Examining the influence of family physician supply on district health system performance in South Africa: An ecological analysis of key health indicators

Background The supply of appropriate health workers is a key building block in the World Health Organization’s model of effective health systems. Primary care teams are stronger if they contain doctors with postgraduate training in family medicine. The contribution of such family physicians to the performance of primary care systems has not been evaluated in the African context. Family physicians with postgraduate training entered the South African district health system (DHS) from 2011. Aim This study aimed to evaluate the impact of family physicians within the DHS of South Africa. The objectives were to evaluate the impact of an increase in family physician supply in each district (number per 10 000 population) on key health indicators. Setting All 52 South African health districts were included as units of analysis. Methods An ecological study evaluated the correlations between the supply of family physicians and routinely collected data on district performance for two time periods: 2010/2011 and 2014/2015. Results Five years after the introduction of the new generation of family physicians, this study showed no demonstrable correlation between family physician supply and improved health indicators from the macro-perspective of the district. Conclusion The lack of a measurable impact at the level of the district is most likely because of the very low supply of family physicians in the public sector. Studies which evaluate impact closer to the family physician’s circle of control may be better positioned to demonstrate a measurable impact in the short term.


Introduction
Strong primary health care systems require primary care teams that consist of an appropriate mix of health workers tailored to the health care needs of the communities they work in. 1 The supply of appropriate health workers is a key building block in the World Health Organization's (WHO) model of effective health systems. 2 In sub-Saharan African countries these primary care teams and their communities are challenged by a mix of health system constraints, socio-economic disparities and disease burdens. 3 Primary care teams are stronger if they contain doctors with postgraduate training in family medicine. 2 The contribution of such family physicians to the performance of primary care systems has been established in high-income countries. 4,5 International studies (mainly in the USA, UK, Canada and Korea) described the public health benefits associated with an increased supply of primary care doctors, especially regarding a reduction in all-cause, infant and chronic disease-related morbidity and mortality. 6,7,8,9,10,11,12,13 Many of these studies applied a broad definition of primary care doctors, by including all clinical specialities that work in primary care (family medicine, general practice, general internal medicine and general paediatrics). 6,7,10,12,14 Some of these studies (notably UK and Canada) focused on family physicians or general practitioners, two terms which apply to the same professional: a primary care doctor with postgraduate training in family medicine or general practice. 9,11,13 Family medicine is a young discipline in Africa, with a number of countries only commencing postgraduate training during the last decade. 3,15,16,17,18,19,20,21,22 Qualitative studies have explored the opinions of African leaders and managers on the potential contribution and possible roles of family physicians in the district health system (DHS). 23,24,25,26 There is, however, little quantitative evaluation of their actual impact to guide policy-and decision-makers on the deployment of family physicians. The uncertainty revolves around their cost-effectiveness and how best to position these family physicians within the different levels and components of the health system. The relationship between family physician supply and DHS performance has not been evaluated in the African context.
In South Africa, family medicine was gazetted as a new speciality during 2007 by the Health Professions Council of South Africa (HPCSA). 5,19 This event paved the way for structured postgraduate training through training posts (registrars) and a consensus on training outcomes. 5 This developmental phase included the creation of new family physician posts within the DHS. These posts are mainly at district hospitals and community health centres, although a few are located at regional hospitals. During this same period, the National Department of Health (NDoH) started implementing primary health care reforms, which included family physicians within district clinical specialist teams that were tasked with strengthening maternal and child health care. 27, 28,29 In addition, the new national policy on human resources for health and the national development plan support the deployment of family physicians within the DHS, but lack sufficient detail to guide managers on how best to utilise these expert generalists. 30 Following further discussions with the national department, a national position paper was published by the leadership of academic family medicine, in order to clarify the contribution of family physicians to the DHS. 5 This consensus statement introduced the 'new' definition of the family physician as an expert generalist in the DHS capable of supporting and leading health care teams through six interwoven roles: competent clinician, consultant to the primary care team, capacity builder, leader of clinical governance, supporter of community-orientated primary care and in some instances a supervisor of under-or postgraduate students.
The first graduates of the new training programmes entered the DHS from 2011. 5 Family physicians from the previous training programmes in South Africa and elsewhere still form the bulk of the available family physicians, as the nine South African training institutions are not yet training to the scale envisaged by the national position paper. 5 The training standards are coordinated through the South African Academy of Family Physicians (SAAFP), and the South African College of Family Physicians (CFP) is responsible for the national exit examination. The nine training institutions, SAAFP and CFP successfully responded to a funding call from the NDoH and EuropeAid to implement a project aimed at strengthening the contribution of family physicians to the primary health care system. 31 This project included an applied research activity, which aimed to evaluate the initial impact of family physicians on the DHS in South Africa. This article presents one of the four complementary studies and looks at the relationship between the supply of family physicians and DHS performance. The other three studies consist of a quasiexperimental comparison of facilities with and without family physicians, a 360-degree evaluation of family physician's impact by their colleagues and qualitative interviews with district managers who employ family physicians.

Aim and objectives
This study aimed to evaluate the impact of family physicians within the DHS of South Africa. The objectives were to evaluate the impact of an increase in family physician supply in each district (number per 10 000 population) on key health system performance indicators, key clinical processes and key health outcomes.

Study design
This ecological study was informed by a pilot study conducted in the Western Cape, South Africa. 32 A retrospective cohort design was used, whereby data were collected for the period 2010/2011 as a baseline and 2014/2015 representing 5 years post-deployment of the new generation of family physicians. The STROBE statement's checklist for reporting cohort studies was used as standard for presenting this research. 33

Setting
This study evaluated all 52 health districts across all nine provinces of South Africa (a national study frame, see Figure 1) for two time periods.

Study population and sampling strategy
All 52 South African health districts were included as units of analysis.

Data collection
A national dataset, the District Health Barometer (DHB), contributed the data on district performance for two time periods: 2010/2011 and 2014/2015. 35,36 The DHB draws data from several data sources provided by the NDoH. Compilation of the DHB is guided by an advisory committee made up of managers from the NDoH, as well as health experts from Health Systems Trust (HST). The DHB is designed to assist the NDoH in monitoring health service delivery at district level for all of South Africa's health districts. Furthermore, the HST encourages providers, managers, researchers and policy-makers to use DHB information by making the publication and its data freely available online on their website. Table 1 presents the list of DHB indicators used. The DHB system of categorising the indicators was used throughout (ranging from financial indicators to clinical process and outcome indicators). The official DHB indicator descriptions are also presented in Table 1.
For the family physician supply, public sector family physicians working in joint appointments (with the universities) or nonjoint appointments and employed at facility-, sub-district and district levels (including district office and district clinical specialist team appointments) were included. Those family physicians employed at regional or tertiary hospitals in fulltime academic positions or in the private sector were excluded. The data on family physician supply per district for these two time periods were obtained from all nine academic institutions involved with postgraduate family medicine training in South Africa and who were familiar with the health system in their catchment area. The absolute numbers of family physicians were converted to family physician supply per 10 000 population (using the DHB population data for the respective time periods).

Data analysis
The DHB data, as well as data on family physician supply, were entered into an Excel sheet and subsequently converted into IBM SPSS version 23 for descriptive and inferential analyses. 37 The data analysis included all 52 units of analysis and commenced with descriptive analysis of the independent and dependent variables. Subsequently, the correlation between change in family physician supply and change in the indicators available for both time periods (37 indicators) was analysed. In addition, a cross-sectional correlation analysis was performed for time period 2 (2014/2015) on the remaining DHB data set (data for 12 indicators were available only for time period 2). Simple scatterplots of the bivariate correlations were inspected to identify the nature of each relationship. A non-parametric test, Spearman's rho, was selected to test for correlation between the independent and dependent variables, because of the non-parametric distribution of the data as well as the presence of outliers (especially in reference to the independent variable). The level of significance chosen was p < 0.05. For those relationships found to be linear and showing at least a lowto-moderate correlation coefficient (see interpretation guide below), further regression analysis was performed using a generalised linear model (GLM), to control for the effect of available confounders, namely province and socio-economic quintile (SEQ) of the districts. Using GLMs with province as covariate created better regression models as opposed to GLMs with SEQ as covariate (using the omnibus test and its likelihood ratio Chi-square value as guide).
Correlation values may be interpreted as: 32

Ethical considerations
This study was approved by the Health Research Ethics Committee, Stellenbosch University (reference S15/01/003) and HST also confirmed their permission for use of the open access data.

Results
Tables 2 and 3 present descriptive statistics for the dependent variables, as well as the results for the non-parametric correlation analysis. The median (and interquartile range) of the independent variable, the supply of family physician per 10 000 total population, was 0.027 (0.000-0.043) for time period 1 and 0.035 (0.016-0.054) for time period 2. The medians (and interquartile ranges) for the absolute numbers of family physicians per district were 2.00 (0.00-4.00) for time period 1 and 2.00 (1.00-5.00) for time 2 (total numbers were 153.5 for time period 1 and 208.5 for time period 2). The majority of correlations were negligible to low and not statistically significant. Two correlations from the change over time correlation analysis were found to be statistically significant (using the initial Spearman's rho analysis): a HIV management indicator, 'Percentage of TB cases with known HIV status' (low negative correlation, rho = -0.351, p = 0.011) and an additional indicator, 'Vaccine expenditure per population under 1 year', a measure of the efficiency of immunisation and not the coverage (low negative correlation, rho = -0.378, p = 0.006). One indicator from the cross-sectional time 2 analysis showed a statistically significant, low negative correlation, namely 'Inpatient crude death rate' (rho = -0.340, p = 0.014). Scatter plots of these correlations are shown in Figures 2, Figure 3 and Figure 4. The influence of the three outlying values was clear on inspection: for example, the scatterplot of 'Percentage of TB cases with known HIV status' (Figure 2) showed a random scatter if one ignores the three outliers.
Regression analysis of these three correlations was performed. After adjusting for province in a GLM, the overall vaccine expenditure became positive in most of the nine provinces (see Table 4). This is a real example of confounding by province. Relative to the Western Cape Province, most of the provinces increased their expenditure on vaccines between time periods 1 and 2. The effect of The provincial and LG district expenditure on DHS per capita (total population) refers to the total amount of money spent on DHS (all sub-programmes except 2.8 Coroner services) per person with and without medical scheme coverage.
Provincial and LG PHC expenditure per capita (total population) a The PHC expenditure per capita (total population) measures the total amount of money spent annually by each district as a percentage of the total population in the district.

ICDR
The ICDR is an impact indicator that refers to the proportion of all inpatient separations because of death. Inpatient separations include inpatient transfers out, deaths and inpatient discharges. The indicator therefore includes deaths from all causes that occur in a health facility.
Delivery care Delivery in facility under 18 years rate a This indicator measures the proportion of all deliveries that occur among women younger than 18 years. The numerator is the number of deliveries among women under 18 years of age, while the denominator represents all deliveries that have been recorded at the health facility. This outcome indicator is used as a proxy to track success in the prevention of teenage pregnancies.
Inpatient ENDR a The inpatient ENDR or inpatient death 0-7 days measures the number of deaths among live born babies that occur within seven completed days after birth per 1000 live births. It only includes neonatal deaths when the foetus is at 26 or more weeks' gestational age and/or weighs 500 g or more.
Maternal mortality in facility ratio a The WHO definition of a maternal death is the death of a woman while pregnant or within 42 days of termination of pregnancy, irrespective of the duration and site of the pregnancy, from any cause related to or aggravated by the pregnancy or its management, but not from accidental or incidental causes. The MMR is the number of maternal deaths per 100 000 live births. This indicator refers to the facility-based (and not the population-based) MMR.
Stillbirth in facility rate a The stillbirth rate measures the number of babies born dead per 1000 total births. The indicator does not differentiate between fresh and macerated stillbirths. Stillbirths should only be counted when the foetus is at 26 or more weeks of gestational age and/or weighs 500 g or more.
Delivery by C-section rate (district hospitals) a The C-section rate measures the proportion of deliveries in hospitals that are carried out by C-section. The numerator is the number of C-sections conducted in the facility, and the denominator is the number of deliveries that took place in that facility over the same time period. It is therefore a facility-based and not a population-based indicator. This chapter focuses on C-sections performed at district hospitals.
Mother postnatal visit within 6 days rate a The mother postnatal visit within 6 days rate indicator monitors access to postnatal care. The numerator for this indicator is the number of postnatal visits by a mother within 6 days of delivery, either at a PHC facility or a postnatal home visit by facility staff. The purpose of the visit is for a postnatal check-up. Only the first visit after delivery should be counted. The denominator is the number of deliveries in facility. Deliveries include deliveries at hospitals and at PHC facilities.
PMTCT Antenatal first visit before 20 weeks rate a Early registration for antenatal care is an important entry point into the health system for pregnant women, allowing them to access health care services (and health information), including PMTCT services. This indicator shows the percentage of pregnant women who have their first antenatal visit before 20 weeks, out of all antenatal clients' first visits (those whose first visit was before and after 20 weeks).  Table 5). The correlation remained negative, but decreased in its strength and became non-significant The cervical cancer screening coverage measures the annual number of cervical smears taken in women 30 years and older as a proportion of the female population 30 years and older, factored for one smear every 10 years. In practice this means that the denominator is 10% of the female population aged 30 years and older.

CYPR (annualised) a
The CYPR indicator measures the percentage of women aged from 15 to 49 years who are protected against unplanned pregnancies for a year using modern contraceptive methods, including sterilisation. The volume of all contraceptives dispensed to clients during a specified period of time (a year) is used to estimate the amount of protection against pregnancy during that particular period. This estimate of protection is called the 'contraceptive year equivalent'. This forms the numerator for the CYPR indicator. Each type of contraceptive method that is distributed is adjusted by a conversion factor (country-specific) to yield an estimate of the duration of contraceptive protection. The denominator for the CYPR is the 'female target population 15-49 years', where females are used as a proxy for couples.  LG, local government.
significance of the provincial covariate was p = 0.810 (Wald Chi-Square test).

Key findings
Five years after the introduction of family physicians this study showed no demonstrable correlation between family physician supply and improved health indicators from the macro-perspective of the district. The lack of a measurable impact at the level of the district is most likely because of the very low supply and deployment of family physicians in the public sector, which makes their impact undetectable.

Discussion of key findings
The family physician supply in the international literature (supply ranging between 4.3 and 12.0 per 10 000 population in countries such as the USA, UK, Canada and Korea) was at least 100 times more than the 0.03 per 10 000 reported here. Our definition of family physician supply, however, differed from the definitions of primary care physician supply in these references, as the international literature generally included all clinical primary care physicians (usually with postgraduate training in specialities such as paediatrics and internal medicine). These international studies were also conducted in less socio-economically deprived settings  where postgraduate training of primary care physicians was well established. It may be more appropriate to compare our family physician supply to that of other BRICS countries (Brazil, Russia, India, China and South Africa): the total family physician supply in South Africa (private and public sector, all levels of health care) was 0.1 per 10 000 in 2015, compared to 0.2 per 10 000 in Brazil and 1.2 per 10 000 in China. 39 The total South African supply of family physician per 10 000 needs to double in order to meet at least Brazil's supply. South Africa's NDoH echoes this by identifying a shortfall of 888 family physicians in their 2011 HR policy document. 30 While some correlations demonstrate a possible trend, the size of these correlations did not exceed 0.5 in either direction. The initial significant correlations disappeared after controlling for the available confounders, especially the provincial covariate. This large degree of heterogeneity between the provinces makes it difficult to assess for an effect of the family physician supply per 10 000 population at a country level.

Strengths and limitations
Our study was limited by our definition of primary care physician supply, by excluding primary care doctors who were not registered as family physicians with the Health Professions Council of South Africa. A further limitation is the exclusion of private sector family physicians who may have an indirect effect on DHS performance, as they are seeing uninsured patients for out-of-pocket consultations. Some private sector family physicians may be contracted into public sector primary care facilities in the NHI pilot districts since 2013.
The study was also limited by the set of DHB variables that were determined by the NDoH and were not specifically intended to measure the impact of the family physician. The DHB data are based on routinely collected data which may lack the rigour required for research, although HST applies statistical methods to clean and improve data quality. Data quality issues of source data were described in the DHB. 35 Furthermore, our analysis was limited by the availability of data for all indicators in both time periods, as an analysis over time is more sensitive to the effect of family physicians as opposed to a cross-sectional analysis.

Implications or recommendations
While this study from a broad macro-level district perspective did not demonstrate an impact of the family physicians on the DHS performance, other studies to be published elsewhere will present additional data from the facility and individual levels. These studies at a meso-level and a micro-level are more likely to demonstrate an impact as they evaluate the family physicians closer to their circle of control and influence. The correlation analysis should be repeated in 5 years, when the family physician supply is greater. It is also recommended that this correlation analysis includes a comparison with a broader definition of primary care doctor supply (all primary care doctors working in the DHS).

Conclusion
It is still too early to demonstrate the impact of an increase in supply of family physicians at the district level on key health system performance indicators, key clinical processes and key health outcomes. Studies which evaluate impact closer to the family physician's circle of control may be better positioned to demonstrate a measurable impact in the short term. A repeat correlation analysis is recommended in 5 years to allow for time (duration of effect) and training output (size of supply). Opportunities to deploy more family physicians within the DHS should be explored and supported.