DOI: https://doi.org/10.21203/rs.3.rs-1719469/v1
Background:
Ensuring access to healthcare services is a key element to achieving the Sustainable Development Goal 3 of promoting healthy lives and well-being through Universal Health Coverage (UHC). However, in the context of low- and middle-income countries, most studies focused on financial protection measured through catastrophic health expenditures, or on health services utilization among specific populations exhibiting health needs (such as pregnancy or recent sickness).
Methods:
This study aims at building an individual score of perceived barriers to medical care (PBMC) in order to predict health services utilization (or non-utilization). We estimate the score on six items: (1) knowing where to go, (2) getting permission, (3) having money, (4) distance to the facility, (5) finding transport, and (6) not wanting to go alone, using individual data from 1787 adult participants living in rural Senegal. We build the score via a stepwise descendent explanatory factor analysis (EFA), and assess its internal consistency. Finally, we assess the predictive validity of the factor-based score by testing its association (univariate regressions) with a wide range of variables on determinants of healthcare-seeking and healthcare services utilization.
Results:
EFA yields a one-dimensional score combining items 3-6 with a 0.7 Cronbach’s alpha indicating good internal consistency. The score is strongly associated – p-values significant at the 5% level – with determinants of healthcare-seeking (including, but not limited to, sex, education, marital status, poverty, and distance to the health facility). Additionally, the score can predict non-utilization of health services at the household level, utilization and non-utilization of health services following an individual’s episode of illness, and utilization of health services during pregnancy and birth. These results are robust to the use of a different dataset.
Conclusions:
As a valid, sensitive, and easily documented individual-level indicator, the PBMC score can be a complement to regional or national level health services coverage to measure health services access and predict utilization. At the individual or household level, the PBMC score can also be combined with conventional metrics of financial risk protection such as CHE to comprehensively document deficits in, and progress towards UHC.
Achieving universal access to healthcare services is a key element to the sustainable development goal of “ensur[ing] healthy lives and promot[ing] well-being for all at all ages” [1]. Specifically, target 3.8 sets to “achieve universal health coverage, including financial risk protection, access to quality essential health-care services and access to safe, effective, quality and affordable essential medicines and vaccines for all” (ibid). The standard metrics for measuring progress towards the financial risk protection aspect of UHC are catastrophic health expenditures (CHE) [2], identifying whether out-of-pocket (OOP) health expenditures represent a “catastrophic” share of the overall household expenditures, usually set at 40% [2–4], or impoverishing health expenditures, which document whether the household’s falling below the poverty line is attributable to health expenditures [5]. These metrics can be easily computed from widely available household surveys.
A recent comprehensive assessment of UHC progress combined CHE prevalence with a measure of service coverage capturing both prevention and treatment indicators at the country level [6]. Service coverage is meant to document the aspects of access which are part of UHC and might be at odds with CHE, especially in the context of low- and middle-income countries (LMICs) where lower OOP might reflect the lower quality of health services [7, 8], unmet health needs [9], or even a younger, healthier population [10]. Indeed, Wagstaff and colleagues found an association between low incidence of CHE and low service coverage in LMICs.
At the population level, access to quality health services is usually measured through health services utilization [11], often within specific populations exhibiting health needs, e.g., children’s immunization records, women with a recent pregnancy, or individuals having experienced a recent or chronic illness. It involves heavy data collection processes and long interviews focusing on specific events in a given timeframe (e.g., two years for recent pregnancy and birth, 12 months for inpatient visits, etc.).
In LMICs, the literature has specifically investigated women’s self-reported barriers to seeking medical care [12, 13], which are collected as part of the Demographic and Health Surveys (DHS) [14]. These questions record perceptions on both the financial (possession of, or perceived ability to obtain monetary resources) and the geographic accessibility (distance and transportation means) as well as barriers pertaining to cultural and social norms (i.e., concerns about obtaining permission and going alone) – thereby covering a wide range of elements which have been identified as determinants to healthcare seeking and health services utilization [15–19].
Existing studies have documented an association between reporting at least one significant barrier and lower maternal and prenatal health services utilization [20–22]. A 2012 study combined socioeconomic, geographical, and psychosocial barriers from the 2003 DHS in Burkina Faso to create a tri-dimensional score of women's perceived ability to overcome barriers to healthcare seeking [23] and validated the score in relation to a select number of socio-demograohic variables (specifically age, education level, poverty status and rural versus urban living) without investigating associations with the utilization of maternal or child services. In addition, all these studies solely focused on women.
This study aimed at building a score of perceived barriers to medical care and assessing its validity to predict health services utilization – or non-utilization— in both men and women, in the context of LMICs.
We employed individual data from the CMUtuelleS survey, a cross-sectional survey conducted in 2019–2020 among 1787 residents of the Niakhar Health and Demographic Surveillance System (HDSS) in rural Senegal [24]. The CMUtuelleS survey aimed at characterizing the implementation of community-based health insurance (CBHI) schemes among voluntary subscribers who paid the full fee, and beneficiaries of the national cash transfer program for poor households (BSF, “Bourse de Sécurité Familale”), whose subscription to the CBHI is supposed to be subsidized [25]. Accordingly, both the subscriber/head of household and their partner were interviewed among three groups: voluntary subscribers (n = 285), BSF recipients (n = 176), and non-enrolled in a CHBI scheme (n = 1326).
In an adaptation from the 2008 Demographic and Health Survey (DHS)’s woman’s questionnaire, both male and female participants were asked “When you are sick, or you want to get medical advice or treatment is any of the following i) not a problem, ii) a small problem, or iii) a big problem:
(1) knowing where to go?
(2) getting permission to go?
(3) getting the money to pay?
(4) the distance to the health facility?
(5) having to take transport?
(6) not wanting to go alone?”.
The CMUtuelleS dataset contained rich self-reported micro-level data on the individuals and their households. In addition to standard socio-demographic variables, including, age, education level, sex and marital status, data reported GPS coordinates, which were used to compute distances between the household and the nearest health facility and CBHI office, respectively. The survey also extensively quantified the household’s expenditures (including monthly consumption expenditures per adult equivalent, and out-of-pocket health expenditures) and included several measures of poverty (specifically, monetary, food, and subjective poverty). Catastrophic health expenditures were computed following Xu et al. [2].
Additionally, the survey documented individual-level health services utilization following health needs (consultation, self-medication, exams, or hospitalization among participants with an episode of illness in the past two months; prenatal consultations and health facility delivery among women who had a live birth in the past two years), as well as unmet health needs at the household level (having forgone healthcare expenses in the past 12 months).
The survey also recorded the participants’ health insurance status and self-reported health (12-Item Short Form Survey questionnaire [26], chronic illness, and handicap). Finally, participants reported perceived quality of care at the local healthcare facility, knowledge of community-based health insurance, willingness to pay for health insurance, risk aversion [27], and generalized trust [28]. All these variables were defined in Appendix A2 in the Supplementary Material.
After checking for sample adequacy using the Kaiser-Meyer-Olkin measure and the Bartlett test of sphericity [29], the score was built using stepwise descendant explanatory factor analysis: starting with the full set of items, each was removed one at a time to test whether any of the reduced form factor analyses provided a better fit to the data. The number of dimensions to retain was selected following scree plot analysis with a conservative Kaiser criterion of eigenvalues > 1.1 [30]. Factors were rotated to provide a clearer pattern of which items loaded on each factor, and only items that contributed to the factors’ dimension (i.e., with factor loadings sufficiently high) were retained to create the final score. The internal consistency of the final set of items was assessed using Cronbach’s alpha [31], and a factor-based score was computed as the average of items.
We assessed the predicted validity of the factor-based score by testing its association with a series of variables, which were grouped into three main categories: (i) determinants of healthcare-seeking, (ii) healthcare services utilization, and (iii) other potentially associated variables. More specifically, we ran univariate regressions of the factor-based score on each variable (logistic, multinomial logistic, linear, and Poisson regressions for binary, polytomous, continuous, and count variables, respectively). In all regressions, standard errors were clustered at the household level to account for intra-household correlation. Regressions were weighted using sampling weights to account for choice-based stratified samples. After each univariate regression, we calculated predictions for the dependent variable at three representative values of the factor-based score: 0 (“not a problem”), 1 (“a small problem”), and 2 (“a big problem”). Predictions are in the form of predicted probabilities for logistic and multinomial logistic regressions, linear predictions for linear regressions, and predicted number of events for Poisson regressions. All estimations were performed using Stata [32].
The score was computed with confirmatory factor analysis [33] in the dataset of the ANRS12356 AmBASS survey, which was conducted in the Niakhar HDSS in 2018–2019 and featured a sample representative of the general population living in the area [34, 35].
Figure 1 presents descriptive results on perceived barriers to medical care. For almost all participants, knowing where to go and getting permission was “not a problem” (98.3% and 98.6% respectively). Having to go alone was “not a problem” either for most participants (88.1%), “a small problem” for about 10% (162 participants), and “a big problem” for only a small share (2.8%). In contrast, over half of the participants (55.1%) reported that having the money to pay was “a big problem”, with an extra 531 participants (29.7%) declaring it as “a small problem”. Distance to the health facility and finding transport was “not a problem” for a majority of participants (57.2% and 61.1% respectively), “a small problem” for about a third (32.5% and 28.5%), and “a big problem” for 264 (14.8%) and 187 (10.5%) participants, respectively.
Our sample passed the Bartlett test of sphericity, rejecting the null hypothesis that variables were not inter-correlated (γ²=2080.857(15), p < 0.001), and gave a value for the Kaiser-Meyer-Olkin measure sufficiently large (0.645) to justify running a factor analysis. Stepwise descendant factor analysis suggested that removing the item “knowing where to go” did not significantly reduce the quality of the factor analysis. Subsequent analyses of the score were therefore performed on items (2)-(6). Following EFA and scree plot analysis, only one dimension was retained (2.13 eigenvalue, explaining 42.6% of variations; detailed results were provided in Appendix A3 in the Supplementary Material). Rotations with weights revealed that only items (3)-(6) significantly contributed to dimension one (loadings > 0.4). The 0.7 Cronbach’s alpha of this reduced set indicated very good internal consistency. We, therefore, built a factor-based score with the average of items (3)-(6). This score of perceived barriers to medical care (hereafter, PBMC score) was comprised between zero and two, with a mean (standard deviation) value of 0.67 (0.47) – and a 0.5 (0.25-1) median (interquartile range) value.
Summary statistics for all variables used were provided in Appendix A4. All univariate regression results are presented in Table 1. Coefficient estimates (CE) are provided for linear regressions, odds ratios (OR) for logistic regressions, incidence-rate ratios (IRR) for Poisson regressions, and relative-risk ratios (RRR) for multinomial logistic regressions. We also provide graphical representations of the univariate regression results for each of the groups of variables; they are displayed in Appendix A5 in the Supplementary Material.
Facing higher barriers to medical care was associated with being a woman, being less formally educated, being unmarried, being poor (whether in terms of monetary, food or subjective poverty, or lower monthly consumption expenditures), being in a smaller household, living further away from the nearest healthcare structure or CBHI office. When it comes to distance, a one-point increase in the PBMC score was associated with living 1.32 km further away from the nearest healthcare structure. More specifically, perceiving barriers to healthcare seeking as “not a problem” was associated with living 2.24 km away from the nearest health structure, while perceiving barriers as “a big problem” was associated with living 4.89 km away from the nearest health structure. Age was the only determinant not significantly associated with the PBMC score.
|
Variable group |
Dependent variable |
Model |
Type of estimate |
Estimate |
Predictions |
N |
||
|---|---|---|---|---|---|---|---|---|
|
At Score = 0 |
At Score = 1 |
At Score = 2 |
||||||
|
Determinants of healthcare-seeking |
Had primary education or higher |
Logistic |
OR |
0.62*** (0.10) |
0.21 (0.02) |
0.14 (0.01) |
0.09 (0.02) |
1787 |
|
Was a woman |
Logistic |
OR |
1.53*** (0.12) |
0.47 (0.01) |
0.57 (0.01) |
0.67 (0.02) |
1787 |
|
|
Was in a union |
Logistic |
OR |
0.68** (0.13) |
0.92 (0.01) |
0.89 (0.01) |
0.85 (0.03) |
1787 |
|
|
Age |
Linear |
CE |
-0.81 (0.92) |
53.41 (0.74) |
52.60 (0.53) |
51.79 (1.31) |
1787 |
|
|
Was poor (monetary poverty, HH level) |
Logistic |
OR |
1.37** (0.19) |
0.46 (0.03) |
0.53 (0.02) |
0.61 (0.05) |
1787 |
|
|
Was poor (food poverty, HH level) |
Logistic |
OR |
1.45*** (0.20) |
0.32 (0.03) |
0.41 (0.02) |
0.50 (0.05) |
1787 |
|
|
Was poor (subjective poverty, HH level) |
Logistic |
OR |
1.77*** (0.26) |
0.21 (0.02) |
0.33 (0.02) |
0.46 (0.05) |
1787 |
|
|
Monthly consumption expenditures per adult equivalent (in CFA) |
Linear |
CE |
-1611.87** (672.56) |
18046.82 (558.64) |
16434.95 (466.11) |
14823.08 (1013.47) |
1787 |
|
|
Number of adult equivalents in the household (HH level) |
Linear |
CE |
-1.12*** (0.39) |
12.30 (0.35) |
11.18 (0.24) |
10.05 (0.54) |
1787 |
|
|
Distance to the nearest healthcare structure (in km) |
Linear |
CE |
1.32*** (0.13) |
2.24 (0.13) |
3.57 (0.08) |
4.89 (0.17) |
1787 |
|
|
Distance to the nearest CBHI (in km) |
Linear |
CE |
0.52*** (0.19) |
5.10 (0.17) |
5.62 (0.11) |
6.14 (0.26) |
1787 |
|
|
Healthcare services utilization |
Forgone medical consultation (HH level) |
Logistic |
OR |
3.10*** (0.45) |
0.20 (0.02) |
0.43 (0.02) |
0.70 (0.04) |
1787 |
|
Forgone medical treatment (HH level) |
Logistic |
OR |
1.30* (0.18) |
0.21 (0.02) |
0.26 (0.02) |
0.31 (0.04) |
1787 |
|
|
Consulted in a health structure following an episode of illness |
Logistic |
OR |
0.40*** (0.09) |
0.70 (0.04) |
0.48 (0.03) |
0.27 (0.06) |
418 |
|
|
Self-medicated following an episode of illness |
Logistic |
OR |
2.09*** (0.47) |
0.20 (0.04) |
0.34 (0.03) |
0.52 (0.07) |
418 |
|
|
Gave birth in a health facility |
Logistic |
OR |
0.46** (0.16) |
0.68 (0.07) |
0.49 (0.05) |
0.31 (0.10) |
197 |
|
|
Number of prenatal consultations |
Poisson |
IRR |
0.87** (0.06) |
3.69 (0.20) |
3.20 (0.12) |
2.78 (0.25) |
197 |
|
|
Other potentially-associated variables |
Had an at least fair knowledge of CBHI |
Logistic |
OR |
0.64*** (0.09) |
0.32 (0.02) |
0.23 (0.01) |
0.16 (0.03) |
1787 |
|
Health insurance status |
Multino-mial logistic |
RRR (Volun-tary) |
0.52*** (0.09) |
0.06 (0.01) |
0.03 (0.00) |
0.02 (0.00) |
1787 |
|
|
RRR (Subsi-dized) |
1.82*** (0.34) |
0.05 (0.01) |
0.10 (0.01) |
0.17 (0.03) |
||||
|
Willingness to pay for CBHI (in CFA francs) |
Linear |
CE |
-1019.09*** (216.71) |
4567.96 (230.46) |
3548.87 (91.52) |
2529.78 (239.92) |
1787 |
|
|
Had a chronic illness |
Logistic |
OR |
1.69** (0.37) |
0.06 (0.01) |
0.10 (0.01) |
0.16 (0.04) |
1787 |
|
|
Had a handicap |
Logistic |
OR |
1.75** (0.46) |
0.03 (0.01) |
0.06 (0.01) |
0.10 (0.03) |
1787 |
|
|
Had a poorer health |
Logistic |
OR |
1.86*** (0.25) |
0.48 (0.03) |
0.63 (0.02) |
0.76 (0.04) |
1787 |
|
|
SF-12 Mental Component Summary |
Linear |
CE |
0.35 (0.49) |
47.65 (0.39) |
48.00 (0.31) |
48.35 (0.72) |
1787 |
|
|
SF-12 Physical Component Summary |
Linear |
CE |
-1.63** (0.65) |
50.66 (0.51) |
49.02 (0.37) |
47.39 (0.93) |
1787 |
|
|
Perception of healthcare quality |
Linear |
CE |
0.16*** (0.03) |
0.40 (0.03) |
0.57 (0.02) |
0.73 (0.05) |
1787 |
|
|
Risk tolerance |
Linear |
CE |
-0.61*** (0.15) |
5.61 (0.13) |
4.99 (0.09) |
4.38 (0.20) |
1785 |
|
|
Generalized trust |
Linear |
CE |
-0.41*** (0.13) |
5.48 (0.11) |
5.07 (0.08) |
4.67 (0.19) |
1786 |
|
|
Catastrophic health expenditures |
Had catastrophic health expenditures, 40% threshold (HH level) |
Logistic |
OR |
1.45 (0.43) |
0.05 (0.01) |
0.07 (0.01) |
0.10 (0.03) |
1787 |
|
Had catastrophic health expenditures, 30% threshold (HH level) |
Logistic |
OR |
1.05 (0.24) |
0.10 (0.02) |
0.11 (0.01) |
0.11 (0.03) |
1787 |
|
|
Had catastrophic health expenditures, 20% threshold (HH level) |
Logistic |
OR |
1.17 (0.21) |
0.16 (0.02) |
0.18 (0.02) |
0.20 (0.04) |
1787 |
|
|
Notes: ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. All variables measured at the individual level, unless when HH-level specified. Robust standard errors (clustered at the household level to account for intra-household correlation) in parenthesis. Regressions were weighted using sampling weights to account for choice-based stratified samples. For linear models, predictions are linear predictions of the dependent variable. For logistic and multinomial logistic models, predictions are predicted probabilities of the dependent variable. For Poisson models, predictions are the predicted number of events. Abbreviations: N = number of observations, HH = household, OR = odds ratio, CE = coefficient estimate, CBHI = community-based health insurance, IRR = incidence-rate ratio, RRR = relative-risk ratio. |
||||||||
Along with the univariate regression results provided in Table 1, Fig. 2 displayed graphical representations of the predictions of health services utilization and non-utilization across the distribution of the PBMC score. The PBMC score was positively associated with the households’ probability of forgoing medical consultation and treatment. For instance, perceiving barriers to healthcare seeking as “not a problem” was associated with a 20% predicted probability of foregoing medical consultation, while perceiving all barriers as “a big problem” was associated with a 50 percentage-point higher probability (i.e., 70%). This was also true – though to a lesser extent—for the probability of foregoing medical treatment, with a 10 percentage-point increase in probability from 21% (“not a problem” for all items) to 31% (“a big problem”).
Among people with a recent episode of illness, perceiving no barriers in seeking medical care predicted a 70% probability of having consulted, versus a 27% probability when perceiving barriers as “a big problem”; conversely, the probability of self-medicating increased from 20–52%.
Among women with a recent pregnancy, the probability of giving birth in a health facility decreased by 37 percentage points (i.e., from 68–31%) when all barriers to medical care were perceived as “not a problem” versus “a big problem”. Similarly, the predicted number of prenatal consultations was 3.69 in women with no perceived barriers, versus 2.79 for those who perceived all barriers as “a big problem”.
Facing higher barriers to medical care was associated with lower odds of knowing about the CBHI scheme, lower odds of having voluntarily enrolled in a CHBI scheme, and higher odds of benefiting from a subsidized CBHI enrollment through the BSF program. The PBMC score was also negatively associated with the willingness to pay for CBHI schemes. Facing higher barriers to medical care was associated with having a chronic illness, a handicap or disability, and poorer self-assessed health. Interestingly, the PBMC score was tied to physical health (negative association with the SF-12 Physical Component Summary score), but independent of mental health (no association with the SF-12 Mental Component Summary score). Finally, reporting higher barriers to medical care was associated with a lower perception of the quality of local healthcare services, lower risk tolerance, and lower generalized trust.
In contrast, catastrophic health expenditures were not significantly associated with the PBMC score. Note that this result was robust to the use of alternative thresholds of catastrophic health expenditures (namely, out-of-pocket health expenditures ≥ 40%, 30%, and 20% of non-food expenditures, respectively – as displayed in Appendix A5.4 in the Supplementary Material).
In the AmBASS dataset, items (3)-(6) yielded a 0.71 Cronbach’s alpha indicating good internal consistency. The CFA model estimated on this reduced set fitted the data well – as indicated by goodness-of-fit measures. The details were reported in Appendix A6 in the Supplementary Material.
As in the 2012 study on women from Burkina Faso [23], we found that obstacles were higher in under-educated, poorer individuals and those living in rural areas (i.e., in our sample, participants living further away from semi-urban – health – facilities). In contast, in both samples used in the present study, the EFA yielded a one-dimensional factor score, whereas Nikiema and colleagues built a second-stage score combining all six items over three dimensions (specifically, psychosocial, socioeconomic, and geographic barriers). However, the Burkina Faso data was from 2005, only among women, a sizeable share of whom was living in urban areas. This suggests that the structure of the score might need to be validated when computed in very different settings or samples.
In line with the literature, we found that perceived barriers were strongly associated with the utilization of prenatal and maternal health services [21, 36]. In our study, the PBMC score’s prediction of health services utilization was robust to the type of health utilization, health need, and population: specifically, the score can be employed to predict the probability of foregoing medical consultation or expenses at the household level, of medical consultation and non-utilization (self-medication) in individuals with a recent episode of illness, and of maternal health services utilization in women who had a live birth the past two years (documented through delivery in a health facility and the number of prenatal consultations).
Unlike measures of access focusing on individuals that experienced an event prompting health services utilization (e.g., individuals with a recent episode of illness or women with a recent pregnancy or birth), the PBMC score can be documented in the general population through simple, and relatively light data collection and data analysis processes.
The factor-based score also has the advantage of being expressed in the same scale as the original items, with values that can be easily interpreted: a 0 score corresponds to having declared “not a problem” to all items, a 2 score indicates that all items were reported as “a big problem”, and values in between reflect increasing levels in barriers. In contrast to studies documenting ‘any’ perceived barrier [12, 21] or focusing on a specific barrier such as distance [20], the PBMC score, therefore, provides a much more precise and sensitive measure of both the intensity and the width of barriers to medical care.
As illustrated by the absence of association with CHE, the PBMC score captures something other than the financial risk protection and is valuable in informing deficits in, and progress towards UHC attainment. There is a wide range of possible uses for the score. For instance, the identification of individual and structural characteristics associated with the intensity of the score can help characterize populations and areas that should be targeted by specific interventions or policies aiming at improving UHC. The score can also be used to evaluate such interventions through the comparison of changes in individual score levels over time (before/after intervention or longitudinal studies) – to name just a few potential applications.
Our study has limitations. The main concern is that it relies on self-reported measures, which can be subject to heterogeneity in reporting associated with psycho-social and socio-economic variables – such biases have been extensively documented in the literature on self-assessed health [37–42]. In addition, our results reveal an association between the PBMC score and psychosocial variables (specifically risk aversion, generalized trust, and perceived quality of the healthcare system), which ought to be accounted and controlled for in potentially future multivariate regressions. However, we provide ample evidence that our score is significantly associated with objective measures and determinants of healthcare-seeking (distance to the health facility, sex, formal education, several measures of wealth and poverty, etc.).
A second limitation is that, though multidimensional, the PBMC score only provides a partial view of access. In particular, it does not include supply-side information on the availability or quality of healthcare services, professionals, equipment, or medications in the area of interest – i.e., the health system’s side of Levesque’s comprehensive framework of patient-centered access to healthcare [43]. Items used to build the PBMC score encompasses the “ability to seek”, “ability to reach” and “ability to pay” of populations defined in this framework, but its scope falls short of abilities to perceive and engage that are instrumental in the populations’ access to healthcare.
A final, and related, limitation is that, by using DHS-based items in a top-down process, the PBMC score may overlook context-specific barriers that are relevant to accessing healthcare goods and services in rural Senegal. Bottom-up approaches to tailoring items to the specific context would gain in internal validity though potentially at the expense of external validity. Indeed, the PBMC score has the ambition of being used in other settings, e.g. through DHS surveys, though data availability is limiting – especially in men.
We used DHS-based items on perceived barriers to medical care to build a one-dimensional score in both men and women living in rural Senegal. This PMBC score is internally consistent and confirmed in a different dataset representative of adult individuals living in the same area. The score is significantly associated with a wide range of determinants of healthcare-seeking (including, but not limited to, sex, education, marital status, poverty, and distance to the health facility). Additionally, the score can predict non-utilization of health services at the household level, utilization and non-utilization of health services following an individual’s episode of illness, and utilization of health services during pregnancy and birth.
As a valid, sensitive, and easily documented individual-level indicator, the PBMC score can be a complement to regional or national level health services coverage to measure health services access and utilization. At the individual or household level, the PBMC score can also be combined with conventional metrics of financial risk protection such as CHE to comprehensively document deficits in, and progress towards UHC.
BSF: Senegalese government family allowance (“Bourse de Sécurité Familale”)
CBHI: community-based health insurance
CFA: confirmatory factor analysis
CHE: catastrophic health expenditures
DHS: demographic and health surveys
EFA: explanatory factor analysis
HDSS: health and demographic surveillance system
LMICs: low and middle-income countries
PBMC: perceived barriers to medical care
OOP: out-of-pocket (health expenditures)
UHC: universal health coverage
WHO: World Health Organization
Ethics approval and consent to participate: The CMUtuelleS survey was approved by the Senegalese National Ethical Committee for Health Research (n°000037/MSAS/DPRS/ CNERS and n°0000118/MSAS/DPRS/CNERS). Informed consent was obtained from all participants. All methods were performed in accordance with the relevant guidelines and regulations, including the Declaration of Helsinki.
Consent for publication: Not applicable.
Availability of data and materials: Data are available from the authors upon reasonable request (contact: Marion COSTE, AMU – AMSE, 5-9 Boulevard Maurice Bourdet, CS 50498, 13205 Marseille Cedex 1, +33652465772, [email protected]). The code is also available from the authors upon reasonable request.
Competing interests: The authors declare that they have no competing interests.
Funding: This research is part of the UNISSAHEL program (Universal Health Coverage in Sahel), funded by the Agence Française de Développement (AFD). This work also received support from the French government under the France 2030 investment plan, as part of the Initiative d'Excellence d'Aix-Marseille Université - A*MIDEX - Institute of Public Health Sciences of Aix-Marseille (AMX-20-IET-014 and ANR-17-EURE-0020) managed by the French National Research Agency.
Authors' contributions: MC and MQB jointly reviewed the literature and performed the statistical analyses. MC wrote the first draft of the manuscript and MQB substantially revised it for important intellectual content. Both authors read and approved the final manuscript.
Acknowledgments: We are grateful to all study participants, the staff at the Niakhar Health and Demographic Surveillance System, and the members of the UNISSAHEL program. A complete list of members of the UNISSAHEL Study Group is given in Appendix A1 in the Supplementary Material. We also thank Bruno Ventelou, Sylvie Boyer, and Mohammad Abu-Zaineh, who co-led the UNISSAHEL economic research program. Our thanks also go to Andrainolo Ravalihasy, Richard Lalou, Jean-Yves Le Hesran, Bruno Boidin, and participants to the AMSE PhD seminar and the UCL Global Health Brown Bag seminar for their help, comments, and suggestions.