Women’s satisfaction with the quality of antenatal care services rendered at public health facilities in Northwest Ethiopia: The application of partial proportional odds model

Objectives The study was aimed: (1) to describe the quality of antenatal care (ANC) at public health facilities in Northwest Ethiopia, including dimensions of the structure, process and outcome; and (2) to assess the relationship between ANC satisfaction and structure and process dimension of ANC quality. Design Cross sectional. Setting Healthcare facilities providing ANC services in Northwest Ethiopia. Participants 795 pregnant women attending the antenatal clinics at 15 public health facilities and 41 health workers working for the surveyed facilities. Outcome measures The outcome variable, women’s satisfaction with ANC, was constructed from multiple satisfaction items using principal component analysis on an ordered, categorical and three-point Likert scale. The key hypothesised factors considered were structural and process aspects of care. Data were analysed using the partial proportional odds model with 95% CI. Results The result revealed that only 30.3% of the pregnant women were highly satisfied, whereas 31.7% had a lower satisfaction level. The findings showed that process quality indicators better predicted client satisfaction. In relation to this, better scores in history taking (aOR 1 =aOR 2; 1.81 (95% CI 1.25 to 2.60)), counselling (aOR 1 = aOR 2; 1.89 (95% CI 1.33 to 2.69)) and screening (aOR 1 = aOR 2; 18.10 (95% CI 11.52 to 28.39)) were associated with achieving higher satisfaction. We also observed a significant but lower satisfaction among women in the late trimester of pregnancy (aOR 1 = aOR 2; 0.87 (95% CI 0.78 to 0.97)). However, we did not see any significant relationship between structural variables and client satisfaction. Conclusions The study demonstrated that women’s satisfaction with ANC was low. The contents of ANC services covered during client-provider interaction were the main factors affecting client satisfaction. This suggests that efforts are required to improve the competencies of health professionals to make them more effective while dealing with clients.

This was a cross-sectional study using a blend of methods (a facility survey, provider interview, direct observation and client exit interview). The study was conducted in five districts of West Gojjam Zone, Northwest Ethiopia from July to August 2018. At the time of the survey, the Zone had a population of 2 611 925 people with women of reproductive age (15–49) making 23.58% of the total population. The zone had 6 public hospitals, 103 health centres and 374 health posts. In addition, it had 114 private clinics and 1 private hospital. All maternal health services, including ANC services, were provided free of charge in public health facilities.31 The population consisted of public health facilities that provided ANC, pregnant women attending ANC clinic, maternal healthcare providers and health authorities working for the surveyed health facilities. This study was part of a large project on the continuum of maternal healthcare that linked health facility data with a household survey. A number of sampling methods can be used to link characteristics of the sampled facilities to those of the serviced population, yet linking the sample areas(clusters) is the best approach32 and has been considered in this study. Full details of the sampling procedure for the project have been reported elsewhere.33 34 In brief, data from household surveys on access to maternity services were linked to health facility data in the same district. The two studies were timed in such a way that one could inform the next. As a first step, a community-based study was conducted using a multistage sampling procedure. For this, the study area was first stratified into 13 rural districts and 2 town administrations. Then, five districts of the zone (four rural districts and one town administration) were selected by a simple random sampling method. Next, 15 kebeles (3 kebeles from each district) were selected using a simple random sampling technique. Kebele is the smallest administrative unit in the Ethiopian context. Hence, in this study, the population-based study was the basis to identify the facilities to be surveyed. Second, the nearby public facilities at which the women sought care were identified during a house-to-house survey. Once the healthcare facilities were identified by type and location where they were found, facility survey was done across all the selected facilities. Providing basic maternity and reproductive health services (including ANC, facility delivery, postnatal care and family planning) during the last 12 months preceding the survey was the eligibility criteria to select health facilities. The health facility survey was conducted in all the 15 public health facilities: 5 primary hospitals and 10 health centres. All the healthcare providers that were engaged in the provision of ANC in the selected facilities during the data collection period were included in the study. This approach was in accordance with recommendations by Turner and colleagues.32 For the satisfaction survey and clinical observation data, a sample size of 824 was calculated using the single population proportion formula based on the following assumptions: 95% confidence level; 5% margin of error, 42% service availability and readiness of public facilities for ANC in Ethiopia,16 design effect of 2, and a 10% non-response rate. Then, the calculated sample size was allocated to each healthcare facility based on the average daily load of ANC attendees for the 2016/2017 fiscal year. From the sampled facilities, 795 women were selected using a systematic sampling procedure with probability to proportional size method. In this study, an attempt was made to measure all dimensions of the quality of ANC services: structure, process and outcome. For each dimension, a set of items were adapted from the WHO guidelines,3 13 the Ethiopian demographic and health survey35 and the list of interventions recommended by the federal ministry of health of Ethiopia.36 The survey comprised four main data collection methods: (1) health facility surveys; (2) provider interview; (3) direct observation of ANC consultations; and (4) exit interview with pregnant women. Structured and pretested questionnaires were used for service providers and client exit interviews, and checklists were used for facility survey and observation. Experienced midwives who were not affiliated with the surveyed healthcare facilities collected the data. In each health facility, a team consisting of two data collectors was assigned. While one of the team members was responsible to carry out the observational study and conduct a healthcare provider interview, the other was responsible for managing the exit interview and doing the facility survey. A facility survey of all the selected facilities was carried out to assess the availability of essential materials and staffing. For staffing, the assessment checklist was based on the national staffing standards for each type of health facility. A non-participatory observation was also made among 41 maternal health workers who were working at the ANC clinic during the time of the survey. This was aimed to evaluate whether healthcare workers conformed in interaction and conducting key ANC tests or examinations. When possible, the health workers who were observed were also interviewed, but when this was not possible, other providers of ANC service were substituted. Following her consultation, each pregnant woman was interviewed at the exit to assess the level of satisfaction on the service she received. Structural quality was calculated at the facility level, while indices for process attributes and client satisfaction were made at individual women level. The outcome variable was ANC client satisfaction, and computed by aggregating women’s responses to a series of questions regarding the ANC visit using a principal components analysis (PCA). During the analysis, important assumptions including Bartlett’s Test of Sphericity, Kaiser-Meyer-Olkin measure of sampling adequacy and communalities scores were checked. The eigenvalue>1 was used to decide the number of latent variables that we did for extracting factors. Initially, 24 variables were considered for the analysis but eventually, 12 variables were dropped as they failed to meet the assumptions of PCA. An index was computed from the original variables retained in the process. Thus, we computed a summed index from the retained items of the components that explained 68% of the total variances. Finally, the overall satisfaction index was developed by categorising the sum of scores into a three-point Likert scale: 0%–50% as ‘low satisfaction’ (coded 1), 51%–79% (coded 2) as having ‘moderate satisfaction’ and 80%–100% (coded 3) as ‘high satisfaction’. A facility inventory checklist with 47 items was used to assess the structural attributes of the health facilities. Each item was scored ‘1’ if the item was available and functional and a score of ‘0’ if this was not the case. The items in each construct were then added together, with equal weights, to generate the following 5 indices: (1) an infrastructure index (7 items); (2) Health Staff index (8 variables); (3) an equipment index (10 items); (4) Index for drugs and vaccines (14 items); and (5) index for lab capacity and supplies (8 items) (online supplemental file 1). Finally, we categorised each index into three categories: poor structure quality, fair and good structure quality. Furthermore, the overall summary score was constructed by aggregating the mean scores of all the five dimensions of care and was set as the structure index. bmjopen-2020-037085supp001.pdf The process attributes comprised interpersonal and technical aspects of the provider–client interaction. Interpersonal aspects included, among others, issues such as greetings, maintenance of privacy and handling of client concerns. Technical aspects included observation of specific services performed, such as history taking, ANC physical examinations, counselling related to pregnancy and laboratory examinations. A scoring system was established to calculate 5 dimensions of process attributes from 46 items: (1) interpersonal communication(6 activities), (2) history taking (12 activities); (3) clinical examination (9 activities); (4) counselling (10 activities); and (5) health screening and preventive measures(9 services; online supplemental file 2). This scoring system categorises whether an accepted standard of quality has been met or not. bmjopen-2020-037085supp002.pdf All the procedures/activities provided were weighted equally and was granted ‘1’ if the activity was observed and performed according to accepted standards of care, and a score of ‘0’ if this was not the case. The scores of the key items for each individual client–provider interaction being observed were added up and averaged to determine a score for each dimension of care. Furthermore, the overall summary score was constructed by aggregating the mean scores of all the dimensions of care and was set as the process index. The total scores ranged between 0 and 46. Accordingly, the process of quality care was scored as follows: low quality <23, moderate quality 23 to <37, and high quality ≥37. Other variables included were: (1) facility type, (2) demographics, (3) socioeconomic factors, and (4) reproductive characteristics of the women. The Cronbach’s alpha was used to measure the internal consistency of a set of items for the three quality dimensions, and a reliability coefficient of 0.70 or higher was considered acceptable.37 The collected data were entered using the Epi-info V.7.0 and analysed using STATA software V.14.0. Owing to the ordinal nature of the outcome variable (low, medium and high satisfaction), a typical approach was to use the standard ordered logit.38 Yet, the Brant test revealed that the proportional odds assumptions were not fulfilled for some independent variables (χ2=63.4 df (16); p<0.001). We, therefore, used a generalised ordered logistic regression with auto fit (also called partial proportional odds model) for assessing the association between satisfaction and explanatory variables. The partial proportional odds model is a hybrid of ordinal regression (same coefficients across the categories) and the default gologit (different coefficients across categories). Hence, since the Brant test was not met, the analysis gave two ORs for an explanatory variable (low vs medium/high satisfaction (OR1) and low/medium vs high satisfaction (OR2)). Whereas for variables that did not violate the proportional odds assumption, a single OR, OR1=OR2 (OR1=2) was reported.38 The bivariable gologit model was used for variable inclusion in the final multivariable model. Accordingly, the independent variables to be included in the multivariable model were selected when the p value was <0.2 in the bivariable model. Before running the multivariable analysis, multicollinearity test between independent variables was done using the Variance Inflation Factor (VIF), and variables were not strongly correlated (the highest value was 2.7). The final multivariable model was applied to define adjusted ORs, measuring the effect of different determinants on being assigned to a higher satisfaction category. Statistical significance for the final model was set at p<0.05. Patients/public were not involved in setting the research question or the outcome measures, and in the design and implementation of the study. Participation was voluntary based and no incentives were provided. The findings of this study will be disseminated to policy-makers and local-level service implementers.

• emiru, amanu aragaw
• alene, getu degu
• debelew, gurmesa tura