High quality care—at a minimum—is a combination of the availability of tangible resources as well as a capable and motivated health workforce. Researchers have suggested that supportive supervision can increase both the performance and motivation of health workers and the quality of care. This study is aimed at assessing the required number of visits and time between visits to bring about improvements in health service delivery. The study employed a primary health care performance improvement conceptual framework which depicts building blocks for improved health service delivery using longitudinal program outcome monitoring data collected from July 2017 to December 2019. The analysis presented in this study is based on 3,080 visits made to 1,479 health centers in the USAID Transform: Primary Health Care project’s intervention districts. To assess the effects of the visits on the repeated measure of the outcome variable (Service-Delivery), multilevel linear mixed model (LMM) with maximum likelihood (ML) estimation was employed. The results showed that there was a significant dose-response relationship that consistent and significant improvement on Service-Delivery indicator was observed from first (β = -26.07, t = -7.43, p < 0.001) to second (β = -21.17, t = -6.00, p < 0.01), third (β = -15.20, t = -4.49, p < 0.02), fourth (β = -12.35, t = -3.58, p < 0.04) and fifth (β = -11.18, t = -2.86, p < 0.03) visits. The incremental effect of the visits was not significant from fifth visit to the sixth suggesting five visits are the optimal number of visits to improve service delivery at the health center level. The time interval between visits also suggested visits made between 6 to 9 months (β = -2.86, t = -2.56, p < 0.01) showed more significant contributions. Therefore, we can conclude that five visits each separated by 6 to 9 months elicits a significant service delivery improvement at health centers.
USAID Transform: Primary Health Care covers a total of 396 districts in the four largest regions of Ethiopia (Amhara, Oromia, SNNP, and Tigray) where a total of 1,880 health centers provide health care to 53 million people. A health center is a health facility at the primary level of the health care system which provides promotive, preventive, curative and rehabilitative outpatient care including basic laboratory and pharmacy services with a capacity for 10 beds for emergency and delivery services. It is staffed with medical doctors, BSc as well as diploma level health science graduates including clinical officers, nurses, midwives, and lab technicians. On average a health center can have 35 direct service providers, and support staff [9]. On average, a health center is designed to provide health care services to 25,000 people residing in its catchment area. A supportive supervision checklist is a set of questions related to reproductive, maternal and child health and health system interventions which was developed by the USAID Transform: Primary Health Care project to guide field level support. The checklist is organized to frame a two-way discussion between the supervisor and the health worker at each institution. Each question has a definition, decision point and a response documentation section for improvement plan. The supervisors responsible for conducting the supervisions and providing technical guidance are—at a minimum- a first degree graduates in health studies, have experience of working at the primary level of care, and have attended a supervision technique training. During each visit, a supervisor is expected to spend at least half a day in the facility. When a supervisor goes to the institutions, s/he is expected to follow the checklist and record the findings and work with the staff and management of the health facility to bring about improvements on the identified problems. During facility support, data collection and entry is conducted onsite using an online electronic system and tablets. The system allows the questionnaires to be programmed and follow skip patterns based on previous responses. On a few occasions, the visit may be carried out by other experts who will use a paper format and then transfer the data to the online system. The study employed a retrospective cohort study. For assessment purposes, a primary health care performance improvement conceptual framework for primary health care (Fig 1) was used to categorize the questions into the major domain. The framework considers the role of service organization and quality of care as important drivers for primary health care performance [10]. As the supportive supervision is targeted to improve service delivery and its management, the Service-Delivery component of the framework was considered. A total of 30 questions were categorized into the five Service-Delivery components—access, availability of effective PHCs, high quality primary health care, population health management, and facility organization and management. The study uses a longitudinal program outcome monitoring data collected from July 2017 to December 2019. The USAID Transform: Primary Health Care project monitoring data is collected from the project intervention woreda health offices, primary hospitals, health centers, health posts, and households during routine and random supportive supervision visits with the objective of providing onsite technical support and producing unbiased data for decision making. During this period, a total of 1,322, 499, 3,080, 4,741, and 23,151 visits were made to woreda health offices, primary hospitals, health centers, health posts, and households respectively. The analysis presented in this study is based on the 3,080 supportive supervision visits made to 1,479 health centers in the project’s intervention districts. The composite measure of the service delivery of primary health care which was the average of the five Service-Delivery components of the PHCPI framework—access, service availability, patient centered care, population health management, and service organization and management—was considered as the outcome variable (Service-Delivery). A high score of this variable suggested the availability of better facility services. The number of visits to health facilities and the interval between consecutive visits were accounted as exposure variable for this study. The study had two levels of control variables. The first group includes the organization of the project support structure—facility distance from cluster office (CLO), average number of woreda per cluster staff, number of low performing woreda in the CLO and region, and the second level was related to health facility factors—number of technical staff, facility infrastructure (water and electricity), catchment population size, facility distance from woreda capital, and facility head’s experience in years. Data were managed using a web-based system, DHIS2 [11], and exported to SPSS version 25 for statistical analysis. Both descriptive and inferential statistics were applied. Descriptive statistics were used to analyze the five service-delivery components. To assess the effect of the control variables on the repeated measure of the outcome variable (service-delivery), multilevel linear mixed model (LMM) with maximum likelihood (ML) estimation was employed. In addition, the effects of access to roads on frequency of visits was also tested using multinomial logistics regression. Since the data had unequal sample sizes, inconsistent time interval, and missing data, applying univariate and multivariate tests of statistics was not recommended [12]. LMM is an appropriate approach when studying individual change as it creates a two-level hierarchical model that nests time within individual [13]. In addition, the study’s interest was on the subject-specific (facilities) interpretation of effects and identifying group variance sources, therefore, LMM was preferred over a generalized estimation equation to fit the data. The overall effect of each control variable on the Service-Delivery was tested through an F-test, while the effect of each category of each factor was tested through t-test with the respective degrees of freedom. To determine the best fit model, first, an unconditional mean model was used. In this model, no predictor was included. This model served as a baseline model to examine individual variation in the outcome variable without regard to time [14]. The model assesses the differences between the observed mean value of each facility and the true mean from the population. If the variation is high, it suggests that certain amount of outcome variation could be explained by the predictors at that level. Then a model containing time (number of visits) as a fixed and random effect was applied. This model tests if time (number of visits) is significant by examining the presence of interindividual difference in trajectory change over time. Finally, a model containing the fixed effects of variables of interest, the random intercept, and the random slopes were fitted. To select the best model, -2 log likelihood ratio test and Bayesian Information Criterion (BIC) were used. Generally, the smaller the statistical value, the better the model fit into the data. In all the statistical tests, significance was refereed at p < 0.05. The study considered aggregate secondary program data. The JSI Institutional Review Board (IRB) has determined that the study does not constitute “human subjects research” under US HHS regulation 45 CFR 46.102(f).