Expanded HIV Screening in the United States: Effect on Clinical Outcomes, HIV Transmission, and Costs FREE

—The Editors

Early detection and timely access to medical care can substantially improve the course of HIV disease among infected persons (1 - 2). Whether they also reduce the risk for transmitting the virus to others (3 - 7) is not clear because survival gains from antiretroviral therapy prolong infectious lifetimes and may lead to complacency toward HIV risk behavior (8). Recent studies report increases in HIV infections, other sexually transmitted diseases, and sexual risk behaviors in vulnerable populations (9 - 11); access to effective antiretroviral therapy may also be associated with sexual risk-taking (12 - 14).

As with any new method of screening for chronic disease (for example, hypercholesterolemia and breast, cervical, prostate, and colon cancer (15 - 19)), the challenge facing both physicians and public health experts is to determine whom to test for HIV infection and how frequently. We address the particular difficulties posed by HIV infection, an infectious disease whose detection and treatment have implications for both the individual being tested and the broader population.

We used a simulation model (20) to project the performance of increasingly frequent HIV screening of all adults using a rapid testing protocol (21 - 23) in communi ties with independently varying levels of prevalence of undetected HIV infection (0.05% to 1.0%) and annual HIV incidence (0.0084% to 0.12%). We considered medical outcomes at the level of the individual HIV-infected patient and transmission at the population level (Appendix). To evaluate transmission-related effects, we used published data on secondary HIV transmission and model-based estimates of lifetime costs and health-related quality-of-life losses attributable to new HIV infections. Following the recommendations of the U.S. Panel on Cost-Effectiveness in Health and Medicine (24), we evaluated outcomes from the societal perspective using a 3% annual discount rate. We expressed comparative value in 2004 U.S. dollars per quality-adjusted life-year (QALY) gained and used multiway sensitivity analysis to examine the effects of uncertainty about the data in the model.

We used a widely published computer simulation, the Cost-Effectiveness of Preventing AIDS Complications (CEPAC) Model, to characterize the progress of HIV disease in an infected individual ((20), (25 - 27)). The “health states” summarize the essential elements of patient status (CD4 cell count and HIV RNA level, history of opportunistic infections, quality of life, and resource use) (28). Effective antiretroviral therapy increases the probability of viral suppression and concomitant CD4 cell count increases, according to clinical trial results. Treated patients receive a sequence of up to 4 therapeutic regimens in which efficacy progressively diminishes. The model tracks each patient's clinical course from entry until death. It then aggregates the simulated clinical courses of individuals to estimate the average quality-adjusted survival and costs for screening and treatment alternatives.

The model's screening simulation accounts for whether and when detection, follow-up, and linkage to HIV care occur (29 - 31). Detection of HIV takes place through 1 of 3 discrete mechanisms: a specific HIV screening program; nonroutine, “background” testing (for example, testing in medical settings, sexually transmitted disease clinics, or correctional institutions or for employment or immigration purposes); and clinical presentation with an AIDS-defining illness. While the model simulates the course of HIV illness for all infected individuals, only patients with detected HIV infection who are successfully linked to care are eligible for antiretroviral therapy and opportunistic infection prophylaxis.

Key input data are ((21 - 23), (32 - 46)) provided in Tables 1 and 2. We considered rapid testing because of its current policy relevance (47 - 48). Rapid testing elicits higher levels of test acceptance, follow-up, and linkage to care (baseline overall likelihood of test acceptance, follow-up, and linkage is 77.6% vs. 32.6% for standard antibody testing (29 - 30)). However, rapid testing may exacerbate the distress associated with false-positive results, since the patients learn the preliminary findings before Western blot confirmation. We conducted extensive sensitivity analysis on the morbidity penalty (base value, 14 quality-adjusted life-days) attributable to false-positive results, ranging from no penalty to 30 quality-adjusted life-days.

We considered all adults (mean age, 33 years) with unknown HIV status in U.S. health care settings. The baseline analysis uses a population (1.0% undetected HIV prevalence, 0.12% annual incidence, and 6.1% lifetime HIV infection risk) that reflects pre–September 2006 guidelines for HIV screening (39). We also simulated the effects of screening the “U.S. general population” (0.1% prevalence, 0.014% annual incidence, and 0.7% lifetime HIV infection risk), using the widely cited estimate of 252 000 to 312 000 undetected, prevalent HIV infections and 40 000 annual infections in a population of 290 million (42). In sensitivity analysis, we considered additional target populations, varying both the prevalence (0.05% to 1.0%) and the annual incidence (0.0084% to 0.12%) as estimated by interpolating and extrapolating the specific values reported here.

To describe secondary HIV transmission, we used the basic reproductive number, R₀, a central concept in infectious disease epidemiology. R₀ can be interpreted as the lifetime number of subsequent infections, regardless of method of transmission, attributable to a single infected individual in a susceptible population. R₀ captures the interaction of 3 factors—HIV transmission efficiency; number of risky contacts; and duration of infectiousness—in producing a summary measure of the power of an infection to emerge and to persist (49). When R₀ is greater than 1, the average infected person generates at least 1 subsequent case and an epidemic can ensue; when R₀ is less than or equal to 1, the epidemic cannot persist.

We applied a baseline R₀ of 1.44 to all cases of undetected HIV infection and to situations where we assume no effect of screening and treatment (Table 2) (43). We also applied this value to cases of HIV infection detected through presentation with an opportunistic infection, an assumption that adopts the conservative view that individuals who decline HIV testing will respond less favorably to behavioral counseling (50). The “favorable transmission impact” scenario reflects the potential virologic benefit of antiretroviral therapy to reduce (R₀ < 1.44) HIV transmission. The value R₀ of 1.27 applies to all individuals identified through HIV screening (43). The “adverse transmission impact” scenario assumes that patients with screening-detected HIV infection are at higher risk for transmitting HIV, presumably because of treatment-related behavioral disinhibition (45 - 46). Lacking scientific evidence, we arbitrarily chose an R₀ value of 1.61, which is as far above the baseline value (R₀ = 1.44) as the “favorable impact” value (R₀ = 1.27) is below it.

Recognizing the critical role played by the transmission impact assumption, we conducted extensive sensitivity analyses by considering values ranging from −1.00 to 1.00 for ΔR₀, the difference between R₀ in the presence and absence of care. ΔR₀ can be interpreted as the lifetime number of secondary HIV infections averted when an HIV-infected person in a susceptible population is identified by screening, counseled, and linked to treatment. By using ΔR₀ to represent the effect of screening on transmission, we could estimate the incremental cost-effectiveness ratios without specifying a base value for R₀.

The expected number of secondary infections under a given HIV screening program is a key outcome measure. To estimate it, we first obtained the proportions of HIV infections identified by each detection mechanism from the simulation model (Table 3) (29). We used these proportions to compute a weighted average of the reproductive numbers for each mechanism for HIV detection (Table 2). This weighted average represents our estimate of the number of secondary transmissions per infected individual. We calculated total transmissions by multiplying this value by the lifetime risk for HIV infection (Table 2) in the target population.

We assigned each secondary infection a survival loss and an economic cost (51). We obtained a survival loss of 30.49 discounted quality-adjusted life-months (QALMs) by comparing a model-based estimate of life expectancy for a secondary infection—assuming current standards of care (52) and HIV-specific quality-of-life weights (53)—with survival without HIV infection. We obtained non-HIV, quality-adjusted survival from U.S. life tables (54) and age-specific, SF-6D utility weights from the Medical Expenditure Panel Survey (55 - 56). (The SF-6D is a health state classification system based on 6 dimensions [“6D”] of the SF-36 health survey.) To determine the $210 100 cost per secondary infection, we reduced a model-based estimate of discounted lifetime costs of HIV patient care (52) to reflect offsetting, non–HIV-related medical costs (57) during the additional life span lived by avoiding HIV infection. We reduced both survival losses and incremental medical care costs to account for a delay of 14 years from the time of HIV transmission until entry into HIV care: 6 years for the average passage of time between primary HIV infection and a secondary HIV transmission and 8 years for the average time between a secondary HIV transmission and eventual entry into HIV care (58). In sensitivity analysis, we eliminated additional discounting (that is, secondary infections were assigned a survival loss of 46.18 QALMs and an economic cost of $318 200).

The National Institute of Mental Health, National Institute of Allergy and Infectious Diseases, National Institute on Drug Abuse, Doris Duke Charitable Foundation, and Centers for Disease Control and Prevention funded this study. The funding sources had no role in the design, analysis, or interpretation of the study or in the decision to submit the manuscript for publication.

When we restricted attention to clinical outcomes affecting only the individual infected patient, current practices of HIV detection (including but not limited to screening) produced discounted quality-adjusted life expectancy of 279.91 QALMs or 23.32 QALYs (Table 4). Discounted lifetime HIV-related costs averaged $7640/person. Adding a single, rapid HIV screening conferred an additional 0.32 QALM (about 10 days in good health) per program participant at an average additional cost of $1000 ($40 for testing and $960 for care). Viewed strictly in terms of individual-level effects, therefore, the addition of a single rapid screening costs $37 100 per QALY gained. Increasing the intensity of screening to every 5 and 3 years would cost $60 100 and $96 800, respectively, per QALY. Annual screening conferred no additional health benefit over screening every 3 years: False-positive results and their associated quality-of-life losses offset the survival gains.

When we broadened the analysis to take into account secondary HIV transmission, current practices resulted in 87.4 cases of secondary HIV transmission per 1000 members of the HIV-uninfected susceptible population. Under current policy, secondary transmission imposes a per capita survival “penalty” of 2.66 QALMs and $18 360 in additional treatment costs. We applied these same survival and cost penalties to all screening strategies under the “no effect of screening and treatment on transmission” scenario. In this scenario, we assumed that expanded HIV detection and treatment improve the course of disease in the individual patient but have no incremental impact—as compared with current practice—on secondary transmissions. The reproductive number was assumed to be constant across all mechanisms of HIV detection (Table 2). Hence, as a consequence of the model structure and these assumptions, secondary transmission effects in this scenario did not affect cost-effectiveness ratios for expanded screening versus current practice.

With no specific screening program and favorable transmission assumptions (R₀ = 1.27 for individuals identified through HIV screening), 81.3 secondary HIV transmissions per 1000 population members occurred. These imposed a per capita survival “penalty” of 2.48 QALMs and $17 070 in additional treatment costs. Adding a single, rapid screening lowered secondary HIV transmission rates to 80.7 per 1000 population, reducing per capita survival and cost penalties to 2.46 QALMs and $16 950. Combining these population-level effects of favorable assumptions about the benefits of screening on transmission with the individual-level outcomes described earlier improved the cost-effectiveness ratio of a single, rapid screening to $30 800/QALY from $37 100/QALY (when we assumed no effect of screening and treatment on transmission). Rapid screening every 5 and 3 years had incremental cost-effectiveness ratios of $32 300/QALY and $55 500/QALY, respectively. The additional benefits of annual rapid screening did not offset the effects of increased false-positive results from screening more often.

Under adverse antiretroviral therapy impact assumptions (R₀ = 1.61 for individuals identified through HIV screening), we observed 93.5 secondary HIV transmissions per 1000 population members, with no specific screening program. One-time rapid screening increased secondary HIV transmissions to 94.1 per 1000 population members, which increased the per capita survival penalty from 2.85 to 2.87 QALMs and increased per capita additional care costs from $19 640 to $19 760. Because there were still survival benefits for the individual patients detected by screening, a single screening nonetheless conferred a positive net health benefit; however, the cost per QALY gained increased from $37 100 in the “no impact” scenario to $44 200. Under the adverse effect scenario, the incremental cost-effectiveness of screening every 3 to 5 years exceeded $100 000/QALY; annual screening produced higher costs and poorer quality-adjusted survival.

When we applied the same screening interventions in a population with 0.1% prevalence and 0.014% annual incidence, a single rapid test had an incremental cost-effectiveness ratio of $72 400/QALY when viewed only in terms of individual patient-level outcomes (“no impact” scenario). That ratio improved to $60 700/QALY under the “favorable impact” scenario; it worsened to $86 200/QALY under “adverse impact” assumptions. With repeated screening in this lower-incidence population, the negative impact of false-positive results on health-related quality of life more than offset any screening-related survival benefits.

Our findings were not sensitive to plausible variation in testing program characteristics, cost structures, discount rates, or health-related quality-of-life valuations. However, we found more favorable cost-effectiveness ratios when we assumed less background testing, higher HIV prevalence and incidence, and a greater impact of screening and treatment on secondary transmission, as measured by ΔR₀ (initial values = −0.17, 0.00, and 0.17 in the favorable impact, no impact, and adverse impact scenarios, respectively). Figure 1 offers recommended HIV screening policies, assuming that society is prepared to pay up to $50 000 to purchase an additional QALY of health for its citizens. Undetected HIV prevalence in the screened population is the principal consideration in choosing to initiate a first screening. If it is assumed that antiretroviral therapy has no impact on secondary transmission, one-time screening is recommended for prevalences greater than 0.28%. With a favorable transmission impact (ΔR₀ = −0.17), the lowest prevalence for which one-time screening is recommended falls to 0.20%; with adverse transmission impact assumptions (ΔR₀ = 0.17), it rises to 0.40%. In formulating a policy for repeated screening, both the prevalence and incidence of HIV infection are important. For testing every 5 years (assuming favorable transmission impact), the threshold population has a prevalence of 0.45% for HIV infection and an annual incidence of 0.0075%.

At prevalences of undetected HIV infection above 1.0%, the curves in Figure 1 are vertical. This suggests that at a higher prevalence of undetected HIV infection in the population, the choice of screening policy no longer depends on the fraction of cases detected; rather, the principal driver of both costs and benefits is the treatment pathway triggered for the comparatively large number of HIV-positive patients identified. The test itself emerges as a critical cost component only at low prevalence (29). Figure 2 illustrates how decision makers might choose between no specific screening program and one-time screening for a range of cost-effectiveness threshold values. If cost-effectiveness ratios up to $75 000/QALY define good value for money, one-time screening is recommended for prevalences above 0.10%, even assuming no impact of screening and treatment on transmission. If society is prepared to pay up to $100 000/QALY, one-time screening is preferred under virtually all plausible scenarios. More frequent screening at lower prevalences would become cost-effective if the 14-day quality-of-life penalty for false-positive reports was smaller (data not shown).

Our findings support routine, rapid HIV testing for all adults in the United States as long as the prevalence of undiagnosed HIV infection is above 0.20%. A single rapid HIV screening in such settings delivers value comparable to many commonly employed screening interventions for chronic disease (59). More generally, the prevalence of undetected HIV infection drives the decision to initiate a first screening while HIV incidence drives the choice of retest frequency. For example, repeated screening every 5 years achieves similar value in a population with a prevalence of 0.45% and an annual incidence of 0.0075%.

This analysis supports the new recommendations of the Centers for Disease Control and Prevention calling for routine HIV screening in all adults and adolescents in U.S. health settings (7). Our findings would lead to stronger recommendations than those of the U.S. Preventive Services Task Force (60), which limits its recommendation to individuals at “increased risk” for HIV infection. The Task Force considers the potential harms associated with screening those without risk factors to be greater than the potential benefits. Our analysis suggests that, from both the clinical and economic perspectives, the benefits of routine HIV testing in all adults in the United States outweigh the likely harms.

These results do not confirm the widely held belief that the preventive benefits of HIV screening for uninfected individuals at risk for acquiring HIV infection exceed the medical benefits to infected patients (61 - 62). We find that transmission effects are less influential for decision making than suggested, for example, by Sanders and colleagues (63): $41 700/QALY (excluding transmission) and $15 100/QALY (including transmission) for one-time HIV screening in populations similar to our baseline. Our results are less optimistic for several reasons. First, we assumed a smaller impact of antiretroviral therapy on HIV infectivity. In our view, the evidence does not support the modeling assumption that high rates of antiretroviral therapy–induced suppression of serum HIV RNA reflect similar rates of eradication of semen or vaginal HIV RNA and, by extension, similar reductions in HIV infectivity. Recent studies report that semen and vaginal fluid contain HIV RNA during treatment and that concentrations of protease inhibitors are much lower in semen or vaginal fluids than in serum (64 - 65), suggesting active viral replication within these compartments (66). Second, our analysis captures current uncertainty about the net effect of antiretroviral therapy on secondary HIV transmission (50); longer survival and therefore increased duration of infectiousness and behavioral disinhibition (plausible but unproven) could dampen—and possibly even reverse—any virologic benefits of therapy on transmission. Published assessments of secondary HIV transmission (51) and model-based estimates of R₀ in the presence of antiretroviral therapy (67) suggest point estimates of −0.17 to −0.10 for the parameter ΔR₀. These are estimates; limited evidence suggests that people who learn their HIV status may reduce risky behaviors and that even larger negative values of ΔR₀—perhaps representing instances where identification of index cases leads to earlier identification of partners (68)—might be achieved (4,69). Finally, because of discounting, our assumption of a long delay between primary infection and the eventual clinical and economic benefits of averting a secondary transmission attenuates the comparative importance of transmission effects.

Our cost and survival findings differ from those we have previously reported (29). Here, we used updated cost and antiretroviral efficacy data (32 - 35) and focused entirely on rapid HIV tests (47). Our current assumption of a large quality-of-life penalty for preliminary false-positive reports highlights the tradeoff between increased rates of detection and increased false-positive penalties with greater retest frequencies.

This study has important limitations. First, we restricted attention to “first-generation” secondary transmissions, which understates the total infections attributable to each infected person. Second, we assigned a fixed survival and economic cost to each secondary infection, which does not fully capture variability in the time, mechanism, or likelihood of HIV detection or referral to care. Third, our model did not use recent evidence suggesting that the risk for HIV transmission varies widely over the course of infection (70). Fourth, we did not account for late antiretroviral-related toxicities that may result in cardiac disease or diabetes. Finally, we compared testing all adults, even in low HIV prevalence settings, with current practice, contrary to the previously recommended strategy of testing high-risk patients in high-risk settings (71).

The Centers for Disease Control and Prevention now recommends routine HIV testing for all patients 13 to 64 years of age in health care settings unless a formal survey documents the prevalence of undiagnosed HIV infection to be less than 0.1%. Our analysis arrived at a slightly higher point estimate for the prevalence threshold (0.2%) but entirely supports the shift from targeted screening based on patient risk factors to routine screening based on prevalence and incidence thresholds. Nevertheless, we recognize the difficulty practitioners face in determining whether the prevalence of undiagnosed HIV infection in their practice setting meets a given threshold. Providers may be able to obtain estimates of local prevalence from their state and local health departments. Ideally, public health departments would do formal seroprevalence surveys and cohort studies that could be analyzed to provide HIV prevalence for specific demographic and geographic target groups. Until then, we recommend that providers initiate routine, voluntary HIV screening for all adults in the United States, unless surveillance data in their particular setting, or in similar settings, show an HIV prevalence below 0.2%. We base this recommendation on the findings presented here and on evidence that the prevalence of screening-detected HIV infection exceeds this threshold in most U.S. health care settings where voluntary HIV screening of all adults has been implemented (72).

The Appendix Figure provides a conceptual overview of the analysis. Individuals drawn from the at-risk population enter the Population Screening Model one at a time. Based on the assumed incidence/prevalence of HIV infection as well as life table–based estimates of life expectancy, a random-number generator immediately determines whether the individual will ever be infected with HIV. A simple “IF/THEN” statement makes this determination; uninfected cases never proceed to the Disease Simulation Model. The small fraction of individuals who do become HIV-infected during their lifetimes proceed to the Disease Simulation Model. However, they are ineligible to receive any kind of HIV clinical care or therapy until and unless their infection is identified. Instances where patients die before their infection is detected are represented by the lower NO branch; instances where patients are identified as infected and become eligible for therapy are represented by the lower YES branch. All patients who proceed to the disease simulation model remain there until death.

Our analysis contains several important modifications from previous models published by our group (notably, a 2005 paper (29)). These include focusing on rapid versus conventional testing; the inclusion of a large quality-of-life penalty for false-positive results; and taking into account the impact of HIV screening on secondary transmission. To assist readers in interpreting our findings in the context of previous models, we have replicated the baseline analysis reported in the current paper but with a few notable changes in the underlying assumptions. The new analyses, which are summarized in Appendix Table 1 and Appendix Table 2, represent stepwise additions of the rapid testing and false-positive assumptions. The analyses are grouped into 3 sets: 1) baseline analysis using conventional enzyme immunoassay antibody testing rather than rapid testing; 2) rapid testing but assuming no quality-of-life penalty for false-positive results; and 3) rapid testing under baseline assumptions (including the 14-day quality-of-life penalty for false-positive results).

Appendix Tables 1 and 2 report the effect on costs and QALYs of adding each feature, both with and without taking into account secondary transmission effects. We highlight the following findings from the analysis.

First, costs and survival benefits are uniformly lower for conventional testing than for rapid testing, reflecting the lower participation/detection/linkage rates achieved with conventional tests. Lower participation/detection via conventional testing also explains differences in the mechanism of detection: Under conventional testing, fewer cases are detected via the “screening” program and more cases are detected either via “background” testing or via presentation with an opportunistic infection.

Second, the quality-of-life penalty has no effect on the costs of rapid testing or on the mechanisms of detection under rapid testing. Its only impact is on quality-adjusted survival—and, by extension, on the costs/QALY cost-effectiveness ratios.

Third, compared with rapid testing without a 14-day penalty, quality-adjusted survival is lower under rapid testing with the 14-day penalty. Compared with conventional testing, the incremental benefit of rapid testing (assuming the 14-day penalty) first rises and then falls. This reflects the initial benefits of improved participation/detection/linkage as well as the increasingly harmful role played by false positives with increased test frequency.

Regardless of whether secondary transmission effects are taken into account, the cost-effectiveness differences among the 3 protocols are surprisingly small. This reflects the observation that costs and benefits typically move in lockstep with increased case detection, whether the participation rate is 67% or 100% or anything in between.

Except at high retest frequencies, the principal driver of both costs and benefits is not the HIV test itself but the increased number of patients receiving expensive care as a result of improved case detection. The cost-effectiveness ratios associated with conventional testing are always more favorable than for rapid testing with no 14-day penalty. Adding the 14-day penalty further diminishes the attractiveness of rapid testing.

Briefly stated, then, the analysis highlights the tradeoff implicit in the switch from conventional to rapid tests: increased rates of detection and linkage versus increased false-positive penalties.

The present analysis uses newer data on cost and efficacy of antiretroviral therapy than those employed in our previous studies. To help readers to understand the impact of these new data, we have reproduced Table 1 from the current manuscript, using the cost and antiretroviral therapy efficacy data used in our 2005 paper (29) (Appendix Table 3). We highlight the following observations about the results.

Overall, there are no striking differences—either quantitatively or qualitatively—between the output obtained with the New England Journal of Medicine input values and the output obtained with updated cost and efficacy data. This is not terribly surprising since the absolute changes in the input data are small and all effects are averaged over large populations comprised predominantly of HIV-uninfected individuals in whom these input data changes have absolutely no effect.

In every instance, the older data produce marginally lower cost and life-expectancy estimates. This reflects the fact that the older data assumed slightly lower efficacy of antiretroviral therapy and slightly lower costs being incurred over slightly shorter lifespans.

Generally speaking, the older data produce less favorable cost-effectiveness ratios. Here again, however, the overall observation is that there is little difference—either in terms of quantitative magnitude or qualitative importance—between the results obtained with the New England Journal of Medicine input values and the results obtained with updated cost and efficacy data.

Appendix Table 4 reproduces the baseline analysis with the addition of a “screen every 24 months” strategy. The performance of this strategy on every dimension—cost, survival, times to detection, CD4 cell counts at detection, percentage detected via screening, and cost-effectiveness—is intermediate to the strategies of screening every 12 months and every 36 months. Similarly, the curve in Figure 1 denoting settings over which screening every 2 years would be preferred is always intermediate to the curves for the “every 12 months” and “every 36 months” strategies. These observations hold for all target populations and all screening protocols, as well.