Social and Behavioral Determinants of Healthcare-Seeking and Satisfaction Among Rural Adults in Bangladesh: A Propensity Score Analysis

Abstract

In rural Bangladesh, financial protection is limited, and care seeking occurs across government and non-government sectors. This study investigated how social and behavioral factors—family monthly income, education, and adherence to prescribed treatment—relate to the usual source of care and patient satisfaction. A cross-sectional survey was conducted among 374 adults in Sirajganj district. Propensity scores were estimated and analyzed using 1:1 nearest-neighbor matching (caliper 0.20) and inverse probability of treatment weighting (IPTW). We used both propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) to reduce confounding and to check the robustness of the estimated associations. Primary outcomes were the usual treatment source and satisfaction with that source; exposures included income, education, and adherence. Adjusted odds ratios (AORs) and 95% confidence intervals (CI) were derived from matched and weighted logistic models. Prespecified interactions and sensitivity analyses assessed robustness. Higher family income strongly predicted use of non-government providers (matching AOR 8.50; weighted AOR 10.11) but showed no significant association with satisfaction. Adherence to prescribed treatment was associated with lower use of non-government care (matching AOR 0.35) and greater satisfaction (matching AOR 3.50), with comparable IPTW estimates. Education above the primary level was not related to provider type but was linked to lower satisfaction (matching AOR 0.30). The positive association between adherence and satisfaction persisted across income and education strata. Sensitivity analyses yielded similar findings. In rural Bangladesh, higher income primarily drives private-sector care use; treatment adherence predicts higher satisfaction and reduced private use; and higher education is associated with lower reported satisfaction. Integrating financial protection with adherence support and experience-of-care improvements may advance equity and patient-reported outcomes toward universal health coverage.

Introduction

Bangladesh’s mixed health system provides care through both government and non-government sectors. ¹ Evidence from Bangladesh and other low-and middle-income countries (LMICs) shows that patient satisfaction is generally higher in private hospitals, which are perceived as more responsive, comfortable, and patient-centered, whereas public hospitals—although critical for poorer populations—often face overcrowding, staff shortages, and resource constraints that contribute to lower satisfaction.^{2 -4} Studies from Bangladesh indicate that type of facility (public vs private) is a key determinant of satisfaction after admission, alongside socioeconomic characteristics of patients.^{1,5 -7}

Socioeconomic position, particularly income and education, appears to shape both healthcare utilization and patients’ experiences with care.^1,8 Lower-income groups in LMICs are more likely to rely on public facilities and to delay or forgo care when faced with out-of-pocket (OOP) payments, and income gradients in healthcare use have been documented across diverse settings.^{3 -5,9,10} Financial barriers extend to medication use; patients with limited resources are more likely to report cost-related non-adherence, such as skipping doses, rationing medicines, or discontinuing treatment, which can undermine control of non-communicable diseases and worsen health outcomes.^{11 -13} Education may influence satisfaction and adherence through its association with expectations of care, health literacy, and ability to understand and follow treatment instructions. Evidence from Bangladesh and other LMICs suggests that higher education can both raise expectations of service quality and improve patients’ capacity to adhere to prescribed regimens, while low education is linked to poorer understanding and adherence.^1,5,14

This study is guided by Andersen’s Behavioral Model of Health Service Use, in which predisposing factors (such as education), enabling factors (such as income), and need-related factors shape health behaviors and subsequent outcomes, including patient satisfaction.^15,16 Within this conceptual framework, income and education are treated as key predisposing/enabling resources, and medication adherence is considered a critical behavioral pathway through which socioeconomic position may influence satisfaction with hospital care. ¹⁷ Although previous studies in Bangladesh ¹⁸ and other LMICs have examined determinants of satisfaction or socioeconomic gradients in healthcare utilization and adherence,^10,12,14 there is limited evidence that jointly analyses income, education, and medication adherence in relation to patient satisfaction, or compares these relationships between public and private hospitals. Addressing this gap, the present study applies Andersen’s model to examine how income and education are associated with medication adherence and patient satisfaction among patients admitted to public and private tertiary care hospitals in Bangladesh.

Locally grounded evidence from rural communities is particularly valuable for district and upazila managers aiming to enhance financial protection, strengthen treatment adherence, and improve service quality in areas of greatest need. However, such evidence from rural Bangladesh remains limited. To date, no study has jointly examined income, educational attainment, and adherence to prescribed treatment, nor investigated how these factors simultaneously influence both the usual source of care and treatment satisfaction within a single rural population. Understanding the determinants of treatment-seeking behavior and satisfaction is critical for evaluating health-system performance.

Against this backdrop, the present study explores how family monthly income, educational attainment, and adherence to prescribed treatment are associated with the usual source of care, government versus non-government and with satisfaction with that source among rural adults in Bangladesh. To address potential confounding, we employ propensity score (PS) approaches to reduce bias from observed covariates and to assess the robustness of findings across alternative model specifications. The previous studies estimated propensity scores using multivariable logistic regression based on preoperative demographic and clinical covariates, then performed 1:1 nearest-neighbor matching with a caliper to create comparable treatment and control groups, and evaluated covariate balance using standardized mean differences.^{19 -21} Unlike previous analyses that rely on conventional regression with limited adjustment for selection bias, our application of PS methods more effectively accounts for the strong correlations of income, education, and treatment adherence with demographic and access-related factors. Residual confounding in standard models can distort associations and undermine causal inference, whereas PS techniques, such as matching and inverse probability of treatment weighting (IPTW), offer complementary strategies to achieve covariate balance and thereby strengthen both the validity and the policy relevance of the estimates.^19,22 In the previous studies, they cannot integrate social and behavioral factors. This study aims to examine how family income, education, and treatment adherence influence provider choice and patient satisfaction in rural Bangladesh.

Methodology

Study Design

This cross-sectional study was conducted between July 2023 and June 2024 in Shahjadpur Upazila (sub-district) of Sirajganj District in Bangladesh. Data were obtained from residents of 3 randomly selected unions (Each union consisted of more than 3 villages) within the upazila. We used a 2-stage cluster sampling procedure. 374 residents aged 18 to 60 years were included in this study using a simple lottery method. Those who had experienced an illness within the preceding 3 months and had attended a healthcare center were included only.

Sample Size Determination

The sample size was calculated based on the following parameters:

Prevalence: p = 0.42 ²³

Confidence level: at 95% confidence level usual Z-score is 1.96.

Margin of error: 5% (E = 0.05).

Using the formula for sample size determination in proportion studies:

$$sn=\frac{(Z^2\times p\times \left(1-p\right)}{E^2}$$

Substituting the values:

$$n=\frac{(1.96^2\times 0.42\times \left(1-0.42\right)}{0.05^2}\approx 374.325$$

Thus, the required sample size was approximately 374. Both male and female respondents (≥18 years old) from the study area were considered. Figure 1 illustrates the comprehensive sampling design used in this study.

Figure 1.

Sampling design of this study.

Ethical Consideration

Prior to collecting the data, ethical approval and permission for the study were first obtained from the Ethical Review Committee of the Faculty of Health and Life Sciences of Daffodil International University, Bangladesh. The ethical approval Ref. No: FHLS-REC/DIU/2023/0027; Date: 28/07/2023. Written informed consent was obtained from all participants prior to enrollment. Confidentiality of all personal information was strictly maintained, and data were anonymized before analysis.

Study Variables

Outcome

The primary outcome variable was satisfaction with treatment received during the most recent visit to a healthcare center, representing participants’ self-reported perception of the adequacy and quality of care, and classified as satisfactory or not satisfactory. The secondary outcome variable was usual treatment source, defined as the type of healthcare facility most frequently accessed for medical needs and categorized as government or non-government.

Exposure

The first exposure variable was family monthly income, self-reported as the total household income in the preceding month and dichotomized at 26 000 BDT: ≤26 000 BDT (below average) versus >26 000 BDT (above average), consistent with national socioeconomic benchmarks and the latest rural mean of 26 163 BDT reported in the HIES 2022. ²⁴ Secondly, educational attainment was categorized as primary (completion of primary school or less) or above primary (secondary or higher), reflecting literacy thresholds relevant to healthcare comprehension in rural Bangladesh. Finally, adherence to prescribed treatment was defined as self-reported compliance covering medications, lifestyle modifications, and other provider-recommended measures, following the most recent healthcare visit, and was classified as Yes (full adherence) or No (partial or complete non-adherence).

Covariates

Covariate identification combined substantive domain knowledge with empirical model selection criteria. A comprehensive list of potential confounders was first compiled from established literature and expert judgment, then refined using the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC), both derived from the maximized log-likelihood but differing in their penalties for model complexity. ²⁵ Covariates jointly minimizing both criteria were retained (Figure 2).

Figure 2.

Process of covariate identification and selection.

The final covariates were grouped into sociodemographic and treatment-related characteristics. Sociodemographic variables included age (<25, 25-34, 35-44, 45-54, and >54 years), gender (female, male), marital status (married, unmarried, divorced/widow/widower), occupation (unemployed, business, job/service, farmer), and type of current residence—kacha (temporary), semi-pucca (tin-shed), or pucca (brick/concrete), use as a proxy for socioeconomic status and environmental health risks. Treatment-related variables comprised the reason for choosing the most recent treatment center, mode of transport, healthcare worker behavior, usual sponsor for treatment, type of recent complications, and main source of health information.

The reason for choosing the treatment center was classified as easy access, affordability, or effective treatment, reflecting key drivers of healthcare utilization. Mode of transport was recorded as vehicle or walking, indicating potential effects on accessibility and timeliness of care. Healthcare worker behavior, based on patient perceptions, was categorized as accessible, friendly, or respectful, capturing the interpersonal dimension of care quality. The usual sponsors for treatment are self, father, husband, or son, which highlights patterns of financial dependency and decision-making. Recent complications were grouped as communicable or non-communicable diseases, enabling disease-pattern assessment. Finally, the main source of health information was recorded as healthcare providers, mass and digital media, pharmacy/self-care, or offline/online social networks, reflecting primary channels of health information access.

Statistical Analysis

The statistical analysis employed propensity score-based methods to minimize confounding and strengthen inference from this observational study. Propensity scores were estimated separately for each exposure, namely family monthly income, education and adherence to prescribed treatment, using logistic regression models that incorporated the same set of identified sociodemographic and treatment related covariates, thereby representing the conditional probability of exposure given baseline characteristics.

These associations were estimated using logistic regression models in both the matched and inverse probability of treatment weighted samples, adjusting for the same covariate set used in the propensity score models.PS matching was implemented using a 1:1 nearest-neighbor algorithm with a caliper of 0.20 standard deviations of the logit of the PS, excluding unmatched individuals to limit residual confounding and improve comparability. This caliper choice is widely recommended because it achieves good bias reduction while retaining an adequate number of matched pairs, as shown by Austin. ¹⁹ Before matching, several covariates showed substantial imbalance between exposure groups. After PSM, all standardized mean differences were below 0.10, indicating good balance. Similar improvements in covariate balance were observed in the IPTW weighted sample. To address missing covariate data, we used inverse probability weighting for missingness under a missing at random assumption. Complementarily, inverse probability of treatment weighting (IPTW) with stabilized weights was applied to create a pseudo-population with balanced covariates across exposure groups, allowing the full sample to contribute to estimation while reducing bias.

Covariate balance between exposure groups was evaluated using standardized mean differences, adopting 0.20 as the threshold for acceptable balance. After propensity score matching and inverse probability of treatment weighting, all covariates showed substantial improvement in balance, with SMDs reduced to at or below these prespecified thresholds, as illustrated by balance plots comparing pre and post adjustment. Additional diagnostics included inspection of propensity score overlap and the distribution of stabilized weights to check for extreme values. Sensitivity analyses included interaction terms between exposures (family income × education, family income × adherence, education × adherence) to explore effect modification across key socioeconomic and behavioral subgroups.

All propensity scores were estimated using the same set of covariates, which was also used in the corresponding outcome models. All estimates are reported as adjusted odds ratios (AORs) with 95% confidence intervals, with statistical significance inferred when the interval did not include 1. Analyses were conducted using R (version 4.5.1), with the dual use of matching and weighting providing methodological triangulation to enhance robustness. A schematic of the analytic workflow, including participant recruitment, exposure and outcome classification, covariate selection, PS estimation, matching, IPTW, covariate balance assessment, sensitivity analyses, and AORs estimation, is presented in Figure 3.

Figure 3.

Schematic representation of the analytic process.

Result

Frequency Distribution

Among 374 participants, 222 (59.4%) reported satisfaction with treatment, while 152 (40.6%) were dissatisfied. Regarding the usual source of care, 207 (55.3%) primarily used government facilities, and 167 (44.7%) relied on non-government sources (Figure 4).

Figure 4.

Characteristics of outcome variables.

We further analyzed the different characteristics of 374 participants focusing on their sociodemographic and treatment-related characteristics.

In terms of Socio-demographic characteristics (Table 1), the average age of the participants was 37.68 years (SD ± 11.80). The largest age cohort was 35 to 44 years (32%), followed by the 25 to 34 age group (22%), while those above 54 years constituted the smallest group (12%).

Table 1.

Description of Sociodemographic Variables.

Variables	Frequency (percentages)
Age
Mean ± SD	37.68 ± 11.80
Below 25	64 (17)
25-34	82 (22)
35-44	118 (32)
45-54	66 (18)
Above 54	44 (12)
Gender
Female	151 (40)
Male	223 (60)
Education
Primary	193 (52)
Above primary	181 (48)
Marital status
Married	297 (79)
Unmarried	60 (16)
Divorced or widow or widower	17 (4.5)
Family monthly income
Mean ± SD	22 104.27 ± 6151.49
Above average	169 (45)
Below average	205 (55)
Occupation
Job/service	35 (9.4)
Business	36 (9.6)
Farmer	102 (27)
Unemployed	201 (54)
House type of current residence
Pucca (building)	10 (2.7)
Semi Pucca (Tin-Shed)	42 (11)
Kacha (temporary)	322 (86)

The sample was predominantly male (60%) and married (79%), with smaller segments being unmarried (16%) or divorced/widowed (4.5%). Educational attainment was nearly evenly split, with 52% having primary education and 48% possessing higher qualifications.

Economically, the mean family monthly income was 22 104.27 (SD ± 6151.49). A slightly larger proportion of households (55%) fell below the average income level compared to those above it (45%). In terms of occupation, more than half of the respondents were unemployed; among the employed, agriculture was the dominant sector (27%), while business and service jobs accounted for approximately 9.6% and 9.4% respectively. Housing conditions indicated that the vast majority resided in Kacha (temporary) structures (86%), with only 2.7% living in Pucca buildings.

Regarding to treatment-related characteristics (Table 2), the primary reasons for choosing a facility were effective treatment (39%), affordability (36%), and ease of access (25%). Most participants (80%) traveled on foot. Interactions with healthcare workers were mostly described as friendly (64%), with 26% accessible and 10% respectful. Self-payment (48%) was the most common source of funding, followed by support from a husband (27%), father (19%), or son (5.6%). Two-thirds reported non-communicable conditions, and the main sources of health information were healthcare providers (47%), mass/digital media (23%), offline/online social networks (23%), and pharmacies/self-care (7.2%).

Table 2.

Description of Treatment-Related Characteristics.

Variables	Frequency (percentages)
Reason for choosing the recently visited treatment center
Easy to access	93 (25)
Affordable	134 (36)
Effective treatment	147 (39)
Mode of transport to the treatment center
Vehicle	80 (21)
Walk	294 (79)
Behavior of healthcare workers of the recently visited treatment center
Accessible	98 (26)
Friendly	238 (64)
Respectful	38 (10)
Usual sponsor for treatment
Self	181 (48)
Father	72 (19)
Husband	100 (27)
Son	21 (5.6)
Type of recent complications
Communicable	129 (34)
Non-communicable	245 (66)
Main source for health information
Health care providers	175 (47)
Mass and digital media	85 (23)
Offline/online social networks	87 (23)
Pharmacy/self-care	27 (7.2)
Adherence to prescribed treatment
Yes	247 (66)
No	127 (34)

Treatment-related outcomes revealed that 66% adhered to the prescribed treatment, and 59% were satisfied with the care. Overall, the sample represents a predominantly married, working-age population with modest education and income, high reliance on walking to government facilities, generally positive provider interactions, but with considerable room for improving satisfaction and adherence to prescribed treatment.

Propensity Score Analyses

Figure 5 illustrates the standardized absolute mean differences (SMDs) of covariates before and after adjustment across exposure variables. Each panel compares covariate balance between before matching (red points) and after matching (blue points). The vertical line at 0.20 represents the conventional threshold for acceptable balance. Prior to matching, several covariates exceeded this threshold, indicating notable imbalance. Post-adjustment, nearly all covariates fell below 0.20, demonstrating effective minimization of confounding by measured variables.

Figure 5.

Standardized mean differences (SMD) between above average and below average in unmatched and propensity score matched samples (first), SMD between primary and above primary in unmatched and propensity score matched samples (second), and SMD between yes and no in unmatched and propensity score matched samples (third).

Matching reduced the available sample but retained sufficient treated individuals for analysis: family income, 113 of 169 (66.9%), education, 130 of 193 (67.3%), and adherence, 108 of 247 (43.7%). These results confirm that PS matching achieved adequate covariate balance while preserving a meaningful proportion of treated individuals, enhancing the internal validity of estimated exposure–outcome associations.

Table 3 presents the associations of family monthly income, education, and adherence to prescribed treatment with usual treatment source and treatment satisfaction among rural adults in Bangladesh, estimated using PS matching and weighting.

Table 3.

Associations Between Social and Behavioral Factors and Healthcare Outcomes After Propensity Score Matching and Weighting.

Propensity score method	Goal	Usual treatment source		Satisfaction with treatment
		AOR	95% CI	AOR	95% CI
Matching	Family monthly income
	Below average	Ref		Ref
	Above average	8.495	[3.532, 22.526]	0.752	[0.341, 1.641]
	Education
	Primary	Ref		Ref
	Above primary	1.243	[0.472, 3.257]	0.295	[0.118, 0.688]
	Adherence to prescribed treatment
	No	Ref		Ref
	Yes	0.346	[0.132, 0.855]	3.499	[1.498,8.583]
Weighting	Family monthly income
	Below average	Ref		Ref
	Above average	10.105	[4.075,25.059]	0.444	[0.150,1.319]
	Education
	Primary	Ref		Ref
	Above primary	0.876	[0.311, 2.465]	0.196	[0.050, 0.760]
	Adherence to prescribed treatment
	No	Ref		Ref
	Yes	0.275	[0.095, 0.791]	4.434	[2.123, 9.264]

AOR = adjusted odds ratio; CI = confidence interval; Ref = reference.

Using matching, participants with above-average income had higher odds of seeking care from non-government/private facilities (AOR 8.50, 95% CI 3.53-22.53) but no significant association with satisfaction (AOR 0.75, 95% CI 0.34-1.64). Education above primary was not associated with treatment source (AOR 1.24, 95% CI 0.47-3.26) but significantly predicted lower satisfaction (AOR 0.30, 95% CI 0.12-0.69). Adherence to prescribed treatment was linked to lower odds of using non-government/private facilities (AOR 0.35, 95% CI 0.13-0.86) and higher satisfaction (AOR 3.50, 95% CI 1.50-8.58).

Using weighting, above-average income remained significantly associated with non-government/private care (AOR 10.11, 95% CI 4.08-25.06) but not with satisfaction (AOR 0.44, 95% CI 0.15-1.32). Higher education continued to predict lower satisfaction (AOR 0.20, 95% CI 0.05-0.76) without a significant effect on treatment source (AOR 0.88, 95% CI 0.31-2.47). Adherence was associated with lower odds of non-government/private care (AOR 0.28, 95% CI 0.10-0.79) and higher satisfaction (AOR 4.43, 95% CI 2.12-9.26).

These findings indicate that income primarily influences choice of provider, education affects satisfaction, and adherence improves satisfaction while reducing reliance on non-government/private facilities, with consistent results across both analytic approaches.

Table 4 presents the associations of exposure and outcome variables including interaction effects, estimated using PS matching and weighting.

Table 4.

Interaction Effects Between Income, Education and Adherence on Healthcare Utilization and Patient Satisfaction.

Method	Goal	Treatment source		Satisfaction with treatment
		AOR	95% CI	AOR	95% CI
Matching	Family monthly income × education
	Below average (vs above average) within primary	0.097	[0.016, 0.456]	1.246	[0.308, 4.908]
	Below average (vs above average) within above primary	0.151	[0.043, 0.541]	2.466	[0.615, 11.167]
	Primary (vs above primary) within below average	0.992	[0.269, 3.777]	0.098	[0.011, 0.590]
	Primary (vs above primary) within above average	1.599	[0.207, 3.078]	0.437	[0.120, 4.090]
	Family monthly income × adherence to prescribed treatment
	Below average (vs above average) within yes adherence to prescribed treatment	0.1403	[0.032, 0.530]	1.156	[0.397, 3.294]
	Below average (vs above average) within no adherence to prescribed treatment	0.169	[0.018, 1.075]	4.884	[0.626, 5.851]
	Adherence to prescribed treatment (yes vs no) within below average	0.638	[0.223, 1.805]	0.300	[0.046, 1.609]
	Adherence to prescribed treatment (yes vs no) within above average	0.021	[0.014, 0.669]	0.099	[0.017, 0.435]
	Education × prescribed treatment maintain
	Primary (vs above primary) within no adherence to prescribed treatment	0.283	[0.035, 1.77]	0.214	[0.006, 3.496]
	Primary (vs above primary) within yes, adherence to prescribed treatment	1.892	[0.908, 2.085]	0.271	[0.069, 0.908]
	Adherence to prescribed treatment (yes vs no) within primary education	3.402	[0.474, 5.833]	0.374	[0.089, 0.701]
	Adherence to prescribed treatment (yes vs no) within above primary education	2.680	[0.358, 6.184]	0.052	[0.001, 0.512]
Weighting	Family monthly income × education
	Below average (vs above average) within primary	0.096	[0.051, 0.496]	1.261	[0.375, 4.237]
	Below average (vs above average) within above primary	0.167	[0.018, 0.520]	1.649	[0.499, 5.451]
	Primary (vs above primary) within below average	1.007	[0.260, 3.895]	0.232	[0.057, 0.935]
	Primary (vs above primary) within above average	1.488	[0.244, 9.072]	0.440	[0.120, 1.613]
	Family monthly income × adherence to prescribed treatment
	Below average (vs above average) within yes, adherence to prescribed treatment	0.247	[0.091, 0.674]	1.525	[0.663, 3.511]
	Below average (vs above average) within no adherence to prescribed treatment	0.063	[0.009, 0.423]	3.951	[0.798, 5.569]
	Adherence to prescribed treatment (yes vs no) within below average	0.062	[0.501, 1.894]	0.313	[0.097, 1.006]
	Adherence to prescribed treatment (yes vs no) within above average	0.063	[0.014, 0.440]	0.177	[0.062, 0.501]
	Education × prescribed treatment maintain
	Primary (vs above primary) within no adherence to prescribed treatment	0.431	[0.076, 2.437]	0.178	[0.019, 1.618]
	Primary (vs above primary) within yes, adherence to prescribed treatment	1.026	[0.455, 1.318]	0.320	[0.125, 0.816]
	Adherence to prescribed treatment (yes vs no) within primary education	3.024	[0.825, 11.074]	0.257	[0.083, 0.796]
	Adherence to prescribed treatment (yes vs no) within above primary education	2.764	[0.439, 7.403]	0.199	[0.068, 0.576]

Using matching, individuals with above-average income and primary education had higher odds of using non-government/private facilities (AOR 0.097, 95% CI 0.016-0.456), though satisfaction was not significantly affected (AOR 1.25, 95% CI 0.31-4.91). Within below-average income groups, education above primary increased satisfaction (AOR 0.098, 95% CI 0.011-0.590). Interactions between income and adherence showed that above-average, adherent individuals had lower odds of private facility use (AOR 0.021, 95% CI 0.014-0.669) and higher satisfaction (AOR 0.099, 95% CI 0.017-0.435). For education × adherence, above-primary education with adherence was associated with lower satisfaction (AOR 0.052, 95% CI 0.001-0.512).

The weighting approach yielded a similar pattern. Above-average, primary-educated participants were more likely to use private treatment (AOR 0.096, 95% CI 0.051-0.496), with no significant effect on satisfaction. Education above primary increased satisfaction among below-average income individuals (AOR 0.232, 95% CI 0.057-0.935). Above-average, adherent individuals had reduced private facility use (AOR 0.063, 95% CI 0.014-0.440) and lower satisfaction (AOR 0.177, 95% CI 0.062-0.501). Above-primary education with adherence predicted lower satisfaction (AOR 0.199, 95% CI 0.068-0.576).

Overall, the results indicate that socioeconomic effects on treatment outcomes are modified by education and adherence, with adherence amplifying the positive impact of higher income on satisfaction. In contrast, higher education sometimes corresponds to lower satisfaction, highlighting the complex interplay of social and behavioral factors in shaping treatment experiences.

Sensitivity Analyses

Table 5 presents a sensitivity analysis examining the robustness of associations between exposure and outcome variables, including their interaction terms.

Table 5.

Sensitivity Analyses Assessing Robustness of Propensity Score-Based Findings. And Add One Extra Column in Each Analytic.

Goal	Treatment Source		Satisfaction with treatment
	AOR	95% CI	AOR	95% CI
Family monthly income
Below average	Ref
Above average	6.805	[2.816, 17.749]	0.812	[0.395, 1.666]
Education
Primary	Ref
Above primary	0.828	[0.335, 1.990]	0.294	[0.123, 0.669]
Adherence to prescribed treatment
No	Ref
Yes	0.314	[0.135, 0.698]	4.089	[1.976, 8.796]
Family monthly income × education
Below average (vs above average) within primary	0.097	[0.016, 0.456]	1.261	[0.395, 4.002]
Below average (vs above average) within above primary	0.160	[0.039, 0.554]	1.649	[0.551, 5.134]
Primary (vs above primary) within below average	1.007	[0.2647, 3.715]	0.232	[0.052, 0.917]
Primary (vs above primary) within above average	1.488	[0.286, 7.793]	0.440	[0.124, 1.471]
Family monthly income × adherence to prescribed treatment
Below average (vs above average) within yes, adherence to prescribed treatment	0.247	[0.082, 0.700]	1.525	[0.671, 3.483]
Below average (vs above average) within no adherence to prescribed treatment	0.063	[0.008, 0.285]	3.951	[0.796, 5.113]
Adherence to prescribed treatment (yes vs no) within below average	0.638	[0.223, 1.805]	0.251	[0.195, 1.365]
Adherence to prescribed treatment (yes vs no) within above average	0.043	[0.004, 0.240]	0.188	[0.524, 0.561]
Education × prescribed treatment maintain
Primary (vs above primary) within no adherence to prescribed treatment	0.675	[0.174, 2.426]	0.178	[0.019, 1.068]
Primary (vs above primary) within yes adherence to prescribed treatment	1.586	[1.175, 1.580]	0.320	[0.118, 0.8211]
Adherence to prescribed treatment (yes vs no) within primary education	3.222	[0.563, 5.691]	0.257	[0.085, 0.723]
Adherence to prescribed treatment (yes vs no) within above primary education	2.680	[ 0.358, 6.184]	0.053	[0.002, 0.529]

For main effects, above-average income was associated with higher odds of using non-government/private facilities (AOR 6.81, 95% CI 2.82-17.75) but not with satisfaction (AOR 0.81, 95% CI 0.40-1.67). Education beyond primary was not significantly linked to treatment source (AOR 0.83, 95% CI 0.34-1.99) but predicted lower satisfaction (AOR 0.29, 95% CI 0.12-0.67). Adherence was associated with lower odds of private facility use (AOR 0.31, 95% CI 0.14-0.70) and higher satisfaction (AOR 4.09, 95% CI 1.98-8.80).

For interactions, above-average income combined with primary education increased the likelihood of using private facilities (AOR 0.097, 95% CI 0.016-0.456), while within below-average income groups, education above primary increased satisfaction (AOR 0.232, 95% CI 0.052-0.917). Income × adherence showed that above-average, adherent individuals had markedly reduced odds of private facility use (AOR 0.043, 95% CI 0.004-0.240). Similarly, adherence among individuals with above-primary education was associated with lower satisfaction (AOR 0.053, 95% CI 0.002-0.529), whereas within primary education groups, adherence increased satisfaction (AOR 0.257, 95% CI 0.085-0.723).

Overall, the sensitivity analysis confirms that the main findings are robust, particularly emphasizing the protective effect of adherence on satisfaction and the modifying influence of education across model specifications. Adjusted odds ratios and 95% confidence intervals from these models are summarized in a forest plot on the log scale to facilitate visual comparison of the direction and magnitude of effects across specifications (Figure 6). Estimates were similar in size and direction, supporting the robustness of the main findings.

Figure 6.

Forest plot of adjusted odds ratios and 95% confidence intervals from main effects and interaction models for treatment source and patient satisfaction.

Discussion

This study examined the impact of sociodemographic position and behavioral factors on health care choice and satisfaction with treatment in a rural area of Bangladesh. Three findings stand out. First, Households above the average income threshold in our sample were substantially more likely to use non-government providers, whereas income showed no clear association with treatment satisfaction. This pattern is consistent with comparative evidence from low and middle-income countries showing that private providers are often preferred for greater timeliness and hospitality, yet the overall literature does not support claims that private care is more efficient, more accountable, or clinically more effective than public care.^4,26 Moreover, studies within this evidence base indicate that formal private services tend to serve more affluent populations, and that the apparent predominance of private sector use at the population level largely reflects care seeking from drug shops and other informal outlets. In Bangladesh, continued constraints in public financing underscore because higher income households may select private or modern providers while poorer households rely on government or lower cost options. ²⁷ Finally, recent comparative reviews and multi-country analyses detect no consistent differences between public and private providers in patient satisfaction or clinical competence, aligning with our null association for satisfaction.^4,26,28

Adherence to prescribed treatment in our matched analysis was consistently associated with higher patient satisfaction, and this relationship did not vary by provider sector. A meta-analysis which reported higher adherence to treatment regimens and higher patient satisfaction among different intervention groups ²⁹ and another meta-analysis highlighted improvements in patient-centered outcomes is associated with medication adherence. ³⁰ More broadly, determinants syntheses underscore that medical care processes and patient communication are among the most critical drivers of satisfaction, reinforcing why adherence supportive interactions are likely to improve patients’ experience even when the source of care differs.^30,31

Third, education beyond primary level was not clearly related to provider type, but was associated with lower satisfaction. Education beyond primary level was associated with lower satisfaction, a pattern echoed in prior studies that report higher expectations and more critical appraisal among better-educated patients. ¹ These 3 patterns were robust across matching and weighting estimators and in sensitivity and interaction analyses, giving confidence that they are not artifacts of a single modeling choice. In rural systems that face constraints in staffing, communication, and amenities, better-informed patients may perceive gaps more acutely, even when clinical outcomes are acceptable.

Interaction results suggest that the beneficial association of adherence with satisfaction is present across income and education strata, though the magnitude varies. In some models, adherence among higher-educated patients coincided with lower satisfaction. A reasonable explanation is that adherent, better-educated individuals monitor care closely, are more attuned to process shortfalls, and may judge quality on dimensions beyond symptom relief, such as communication, privacy, or wait times. These nuances suggest the need for differentiated quality improvement, with an emphasis on patient communication, service experience, and clinical effectiveness.

Our findings align with a well-documented pattern in South Asia and other low and middle income regions, where higher income predicts private sector use and where patient satisfaction is often higher in private facilities, largely due to amenities and responsiveness rather than superior clinical quality. The salience of affordability and out-of-pocket burden for provider choice is consistent with evidence that cost barriers steer poorer households toward public facilities and toward delaying or foregoing care.^2,6,7 The positive association between adherence and satisfaction coheres with work linking cost-related nonadherence to worse experiences and outcomes, particularly among low-income patients. ³² Finally, the inverse association between education and satisfaction mirrors studies where higher education correlates with higher expectations and more critical assessments of service quality. ⁸

Our study contributes in 3 ways. First, we designed for covariate balance, not only statistical adjustment. We implemented PS matching with a caliper and inverse probability weighting, checked standardized mean differences, assessed overlap, and trimmed observations with poor common support. This reduces confounding by observed factors and improves interpretability compared with conventional regression studies that omit balance diagnostics. Second, we prespecified and tested heterogeneity by interacting income, education, and adherence, rather than reporting only average associations. This reveals subgroup patterns that are more actionable for policy and practice. Third, we conducted robustness checks across alternative PS specifications, different matching ratios, and re-estimation with weighting, and the direction and magnitude of effects were consistent. We were attentive to common threats to validity, including unmeasured confounding, measurement error in self-reported adherence and satisfaction, and selection into provider type. We addressed these by including rich covariates, verifying post-adjustment balance, examining overlap, and running sensitivity analyses that quantify how strong an unmeasured confounder would need to be to overturn the results. Together, this design increases credibility and transportability relative to typical cross-sectional analyses that report only regression coefficients without balance checks or explicit heterogeneity assessment.

Taken together, these strengths indicate that the study provides credible, policy-relevant associations for rural Bangladesh, while the limitations suggest that the results should inform the prioritization and design of interventions rather than be interpreted as definitive causal effects. Where possible, future work should incorporate validated adherence scales, more comprehensive facility-level quality measures, longitudinal follow-up to establish temporal order, and design-based estimators with full survey weighting and clustering to enhance generalizability and causal interpretation. The design is observational and cross-sectional, so temporal order cannot be established, and estimates are associational. Residual confounding may persist despite the use of PS matching and weighting. Adherence and satisfaction are self-reported, which introduces recall and expectation bias. Provider categories pool heterogeneous private and informal facilities, which can mask variation in quality. Some subgroups may have limited common support, which narrows generalizability to populations with adequate overlap.

Conclusion

Using propensity score matching and inverse probability weighting, this study provides novel evidence that social and behavioral factors collectively shape healthcare utilization and patient experience in rural Bangladesh. Higher household income is associated with increased use of non-government providers, whereas adherence to prescribed treatment consistently predicts greater patient satisfaction and reduced reliance on private care. Education beyond the primary level tends to correspond with more critical evaluations of healthcare services. These findings were robust across PS matching and inverse probability weighting, with satisfactory post-adjustment covariate balance, reinforcing confidence in the validity of the results. However, the cross-sectional design and reliance on self-reported measures may introduce recall and social desirability bias, so the findings should be interpreted with appropriate caution.

In essence, income primarily influences provider choice, while adherence and education significantly affect satisfaction. The results underscore actionable policy levers: enhancing financial protection to mitigate income-driven shifts toward private care, strengthening adherence support through counseling, reliable medication provision, and follow-up reminders, and improving communication and service quality in public facilities to meet the expectations of better-educated patients. These measures are operationally feasible within existing systems and offer potential improvements in patient experience without requiring substantial capital investment. Therefore, integrating adherence support with financial protection may ultimately enhance equity and patient satisfaction within Bangladesh’s rural health system.