Efficient Monte Carlo Estimation of the Expected Value of Sample Information Using Moment Matching

Introduction The 2006 World Health Organization (WHO) guidelines on antiretroviral therapy (ART) established a worldwide standard of care for patients with HIV infection [1]. Since this publication, new evidence has emerged on how to treat patients infected with HIV, and this evidence formed the basis for the WHO 2010 ART guidelines update [2]. These revisions aim to better align global standards with those already adopted in well-resourced countries [3],[4]. Specifically, revised guidelines recommend an increased number of sequential ART regimens, routinely available CD4 count monitoring, earlier ART initiation thresholds (CD4 100,000 copies/ml 42.5  30,001–100,000 copies/ml 28.3  10,001–30,000 copies/ml 17.9  3,001–10,000 copies/ml 7.8  501–3,000 copies/ml 2.3   30,000 6.4  10,001–30,000 5.4  3,001–10,000 4.6  501–3,000 3.7 Monthly risk of severe opportunistic infections (%)a [20]  Bacterial 0.08–0.71  Fungal 0.02–2.22  Tuberculosis 0.21–1.96  Toxoplasmosis 0.00–0.06  Nontuberculosis mycobacteriosis 0.00–0.30  P. jiroveci pneumonia 0.00–0.12  Other severe opportunistic infections 0.25–2.57 Monthly risk of mild opportunistic diseases (%)a [20]  Fungal 0.59–3.51  Other 2.51–3.10 Efficacy of co-trimoxazole (% reduction in probability of infection) Severe bacteria 49.8 [22],[23] Mild fungal infections −46.4b [22],[23] Toxoplasmosis 83.2 [22],[23] P. jiroveci pneumonia 97.3 [22],[23] Other WHO stage IV defining diseases 17.9 [22] Efficacy of ART (range examined) First line: stavudine-based regimen [19]  HIV RNA suppression 75% at 24 wk  CD4 count increase 136 cells/µl at 48 wk Probability of later failure (monthly, after 24 wk) 0.02c (0.01–0.02) First line: tenofovir-based regimen  HIV RNA suppression 85% at 24 wk (85%–95%) [24]  CD4 count increase 136 cells/µl at 48 wk [19] Probability of later failure (monthly, after 24 wk) 0.01d (0.005–0.01) [24] Second line: lopinavir/ritonavir-based regimen [16]  HIV RNA suppression 78% at 24 wk (40%–88%)  CD4 count increase 151 cells/µl at 48 wk Probability of later failure (monthly, after 48 wk) 0.03e (0.01–0.06) Toxicity (one-time probability [%]) Stavudine-based regimen (range examined)  Severe lactic acidosis 1.7 (1.7–3.4) [36]  Lipodystrophy 1.3 (1.3–2.6) [36]  Neuropathy 2.6 (2.6–5.2) [36] Tenofovir-based regimen (range examined)  Nephrotoxicity 1.6 (1.6–3.2) [37],[38]  Anemia 0.4 (0.4–0.8) [24] Discount rate 3% [12] Costs (2008 US dollars) (range examined) Co-trimoxazole prophylaxis (monthly) 1.03 [23] Stavudine-based first-line ART (monthly) 8.33 [26] Tenofovir-based first-line ART (monthly) 17.00 (10.00–17.00) [26] Lopinavir/ritonavir second-line ART (monthly) 55.75 (8.36–55.75) [26] Routine care (range by CD4, monthly)a 9.99–131.23 [20],[39],[40] Inpatient hospital care, per day 224.25 [39] Outpatient hospital care, per visit 10.87 [39] CD4 count test US$25 (25–75) [27],[28],[41] “Range examined” indicates that we examined both extreme and intermediate values within the specified ranges. a Range indicated by CD4 count; details by CD4 strata are presented in the Text S1. b The percent monthly risk of mild fungal infections is increased by 46.4% in the presence of co-trimoxazole [22]. c Projected using published 24-wk data [19]. d Estimated from published 24- and 48-wk data [24]. e Estimated from published 24- and 48-wk data [16]. Cohort characteristics We define an ART-naïve cohort of patients with HIV in South Africa, with mean age 32.8 y [20]. We intentionally choose an initial cohort with a relatively high mean CD4 cell count of 375 cells/µl (SD, 25 cells/µl). A cohort with a lower mean CD4 cell count would not clearly demonstrate the benefits of an ART initiation threshold of CD4 100,000 copies/ml (Table 1) [21]. In the model, this ART-naïve cohort is then subject to the policies of ART initiation and drug availability as indicated by each of the 12 strategies. In the absence of ART, the model tracks the patients' natural history of disease for use in comparing the incremental clinical benefits and costs. Figure S3 illustrates the internal validation of South African data used to derive critical model input parameters such as monthly mortality and opportunistic disease incidence rates, stratified by CD4 count. Opportunistic disease prophylaxis and ART efficacy All patients at model entry are provided co-trimoxazole prophylaxis, conferring protection against mild and severe bacterial infections, P. jiroveci̧ and toxoplasmosis [22],[23]. We assume a non-nucleoside reverse transcriptase inhibitor–based ART regimen that includes stavudine. This regimen results in a 24-wk virologic suppression rate of 75% with a mean 48-wk CD4 count rise of 136 cells/µl among those with suppression [19]. The monthly probability of virologic failure after 48 wk is 0.02. When stavudine is replaced with tenofovir in first-line regimens, in the absence of reliable efficacy data for a tenofovir-based regimen in resource-limited settings, we use a virologic suppression rate of 85% at 24 wk, as reported in clinical trials [24],[25]. Despite the improved rates of virologic suppression, we want to maintain conservative assumptions with regard to CD4 benefit among those suppressed, so we use the same benefit (136 cells/µl) as that used for the stavudine-based regimen [19]. From these studies, the monthly probability of failure of tenofovir-based ART after 48 wk is 0.01 [24],[25]. When second-line ART is available, it is a lopinavir/ritonavir-based regimen with a 24-wk suppressive efficacy of 78%, a resultant CD4 count increase of 151 cells/µl, and a 0.03 monthly probability of virologic failure after 48 wk [16]. In sensitivity analyses, we examine the impact of improved efficacy of first-line ART associated with the use of tenofovir and the impact of alternative second-line ART efficacies (Text S3). Costs We consider HIV-associated direct medical costs, including inpatient days, outpatient visits, medication costs, and laboratory tests, when available (Table 1). Direct non-medical costs and indirect costs are excluded. Costs attributable to inpatient hospitalization resulting from an opportunistic infection are calculated as the mean cost of each inpatient day multiplied by the mean length of stay for any given opportunistic disease. Outpatient care costs include the mean cost of each visit, inclusive of standard laboratory tests and procedures. Routine care costs are stratified by CD4 cell count to account for the increased frequency of visits that may be attributable to lower CD4 cell counts (Table 1). The stavudine-based first-line regimen costs US$100 per person-year (stavudine component  =  US$36), and the tenofovir-based regimen costs US$204 per person-year (tenofovir component  =  US$135) [26]; all other first-line regimen costs are identical. Second-line ART regimens, when available, cost US$669 per person-year [26]; CD4 count tests cost US$25 each [27],[28]. Tenofovir, second-line ART, and CD4 monitoring costs are each varied in sensitivity analyses. Results Prioritization by Survival Benefits (Undiscounted) An untreated HIV-infected South African cohort starting with a mean CD4 count of 375 cells/µl (SD, 25 cells/µl) has a mean undiscounted life expectancy of 47.9 mo. A single-line stavudine-based ART regimen, initiated on development of WHO stage III/IV disease (“reference strategy”; stavudine/WHO/one-line) increases life expectancy to 99.0 mo. Table 2 provides the projected 5-y survival and life expectancies of alternative stepwise progressions toward the 2010 WHO recommendations. Compared to stavudine/WHO/one-line (step 1), 5-y survival is largest (87% survival) with the addition of CD4 count availability and ART initiation at CD4<350 cells/µl (stavudine/<350/µl/one-line). In this initial step, tenofovir/WHO/one-line (66%), stavudine/<200/µl/one-line (80%), or stavudine/WHO/two-lines (66%) each yield lower projected short-term survival. Considering each of the guideline components, stavudine/<350/µl/one-line also produces the greatest anticipated life expectancy increase, Δ25.3 mo. With stavudine/<350/µl/one-line (step 2), adding a second-line regimen results in the next largest life expectancy increase (stavudine/<350/µl/two-lines, Δ53.3 mo). The final step replaces stavudine with tenofovir (tenofovir/<350/µl/two-lines, Δ16.0 mo, step 3), resulting in a comprehensive strategy concordant with the 2010 WHO guidelines, a 5-y survival of 91%, and a projected life expectancy of 193.6 mo (Table 2). 10.1371/journal.pmed.1000382.t002 Table 2 Projected life expectancies associated with alternative choices in the stepwise progression toward full implementation of the 2010 WHO HIV treatment guidelines. Step 5-y Survival (%) Projected Life Expectancy (Months) Δ Projected Life Expectancy (months)a Step 1: begin with stavudine/WHO/one-line (reference strategy) (four options) 65 99.0 — (1) Switch from stavudine to tenofovir, or 66 112.9 13.9 (2) Add CD4 monitoring capacity, initiate ART at CD4<200 cells/µl, or 80 115.6 16.6 (3) Add second-line ART regimen, or 66 121.4 22.4 (4) Add CD4 monitoring capacity, initiate ART at CD4<350 cells/µl 87 124.3 25.3 Step 2: begin with stavudine/<350/µl/one-line (two options) 87 124.3 — (1) Switch from stavudine to tenofovir, or 89 144.8 20.5 (2) Add second-line ART regimen 91 177.6 53.3 Step 3: begin with stavudine/<350/µl/two-lines (one remaining option) 91 177.6 —  (1) Switch from stavudine to tenofovir 91 193.6 16.0 We use the following nomenclature to define the strategies: nucleoside analog used in first line/ART initiation criterion/number of available regimens. All strategies with initiation criteria indicated by a CD4 count threshold assume availability of CD4 count monitoring. For each step, the option that maximizes survival is shown in bold. a Change relative to the program selected in the previous step. Model-generated survival curves are provided for no ART, the reference strategy, and the three steps in Table 2, which act stepwise to maximize life expectancy (Figure 2). Marked differences in early survival are attributable to earlier ART initiation thresholds; differences in survival later in the disease course are associated with second-line ART availability. 10.1371/journal.pmed.1000382.g002 Figure 2 Model-projected survival curves. Model-projected survival curves (undiscounted) of the reference strategy (stavudine/WHO/one-line) and the three strategies projected to maximize life expectancy in stepwise progression toward the 2010 WHO guidelines (see Results and Table 2 for details). Curves highlighting outcomes over the next 5 y are provided in Figure S4. The 20-y horizon is presented here, not to imply that HIV treatment will remain unchanged over this time horizon, but rather to demonstrate when different interventions will have meaningful survival impacts. Median survival increases from 90 mo with stavudine/WHO/one-line (reference strategy) to 121 mo with the addition of CD4 monitoring and ART initiation at CD4<350 cells/µl (stavudine/<350/µl/one-line, step 1) to 177 mo with the addition of a second-line ART regimen (stavudine/<350/µl/two-lines, step 2). A subsequent switch from stavudine to tenofovir results in a comparatively modest survival advantage, with a median survival increase to 196 mo (tenofovir/<350/µl/two-lines, step 3). The survival curve of step 3 represents what might be expected when allthe 2010 WHO treatment guidelines are fully implemented. Prioritization by Cost-Effectiveness Incremental cost-effectiveness analysis (Table 3) reveals three non-dominated strategies (i.e., strategies that attain a given survival level by the least costly means): (1) stavudine/<350/µl/one-line (US$610/YLS), (2) tenofovir/<350/µl/one-line (US$1,140/YLS), and (3) tenofovir/<350/µl/two-lines (US$2,370/YLS). All other strategies are “dominated”—i.e., they are more expensive and confer less survival benefit than some other combination of strategies. Figure 3 (upper panel) maps the 13 strategies on a discounted cost and life expectancy plane. The line connecting the non-dominated strategies designates the “efficient frontier,” which represents both the least expensive way to attain a given survival and the maximum achievable survival for any given cost [12]. 10.1371/journal.pmed.1000382.g003 Figure 3 Clinical and economic outcomes of each of the scale-up interventions. The clinical and economic outcomes of all combinations of scale-up interventions are examined. The efficient frontier (marked by the line) connects the non-dominated strategies in the cost-effectiveness plane. Strategies below and to the right of the efficient frontier are those that are either strongly or weakly dominated by other options (see Methods). As illustrated in the upper panel, strategies based on clinical criteria (WHO stage III/IV) alone fall far below the efficient frontier (lower right oval), indicating their relatively high cost for the comparative benefit gained. Strategies in the upper left oval are those representing CD4 monitoring and one line of ART. Strategies incorporating a second-line regimen (upper right oval) all confer large survival benefits but at increased costs. The lower panel examines potential country situations. For instance, a country with a current stavudine/WHO/one-line policy could switch to a tenofovir/<350/µl/one-line policy (open arrow) and both decrease projected per-person lifetime costs and improve survival. A country with a stavudine/<200/µl/one-line policy could decrease per-person costs and also improve outcomes by changing to a stavudine/<350/µl/one-line policy (solid arrowhead). Countries with a stavudine/<200/µl/two-lines policy would require increased per-person expenditures to achieve the survival benefits associated with tenofovir/<350/µl/two-lines (dotted arrow). To illustrate the impact of a policy requiring that all persons receive the same intervention, we examine the arbitrary affordability threshold of US$11,500 per person. The bracket (upper right) denotes the per person survival loss (14.5 mo) attributable to a policy requiring that all persons receive the same intervention. 10.1371/journal.pmed.1000382.t003 Table 3 Life expectancy, costs, and incremental cost-effectiveness ratios of the 12 possible stepwise combinations (and no ART) from the reference strategy to full implementation of 2010 WHO HIV treatment guidelines. Strategya Discounted Cost Discounted Per-Person Life Expectancy (Undiscounted) Months Incremental Cost-Effectiveness Ratio (US$/YLS) No ART 2,540 44.9 (47.9) Stavudine/<350/µl/one-line (step 1) 5,550 104.3 (124.3) 610 Stavudine/<200/µl/one-line 5,740 97.3 (115.6) Dominatedb Tenofovir/<350/µl/one-line 6,870 118.3 (144.8) 1,140 Tenofovir/<200/µl/one-line 6,930 109.9 (133.9) Dominatedb Stavudine/WHO/one-line (reference strategy) 7,440 84.5 (99.0) Dominated b Tenofovir/WHO/one-line 8,400 93.9 (112.9) Dominatedb Stavudine/WHO/two-lines 10,140 98.8 (121.4) Dominatedb Tenofovir/WHO/two-lines 10,640 105.0 (131.2) Dominatedb Stavudine/<200/µl/two-lines 11,460 127.0 (161.3) Dominatedc Tenofovir/<200/µl/two-lines 11,930 135.3 (175.5) Dominatedc Stavudine/<350/µl/two-lines (step 2) 12,270 138.7 (177.6) Dominated c Tenofovir/<350/µl/two-lines (step 3) 12,820 148.4 (193.6) 2,370 The reference strategy and the strategies selected in the stepwise progression in Table 2 are shown in bold. a We use the following nomenclature to define the strategies: nucleoside analog used in first line/ART initiation criterion/number of available regimens. All strategies with initiation criteria indicated by a CD4 count threshold assume availability of CD4 count monitoring; WHO indicates WHO stage III/IV disease. b Strongly dominated (more expensive but confer less clinical benefit than some other strategy) [12]. c Weakly dominated (more expensive but confer less clinical benefit than some combination of other strategies) [12]. Thus, a country with a current stavudine/WHO/one-line policy (Figure 3, lower panel) could switch to a tenofovir/<350/µl/one-line policy (open arrow) and thereby simultaneously decrease projected per-person lifetime costs and improve survival. Similarly, a country with a stavudine/<200/µl/one-line policy could decrease per-person costs and also improve outcomes by changing to a stavudine/<350/µl/one-line policy (solid arrowhead). Countries with a stavudine/<200/µl/two-lines policy would require increased per-person expenditures to achieve the survival benefits associated with tenofovir/<350/µl/two-lines, as suggested in the revised WHO guidelines (dotted arrow). Evaluating the Cost of Equity Of the three efficient programs (Table 3; Figure 3), tenofovir/<350/µl/one-line has a projected per-person lifetime cost of US$6,870, and tenofovir/<350/µl/two-lines has a projected lifetime cost of US$12,820. An HIV program budget that allows for a per-person cost between US$6,870 and US$12,820 might be achieved in several ways; two are illustrative. The first would be to proportionately divide the cohort between two of the programs along the efficient frontier, so that part of the cohort receives tenofovir/<350/µl/one-line and the rest receives tenofovir/<350/µl/two-lines. An alternative would be to provide everyone in the cohort a third program—one that lies below the efficient frontier. The opportunity cost (e.g., the anticipated net loss in discounted life expectancy associated with an alternative strategy choice) of any non-efficient strategy may be quantified by measuring its vertical distance from the efficient frontier. To illustrate this opportunity cost, we take an arbitrary affordability threshold of US$11,500 per person. In the example of a program that can afford no more than US$11,500 per person (stavudine/<200/µl/two-lines; Figure 3, lower panel), the opportunity cost of uniformity in care (“equity”) is 14.5 mo per person of survival (shown by the bracket in Figure 3, lower panel). Sensitivity Analyses Clinical parameters In sensitivity analyses, we examine changes in clinical input data required to alter the stepwise ordering of program additions. Modest reductions in the mean CD4 count of the cohort (to 250 cells/µl) show decreased clinical benefits to earlier ART initiation but no substantial changes in cost-effectiveness. When the mean CD4 count of the cohort is less than 100 cells/µl, the benefits of a policy change to earlier ART initiation are largely irrelevant (Text S2). This is because the majority of the cohort is already ART-eligible regardless of the initiation criterion (WHO stage III/IV disease, CD4<200 cells/µl, or CD4<350 cells/µl). Although CD4 monitoring still improves cohort survival compared to clinically based ART initiation, in populations with mean CD4 counts far below the policy-relevant ART initiation criteria, the addition of a second-line regimen becomes the most clinically beneficial intervention. For the anticipated life expectancy benefits of tenofovir/WHO/one-line to exceed those expected with stavudine/<350/µl/one-line, replacement of stavudine with tenofovir would have to increase the 24-wk suppressive efficacy from 85% to 95% and simultaneously decrease the monthly probability of later virologic failure by 50% (from 0.01 to 0.005) (Text S3) [24]. Second-line ART maintains its position in the stepwise order (step 2) as long as its 24-wk viral suppression rate remains between 40% and 88%, even with a 3-fold increase in the rate of late failure when efficacy decreased to 40% (Text S3). Increasing stavudine toxicity by 2-fold alters life expectancy estimates by less than 1 mo and does not change the recommended stepwise additions (Text S3). Similarly, changes in the gender distribution of the cohort have little impact on the results (Text S3). Cost parameters Holding efficacy constant, results are very sensitive to the price of tenofovir; a decrease in the cost of tenofovir from US$135 to US$51 per person per year would make tenofovir both more effective and less costly than stavudine. Results are less sensitive to the costs of second-line regimens (15% of base case) and CD4 monitoring (three times base case), neither of which produced meaningful changes in cost-effectiveness results (Text S2). In two-way sensitivity analyses, where the cost of tenofovir is decreased and its efficacy increased, tenofovir/<350/µl/one-line dominates stavudine/<350/µl/one-line when the tenofovir regimen costs are US$153 annually (75% of the base case) and its 24-wk suppressive efficacy is 90% (5% increase from the base case). Additional sensitivity analyses Further sensitivity analyses are detailed in the Texts S2 and S3. In Text S2, we present the 1- through 5-y survival rates for all 12 strategies examined, as well as the survival curves of the stepwise strategies selected on a 5-y, rather than a 10-y, horizon (Figure S4). Text S2 also provides the details of analyses under conditions of alternative mean CD4 counts for the cohort and alternative costs of both second-line regimens and CD4 monitoring. Further analyses (Text S3) offer additional comprehensive analytic variations in cohort gender distributions, ART initiation criteria, first- and second-line ART efficacies, stavudine-related toxicities, and costs. Within plausible ranges, these sensitivity analyses, other than those reported above, had little impact on clinical- or policy-relevant results. Discussion The new 2010 WHO ART guidelines aim to promote public health interventions that “secure the greatest likelihood of survival and quality of life for the greatest number” of individuals with HIV. The reported guiding principles in the revision process include: (1) do no harm, (2) ensure access and equity, (3) promote quality and efficiency, and (4) ensure sustainability. Motivated by these tenets, the new guidelines recommend a single CD4-based ART initiation criterion for all populations, a switch from stavudine to tenofovir, and universally available second-line regimens [2]. We find that in settings where immediate implementation of all of the new WHO treatment guidelines is currently not feasible, ART initiation at CD4<350 cells/µl provides the greatest short- and long-term survival advantage and is very cost-effective. In countries that are already initiating stavudine at CD4<350 cells/µl and have access to CD4 monitoring, switching from stavudine to tenofovir increases survival and is also cost-effective. Access to second-line ART provides more clinical benefit than access to tenofovir but at substantially greater costs. The additional outlays implied by the new guidelines stand in stark contrast to the resource-constrained reality encountered on the ground. Many countries are still striving to meet goals set by the now-superseded 2006 guidelines. The WHO estimates the current ART coverage rate across low- and middle-income countries to be 42% [5],[29]. Meanwhile, the new guidelines recommend access to CD4 count monitoring, call for treatment of almost double the 3–5 million people already requiring treatment based on the previous guidelines [30], and suggest replacement of the most widely used antiretroviral drug with one that costs nearly US$100 per patient-year more [26]. In most resource-limited settings, the relevant policy questions are: What is feasible now? and What to do first? Based on projected short- and long-term survival and cost-effectiveness results, we identify three critical messages. First, countries with very limited resources and still only one line of ART available should focus first on access to CD4 count monitoring and ART initiation at CD4<350 cells/µl. These should be implemented before switching from stavudine to tenofovir and prior to providing second-line ART. Although advising to use stavudine in the first-line regimen—with its inherent toxicities—may be seen as conflicting with the primary WHO principle “first, do no harm,” the switch from stavudine to tenofovir is the recommendation that provides the least overall increase in survival, according to the results presented here. Initiating stavudine-based ART at CD4<350 cells/µl, compared with clinically based ART initiation, provides immediate and substantial short-term survival benefits, yields the greatest life expectancy compared to other guideline components, and is cost-effective by international standards. In cases where most patients present to care with CD4 counts far below the ART initiation threshold (e.g., CD4<100 cells/µl), a policy of earlier ART initiation is neutral at worst—both in terms of cost and clinical outcomes—as it serves only to increase life expectancy among patients with less advanced disease. Second, countries with currently one line of ART available but more resources should ensure access to CD4 count monitoring with ART initiation at CD4<350 cells/µl and then switch from stavudine to tenofovir, before making second-line ART available. Indeed, some countries have already responded to the 2010 WHO guidelines and have made plans to phase out stavudine [31]. Reductions in the price of tenofovir could resolve the ongoing debate surrounding the role for stavudine in resource-limited settings. At an annual cost of US$51, tenofovir would be both less costly and more effective than stavudine. Third, in countries with sufficient budgets to provide second-line ART, it is neither effective nor cost-effective to maintain stavudine in first-line regimens. Second-line ART may offer additional efficiencies by decreasing the prevalence of resistant virus and leaving future drug regimen options available. Once countries have the capacity to provide early ART initiation, tenofovir, and second-line regimens, there will be additional clinical and policy questions. Policy makers will be addressing what to do upon second-line failure; optimal third-line regimens will be in question. Expanded ART regimen availability leads to clinical need for timely ART switches and forces the issue of HIV RNA laboratory availability. Finally, timely ART initiation is currently limited by late presentation to care [32],[33]. Concurrent with scaling up to achieve the 2010 WHO ART guidelines, there should be a concerted effort to achieve the 2007 WHO HIV screening guidelines [34]; without earlier case detection, a policy of ART initiation at CD4<350 cells/µl will never be effectively realized. It is important to highlight that full and immediate implementation of the comprehensive set of new guidelines is cost-effective by South African standards. But, while it is helpful to critically examine the survival and economic efficiency of alternative programmatic choices, “cost-effective” does not mean “affordable.” In the setting of clear budget constraints, the question of affordability may conflict with the political imperative that all persons receive the same care package. In this case, prioritization of equity over efficiency decreases mean life expectancy—sometimes by more than 1 y per person—for the same healthcare expenditure (Figure 3, lower panel). This analysis has several limitations. We report results from a cohort of HIV-infected individuals initiating ART. Although we believe the overall results would be consistent, this analysis does not specifically address ART programs with patients already in alternative stages of care, including some on first-line regimens, some on second-line regimens, and some who have previously accumulated drug-related toxicities. Such diversity within a cohort would require more individualized analyses. Additionally, a full budget impact analysis would be required to examine the number of patients in need of care, and to project the implications of each component of the WHO recommendation on program budgets over alternative time horizons. Despite its limitations, this analysis represents the only systematic, scientific effort we are aware of that marshals the evidence base in support of implementing the WHO guidelines. The most unfortunate outcome upon release of the revised WHO guidelines would be either their complete dismissal on cost grounds alone, or the execution of more expensive—though easier to implement—interventions that offer less overall health benefit than other interventions. In cases where the simultaneous implementation of all components of the 2010 WHO ART guidelines is beyond the reach of programs or countries, important prioritization questions emerge. This analysis suggests that CD4 count monitoring and ART initiation at CD4<350 cells/µl are the critical initial priorities. Replacing stavudine with tenofovir would further increase survival and would also be cost-effective. Adding a second-line ART regimen would provide large survival benefits, but with substantial increases in the necessary budgets. Supporting Information Figure S1 ART scale-up strategies. (0.62 MB DOC) Click here for additional data file. Figure S2 Course of disease. (0.22 MB TIF) Click here for additional data file. Figure S3 Validation of South African natural history data in the CEPAC model. (0.33 MB TIF) Click here for additional data file. Figure S4 Patient survival in the first 5 y after model entry. (0.41 MB TIF) Click here for additional data file. Text S1 WHO priorities: Technical appendix. (0.27 MB DOC) Click here for additional data file. Text S2 WHO priorities; sensitivity analyses addendum, part 1. (0.26 MB DOC) Click here for additional data file. Text S3 WHO priorities; sensitivity analyses addendum, part 2. (0.12 MB XLS) Click here for additional data file.

Conventional estimators for the expected value of sample information (EVSI) are computationally expensive or limited to specific analytic scenarios. I describe a novel approach that allows efficient EVSI computation for a wide range of study designs and is applicable to models of arbitrary complexity.