We used data from the Service Provision Assessment, which is a census survey of health facilities conducted in Haiti in 2013 by the Demographic and Health Survey Program. The census included a facility assessment, a questionnaire for health-care providers, observations of sick child, antenatal care and family planning visits, and exit interviews with observed clients. We limited our analysis to the data collected on outpatient primary care facilities, i.e. dispensaries and health centres with or without beds.22
We also used WorldPop maps to obtain estimates of the 2015 population density of Haiti, at a resolution of 100 m2.23
Measuring primary care quality
We developed metrics of service delivery quality following the Primary Health Care Performance Initiative’s framework. Several modifications were required to adapt the framework for health facility assessment (Fig. 1). We excluded the domain “population health management”, because of a lack of relevant facility-related data. For clarity, we also altered the labels for two of the domains, using “effective service delivery” for the availability of effective services and “primary care functions” for high-quality primary health care.14
Fig. 1. Conceptual framework of quality in primary health care
Fig. 1. <b>Conceptual framework of quality in primary health care</b>
Source: Adapted from the Primary Health Care Performance Initiative’s framework,13 for use in Haiti.
Figure 1 - full screen
We reviewed the data available in the survey and selected 28 indicators that most appropriately matched each of the quality subdomains included in our analysis. For this selection, we were guided by the Primary Health Care Performance Initiative’s method note.13 Each indicator is a proportion or an index that ranges from 0 to 1. For example, the indicator “sick child did not first visit traditional healer” measures first-contact access to a facility as the proportion of sick children who came to the facility for care without first visiting a traditional healer. All selected indicator definitions are available from the corresponding author. Within the survey data, we were unable to find relevant indicators for two of the subdomains that we wished to investigate: geographical access and the organization of team-based care. As people need to be able to access health facilities to benefit from quality care, we used the WorldPop maps to determine geographical access to facilities.
For each primary care facility, we calculated a score for each of four service delivery domains: (i) accessible care; (ii) effective service delivery; (iii) management and organization; and (iv) primary care functions. Each of these scores, which could range from 0 to 1, was the mean of all the indicators under the domain. As we considered the four domains to be equally important elements of quality primary care, we took the mean of the four scores calculated for each facility as the overall measurement of the quality of the facility’s service delivery for primary care.
Although the census covered all but two of the health facilities in Haiti in 2013, two of the survey tools, i.e. clinical observations and patient interviews, were applied only in a selected subset of facilities. For each indicator included in our analysis, we used multiple imputation to generate five versions of a completed data set for all quality indicators. We based the imputation on observed covariates, e.g. management type and urban, and the non-missing indicators.
Finally, we assessed the distribution of indicators across facilities and sought valid groupings of better and worse quality. Given the lack of universally defined minimum quality thresholds and the rudimentary nature of many of the indicators included in our analysis, we divided the facility scores into three categories of quality. Scores of less than 0.50, 0.50–0.74 and at least 0.75 were considered indicative of poor, fair and good quality, respectively.
We defined each 100 m2 block of population as an urban or rural population using the census’ urban or rural classification of the facility nearest to the centre of the block. As a sensitivity check, we also defined an urban population as one in which there were at least five people per 100 m2 block.
We calculated descriptive statistics of the primary care facilities with non-response weights. We summarized mean values and uncertainty intervals for each indicator, domain and overall quality score for service delivery. As the data we analysed provided a census of the primary care facilities in Haiti in 2013, the uncertainty intervals that we calculated indicate the measurement error attributable to missing data.24 Using inverse distance-weighted interpolation, we mapped, across Haiti, the quality of the primary care available to a nearby population. In the resultant map, the colour of each 100 m2 block indicates whether the quality of the nearest primary care facility was poor, fair or good. We used the global Moran’s I statistic, which tests for the presence of spatial autocorrelation,25 to investigate whether facilities of good or poor quality, in terms of each of the four domains of interest, were clustered geographically. Moran’s I can range from −1 to 1. In our analyses, positive I values would indicate that primary care facilities of similar quality were clustered together. We defined proximity using an inverse-distance weight matrix.26 In keeping with prior research on physical access to care in Haiti,27 we calculated the percentages of the entire Haitian population, rural population and urban population living within 5 km of any facility and within the same distance of a facility with a good overall score. Finally, we mapped the areas that lay within 5 km of any facility and a facility with a good overall care score.
Multiple imputation was conducted in R 3.2 (R Core Team, Vienna, Austria). All other analyses were conducted in Stata version 14.0 (StataCorp, LP, College Station, United States of America). We used QGIS version 2.1228 to map the data.
The Harvard University Human Research Protection Program categorized this secondary analysis of data as exempt from human subjects review.
The survey obtained detailed data from 905 (99.8%) of the 907 health facilities in Haiti in 2013, 786 of which were primary care facilities and included in the analysis (Table 1). Most primary care facilities were classified as rural, although there was a high concentration of primary care facilities in and around Port-au-Prince. Fig. 2 summarizes the performance of the primary care facilities across the four domains of primary care service delivery. At the average facility, 86% and 94% of clients, respectively, stated that they did not find wait times or the costs of care to be a problem, even though about half of all primary care services required payment and over half of the primary care facilities had mean wait times in excess of one hour. Large gaps in quality were evident in the metrics for the availability of effective services. The indicators for provider motivation and safety were found to be especially low. Basic elements of clinical care were not universally followed. For example, at the average facility only 57% of the providers asked about maternal age at a first visit for antenatal care. Low quality scores for primary care functions were partially attributable to poor provider communication. Under management and organization, only 2% (18) of the primary care facilities had a system for gathering feedback from their clients and nearly three-quarters (577) did not have routine quality assurance processes. For their overall quality of service delivery, the primary care facilities in Haiti achieved a mean score of 0.59.