The ART service based at the Gugulethu Community Health Centre, in Cape Town, South Africa has previously been described 202122. The district it serves has a predominantly African population of over 300,000, the vast majority of whom live in conditions of low socioeconomic status. In 2003 the antenatal HIV seroprevalence was 28%. Patients are referred from primary care HIV clinics in the community to the ART programme. Treatment criteria are based on the WHO 2002 recommendations 23, which include those with a prior AIDS diagnosis (WHO stage 4 disease) or a blood CD4 cell count <200 cells/μl.

The mean time between enrolment of a patient in the service and initiation of ART is approximately one month, permitting thorough evaluation of patients (including screening for tuberculosis) and preparation for treatment as described previously 7112021. Blood samples are routinely drawn on all patients at a standard clinic visit during this period to measure CD4 cell counts by flow cytometry (FACSCount™, Becton Dickinson, Franklin Lakes, NJ, USA) and plasma viral load using Versant™ HIV-1 RNA 3.0 branched chain DNA assay (Bayer HealthCare, Leverkusen, Germany). Excess EDTA plasma is stored at ≤ -80°C.

First-line ART, comprised of stavudine, lamivudine plus a non-nucleoside reverse transcriptase inhibitor (efavirenz or nevirapine), was supplied free of charge. Treatment adherence and viral load suppression <400 copies/ml in this cohort both exceed 90% at all 16 weekly follow-up time-points 20. All patients with CD4 counts <200 cells/μl received prophylaxis with daily cotrimoxazole or dapsone as an alternative but isoniazid prophylaxis was not used. In addition to scheduled clinic appointments at 4, 8, and 16 weeks and 16-weekly thereafter, patients had open access to the clinic for medical problems.

Assignment of each patient to a community-based therapeutic counsellor promotes high levels of treatment compliance and also high levels of data-completeness so that final outcomes are known for the vast majority of patients. A proportion of patients did not initiate treatment for a variety of reasons other than death and such individuals were deferred from the programme and follow-up was censored at the time they exited the programme. Principal causes of death in this cohort have been previously been described in detail and include wasting syndrome, acute sepsis, tuberculosis, malignancy and immune reconstitution disease (mainly associated with cryptococcal meningitis) 71924.

Structured clinical records were maintained on all patients screened on entry to the ART programme and this information was transferred on a weekly basis to a computer database. All patients enrolling in studies at this ART clinic provide written informed consent and this study was approved by the Research Ethics Committee of the University of Cape Town.

Total plasma suPAR concentrations were measured at a single time-point in plasma samples that had been stored at the time of programme enrolment. In order to assess this prognostic marker a patient population with high mortality risk, we selected all patients who enrolled into the programme between September 2002 and February 2005 and who had a CD4 cell count ≤100 cells/μl.

Plasma suPAR concentrations were measured using a commercially available enzyme-linked immunosorbent assay (ELISA) (suPARnostic,™ ViroGates, Lyngby, Denmark) following the manufacturer's instructions. This is a simple double monoclonal antibody sandwich assay that measures total suPAR, including both full-length and cleaved forms of the receptor. In brief, a standard control curve (range 0.6 – 19.3 ng/ml), positive control, and test samples were incubated in duplicates in a 96-well plate pre-coated with anti-suPAR antibody. Following further incubation with a secondary peroxidase-conjugated antibody, the assay was developed by addition of a tetramethylbenzidine (TMB) chromogenic substrate. The reaction was terminated by addition of sulphuric acid and absorbance at 450 nm was determined using a microtitre plate reader. The linear standard curve was used to determine concentrations in positive control and test samples. Samples with concentrations exceeding the highest standard (19.3 ng/ml) were reanalysed using a further 5-fold sample dilution.

Outcomes for all patients were determined from the time of entry to the programme until completion of 4 months ART or exit from the programme due to death or other causes. Data were analysed using STATA Version 9.0 (College Station, Texas, USA) and GraphPad Prism 4.0 (GraphPad Software Inc. San Diego, CA, USA). As the frequency distribution of values was highly right-skewed, the suPaR values were log₁₀-transformed for bivariate analyses (based on the Mann Whitney U or Kruskall Wallis tests to compare medians) and multivariate modelling. We constructed multiple linear regression models to predict suPAR values at programme enrolment, and used Cox's proportional hazards to examine the associations between suPAR concentrations and the relative hazard of mortality. For the latter set of models, time at risk was calculated from entry into the programme to the first of death, transfer out of the programme, loss to-follow-up, or the 4 month follow-up visit on ART. Receiver operating characteristic (ROC) curves were constructed to examine the effect of different cut-points of log₁₀ suPAR values on the sensitivity and specificity of predicting mortality during the study period. Confidence intervals for the area under the ROC curves were calculated using the method described by Bamber 25.

Of patients who enrolled into the ART programme between September 2002 and February 2005, 354 had a blood CD4 cell count ≤100 cells/μl. Of these, a stored plasma sample was available for measurement of suPAR concentration in 293 (83%) patients and these were included in the study. These patients had a median age of 33 years and approximately two-thirds were female (Table 1). 89% had either WHO stage 3 or 4 disease. A total of 43% of patients had either an active diagnosis of TB prior to initiation of ART or developed symptoms during the first 4 months of ART that were due to confirmed incident TB.

Twenty-four (8%) patients did not start ART due to death before starting treatment (n = 13) or other reasons (n = 11), including treatment refusal, loss to follow-up or access to treatment at another clinic. A total of 269 (92%) patients did start ART. Of these, 218 (82%) were retained within the programme and were alive after 16 weeks of treatment. Twenty six (9.7%) patients died within the first 16 weeks of treatment. Other programme losses during ART included 15 (5.6%) patients who moved out of area and 10 (3.7%) who were lost to follow-up. The median (IQR) total period of follow-up was 145 days (139–173) and was comprised of a median pre-ART period of 34 days (28–56) and a median of 112 days (112–119) on ART.

Detectable levels of suPAR were measured in plasma samples from all 293 patients. The standard curves for each run were linear (mean r² = 0.995; SD = 0.004) and all positive control readings were consistent with the expected value. The median suPAR concentration in the patient plasma samples was 4.05 ng/ml (IQR, 3.03–5.77; range 1.14–52.4 ng/ml) or 0.607 log₁₀ ng/ml (IQR = 0.481–0.761).

Associations between baseline patient characteristics and log₁₀ suPAR concentrations were examined (Table 2). In unadjusted analyses, low baseline CD4 cell counts and WHO stage 4 disease were strongly associated with higher log₁₀ plasma concentrations of suPAR. There was also a trend towards patients with active TB diagnoses having higher suPAR concentrations. When the cases of tuberculosis were subdivided into those with active TB at the time of the blood sample and those who developed later tuberculosis during ART, neither group was significantly associated with suPAR concentration. Other baseline characteristics, however, were not significantly associated with suPAR concentrations. In adjusted analyses, the association with CD4 cell count and WHO stage remained, but female sex was also found to be significantly associated with lower suPAR concentrations (Table 2).

To examine the association between log₁₀ serum suPAR concentrations and clinical outcome, patients were divided into three groups according to final outcome: (i) received ART and alive after 4 months treatment (n = 218, 74%); (ii) died during the analysis period (n = 39, 13%); or (iii) other programme losses (n = 36, 12%). Among those who died, 13 died before starting ART and 26 died during the first 4 months of ART.

The median log suPAR concentration among patients who died (0.761 ng/ml, IQR = 0.548–0.949 ng/ml) was significantly greater than that of patients who remained alive on ART (median = 0.583 ng/ml, IQR = 0.479–0.715 ng/ml) or who were lost to the programme for reasons other than death (median = 0.609 ng/ml ng/ml; IQR = 0.429–0.833 ng/ml) (Figure 1). However, concentrations did not differ significantly comparing those who remained alive with non-death losses. Thus, higher suPAR concentrations were strongly associated with mortality. However, median suPAR concentrations were almost identical comparing those who died pre-ART with those who died during the first 4 months of ART (5.92 and 5.93 ng/ml, respectively) and suPAR concentrations were not associated with causes of death (data not shown).

We next examined predictors of mortality using multivariate analysis. In an initial model that included baseline patient characteristics but excluded suPAR concentration, only baseline CD4 cell count was significantly associated with mortality. There was also a strong trend towards an association with WHO clinical stage (Table 3a). However, when the log concentration of plasma suPAR was included in the model, the strength of the associations with CD4 cell count and WHO stage were diminished and log suPAR concentration was found to be the strongest independent predictor of mortality (Table 3).

Having demonstrated that plasma suPAR concentration was strongly associated with mortality at a group level, we next examined the utility of this as a prognostic test to be applied to individual patients. To do this, receiver operating characteristic (ROC) curve analysis was done (Figure 2). The area under the ROC curve was moderate (0.681; 95% CI 0.582–0.781) and yet a clear cut-point that maximised the ability of suPAR concentration to distinguish between patients who died or survived was lacking. Using cut-points of ≥5.0 ng/ml, ≥10.0 ng/ml or ≥15 ng/ml, the corresponding sensitivity and specificity values showed that none of these cut-offs provided sufficient sensitivity and specificity that would be clinically useful in this patient population (Figure 2).