Impact and Cost-Effectiveness of Point-Of-Care CD4 Testing on the HIV Epidemic in South Africa

Rapid diagnostic tools have been shown to improve linkage of patients to care. In the context of infectious diseases, assessing the impact and cost-effectiveness of such tools at the population level, accounting for both direct and indirect effects, is key to informing adoption of these tools. Point-of-care (POC) CD4 testing has been shown to be highly effective in increasing the proportion of HIV positive patients who initiate ART. We assess the impact and cost-effectiveness of introducing POC CD4 testing at the population level in South Africa in a range of care contexts, using a dynamic compartmental model of HIV transmission, calibrated to the South African HIV epidemic. We performed a meta-analysis to quantify the differences between POC and laboratory CD4 testing on the proportion linking to care following CD4 testing. Cumulative infections averted and incremental cost-effectiveness ratios (ICERs) were estimated over one and three years. We estimated that POC CD4 testing introduced in the current South African care context can prevent 1.7% (95% CI: 0.4% - 4.3%) of new HIV infections over 1 year. In that context, POC CD4 testing was cost-effective 99.8% of the time after 1 year with a median estimated ICER of US$4,468/DALY averted. In healthcare contexts with expanded HIV testing and improved retention in care, POC CD4 testing only became cost-effective after 3 years. The results were similar when, in addition, ART was offered irrespective of CD4 count, and CD4 testing was used for clinical assessment. Our findings suggest that even if ART is expanded to all HIV positive individuals and HIV testing efforts are increased in the near future, POC CD4 testing is a cost-effective tool, even within a short time horizon. Our study also illustrates the importance of evaluating the potential impact of such diagnostic technologies at the population level, so that indirect benefits and costs can be incorporated into estimations of cost-effectiveness.

Details of the calibration process are given in S1.5 of S1 Supporting Information. Of the 2 million parameter sets simulated, we obtained 71 acceptable fits according to our calibration procedure, whereby prevalence was constrained to lie within twice the UNAIDS prevalence estimates confidence intervals and 2 million were required to be on ART by mid-2012. HIV prevalence and incidence curves for all simulations are given in Fig   Year Incidence (%) The logistic regression showed that the key parameters in simulating the epidemic were: epidemic start date, proportion in higher and middle risk groups, higher risk group partner change rate, basic infectivity, assortativity and second line ART drop out (p < 0.001 in all cases except second line ART dropout for which p = 0.048). Due to the borderline significance for the second line drop-out rate, this parameter was varied in the cost-effectiveness analysis, along with all the remaining parameters (see S1 Supporting Information, Table A) not shown to be instrumental in determining whether the epidemic curve plausibly represented South Africa.

S2.1.1 Results after calibrating more closely to UNAIDS estimates
The prescription of fitting within twice the UNAIDS estimates was used because the confidence intervals are extremely narrow before 1998 and fitting within only 1× confidence intervals over the whole period lead to only 4 parameter sets satisfying the calibration criteria. Even if this criterion is relaxed such that fits are only required to be within the 1× confidence intervals after 1998 (and are required to be within 2× confidence intervals before this) still only 15 parameter sets were accepted, see Fig D. To test whether only including those fits that did pass closer to the UNAIDS mean prevalence values would affect the outcomes, we removed all fits that passed outside of the 1x confidence intervals after 1998. This corresponded to including 4618 (26%) of the simulations originally performed. Doing so we recalculated the probability of the Prevalence and incidence of calibrated parameter sets. Only shown are the best-fitting simulations, defined as those that remain within the UNAIDS prevalence estimates' confidence intervals after 1998 (shown as circles on the left panel); simulations may be within twice the confidence intervals up to 1998.

S2.2 Sensitivity analysis
The standardised regression coefficients and proportion of variance explained by the varied parameters for the 1 and 3 year projections are shown in Table F and Table G respectively. Using non-standardised regression coefficients (to keep parameters in their original units) linear relationships between infections averted (IA) and ICER are listed (from the linear regressions); only those variables that contribute greater than 10% to the R 2 are included in the relationship. In general this leads to including treatment effectiveness and probability of getting CD4 results in the linear approximation to the IA and probability of getting CD4 results, cost of ART (including drug cost and delivery cost with logistic markup) and (in some cases) cost of CD4 test in the linear approximation to the ICER.  Cost-effectiveness acceptability curves. These curves show the probability that introduction of POC CD4 testing compared to laboratory CD4 testing is cost-effective at a range of decision rule thresholds for the 1 year projection (left) and 3 year projection (right) for the subset of fits that fall within the 1x UNAIDS confidence intervals after 1998. The colours correspond to (left to right within each plot): grey -current care (CC) context, orange -enhanced counselling and testing (ECT), blue -universal test and treat (UTT Var. Sl. is the standardised regression coefficient (slope) for the particular parameter; Var. is the % variation due to that parameter (adjusted R 2 Var. Sl.

Median return to
Var. Sl.

Var.
Median return to ART time Sl. is the standardised regression coefficient (slope) for the particular parameter; Var. is the % variation due to that parameter (adjusted R 2 ).