Date Published: June 7, 2019
Publisher: Public Library of Science
Author(s): Lindsay Mallick, Gheda Temsah, Wenjuan Wang, Jacobus P. van Wouwe.
Measuring quality of care in family planning services is essential for policymakers and stakeholders. However, there is limited agreement on which mathematical approaches are best able to summarize quality of care. Our study used data from recent Service Provision Assessment surveys in Haiti, Malawi, and Tanzania to compare three methods commonly used to create summary indices of quality of care—a simple additive, a weighted additive that applies equal weights among domains, and principal components analysis (PCA) based methods. The PCA results indicated that the first component cannot sufficiently summarize quality of care. For each scoring method, we categorized family planning facilities into low, medium, and high quality and assessed the agreement with Cohen’s kappa coefficient between pairs of scores. We found that the agreement was generally highest between the simple additive and PCA rankings. Given the limitations of simple additive measures, and the findings of the PCA, we suggest using a weighted additive method.
Experts theorize that high-quality care for family planning has the potential to influence reproductive and fertility intentions [1, 2] and can consequently affect contraceptive use [3, 4]. Research in the last several decades has focused on both defining the critical elements of quality of care and assessing the impact of those elements [5–7].
Our study examines data from health facilities assessed by SPA surveys carried out in Haiti in 2013, Malawi in 2013–14, and Tanzania in 2014–15. These three countries had well timed SPA and Demographic and Health Surveys (DHS) surveys that allowed for ecological linkage between the surveys in each country for the technical report from which this manuscript is based . The SPA surveys comprise 4 separate questionnaires that assess service availability and readiness of facilities, health worker demographics and training, observation of selected client visits, and an exit interview assessing client demographics and perception of the visit.
This study uses secondary data originally collected through The Service Provision Assessment (SPA) survey by The Demographic and Health Surveys (DHS) Program. The SPA survey protocols were reviewed and approved by the ICF Review Board and the Ethics Review Committee of respective countries included in this study (Haiti, Malawi, and Tanzania). Data include no personally identifiable information. The methods described here are described in more detail in the corresponding technical report .
Hospitals constituted less than 15% of all facilities providing family planning services in Haiti, Malawi, and Tanzania (Fig 1). The majority of the facilities were health centers, dispensaries, and clinics. In Haiti and Malawi, the government or private entities managed a similar proportion of facilities. In contrast, in Tanzania nearly 90% of facilities were publicly managed. Facilities in Tanzania were disproportionately located in rural areas, at around 80%. S1 Table shows the distribution of facility characteristics among facilities with family planning clients observed on the day of the survey.
Simple additive scores are critiqued for being unsophisticated, failing to consider the greater relative importance of some indicators over others. Others adopt more complex scoring mechanism, such as applying weighting schemes based on predetermined weights or from multivariate analyses such as PCA. In our study, the PCA yielded results that suggest that the most critical indicators of quality of care vary across country. The low loadings produced by the PCA created scores most similar to simple additive measures in two of the three countries, resulting in consistency between simple additive scores and PCA-based scores in the way they rank the quality of care at health facilities. Moreover, the PCA results confirm that the construct of quality is multidimensional, the implications being that researchers should consider the use of sub-scales if possible. If a summary index must be used, we suggest a weighted additive summary measure, as it could be more useful and intuitive from the perspective of program planning. Simpler to construct and easier to interpret than a PCA-based summary measure, weighted additive measures can overcome shortcomings of the simple additive measures by accounting for issues of dimensionality and collinearity.