Study location
There are seven administrative regions in the City of Johannesburg in Gauteng Province, South Africa. Each municipal region is responsible for the delivery of primary and environmental healthcare services to its inhabitants [11]. We used purposive sampling to choose three public healthcare clinics within one municipal region of the City. All were under the jurisdiction of the City (as opposed to the Province). Clinic 1 was a relatively large clinic, which previously housed a CCMT on its grounds, and was located in an under-served, peri-urban, densely populated area. Clinic 2, which was smaller than Clinic 1, was located about 2 km away and served the same population. Clinic 3—also a small clinic—was located about 12 km away from Clinic 1 in a more developed urban area which was home to most of the approximately 250,000 residents of the region [11].
Data sources
We conducted a retrospective evaluation using healthcare utilisation and expenditure data from several sources: South Africa’s District Health Information System (DHIS), Tier.net data, clinic- and province-level expenditure reports, and staff records supplied by the municipal regional health authority.
The DHIS is South Africa’s official system for tracking service delivery metrics in the public health sector. It is used for recording the number of routine health services rendered by public health facilities on a monthly basis. Data collected at the primary healthcare level include the number of individuals who receive HIV testing, ART, tuberculosis testing and treatment, child health, immunisations, contraception, and other women’s health services. The DHIS also contains patient headcounts, which represent the total number of patients who have visited a clinic in a given month. The source documentation for the DHIS is paper registers maintained by staff in clinics, and although the data are checked at various stages during compilation, they are prone to errors [12, 13]. Nonetheless, we obtained monthly DHIS data for the period of 2004 to 2016 for this analysis from the City of Johannesburg.
For presentation of contraceptive data from the DHIS, we summed the number of visits reported for delivery of subdermal contraceptive implants, intra-uterine contraceptive devices (IUCDs), medroxyprogsterone injections, norethisterone enanthate injections, and oral contraceptive pills. Condoms visits were excluded to avoid double counting as they were often distributed at the same time as several other services. We also obtained immunisation data from the DHIS. These data represent the number of under-1-year-olds reported to have been fully immunised at each clinic—meaning they presented for their last under-one vaccination and were found to have had all previously required doses as well.
Tier.net is an electronic database that captures patient-level information regarding ART services rendered, including visit dates, viral load results, and drugs administered. As with the DHIS, the original source of this information is paper-based. Tier.net was implemented in 2014. At that time, older data from a prior electronic system (Therapy-Edge) were migrated into Tier.net in the clinics under study. Thus we were able to obtain monthly Tier.net data for the period of 2004 to 2016 from the study clinics.
We also obtained annual staffing data from the municipal regional authority and the implementing partner that ran the CCMT at Clinic 1. Staff workloads per annum were estimated by dividing the number of clinical staff members (i.e. doctors, nurses, dieticians, etc.) by the patient headcounts derived from the DHIS.
We obtained monthly public sector laboratory cost data from the National Health Laboratory Service (NHLS), the sole provider of laboratory services for the public health sector. These data are presented as NHLS charges to the state expressed in 2016 ZAR. Using charges is not ideal if they do not reflect actual costs or expenditure. However, as the analysis is meant to reflect the perspective of the health care system, i.e. the clinics where care was provided, then charges do in fact reflect the resource requirements for providing care.
Finally, we were unable to obtain pharmacy expenditure data (i.e. medication costs) for this analysis. Also, while we obtained monthly clinic-level expenditure data from South Africa’s Basic Accounting System for 2009 onward, these data were found to be “lumpy” in nature—i.e. representing bulk expenditure at discreet time points, rather than real time expenditure—and so were deemed inappropriate for this analysis.
Data management and analysis
All data were first manipulated and reviewed for inconsistencies in Excel (2013, Microsoft Corporation). This involved producing simple graphic representations of the data to view changes over time. Discrepancies, if found, were discussed with the source of the data and corrected.
We then compiled final datasets in Excel and exported the data to Stata (StataCorp. 2015. Stata Statistical Software: Release 14. College Station, TX: StataCorp LP). Using recommended methods for Interrupted Time-Series Analysis (ITSA) [14] and visual analysis [15], we compared resource data at all three study clinics across four comparison periods: (1) before establishment of the CCMT (2004–2007), (2) CCMT operational (2007–2012), (3) CCMT close-out occurring (2012–2014), (4) CCMT closed (2014–2016). We used ITSA where monthly data were available, allowing for the required minimum of 10 observations per period [11]. Where annual data were available, we used visual inspection only due to the limited number of observations.
In an ITSA, outcome variables are compared over multiple time periods before and after the introduction of one or more interventions that are expected to interrupt the variable’s level and/or trend [14]. An ITSA model assumes linear trends over time, and cannot be used to determine whether fluctuations in the data represent pre-emptive reactions to forthcoming interventions or delayed responses to past interventions. These problems have been mitigated in this analysis, following recommended practice [15], by conducting the ITSA in parallel with visual inspection of the data. Visual analysis should involve examination of within- and between-phase data patterns. Within-phase analysis involves an examination of levels, trends, and variability. Between-phase analysis involves the consideration of immediacy of effect, overlap, and consistency of effect [15].
For this analysis, the ITSA assumed the following functional form:
$$\begin{aligned} {\text{Y}}_{\text{t}} & = \upbeta_{0} + \upbeta_{1} {\text{T}}_{\text{t}} + \upbeta_{2} {\text{X}}_{2007} + \upbeta_{3} {\text{X}}_{2007} {\text{T}}_{2007} \hfill \\ & \quad + \upbeta_{4} {\text{X}}_{2012} + \upbeta_{5} {\text{X}}_{2012} {\text{T}}_{2012} + \upbeta_{6} {\text{X}}_{2014} \hfill \\ & \quad + \upbeta_{7} {\text{X}}_{2014} {\text{T}}_{2014} + \upepsilon_{\text{t}} \hfill \\ \end{aligned}$$
where Yt is the aggregate outcome variable measured at each equally spaced time point t, Tt is the time since the start of the study, Xt are dummy variables representing the intervention, and XtTt are interaction terms. β0 is the intercept, or starting level of the outcome variable at the beginning of period 1 (i.e. pre CCMT), and β1 is the slope of the outcome variable throughout period 1. β2, β4, and β6 represent immediate treatment effects [14]. These represent comparisons of the level of the outcome at the start of each period. For example, β2 represents a comparison of the intercept (i.e. the start of period 1) to the outcome at the start of period 2 when the CCMT became operational. Likewise, β4 represents the change in the outcome comparing the start of period 2 (i.e. CCMT operational) to the start of period 3 (i.e. CCMT close-out), and β6 represents the change in the outcome comparing the start of period 3 (i.e. CCMT close-out) to the start of period 4 (i.e. CCMT close-out completed). β3, β5, and β7 represent treatment effects over time [14]. Specifically, β3 is the change in slope from period 1 to period 2; β5 represents the change in the slope between period 2 and period 3; and β7 represents the change in slope between periods 3 and 4. ℇt represents an error term.
The study received prior approval from the Human Research Ethics Committee at the University of the Witwatersrand and from the City of Johannesburg, the municipal regional authority, and the study facilities.