Menu
Menu
Menu
Menu
Objective: To examine the associations between perceived quality of care and patient satisfaction among HIV and non-HIV patients in Zambia. Setting: Patient exit survey conducted at 104 primary, secondary and tertiary health clinics across 16 Zambian districts. Participants: 2789 exiting patients. Primary independent variables: Five dimensions of perceived quality of care (health personnel practice and conduct, adequacy of resources and services, healthcare delivery, accessibility of care, and cost of care). Secondary independent variables: Respondent, visit-related, and facility characteristics. Primary outcome measure: Patient satisfaction measured on a 1-10 scale. Methods: Indices of perceived quality of care were modelled using principal component analysis. Statistical associations between perceived quality of care and patient satisfaction were examined using random-effect ordered logistic regression models, adjusting for demographic, socioeconomic, visit and facility characteristics. Results: Average satisfaction was 6.9 on a 10-point scale for non-HIV services and 7.3 for HIV services. Favourable perceptions of health personnel conduct were associated with higher odds of overall satisfaction for non-HIV (OR=3.53, 95% CI 2.34 to 5.33) and HIV (OR=11.00, 95% CI 3.97 to 30.51) visits. Better perceptions of resources and services were also associated with higher odds of satisfaction for both non-HIV (OR=1.66, 95% CI 1.08 to 2.55) and HIV (OR=4.68, 95% CI 1.81 to 12.10) visits. Two additional dimensions of perceived quality of care-healthcare delivery and accessibility of care-were positively associated with higher satisfaction for non-HIV patients. The odds of overall satisfaction were lower in rural facilities for non-HIV patients (OR 0.69; 95% CI 0.48 to 0.99) and HIV patients (OR=0.26, 95% CI 0.16 to 0.41). For non-HIV patients, the odds of satisfaction were greater in hospitals compared with health centres/ posts (OR 1.78; 95% CI 1.27 to 2.48) and lower at publicly-managed facilities (OR=0.41, 95% CI=0.27 to 0.64). Conclusions: Perceived quality of care is an important driver of patient satisfaction with health service delivery in Zambia.
The exit interviews were conducted between December 2011 and May 2012 across 16 Zambian districts as part of the Access, Bottlenecks, Costs, and Equity (ABCE) project. The details of this project are documented elsewhere and available online.26 A two-step stratified random sampling process was used to select health facilities. First, Zambia’s districts (72 at the time, currently 103) were stratified on the basis of average household wealth, population density and skilled birth attendance (SBA) coverage. One district was randomly selected from each wealth–population–SBA category, in addition to the capital, Lusaka. In each selected district, we selected all hospitals, two urban health centres, three rural health centres, and a quota of associated health posts. The exit interviews were conducted at a subset of the facilities selected for the overall ABCE project. Our study reports on interviews conducted at 104 facilities. Compared with all facilities in Zambia, we oversampled hospitals and urban health centres and undersampled rural health centres and health posts to allow for platform-specific analyses (see online supplementary appendix table 1). Our sample is representative of the Zambian population and health delivery system, except that we oversampled hospitals to allow for separate analyses of hospital data. The sample of patients who sought care was also skewed towards females, which is expected due to several factors including women seeking maternal health services and a higher HIV prevalence among women (15.1%) than men (11.3%).11 At each facility participating in the exit survey, 30 patients were systematically sampled as they exited. Sampling intervals varied from every patient to every four patients, depending on the patient volume reported by the facility manager. The sample size of 30 patients at each facility was estimated using the Kish method with the following assumptions: patient satisfaction rate of 10%, precision of 5%, α of 1%, design effect of two, and non-response rate of 20%. The estimated sample from the Kish method was further adjusted to allow for robust subgroup analyses (eg, HIV vs non-HIV; hospital vs health clinic; rural vs urban). Interviews were conducted over at least two days at each facility. Patients were required to be 15 years or older and in an appropriate physical and mental state to be eligible to complete the survey. If a patient was too young or otherwise ineligible, an eligible attendant was asked to answer on their behalf when possible. Verbal consent was obtained from all respondents, and surveys were conducted in a location where the facility staff and other patients were not present. Trained research assistants recorded exit interview responses electronically using the DatStat data collection software. On a daily basis, data were uploaded to a database accessible from Seattle, where they were continually verified for quality during the collection process. The median interview time was nine minutes. At each health facility, research assistants interviewed facility administrators to collect information about facility resources, staffing, management and practices. Facility level and management were verified against a facility roster provided by the Ministry of Health (MOH). The exit instrument drew questions from established patient exit and household surveys, which in-country partners tested and modified to fit the Zambian context. Demographic questions were based on the Zambian DHS.27 Questions about visit circumstances and costs were adapted from the World Health Survey.28 We measured patients’ overall satisfaction with the facility with the following question from the Consumer Assessment of Healthcare Providers and Systems Adult Visit questionnaire: Using any number from 1 to 10, where 1 is the worst facility possible and 10 is the best facility possible, what number would you use to rate this facility?29 30 The survey also captured how patients perceived the quality of specific aspects of the facility and its providers, based on a validated questionnaire developed by Baltussen et al31 that has been used in other developing settings.32 33 Patients were asked to rate 25 aspects of the facility on a five-point Likert scale: very bad, bad, moderate, good or very good. The majority of questions were answered by over 95% of patients, but we excluded five questions to which over 10% of patients responded ‘not applicable’, ‘don’t know’, or ‘decline to respond’. These five questions concerned: adequacy of doctors for women, ease of making payment arrangements, time doctors allow for patients, availability of good doctors, and provider’s follow-up with patients. We then used principal component analysis (PCA) with orthogonal rotation to examine the structure of the remaining 20 perceived quality questions (see online supplementary appendix table 2). The analysis identified five components with eigenvalues ranging from 0.94 to 7.8, which explained 62% of the variance. The factors aligned with theoretical domains and can be interpreted as: (1) health personnel practices and conduct, (2) adequacy of resources and services, (3) healthcare delivery, (4) accessibility of care, and (5) cost of care. The specific questions under each domain are listed in online supplementary appendix table 2. The factor with an eigenvalue under 1 (accessibility of care) was retained because the variables it contained were theoretically grouped and not otherwise represented. Cronbach’s α coefficients for each grouping ranged from 0.70 to 0.90, which met the generally accepted threshold of 0.70 and was comparable to or better than studies conducting similar exercises.34 To condense the information for each domain, we created a new variable that was the per cent of questions within the domain which the respondent rated ‘good’ or ‘very good’. We opted to examine the responses in this categorical manner rather than as continuous averages because (1) Likert scales from very bad to very good are not truly continuous and (2) research shows that patients typically rate facilities favourably, and therefore the important distinction is achieving the very highest ratings.35–37 If a patient did not answer a given question, we took the per cent among the questions that were answered. We used random-effects ordered logistic regression models to examine how overall satisfaction (rated from 1 to 10) was related to objective patient, facility and visit factors, as well as patient perceptions of specific aspects of care (measured with the 5-point Likert scale). The unit of analysis was the patient, and the outcome for all models was the patient’s overall rating of the facility out of 10 (described above in measuring satisfaction). An ordered model was selected because the outcome scale was ordered but not truly continuous. Additionally, since the outcome variable was skewed towards higher ratings, we grouped all responses below six into a single category for the purpose of the regression models (see online supplementary appendix figure 1). The first model examined how facility, patient and visit characteristics were associated with overall satisfaction. Independent variables were selected a priori based on relationships previously identified in the literature. Facility variables included facility type (hospital or health centre/post), location (urban or rural), and management (public or non-governmental organisation [NGO]/private). Demographic variables included age, self-rated overall health, ethnicity, sex, education level, and a binary indicator of whether the respondent was the patient or an attendant. Variables surrounding visit circumstances included travel time, wait time, and type of provider seen. We did not include whether or not the patient paid a user fee as this was largely determined by facility management—public and NGO facilities typically offer free services while private facilities often charge fees. The second model looked at how patients’ perceptions of particular domains of care related to their overall perception, to identify which aspects are most influential. The predictor variables in this case were the five summary perceived quality variables (described above in condensing perceived quality responses): health personnel practices and conduct, adequacy of resources and services, healthcare delivery, accessibility of care, and cost of care. Our final combined model included all of the facility, patient, visit, and perceived quality predictors from the first and second models. This allowed us to examine whether any facility, patient or visit characteristics were associated with overall satisfaction independent of how the patient rated specific aspects of care. All models included facility random effects to account for unmeasured facility characteristics, and we estimated robust standard errors (SEs) to account for intragroup correlation within facilities. Patients missing one or more covariates were excluded from all regression analyses. Our sample contained a substantial number of patients receiving HIV-related services; we analysed these patients separately from those receiving other services because HIV care may involve specialised staff, equipment and drugs, and because HIV often receives unique policy attention based on the large burden it poses in Zambia. We additionally conducted sensitivity analyses to test for effect modification by facility management, facility location, facility level and respondent identity (patient or attendant). To do this, we conducted the same analyses described above, stratified by the characteristic of interest, rather than by the HIV visit or not. Data management and analysis were conducted in Stata V.13.1.
N/A
