Women’s preferences for maternal healthcare services in bangladesh: Evidence from a discrete choice experiment

Despite substantial improvements in several maternal health indicators, childbearing and birthing remain a dangerous experience for many women in Bangladesh. This study assessed the relative importance of maternal healthcare service characteristics to Bangladeshi women when choosing a health facility to deliver their babies. The study used a mixed-methods approach. Qualitative methods (expert interviews, focus group discussions) were initially employed to identify and develop the characteristics which most influence a women’s decision making when selecting a maternal health service facility. A discrete choice experiment (DCE) was then constructed to elicit women’s preferences. Women were shown choice scenarios representing hypothetical health facilities with nine attributes outlined. The women were then asked to rank the attributes they considered most important in the delivery of their future babies. A Hierarchical Bayes method was used to measure mean utility parameters. A total of 601 women completed the DCE survey. The model demonstrated significant predictive strength for actual facility choice for maternal health services. The most important attributes were the following: consistent access to a female doctor, the availability of branded drugs, respectful provider attitudes, a continuum of maternal healthcare including the availability of a C-section delivery and lesser waiting times. Attended maternal healthcare utilisation rates are low despite the access to primary healthcare facilities. Further implementation of quality improvements in maternal healthcare facilities should be prioritised.

The objective of this study was to investigate the relative significance of the characteristics of maternal health services to Bangladeshi women by using the DCE method when choosing a health facility to deliver their baby. A hypothetical healthcare facility was thus developed to explore how different characteristics and levels of healthcare influence the demand for maternal healthcare services among Bangladeshi women. The study was conducted in four selected catchment areas of the NGO Health Service Delivery Project (NHSDP) in Smiling Sun (or Surjer Hashi) health facilities in Bangladesh. The Smiling Sun franchise program or network is funded by the United States Agency for International Development and the United Kingdom’s Department for International Development and is intended to complement government health facilities. The study research team worked with NHSDP staff to identify eligible households in the catchment area in four different settings: Harirampur (in the administrative district of Manikganj), Gazipur district, Keraniganj (in the Dhaka district) and Tejgoan areas (in Dhaka city)—all in the Dhaka Division of Bangladesh. The study used a mixed-methods approach, i.e., both qualitative and quantitative. Qualitative methods (i.e., expert interviews, focus group discussions) were initially employed to identify and develop the characteristics, which influence women’s decision making most when selecting a maternal health service facility (Figure 1). These characteristics formed the basis for the quantitative methods (DCE and household survey) that were used to examine women’s preference for maternal healthcare services. Figure 1 below describes the methods that have been applied throughout the implementation of the study. The study flowchart. The literature search strategy was designed to identify the most desired attributes of maternal health facilities including quality of services, patient-provider relationships, accountability, affordability and referral services (Figure 2). The attributes were categorized into five groups, which were further examined in expert interviews as well as in focus group discussions (FGDs). The attributes identified during the literature review. An expert interview guideline was developed based on the findings from the literature review. A total of ten expert interviews were conducted among maternal and child health experts. The experts were selected from different organizations or institutes including: international research organizations (e.g., Save the Children, icddr,b), NGOs (e.g., BRAC, NGO Health Service Delivery Project), government institutes (e.g., Directorate General of Health Services and Directorate General of Family Planning, Ministry of Health and Family Welfare, Bangladesh), and medical research professional. Experts raised specific issues that they felt were likely to influence women’s preferences in choosing a health facility. Healthcare accountability, financing, and quality of services were common arguments with experts from government organisations or institutes, academia, and program implementers. However, healthcare financing was raised as a significant concern as financial hardship is an essential barrier to accessing healthcare. There have been recent initiatives such as vouchers, pay-for-performance schemes, and pre-payment mechanisms to reduce financial barriers to accessing healthcare. Accountability, in particular, community involvement, in the running health facilities was also perceived to be a significant issue. Other issues that were raised by experts included a designated referral system and the availability of healthcare and the associated distribution of sufficient human resources to meet demand. The guidelines for the Focus Group Discussions (FGDs) were developed based on the literature review and expert interviews in order to obtain information from potential survey respondents to validate the identified characteristics for the survey. The FGD guideline covered several topics: health-seeking behaviours, perceptions of the quality of healthcare and facilities, the main challenges of households’ appearance when in need of healthcare, and perceptions of health facilities and the payments for that. Four FGDs were conducted with 8 to 10 women in each to collate information for survey instrument development. Health service promoters organised these groups from four NHSDP clinic catchment areas. The participants of FGDs were excluded for the DCE and household surveys. Based on the participant’s responses, a structure content analysis was performed and re-structured by clustering within a similar group of maternal healthcare parameter. The main results arising were: (1) Facility choice: women select health facilities based on their health needs or demands, not based on distance or perceived health facility characteristics; (2) Paying for services: Respondents reported that they were satisfied in receiving healthcare services from the private facilities on payments or NGO clinics whereas they had to pay out-of-pocket and tended to prefer them over to public health facilities. The destitute mothers sought healthcare at public facilities generally because it is free but the quality of services was perceived to be higher at non-public facilities, which are chosen if payment could be made. Although the price of services is an issue, respondents traded it off in return for higher quality of services; (3) Quality of care: Participants also considered a broad range of health facility characteristics when they were asked how they thought of quality healthcare, such as immediate service, availability of healthcare providers (e.g., doctors/nurses), branded drugs and the availability of diagnostic services, and consistent attention and monitoring by staff during the time they are in the facility, courteous attitudes of facility staff, flexible opening hours, and the availability of referral or emergency services. Upon completion of expert interviews and FGDs, the research team had identified 15 attributes that appeared to be most influential for the utilisation of maternal health services. To review and prioritise the attributes, a final workshop was held in May 2014 with NHSDP staff including clinic managers and service promoters. The study method and preliminary findings were presented, and respondents were asked to review, rank and prioritise the attributes. In the DCE methodology, the characteristics of the health facility were referred to as attributes and the specific domains of those attributes or characteristics including attribute levels. The final set of nine attributes and their ranking are shown in Table 1. Given the high level of illiteracy among respondents, a pictorial guide was developed to represent the attribute and attribute levels. The attributes of maternal healthcare services. We conducted two cross-sectional surveys; a household survey to capture household-level characteristics and healthcare-seeking behaviour for maternal healthcare in 601 households and then the DCE survey among with its development. The DCE approach is a system of conjoint analyses or choice-based conjoint (CBC) analyses [21,23,24,25]. The CBC methods are effective in identifying preferences for services or non-market goods where the design contains information about the combinations of attribute levels to test for ensuring its efficiency [21,22,26,27,28]. The DCE method has been applied successfully to quantifying patient’s or client’s preferences in different health settings such as obstetric care [29], delivery care [30,31], cancer treatments [32,33], asthma medications [34], diabetes treatment and prevention [35], mental health [36], weight-loss programs [37]. The Sawtooth software was used to design the experiment so that the number of attribute levels was selected for the hypothetical choice set [38]. A range of randomly generated hypothetical choice scenarios was developed whereas each level of the attribute has an equal chance of selecting in the choice set developed. A 12 different alternative versions of the DCE survey choice sets were designed, each having nine questions. An example of a single choice scenario presenting three hypothetical health facilities along with an option NONE, i.e., “I wouldn’t choose any of these”, is shown in Figure 3. Each option refers to a hypothetical health facility with seven attributes that were designated pictorially and with text in the local Bangla language and English version. Each hypothetical scenario shows six attributes, namely, a health facility with a service provider, attitude of providers, cost of service, continuum of care, availability of branded drugs, and availability of diagnostic services. The attributes for facility environment, availability for complaints, and waiting times were ‘cycled’ through so that only one was presented in each scenario, meaning that respondents could consider seven attributes at a time. The respondent was asked to observe the scenarios and select the most preferable one that denotes the facility they would choose. Before conducting the DCE survey, women reviewed the pictorial guideline (Table A1) and chose scenarios that explained three different hypothetical healthcare facilities using nine attributes. Eight of these attributes had answers that were selected in the dataset for analysis. One of the questions, nevertheless, was a ‘fixed’ select attribute signifying that the attributes in each of the three hypothetical scenarios shown in the scenarios were indistinguishable across the surveys that included a combination of attribute and attribute levels. The question was selected, one of the best options was a health facility fulfil all the attributes that could be considered most anticipated a priori. In this empirical experiment, the first facility scenario option considered a female medical doctor with a polite attitude, had free services, had branded drugs available, and a standard continuum of care from antenatal care to C-section delivery facility or referral with ambulance services. The fixed choice question was set so that each rational participant should choose this option, where this question was not an item that was unique or could be analysed. This system of the experiment process can be restructured if a high (<10%) percentage of participants perform irrationally or uncontrollably due to a lack of rationality. The hypothetical scenarios of the health facility. A total of 566 individuals was required in the study. Assuming a 25% non-response rate, 588 participants were assessed, with 720 households visited, and 601 respondents agreed to participate in this study. Examining the equation for sample size provides an explanation, Nk=(Tk2× SEk2)betak where Nk is the sample size, Tk2 is the t-statistic required for significance, SEk is the standard error for the prior parameter and betak is the prior parameter [38]. Therefore, as beta approaches zero, the sample size needed to detect statistical significance increases. The sample participants were selected using the probability proportion sampling technique [39]. The probability of selection for a sampling unit was directly proportional to the size measure. The study participants were selected randomly in each catchment area from the eligible couples list of each facility. The design of the experiment was verified by using the Sawtooth Software to confirm an adequate sample size considering the number of attributes and attribute levels that were to be explored [40]. Data collection was completed during June–July 2014. The surveys were conducted with women aged 18 years or older who had delivered a baby in the past two years and had one or more child under five years. Written informed consent was taken from study participants and an explanation of the study aims and objectives was provided prior to the interview. Training was given to the data collectors on the objective of the study, confidentiality of information, respondent rights and techniques for conducting the interview. During data collection, checks were done by the study research supervisors to ensure the quality of the collected data. This study protocol, discussion guides, and survey questionnaires were approved for human subject research from the Institutional Review Board (IRB) of BRAC James P Grant School of Public Health, BRAC University, Dhaka, Bangladesh and Abt Associates International (ethical reference no-34). A descriptive univariate analysis for socio-demographic variables and a demand for maternal healthcare related variables were considered. Averages and percentages were showed to describe the study population including age, gender, socioeconomic status (SES), education status and health-seeking behaviours. Descriptive statistics were analysed using Stata 13 (StataCorp., College Station, TX, USA). The Sawtooth statistical package (Sawtooth Software Inc., Sequim, WA, USA) was used to measure individual utilities at the attribute level. In the Choice-Based Conjoint (CBC) with Hierarchical Bayes (HB) estimation, it is hypothesized that individuals’ utility scores for all attribute are explained by using the multivariate statistical technique [40]. Choice-Based Conjoint with Hierarchical Bayes assumes that the participants’ responses choice sets based on a Multinomial Logit Model (MNL) [41]. MNL considers the probability of the specific alternative being chosen related to the proportion of the total utility for that concept relative to the total utility for all the concepts. This distribution is described using a mean vector and variance and co-variance matrix for an individual’s characteristics. At a minimum level, it is supposed that an individual’s probability of selecting particular alternatives are constituted by a multinomial logit regression model [41]. The Markov Chain Monte Carlo method was used to iteratively estimate the means and variance. Averages across all the participants are offered in the results section as ‘average utilities’ for each attribute.

N/A

• Dima Health