The value of confirmatory testing in early infant HIV diagnosis programmes in South Africa: A cost-effectiveness analysis

Background: The specificity of nucleic acid amplification tests (NAATs) used for early infant diagnosis (EID) of HIV infection is <100%, leading some HIV-uninfected infants to be incorrectly identified as HIV-infected. The World Health Organization recommends that infants undergo a second NAAT to confirm any positive test result, but implementation is limited. Our objective was to determine the impact and cost-effectiveness of confirmatory HIV testing for EID programmes in South Africa. Method and findings: Using the Cost-effectiveness of Preventing AIDS Complications (CEPAC)–Pediatric model, we simulated EID testing at age 6 weeks for HIV-exposed infants without and with confirmatory testing. We assumed a NAAT cost of US$25, NAAT specificity of 99.6%, NAAT sensitivity of 100% for infants infected in pregnancy or at least 4 weeks prior to testing, and a mother-to-child transmission (MTCT) rate at 12 months of 4.9%; we simulated guideline-concordant rates of testing uptake, result return, and antiretroviral therapy (ART) initiation (100%). After diagnosis, infants were linked to and retained in care for 10 years (false-positive) or lifelong (true-positive). All parameters were varied widely in sensitivity analyses. Outcomes included number of infants with false-positive diagnoses linked to ART per 1,000 ART initiations, life expectancy (LE, in years) and per-person lifetime HIV-related healthcare costs. Both without and with confirmatory testing, LE was 26.2 years for HIV-infected infants and 61.4 years for all HIV-exposed infants; clinical outcomes for truly infected infants did not differ by strategy. Without confirmatory testing, 128/1,000 ART initiations were false-positive diagnoses; with confirmatory testing, 1/1,000 ART initiations were false-positive diagnoses. Because confirmatory testing averted costly HIV care and ART in truly HIV-uninfected infants, it was cost-saving: total cost US$1,790/infant tested, compared to US$1,830/infant tested without confirmatory testing. Confirmatory testing remained cost-saving unless NAAT cost exceeded US$400 or the HIV-uninfected status of infants incorrectly identified as infected was ascertained and ART stopped within 3 months of starting. Limitations include uncertainty in the data used in the model, which we examined with sensitivity and uncertainty analyses. We also excluded clinical harms to HIV-uninfected infants incorrectly treated with ART after false-positive diagnosis (e.g., medication toxicities); including these outcomes would further increase the value of confirmatory testing. Conclusions: Without confirmatory testing, in settings with MTCT rates similar to that of South Africa, more than 10% of infants who initiate ART may reflect false-positive diagnoses. Confirmatory testing prevents inappropriate HIV diagnosis, is cost-saving, and should be adopted in all EID programmes. We used the Cost-Effectiveness of Preventing AIDS Complications (CEPAC)–Pediatric model to simulate a cohort of HIV-exposed infants in South Africa undergoing 2 EID algorithms: 6-week EID testing without confirmatory testing and 6-week EID testing with confirmatory testing. The CEPAC–Pediatric model is a first-order Monte Carlo simulation model of paediatric HIV infection, disease progression, diagnosis, and treatment [19–21]. For this analysis, we simulated HIV-exposed infants (born to women living with HIV) from birth through death. Risk of intrauterine or intrapartum HIV infection was modelled as a 1-time risk, based on 3 key maternal characteristics: the probability a mother was aware of her HIV diagnosis during pregnancy; the probability that she received ART during pregnancy, reflecting prevention of MTCT (PMTCT) coverage; and maternal CD4 count for women not receiving ART, reflecting disease stage. Uninfected infants faced a monthly risk of postpartum transmission based on these same characteristics until complete cessation of breastfeeding. All simulated patients faced monthly risks of non-HIV-related mortality. After HIV infection occurred, patients faced additional risks of opportunistic infections (OIs) and HIV-related mortality based on their age, CD4 percent (age < 5 years) or CD4 count (age ≥ 5 years), retention in care, and ART use. Full details of the CEPAC–Pediatric model structure are available in S1 Text and S2 Table and at http://web2.research.partners.org/cepac/model.html. This work was approved by the Partners Human Research Committee, Boston, MA, US. In the base-case analysis, we simulated HIV-exposed South African children presenting to care at 6 weeks of age. EID in South Africa is currently directed at infants with known HIV exposure [14]; we therefore included only infants born to women identified as living with HIV during pregnancy. South African guidelines currently recommend ‘Option B+’, or lifelong ART for all pregnant/breastfeeding women identified as living with HIV [14]. We assumed that 90% of women had access to ART during pregnancy and breastfeeding [1]. Based on early data after the release of new infant feeding guidelines in South Africa, we assumed 80% of the cohort was breastfed, for a mean duration of 12 months [6,14,22]. The model records true infection status for all infants, as well as the results of each administered assay, allowing the direct reporting of true-positive, true-negative, false-positive, and false-negative diagnoses. The primary model outcomes were the number of infants with false-positive diagnoses linked to care as a proportion of total ART initiations, PPV (defined as the proportion of positive test results due to truly HIV-infected infants), life expectancy (LE, in years), and average per-person lifetime HIV-related healthcare costs, from the perspective of the healthcare provider. We projected survival, LE, and costs separately for HIV-infected infants and for the complete birth cohort of HIV-exposed infants, which included both HIV-infected and HIV-uninfected infants. Costs are presented in 2013 US dollars; costs and life expectancies were modelled both undiscounted and discounted at a rate of 3%/year [23]. We first calculated per-person outcomes, then translated these to population outcomes for all 350,000 HIV-exposed infants born in South Africa in 1 year [24,25]. Where clinical outcomes were equal for both strategies, calculation of incremental cost-effectiveness was not necessary; we considered alternative strategies as either cost-saving or more costly in these cases. The strategy ‘without confirmatory testing’ simulated all HIV-exposed infants receiving a NAAT at 6 weeks of age; in this strategy, all infants who received positive results initiated ART upon result return (mean turnaround time: 1 month). The strategy ‘with confirmatory testing’ simulated all HIV-exposed infants receiving a NAAT at 6 weeks of age; all infants who received positive results on the first test initiated ART at result return, but a second blood sample was drawn before ART initiation and sent for a confirmatory NAAT [6,14]. We assumed conditional independence of the primary and confirmatory NAAT results, because WHO recommends that a second specimen be used for confirmatory testing, and most false-positive NAAT results are likely consequences of specimen handling error rather than biological phenomena [6,26,27]. Infants with negative confirmatory tests underwent a third test per WHO guidelines before HIV infection was ruled out and ART stopped. For infants who were truly uninfected but initiated ART and therefore entered HIV care after a false-positive diagnosis (both strategies), we assigned costs for 10 years of routine HIV care, ART, and laboratory monitoring. We varied this duration widely in sensitivity analysis. For truly infected infants diagnosed and linked to care, the model includes lifetime costs for routine care, ART, and laboratory monitoring, as well as care for OIs. We conservatively excluded clinical harms from incorrect ART initiations, such as medication toxicity and stigma. We modelled cohort characteristics, PMTCT coverage and MTCT risks, disease progression, and ART outcomes using data from published trials and cohort studies in sub-Saharan Africa (Table 1; Table A in S1 Text; S2 Table) [28–32]. The specificity of NAATs was modelled as 99.6%, from a 2015 WHO meta-analysis [33]. We modelled NAAT sensitivity as a function of time since infection, reaching 100% among infants infected during pregnancy or at least 4 weeks prior to testing (Table 1) [34]. In the base case, laboratory-based NAAT cost was US$25, which included assays, reagents, and human personnel resources associated with specimen processing and result return. To describe the full potential impact of each strategy, in the base-case analysis we modelled guideline-concordant (100%) probabilities of presentation to testing, result return (first NAAT: 1 month), and ART initiation after an HIV diagnosis; we varied these widely in sensitivity and scenario analyses to reflect implementation across a range of settings [1,35–37]. A full description of all model input parameters, as well as ranges for sensitivity analyses and uncertainty analyses, is provided in Tables A and B in S1 Text. aOf the total population of breastfed infants (80% in the base case), for the first 6 months of life: exclusive breastfeeding in 55%; mixed breastfeeding in 25%; replacement feeding from birth in 20%. After 6 months of age, all infants still breastfeeding are assumed to have complementary feeding (breastmilk and other liquids/solids). bLopinavir, ritonavir, abacavir, and lamivudine. cEfavirenz plus zidovudine and lamivudine. ART, antiretroviral therapy; CEPAC, Cost-effectiveness of Preventing AIDS Complications; EID, early infant diagnosis; IP, intrapartum; IU, intrauterine; NAAT, nucleic acid amplification test; OI, opportunistic infection; PMTCT, prevention of mother-to-child transmission; POC, point-of-care; PP, postpartum; SD, standard deviation. We incorporated published South African healthcare costs for HIV-infected children less than 5 years of age. Costs associated with OI care for children 5 years of age and older were calculated from South African adult resource utilisation (outpatient visits, inpatient days, and laboratory testing) multiplied by South African unit costs [38,39]. Routine HIV care costs (all ages) ranged by age from US$20 to US$165 per month [40,41]. First-line ART costs were from Clinton Health Access Initiative price lists and WHO weight-based dosing recommendations, ranging by age and weight from US$7 to US$40 per month (Table 1) [42,43]. Following ISPOR–SMDM Good Research Practices Task Force (S1 Table) guidelines on uncertainty in model-based analyses, we conducted extensive univariate and multivariate sensitivity analyses (Tables B and D in S1 Text) [71]. We first conducted univariate sensitivity analyses, in which we varied NAAT specificity and sensitivity, infant HIV prevalence via PMTCT coverage and MTCT risks, the probability of initiating ART after a positive NAAT result, the amount of time infants with a false-positive diagnosis spent on ART, the costs of NAATs, routine HIV care and OI care costs, and ART treatment costs (S1 Table; Task Force recommendation VI-9). Holding all other parameters at their base-case values, we identified the threshold value for each parameter at which the comparison between with and without confirmatory testing would change (S1 Table; Task Force recommendation VI-12). We next performed multivariate sensitivity analyses to evaluate the impact of simultaneous variation in multiple parameters (S1 Table; Task Force recommendation VI-10). We also examined the assumption of conditional independence of the first and confirmatory assays by varying their specificities simultaneously, and assigned a combination of assay cost, sensitivity, specificity, result return time, and result return probability to reflect emerging point-of-care (POC) EID assays (Table 1). We conducted 3 scenario analyses to examine important issues in EID implementation (S1 Table; Task Force recommendation VI-10). First, we simulated lower rates of implementation of 6-week EID testing incorporating input parameters from current testing and treatment cascades in resource-limited countries. We modelled (a) probability of presenting to testing as 73% and probability of ART initiation as 71% of infants with a positive test result and (b) probability of presenting to testing as 95% (2016 UNAIDS estimates for EID coverage in South Africa), probability of result return to caregiver as 80%, and probability of ART initiation as 71% [1,35–37]. Second, we examined programmes offering routine EID testing at both birth and 6 weeks of age to all HIV-exposed infants. Birth tests were modelled as standard laboratory-based NAAT tests with a 1-month result return time (Table 1). Third, WHO strongly recommends prompt ART initiation after a first positive test result (as in our base-case analysis) to avoid delays during a period of high mortality without treatment, but that NAAT testing be repeated on a second specimen drawn prior to ART initiation, to confirm the initial diagnosis [6,7]. Because some providers may be reluctant to initiate ART before receiving a confirmatory result, we also examined the impact of postponing ART until return of the confirmatory test result (1–2 months). The univariate sensitivity analyses described above reveal the sensitivity of policy conclusions to variations in key parameters through a wide range of plausible values. To reflect the uncertainty in the primary data estimates, we also varied key parameters through reported 95% confidence intervals (or range or interquartile ranges, if 95% confidence intervals were not available; S1 Table; Task Force recommendations VI-6 and VI-7). For most parameters, this interval fell well within the range examined in sensitivity analyses. We used the results of previously reported model calibration and validation analyses [19,20] to examine the impact of parameter and model structural uncertainty on the comparison between without and with confirmatory testing (S1 Table; Task Force recommendation VI-14).