The impact of provider-initiated (opt-out) HIV testing and counseling of patients with sexually transmitted infection in Cape Town, South Africa: a controlled trial

The study was performed in the public sector, primary healthcare services in Cape Town, South Africa. STI services are free of charge and are delivered mostly by trained nurse practitioners. VCT is the standard approach and is provided by trained lay health counselors. At the start of the intervention in 2006, the total new STI caseload for Cape Town was more than 5,000 patients per month, with an estimated 30% HIV testing rate. Although Cape Town has amongst the lowest HIV prevalence in the country, the prevalence varies dramatically between sub-districts. In 2005, the average HIV prevalence for pregnant women was 12.7% [29], whilst some of the poorest sub-districts had rates of over 30% which are amongst the highest in the country [30].

The health authority gave permission for the evaluation, and ethical approval was obtained from the University of Cape Town Research Ethics Committee. A confidentiality agreement was signed between the health authorities and the research team to ensure additional protection of patient information.

The design was a pragmatic cluster non-randomised controlled trial, in which seven clinics were selected to receive the intervention, and 14 clinics served as control clinics. The trial aimed to test the effect of the intervention under normal operational and management conditions. Randomization was not feasible because the intervention clinics were selected before the evaluation was planned. The assessment of the intervention was facilitated by the health management team, which also made it impractical to blind the implementers and the assessors.

The primary outcome was the HIV testing rate amongst new STI clients. Secondary outcomes were the proportion of STI clients offered HIV testing (irrespective of whether they accepted or not) and the proportion that declined HIV testing, once offered. We also describe the sample in terms of the participant's gender, age and proportion diagnosed with HIV.

A pool of 24 clinics was identified by the health services as eligible to participate, based on criteria of geographical representation (one clinic from each of the health sub-districts), nurse-patient ratio, and a minimum STI caseload of 30 new STI patients per month. A project steering committee, in consultation with local district managers, selected one representative clinic from each district. Nine intervention sites were selected initially, but two clinics declined participation shortly before implementation, citing operational difficulties. One control clinic opted out, citing similar reasons. The remaining fourteen eligible clinics became the comparison group. There was no matching of clinics, but statistical comparison of clinics at baseline was conducted using general routine administrative data that were collected retrospectively.

A sample size validation was done, given that the intervention arm would consist of seven intervention clinics. With this restriction, the study would be able to show an increase of 20% in testing rate from an estimated baseline of 30%, with 80% power, at a 5% significance level and using an intra-class correlation coefficient (ICC) of 0.08 and a cluster size of 90. In a recent review of ICCs in 188 health systems research studies, the median ICC used was 0.051 (IQR 0.011 to 0.094). The ICC of 0.08 in this study is closer to the conservative end of this range [31]. We doubled the control group to 14 clinics to increase the power of the study. Based on available data, we assumed a cluster size of 90 new STI patients per quarter per clinic.

This intervention is an adapted version of the 'ACTS' approach which includes four brief steps: assess, get consent, test, and provide supportive services [32]. In this PITC intervention, the STI nurse offered HIV testing as a standard part of STI care for all STI clients, and the client had to decline or 'opt-out' of this testing. According to policy in South Africa, written consent was required (although the WHO guidelines for PITC allow for only verbal consent). Abbreviated pre-test counseling consisted of informing patients that HIV is an STI and recommending that they test for HIV at this consultation. If they agreed, the nurse would do a brief test readiness assessment, obtain written informed consent, and perform the rapid test along with other routine blood tests such as those for syphilis.

By using clinical staff to deliver the intervention as an integrated part of the STI consultation, the intervention departed significantly from the standard approach in South Africa where VCT is predominantly provided by lay counselors. The shift required a reconfiguration of roles for nurses and lay counselors, with nurses taking on some of the tasks previously performed by lay counselors. It was anticipated that more patients would be tested because the clinical care setting provided the opportunity to offer HIV testing to all patients, and the setting may also increase patient willingness to test. Because this was a demonstration project to evaluate also the feasibility and acceptability of using clinical staff for HIV screening (instead of lay workers), the intervention needed to address concerns about the possible overloading of nurses. To save nurse time, nurses were allowed to refer the patient to the clinic lay counselor for the test result and post-test counseling once they completed the STI consult and HIV testing.

STI nurses from each of the intervention sites received a two-day training course on PITC and how to implement it. Lay counselors also received training to emphasize follow-up support for HIV positive clients. The implementation was standardized in terms of the main components of integrated nurse-initiated testing, but clinics were allowed to adapt their patient flow arrangements to suit their clinic. No additional nursing resources were provided to implement the intervention at clinic level, but a part-time project manager from within the health service allocated 30% of her time to oversee the implementation, supervision, and monitoring of the intervention. Intervention clinics received quarterly feedback about their performance.

In control sites, the routine VCT approach continued. The VCT service is available at the same site as the STI service, and any person can access the VCT service on his or her own initiative or via medical referral. The majority of patients using the VCT service would be required to initiate testing themselves, whilst a few would be medically referred if they had HIV-related symptoms or if considered high risk. In addition to patient barriers to access such as low perceived risk and poor perception of testing services, the lengthy pre-and post-test counseling and shortages of counselors are potential barriers to VCT access in these settings [2‐5].

In the VCT service, lay counselors provided pre- and post-test counseling that lasted an average of 45 minutes [33]. This excluded waiting time which can be considerable in public sector clinics. Due to limits in the scope of practice of lay workers, a nurse had to perform the rapid and confirmatory tests, which added to the duration of VCT and also fragmented the VCT service. In this study, the pre-test counseling session offered by the lay counselor in the VCT approach and the abbreviated pre-test counseling offered by the nurse in the PITC intervention are both captured by the outcome 'offered HIV testing'.

The proportion of patients offered testing, tested, and declined per clinic was calculated. For the variables 'HIV tested', 'not tested' and 'offered HIV testing' the denominator was the total number of new STI patients treated in these clinics. For variable 'declined HIV testing' the denominator was the total number of new STI patients who were offered HIV testing, in other words, a subgroup of the total new STI population. These proportions were utilized as clinic level outcomes and used to compare the two groups. This clinic level analysis accounts for the clustering within clinics [34].

The statistical comparison was done using STATA 10. A two-sample t-test was used for the comparison of the study outcome and baseline data between the intervention and control group. We checked the normality assumption of the t-test. A formal test for the equality of variances in the two groups was done using the F-test. The appropriate t-test was subsequently performed based on the homogeneity status of the variances. Multinomial logistic regression analysis with adjustment for clustering was done on the composite testing outcome ('tested', 'not tested 'and 'not offered') to confirm the results of the clinic level analysis. Only the clinics that were operationally able to implement the protocol and provide outcome data to the trial were included in the analysis because we wanted to know the effectiveness of the PITC intervention in clinics that actually implemented the intervention. The statistical adjustment for baseline differences is not reported and this is considered in the results and in the study limitations.

Baseline data on clinic demographics and service delivery from the start of the intervention in 2005 are provided in Table 1. There were no significant differences between the intervention and control groups for either their STI and HIV testing service delivery characteristics or their TB treatment outcomes, except for the HIV test acceptance, VCT variable (p = 0.03).

Clinic level raw data and results are shown in Table 2, with the proportions for the main outcomes averaged over clinics. A total of 9,080 new STI patients were treated during the study period; 3,053 in the intervention and 6,027 in the control group. The main results are shown in Figure 2 and the corresponding Table 3. The two bars in Figure 2 represent 100% of the new STI patients seen in the period of January to June 2007 in both intervention and control clinics. A significantly greater proportion of new STI patients were tested for HIV in the intervention compared to the control group (56.4% intervention versus 42.6% control, p = 0.037). The increase of 13.8% in test uptake with the new intervention means that for every seven new STI patients treated, one additional HIV test was achieved (NNT = 7.3).

Patients were also more likely to be offered HIV testing in the intervention clinics: intervention clinics offered the HIV test to 2,326 (76.8%) of new STI patients compared with 3,406 (50.7%) in the control group (p = 0.0029). In Figure 2, the proportion of patients offered testing is made up of a combination of the blue ('HIV tested') and red ('not tested') sections of the graph. The 'not tested' variable represents the patients in the CT register who were offered testing but who declined this. The proportion of patients who declined testing ('not tested') in the figure is calculated as a proportion of all new STI patients.

Another way to measure those who declined testing is as a proportion of those offered testing. This alternative calculation represents test refusal within the context of those offered the test rather than all new STI patients and is reflected by the variable 'declined HIV testing' in Figure 1, Table 2 and Table 3. Of those patients offered HIV testing, a significantly greater proportion (26.7%) declined testing in the intervention group than in the control group (13.5%), p = 0.0086.

As mentioned earlier, the adjusted baseline analysis is not used. The inverse association observed between the baseline variable of a broader population and the much narrower study population may lead to a distorted estimate of the intervention effect. The unadjusted results reported in the manuscript is a more conservative estimate of the intervention effect when compared to the adjusted estimate given in this analysis (13.8% compared to 22.2%). We feel that the unadjusted result is a more reasonable result with respect to the health system setting in which this was done.

From the multinomial regression (adjusted for clustering) on the composite testing outcome (tested, not tested, not offered) the odds ratio for being tested is 2.24 (95% CI: 1.12 to 4.46), p = 0.022 for the intervention group compared to the control group with the reference category being 'not offered'. This result confirms the outcome specific comparison reported in Figure 2 and Table 3.

Although not a pre-specified outcome measure, the data suggested wide variability in range for outcomes measured in the clinics, and this was compared statistically using the F test. As shown in Table 3, there was significantly less variation in clinic performance in the intervention group for both 'HIV tested' (intervention: 39% to 71%; control: 8% to 81%; p = 0.033) and 'HIV test offered' rates (intervention: 63% to 84% control: 7.8% to 92%; p = 0.0076). Of note is that two control clinics (clinic number 13 and 21 in Table 2) achieved higher testing rates than the intervention sites (80.6% and 77.5%, respectively, as compared to the highest rates of 71,1% in the interventions clinics).

There were no significant differences in the male to female ratio between the intervention and control groups on any of the outcomes examined (Table 4). Of those who tested, a similar proportion was male (43%) in both groups. There was no significant difference between intervention and control clinics in the proportion of patients who tested HIV positive (18,6% in the intervention and 21,4% in the control group, p = 0.147).

As a pragmatic trial, this intervention was implemented using existing management and clinical staff resources and some internal project management support. This situation, where the service managers initiate and run the intervention and where there is little additional resources added, is an ideal context for an effectiveness evaluation as it reflects the reality of implementation in resource-constrained settings [38, 39]. The estimated baseline testing rate for STIs of 30% in this study may be higher than in other low resource settings, and the staff configurations for HIV service delivery may also differ. These factors could influence the size of the impact of the intervention in other settings.

The lack of randomization of clinics to intervention and control groups introduces the risk of bias for the study outcomes. While the baseline comparison showed few differences between intervention and control sites, we cannot exclude other unknown sources of selection bias. Also, we were not able to obtain baseline measures for the main outcome. The one baseline testing variable (HIV test acceptance, VCT) that was significantly different across intervention and control sites was inappropriate as a measure of baseline differences for the following reasons: this variable is not strictly comparable to the main outcome in this study because it refers to the proportion of general clinic patients who were tested as a proportion of those who received pre-test counseling by lay counselors in the VCT service; and it covers a different and much larger general patient target group (and not STI patients only) and refers to a different testing service (VCT), delivered by lay counselors and not clinicians. However, given the lack of baseline data on the main outcome, we felt that this 'HIV test acceptance, VCT' variable should be examined in the baseline comparison of the two groups, and that unadjusted outcome measures should be presented. In the absence of blinding, factors such as staff enthusiasm and increased monitoring could have had a modifying effect. As mentioned earlier, we cannot be certain about how the unforeseen management efforts to increase HIV testing rates, implemented in parallel to this intervention, may have affected the study results.