track,2 3 and UHC tracking indicators show no significant gains for children between 2010 and 2020.4 The COVID-19 pandemic will exacerbate the limited progress.5 Innovative approaches are needed to overcome persistent barriers to catalyse progress to achieve universal healthcare access.
Common barriers are accessibility, availability and acceptability.6 Determinants inside and outside of households influence when a potential barrier manifests as an actual barrier. One method to prioritise determinants for action is to investigate healthcare seeking behaviours starting with illness recognition (‘is the family member sick?’) and response (‘what should the family do?’). Determinants associated with ‘recognition’ and ‘response’ include biomedical understanding of illness,7 ability to recognise danger signs,8 caregiver’s perception of illness severity,9–11 age,12 level of education7 8 and marriage status,11 number of symptoms,11 gender,13 14 rural/urban location,7 15 intra-household relationships,16 17 distance,15 18 19 finances10 17 19 and wait times.18
Once a decision is made to seek care, factors that influence provider selection reveal additional barriers. LMIC healthcare systems are poorly defined with a range of conventional and non-conventional providers; alternative terminologies are qualified/unqualified or formal/informal.20 Parallel and differential access to conventional and non-conventional providers manifests in chaotic systems.21 Conventional providers can be defined as licensed doctors and nurse practitioners at governmentally recognised public, non-governmental organisation (NGO) or private facilities. Non-conventional providers can be defined as traditional healers, medication vendors, unlicensed practitioners and pharmacists unlicensed to independently write prescriptions. Provider selection exposes conflicts between a patient’s intention (‘would do’) and behaviour (‘did do’).22 23 These conflicts represent an opportunity to reveal unanticipated solutions to improve access to care.
The associated networks within pluralistic healthcare systems rely on the relationships among and between conventional and non-conventional providers.24 25 Conventional primary care providers, including community health workers, are assumed to be the first access point into healthcare systems,26 however, non-conventional providers have an important and underappreciated role.27 Non-conventional providers often practice in parallel without disclosure to conventional providers which creates alternative pathways for seeking care.28 These complex relationships are not adequately understood, yet are essential to improving healthcare access.25 29 30
To address these knowledge gaps Haiti was chosen as a setting generalisable to areas with healthcare systems stressed by factors such as high rates of poverty, volatile political leadership, economic instability, security concerns and limited investment in physical infrastructure. Access to high-quality healthcare in Haiti is low at 23% nationwide and 5% among the rural population.31 Non-conventional healthcare services and traditional medicine are common.25 32–34 Child health and wellness is substandard with Haiti ranking 150 out of 180 countries.35 The under 5-year mortality rate is 67/1000 live births compared with 41/1000 live births globally.36 Healthcare seeking behaviour in Haiti has been evaluated for mental health,37 prenatal care,38 childbirth location39 and cost.40 41 Our objective was to identify actionable knowledge gaps on how to improve access to quality healthcare by identifying determinants of intended versus actual care-seeking behaviours in a pluralistic healthcare system that is reliant on both conventional and non-conventional providers.
Study design and participants
In this cross-sectional study, consenting participants were enrolled at the household and healthcare provider levels. Household inclusion criteria were households with an adult (18 years or older) head-of-household (HoH) and at least one child under 5 years. Provider inclusion criteria were adult healthcare providers identified by an enrolled household and located within the study area. The 477 km2 study area, encompassing the communes of Gressier and Leogane Haiti, was divided into 2 km2 grid-cells (figure 1A). Population density was determined by approximating the number of structures within the grid cell using satellite data from the Oak Ridge National Laboratory (USA). Density per grid cell was categorised by structure quartiles; low (0–112 structures), medium (113–397 structures) and high (398–4664 structures). Each grid cell was divided into Thiessen polygons using ArcGIS Online (Esri V.18.0.3) with divisions that reflected the number of households targeted for enrolment; low=18, medium=24 and high=53 per grid cell. The first house encountered in a polygon was recruited. Grid-cells were surveyed sequentially (eg, low, medium, high) and randomly ordered within each category across a 12-month study period (August 2018–July 2019). Although ten grid-cells of each density were designated for survey, six of each were sampled because of logistical limitations.
Data were collected by one Haitian enumerator and one Haitian nurse. Survey instruments included two 30-minute in-person questionnaires administered with REDCap mobile V.9.1.1 and were piloted with non-participant households. A household questionnaire collected GPS coordinates, demographic and socioeconomic data and healthcare seeking behaviour for both standardised cases and health events (online supplemental file). The standardised respiratory and diarrhoeal cases consisted of hypothetical scenarios involving an ill child at 22:00 with typical symptoms of acute respiratory infection or diarrhoea, illnesses chosen for their commonality in this population. A health event was defined as any illness a household member experienced in the previous month, regardless of whether care was sought. A provider questionnaire captured GPS location, details about the facility/business, personnel qualifications and resources available (online supplemental file). Conventional providers were licensed persons who worked at a licensed facility. For large facilities with greater than 200 patients per month, the facility itself was defined as a ‘provider’. Non-conventional providers were licensed or non-licensed providers at non-licensed facilities, or mobile non-licensed providers; the definitions for conventional and non-conventional were established a priori.
Household demographics and health event characteristics were described by proportions for categorical variables and medians with IQRs for continuous variables. Socioeconomic status (SES) index was generated using a summative score of the following components: land, animal, bank account and phone ownership, HoH education level, household electricity source, fuel source, water source, primary source of income, primary transportation method, sanitation type, household floor, wall and roof type, and grid cell density. Provider characteristics were described similarly. The two primary analyses compared (i) households that did and did not seek care for health events and (ii) households that did and did not seek care from a conventional provider. For care-seeking, we used bivariate and multivariate logistic regression models with seeking care as the dependent variable and expressed the results as ORs and adjusted ORs (aORs) for care sought. Similar methods were used to compare the selection of seeking care from a conventional versus non-conventional provider. Health events were analysed as discrete variables. Two-mode whole networks42 43 of households and providers compared provider types identified/sought in standardised cases to health events; subanalyses were performed for respiratory infection (‘cough’ or ‘cold’ with fever), diarrhoeal illness (‘diarrhoea’ with/without blood) and ‘other’ cases (all other illness types not categorised as respiratory or diarrhoea). Ego networks42 44 were generated for commonly identified providers in the whole-network analysis. One-mode provider referral networks42 were generated. Statistical significance was defined at α=0.05% and 95% CIs are provided. Missingness that was not excluded was for provider types identified by households but not located and enrolled; for regression models, case-wise deletion was used. We chose not to use imputation because we did not want to risk introducing new bias into the analysis.45 46 Analyses were completed in Stata (V.11) and the igraph, and in R (R Foundation for Statistical Computing; packages included ‘sf’ by E. Pebesma and R. Bivand 2018).47 48 The deidentified data are available in online supplemental file 1 “Providers_networks_DI.csv” and “Household_DI.csv”.
Patient and public involvement
Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.
A total of 868 households were randomly screened and 568 (65%) were enrolled with a distribution of 112, 144 and 312 between low, medium and high-density grid-cells (figure 1A); 24% were excluded because the household had no children under 5 years (figure 1B, table 1). The average household had 5.2 members, and among the 2900 household members, 58% were under 21 years. The sampling distributions per grid cell density were as follows: (i) 70% of households in the population were in high-density areas and 56% of the households surveyed were in a high-density area; within a high-density grid cell the goal was to sample 5% of households, (ii) 26% of households in the population were in medium-density areas and 25% of the households surveyed were in a medium-density area; within a grid cell the goal was to sample 10% of households, (iii) 4% of households in the population were in low-density areas and 19% of the households surveyed were in a low-density area; within a grid cell the goal was to sample 20% of households. These varied distributions were intentional to enable networks to be identified within grid cells of different densities.
Health event characteristics
A total of 636 health events were identified (table 2 and online supplemental tables S1 and S2); 22%, 50% and 28% of households reported 0, 1 or at least 2 health events in the previous month, respectively. The most common symptoms were congestion, fever and cough, present in 61%, 52% and 49% of health events, respectively. The most common perceived cause of health events was humoral pathology49 (41%), considered an imbalance of ‘hot’ and ‘cold’ within the body caused by environmental exposures. Forty per cent of health events were considered severe. Half of all health events started at nighttime (55%), of these, 36% were severe (online supplemental table S1). Respiratory infection symptoms started more often at night-time (66%) compared with diarrhoea that more often started during the daytime (65%) (online supplemental table S2).
Determinants to seek care for health events
Healthcare was sought outside the household for 35% (n=220/636) of health events, and most often for children between 0 and 4 years (65%)(table 3). The most common reason cited for not seeking care was preference to treat at the household with home remedies (67%; n=277/416) followed by the health event did not necessitate treatment (24%; n=99/416). The majority of households delayed seeking care (67%; n=138/201), most commonly citing monetary and transportation (46%; n=63/138) barriers. Among those who sought care without delay, 68% (n=43/63) chose to do so because of severity. In the bivariate analysis, decreased odds of seeking care was associated with nighttime events (OR=0.69; 95% CI 0.50 to 0.96; p=0.029) and informal transport (foot, bicycle or donkey) (OR=0.53; 95% CI 0.36 to 0.79; p=0.002). Increased odds of seeking care were associated with illness severity (OR=3.20; 95% CI 2.28 to 4.50; p<0.001), high-density grid-cells (OR=1.66; 95% CI 1.18 to 2.34; p=0.004) and high SES (OR=1.45; 95% CI 1.03 to 2.03; p=0.034). In the multivariate analysis, the odds of seeking care increased by 217% with severe illness (aOR=3.17; 95% CI 1.99 to 5.05; p<0.001).
Households identified 140 providers and 65 (46%) were located and enrolled (figure 1C; online supplemental table S3). Failure to enrol providers was most commonly due to the provider being located outside the study area (21%; n=29) or an inability to locate the provider/had ceased operation (14%; n=19). Details from the household surveys enabled the determination of provider type and setting (online supplemental table S3). Among surveyed providers, 68% (n=44) had a business license. Variation was observed by grid cell density; only four providers were identified in the low-density grid-cells and none were available after 20:00 (online supplemental table S3).
Determinants to seek care from a conventional provider
Care was sought from a conventional provider for 83% (n=180/218) and a non-conventional provider for 17% (n=38/218) of health events (online supplemental table S4). For 24% (52) of health events, households prioritised quality of care/trust/reputation in provider selection, among these 92% were conventional and 8% were non-conventional. Bivariate analysis found decreased odds of seeking care from a conventional provider was associated with informal transport (OR=0.33; 95% CI 0.15 to 0.71; p=0.005) and HoH with no formal education (OR=0.32; 95% CI 0.12 to 0.82; p=0.018). The odds increased in association with SES (OR=2.78; 95% CI 1.25 to 6.25; p=0.12). In multivariate analysis, increased odds of seeking care from a conventional provider was associated with increased travel distance (aOR=1.37; 95% CI 1.06 to 1.79; p=0.016) and HoH education at or above secondary school (aOR=5.00; 95% CI 1.05 to 25.00; p=0.044). These findings are consistent with responses that 53% of families (n=116/220) cited distance as the most important factor in provider selection. Households that cited distance travelled a median of 1.5 km (IQR 0.7 km–3.9km) in comparison to 3.6 km (IQR 1.5 km–7.8km) for households that did not cite distance. Households travelled further (Mann-Whitney, p<0.0001) to seek care from conventional providers (median 2.6 km, IQR 1.1 km–5.9km) than non-conventional providers (median 0.9 km, IQR 0.0 km–2.3km). The difference in cost was less when households sought care from non-conventional providers (median US$1.28, IQR US$0.16–US$2.56) compared with when households sought care from conventional providers (median US$4.49, IQR US$1.92–US$7.85; Mann-Whitney, p<0.0001).
Congruence between care seeking intentions and behaviours by provider type
Comparisons of intentions versus behaviours for the type of provider sought was conducted using data from the household standardised cases (‘where would you go?’) and the household health events (‘where did you go?’). Among all health event types, there was a lack of congruence between intentions and behaviours (OR=0.15; 95% CI 0.04 to 0.43; p<0.001); 27 of the 31 non-congruent events were attributed to households that intended to seek care from a conventional provider but switched to non-conventional (online supplemental table S5). In bivariate analysis, respiratory infection with diarrhoea and informal transportation were associated with increased non-congruence. In contrast, HoH education level of secondary and above and high-density grid-cells were associated with congruence. Multivariate analysis found no significant factors (online supplemental table S6).
Network analysis of care-seeking intentions versus behaviors
Two-mode whole networks were used to evaluate the relationships between households and providers identified through standardised cases and health events (figure 2A–C). The number of households linked to each provider per health event was statistically different compared with standardised respiratory (Mann-Whitney, p=0.014) and diarrhoeal (Mann-Whitney, p=0.007) cases; distance travelled was statistically different compared with diarrhoeal cases (Wilcoxon, p=0.001). Subanalysis comparisons for respiratory (online supplemental figure S1A,B) and diarrhoeal (online supplemental figure S1C,D) specific health events were similar. Two-mode ego network analyses focused on a public clinic (C1) and a private hospital (H1) commonly identified in the whole-network analysis as the intended provider (figure 2D–I). However in practice, households sought care from these providers less often (Student’s t-test; both p<0.001). To analyse relationships among providers, we used one-mode network analysis to examine provider referral networks. Providers identified 33 conventional and 0 non-conventional referral providers that formed a three-tiered network (online supplemental figure S2A); 45% (n=15) received one referral at the lowest tier, 42% (n=14) received between 2 and 10 referrals at the middle tier and 12% (n=4) received 10 or more referrals at the highest tier. Sixty-eight per cent of households sought care from providers receiving zero referrals. Referral networks from standardised respiratory and diarrhoeal cases were similar (online supplemental figure S2B,C).
Should care be sought, and from whom, are fundamental questions when medical problems arise in households. The decisions may trigger entry into a linear healthcare system with primary, secondary and tertiary conventional providers or a pluralistic system that includes non-conventional providers. In this cross-sectional study, we found that approximately one-third of household health events were not treated, one-third were treated at the household and one-third sought care. When care was sought, it was for severe illnesses and from a mixture of conventional and non-conventional providers, largely disconnected from healthcare networks, despite intentions to seek care from connected conventional providers. The findings reveal insights on how to catalyse improved healthcare access in seemingly change resistant healthcare settings.
Households living in low population density regions had the lowest SES compared with those in high population density regions. Households approached health events with a pragmatic mindset aware of healthcare access barriers. Barriers to seeking care included night-time presentation, limited access to motorised transport and poverty. Disease severity was the major determinant to seek care. Families delayed seeking care because of financial and transport barriers, unless the event was severe or not resolving. These data document that poverty and geographic isolation in rural communities are barriers to seek care.
Providers were identified by households in both the standardised cases and the health events. Increased SES correlated with seeking care from a conventional provider. Multivariate analysis identified increased distance, education and higher population density as correlates of seeking care from a conventional provider. These findings highlight that once a household decides to seek care, poverty and geographic isolation again influence the decision to seek care from a conventional versus non-conventional provider.
Intentions (‘would do’) and behaviours (‘did do’) for care-seeking and provider selection were compared by analysing standardised cases against the health events. Conflict between household intentions and behaviours for seeking care from a conventional provider was found. The directionality was from an intention to seek care from a conventional provider to care sought from a non-conventional provider. These analyses highlight the conflict that both poor and geographically isolated households confront between intentions and behaviours. In addition, households that prioritised quality of care were more likely to seek a conventional provider. The data suggest that poor and isolated households are not content with the choice they are forced to make. Network analysis compared intentions versus behaviours for provider selection. A stark contrast was revealed between an intention to seek care from commonly shared providers yet a behaviour to seek care from providers that were less commonly shared. The care sought from these disconnected providers was likely lower quality because it is more difficult to standardise care, train providers, enforce regulations and disseminate healthcare resources to disconnected providers. Access to information about provider types, costs and methods to contact the provider are likely additional barriers to seeking care from intended providers.
Disparity between intentions and behaviours can be found broadly across the domain of healthcare. However, this study provides insight that when faced with poverty and isolation, households are forced to turn to nearby providers that are often non-conventional. Methods to reduce non-congruence between intentions and behaviours are needed. We advocate that governments and NGOs develop initiatives to investigate how to link disconnected non-conventional providers into the healthcare networks with conventional providers. The main barrier to integration is the resources needed to standardise quality of care. Non-conventional providers, by definition, do not have the formal education/training or facilities that are associated with conventional healthcare providers.
At the household-provider level, our recommendation is that with a careful strategy providers should be connected into the network through training and referral mechanisms, which might also create an opportunity to professionalise non-conventional providers that carry a significant portion of the healthcare case-load. Continuing marginalisation will likely continue to fail to improve healthcare access (figure 3A). In parallel, households may desire and need an innovative alternative mechanism to ‘bypass’ barriers they face when attempting to seek care from a conventional provider (figure 3B). A new model may include improved access to centralised services, or extending access to/near households by mobile healthcare services either physically or electronically through telemedicine.
At the systems level, improvement and regulation strategies for connected versus non-connected providers will likely be different. The finding that a majority of households seek care from providers who received zero referrals reveals that households commonly seek care from front-line providers. However, there is a diversity of these ‘primary providers’, including those that consult very few patients per week and are disconnected from the healthcare network. While conventional and non-conventional providers were identified in the network, conventional providers were more central in the network. This distinction between non-conventional providers tending to be first-line providers, whereas conventional providers are sought when a referral is required, is a crucial systems level insight and suggests multi-dimensional strategies are needed to improve access and quality in pluralistic healthcare systems like the one studied herein.
These findings should be viewed within the context of the study limitations. First, the standardised case questionnaires were designed a priori to investigate the decision-making process to seek care for children with respiratory or diarrhoeal disease at night. The analytic strategy to compare standardised cases as intentions and health events as behaviours was established post-hoc. The limitation is that the specific standardised cases were limited to children who developed symptoms at night which reduced the strength of comparison to the diversity of health events despite subanalysis to compare cases by chief complaint. The sample size of health events with only respiratory or diarrhoeal complaints alone also limited the strength of comparisons with standardised cases. Second, adjustments for household clustering were not performed because 72% of households had zero or one health event. Third, providers identified by households were difficult to locate and only a portion of providers were located. This resulted in missingness and an information bias such that mobile drug vendors were interviewed less frequently. To address this limitation, key provider data (eg, type and distance) used in the analysis were obtained for many unlocated providers by leveraging the household surveys. However, the remaining missingness for providers in the study area that we could not contact may have resulted in an underestimation of households referred from the periphery of the healthcare network, and for providers outside the study area, this missingness represents an element of mobility of household member(s) leaving the area to seek care which we feel is incorporated in the analysis. Finally, there was risk of recall and response biases. The questionnaires relied on 1 month and 1 week recall by households and providers, respectively, and families may have favoured providing socially acceptable responses. Despite these limitations, the approach and findings represent a meaningful contribution.
The study supports a model that households are likely to treat illnesses at home without assistance. When care is sought, it is for severe illnesses and often from a mixture of disconnected conventional and non-conventional providers despite an intention to seek care from a conventional provider. The findings support bridging geographic barriers, possibly with technology solutions, to meet healthcare seeking intentions, while developing approaches to connect non-conventional providers into healthcare networks through training and supervision when geographic barriers cannot be overcome.
Data availability statement
All data relevant to the study are included in the article, uploaded as supplementary information and uploaded to the Dryad data depository. Extra data can be accessed via the Dryad data repository at http://datadryad.org/ with the doi: 10.5061/dryad.wpzgmsbkm.
Patient consent for publication
Ethical approvals were obtained from the Comité National de Bioéthique (National Bioethics Committee of Haiti; 1718-35) and the University of Florida (IRB201703246). Adult participants provided informed written consent.
We thank the households and providers that agreed to participate in this study, and diligent field research team who made this study possible. We are grateful to R Autrey, K Berquist, D Schatz and S Rivkees at the University of Florida for their support and helpful discussions. We thank M Gurka, R Vacca and C McCarty at the University of Florida for guidance and insight on study design and social network analysis. We also express our gratitude to the Christianville Foundation (Gressier, Haiti) for their shared commitment to healthcare access and providing infrastructure for implementation.
MK and JS are joint first authors.
Contributors EN and MK conducted the literature search. The study was designed by EN and MK. MK and YC implemented the study and collected data with guidance from SM. SM provided the approach and tools for geospatial mapping. Data analysis, interpretation, figure design and writing was conducted by JS, MK and EN. Clinical oversight in Haiti was governed by CB with assistance by VMBdR. Communication with IRBs was performed by CB, VMBdR and EN. IRB documentation was written and managed by MK and EN. EN is the guarantor of the study.
Funding This work was supported by the National Institutes of Health [DP5OD019893] to EN. Internal support was provided by the University of Florida.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.