The influence of family physicians within the South African district health system: A cross-sectional study

PURPOSE Evidence of the influence of family physicians on health care is required to assist managers and policy makers with human resource planning in Africa. The international argument for family physicians derives mainly from research in high-income countries, so this study aimed to evaluate the influence of family physicians on the South African district health system. METHODS We conducted a cross-sectional observational study in 7 South African provinces, comparing 15 district hospitals and 15 community health centers (primary care facilities) with family physicians and the same numbers without family physicians. Facilities with and without family physicians were matched on factors such as province, setting, and size. RESULTS Among district hospitals, those with family physicians generally scored better on indicators of health system performance and clinical processes, and they had significantly fewer modifiable factors associated with pediatric mortality (mean, 2.2 vs 4.7, P =.049). In contrast, among community health centers, those with family physicians generally scored more poorly on indicators of health system performance and clinical processes, with significantly poorer mean scores for continuity of care (2.79 vs 3.03; P =.03) and coordination of care (3.05 vs 3.51; P =.02). CONCLUSIONS In this study, having family physicians on staff was associated with better indicators of performance and processes in district hospitals but not in community health centers. The latter was surprising and is inconsistent with the global literature, suggesting that further research is needed on the influence of family physicians at the primary care level.

We conducted a cross-sectional, observational study to compare community health centers (primary care facilities) and district hospitals with vs without family physicians. Use of family physicians was not randomized as the creation and filling of family physician posts were predetermined by local policy and service requirements. The group of facilities with family physicians had had a family physician in a designated post for a minimum of 2 years. The comparison group consisted of facilities that did not have a family physician post on staff or any other exposure to a family physician. We used the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement’s checklist23 to guide this research. We created a conceptual framework (Figure 1) to inform our approach to designing the study and determining the data collection instruments. In this framework, structure refers to issues of governance and economics that are largely affected by changes in policy. Health service processes are subdivided into generic (cross-cutting organizational processes), targeted (aimed at a specific program or condition), or clinical (services at the level of the patient). Generic and targeted processes can affect health system performance, which also influences the quality of clinical processes that in turn affect clinical outcomes. Family physicians were seen as a generic intervention as they were not limited to a specific program or condition and could have impact broadly on health system performance and clinical processes. We assessed 4 key aspects of primary health system performance: accessibility, coordination, comprehensiveness, and continuity.26 The key clinical processes were drawn from South Africa’s quadruple burden of disease and public health issues: HIV/AIDS and tuberculosis; violence and injury; maternal and child health; and noncommunicable diseases.17 Conceptual framework of the study (a modified Donabedian causal chain).24,25 This study was conducted in the district health system of the South African public sector in 7 of the country’s 9 provinces. (See Supplemental Appendix 2 at http://www.AnnFamMed.org/content/16/1/28/suppl/DC1 for a brief description of the South African district health system.) We used a clinical process indicator (the diabetes management score) and a health outcome indicator (the facility-based perinatal mortality rate) for calculation of study sample size, given that family physicians are reported to have a positive impact on these indicators, with an earlier study providing standard deviations and estimates of likely effect size.27 A sample size of 14 community health centers in each group gave 80% power to detect an effect size of 10% in the diabetes management score (SD, 13%) with a possible 5% type 1 error.27 A sample size of 14 district hospitals in each arm gave 80% power to detect an effect size of 8.4 perinatal deaths per 1,000 births in perinatal mortality rate (SD, 7.91) with a 5% type 1 error.27 We therefore chose a final sample size of 15 district hospitals and 15 community health centers in each group (with and without family physicians, for 60 facilities in total) to have sufficient power and to allow for some loss of facilities or incomplete data collection. Seven out of the 9 provinces were included in the study as determined by the educational footprint of the 6 participating universities that train family physicians in South Africa. We obtained a complete list of district hospitals and community health centers from the National Department of Health. With the assistance of the participating universities, this list was split into lists of facilities with and facilities without family physicians, which were then randomly reordered. Starting at the top of the randomly ordered lists, we selected 2 district hospitals and 2 community health centers with family physicians from each province to give 14 district hospitals and 14 community health centers. Each was then matched with a facility without family physicians from the other list using criteria shown in Table 1. One additional facility for each group was selected from the Western Cape Province, where the study was based, to arrive at the intended number of 15 facilities per group. (Supplemental Appendix 3, at http://www.AnnFamMed.org/content/16/1/28/suppl/DC1, shows the facility sampling selection process.) Matching Criteria by Facility Type For study outcomes, we selected a set of indicators that we expected would reflect the influence of the family physician on clinical processes, health system performance, and clinical outcomes (Figure 1). The selection of corresponding data collection instruments/ tools (Table 2)28–35 was dependent on the availability of reliable and valid routinely collected data or existing tools, the feasibility of collecting data, the different scope of practice in district hospitals and community health centers, and an a priori consensus between the researchers in the participating academic departments. Instruments/Tools Used for Data Collection CDM = chronic disease management; COPD = chronic obstructive pulmonary disease; MRC = Medical Research Council of South Africa; PCAT = Primary Care Assessment Tool; PIP = Problem Identification Program; WHO = World Health Organization. We trained 4 teams with a total of 16 fieldworkers (11 health professionals and 5 assistants with previous experience in research data collection) to collect data in the 7 provinces according to a detailed fieldwork protocol (Supplemental Appendix 4 at http://www.AnnFamMed.org/content/16/1/28/suppl/DC1). Fieldworkers were interviewed before appointment. Training was facilitated by the lead investigator (K.B.vP.) over 2 to 3 days, and consisted of face-to-face training, role playing, and practical evaluation in the field. Each team was led by a health professional and supervised by an academic family physician attached to a participating university. The teams also interacted remotely with the lead investigator (K.B.vP.) via telephone, e-mail, and a communication application (WhatsApp). Facility-level data were collected between June 2015 and March 2016, and then captured with EpiData version 3.1 (EpiData Software) via a double-entry method and using checks to minimize data entry errors.36 We then imported the data from EpiData into Microsoft Excel (Microsoft Inc) and used SPSS version 23 (IBM Corp)37 to conduct the analysis in consultation with a biostatistician. Data analysis commenced with descriptive statistics for the facilities. Subsequently, the independent samples t test for equality of means was used to compare means between the groups with and without family physicians (continuous dependent variables, see Table 2 for detail on the data collected). For those means found to be significantly different, we performed regression analysis using a generalized linear model to control for the effect of confounders. Confounding variables for health system performance were levels of staff (professional nurses, junior and senior physicians) and distance from a referral hospital. Confounding variables for clinical processes were the presence of outreach to the district health system facility (from the general specialties at the referral hospitals) and bed utilization rate (as proxy of district hospital inpatient workload). This study was approved by the Health Research Ethics Committee (Medical), Stellenbosch University (reference S15/01/003) and by each partner institution. The 7 provincial health authorities and research committees also gave permission to access facilities across the study setting (Supplemental Appendix 5 at http://www.AnnFamMed.org/content/16/1/28/suppl/DC1).

N/A

• von pressentin, klaus b.
• mash, robert
• baldwin-ragaven, laurel
• botha, roelf petrus gerhardus
• govender, indiran
• steinberg, wilhelm j.
• esterhuizen, tonya m.