Intended for healthcare professionals

CCBYNC Open access
Research

# CD4 cell count and viral load monitoring in patients undergoing antiretroviral therapy in Uganda: cost effectiveness study

BMJ 2011; 343 (Published 09 November 2011) Cite this as: BMJ 2011;343:d6884
1. James G Kahn, professor1,
2. Elliot Marseille, principal2,
3. David Moore, research scientist3,
4. Rebecca Bunnell, acting director, division of community health45,
5. Willy Were, medical epidemiologist4,
6. Richard Degerman, data management adviser4,
7. Jordan W Tappero, director, health systems reconstruction office45,
8. Paul Ekwaru, statistician4,
9. Frank Kaharuza, chief, epidemiology branch4,
10. Jonathan Mermin, director, division of HIV/AIDS prevention45
1. 1Philip R Lee Institute for Health Policy Studies and Department of Epidemiology and Biostatistics, University of California, San Francisco, CA, USA
2. 2Health Strategies International, Oakland, CA
3. 3Department of Medicine, Faculty of Medicine, University of British Columbia and British Columbia Centre for Excellence in HIV/AIDS, Vancouver, Canada
4. 4CDC-Uganda, National Center for HIV, Viral Hepatitis, STD and TB Prevention (NCHHSTP), Centers for Disease Control and Prevention, Entebbe, Uganda
5. 5Centers for Disease Control and Prevention, Atlanta, GA, USA
1. Correspondence to: J G Kahn, University of California, 3333 California Street, Suite 265, San Francisco CA, US 94118 jgkahn{at}ucsf.edu
• Accepted 21 September 2011

## Abstract

Objective To examine the cost and cost effectiveness of quarterly CD4 cell count and viral load monitoring among patients taking antiretroviral therapy (ART).

Design Cost effectiveness study.

Setting A randomised trial in a home based ART programme in Tororo, Uganda.

Participants People with HIV who were members of the AIDS Support Organisation and had CD4 cell counts <250 ×106 cells/L or World Health Organization stage 3 or 4 disease.

### Cost effectiveness model

Our model was designed to assess the cost and health value of each incremental use of resources for ART monitoring. This incremental cost effectiveness ratio (ICER) is the difference in cost between two monitoring options, divided by the difference in DALYs averted. We compared each monitoring option with the next less expensive alternative. The clinical option was the least expensive, followed by clinical/CD4, and clinical/CD4/viral load. We did not calculate cost effectiveness of clinical/CD4/viral load monitoring compared with clinical monitoring as this would numerically blend the two laboratory monitoring strategies and thus obscure their independent value.

The increase in costs between monitoring options reflects differences in costs per person year for the monitoring tests themselves (that is, CD4 and viral load); differences in costs of antiretroviral regimens (because of unequal rates of progression to more costly second line treatment); and outpatient and inpatient care. These were estimated based on use during the three years’ median follow-up during the trial. Mortality influences total costs by affecting the number of person years of care in each arm. We projected costs of future HIV care associated with the lives saved during the trial to a total of 15 years from start of treatment, using the observed arm specific costs and rates of clinical outcomes during the trial.

The increase in health benefits (that is, DALYs averted) between monitoring options reflects differences in mortality, severe morbidity, and the DALYs incurred with these clinical events. Specifically:

DT=NDA×DD+Σ (NMA×DM)

where DT=total DALYs averted; NDA=number of deaths averted; DD=DALYs averted per death averted; NMA=number of severe morbid events averted; and DM=DALYs averted per morbid event averted.

The DALYs from morbidity are summed across 14 diagnoses. The DALYs associated with mortality are calculated for the three year trial and then over the subsequent 12 years.

### Data inputs

We relied mainly on data from the trial to determine the value of health and intervention cost inputs,11 16 17 supplemented by published sources and expert opinion within the trial. Table 1 summarises the data inputs, which are discussed below for the base case. Alternative input values and assumptions are described for the sensitivity analyses.

Table 1

Model data input values for cost effectiveness analysis, antiretroviral monitoring study, Tororo and Busia Districts, Uganda, 2003-7. All costs are in US $ParameterBase case valueSource(s) Clinical/CD4/viral loadClinical/CD4Clinical Frequency of health events: Deaths (per 100 person years)*3.74.15.8Trial11 Severe morbid events (per 100 person year)*1.72.74.9Trial11 Disability adjusted life years per clinical event during trial: Deaths2.32Trial,11 Global Burden of Disease18 Severe morbid events (range)0.019-0.75Global Burden of Disease,18 expert opinion Unit costs of monitoring tests: CD4$4.68$4.68Downing (Clinton Fund); trial11 Viral load$29.64COBAS Amplicor; trial11
% Change in antiretroviral drug regimen (annual)0.7%0.4%1.8%Trial11
Costs of use during trial (per person year):
Laboratory monitoring of CD4s and viral loads$137.30$18.74$0Trial11 Antiretroviral drugs (discounted; observed mix of first and second line regimens during trial)$160$156$175Trial,11 Médecins Sans Frontières19
Diagnostic tests$168$168$164Trial11 Opportunistic infection treatment including tuberculosis (outpatient)$122$123$127Trial11
Inpatient costs$2.05$1.96$2.87Trial11 *From intention to treat analysis, adjusted for multivariate regression findings (see text). Table 1 Model data input values for cost effectiveness analysis, antiretroviral monitoring study, Tororo and Busia Districts, Uganda, 2003-7. All costs are in US$

#### Health inputs

Health events for the baseline analysis reflect the trial’s intention to treat analyses. Point estimates for each monitoring strategy are used, even if differences between study arms were not significant at P=0.05. When multivariate analyses were conducted in the trial, we calculated event rates reflecting the adjusted differences between monitoring arms.

Deaths during the trial were most common in the clinical monitoring arm (5.8 per 100 person years, adjusted), followed by clinical/CD4 monitoring (4.1 per 100 person years, P=0.109 versus clinical), and clinical/CD4/viral load monitoring (3.7 per 100 person years, P=0.698 versus clinical/CD4). Severe morbid events (without death) followed a similar pattern, though at lower absolute rates.

Each death was estimated to cause 2.32 DALYs during the trial, representing two years of lost life on average (deaths were more common toward the start of the trial) and an age disability weight of 1.16 given the average age in this cohort of 39. For severe morbid events (14 different diagnoses), we multiplied the rates in the trial by disability weights18 and by duration (estimated by project clinicians), yielding a wide range in DALYs per event (0.019 to 0.75).

Future longevity and DALYs were modelled from trial completion to 15 years by using arm specific annual mortality rates from the trial. We estimated mean longevity during this 12 year period at 6.43 years (discounted) for clinical, 7.12 for clinical/CD4, and 7.28 for clinical/CD4/viral load. The number of DALYs per arm after the trial quantifies lost health from mortality that occurs both during and after the trial.

#### Cost inputs

The cost of monitoring is predominantly in the consumable test kit. For CD4 counts, the kit constitutes $3.80 (£2.40, €2.70) of the total$4.68 unit cost and for viral load $27.20 of the total$29.64. The remaining costs are capital, primarily for the test machines, staff, and other costs—such as labelling stickers.

Rates of change of ART from first to second line treatment affect the cost of monitoring strategies. From the trial, the rate of change is highest for clinical (1.8% per year) and similar for clinical/CD4 (0.4%) and clinical/CD4/viral load (0.7%). The annual cost of treatment is $104 for first line ART (stavudine 150 mg, lamivudine 40 mg, and nevirapine 200 mg in a fixed dose pill) and$995 for second line ART (tenofovir 300 mg, didanosine 400 mg, and lopinavir 133.3 mg, and ritonavir 30 mg in a fixed dose pill).19 Different antiretroviral treatment for people coinfected with tuberculosis and the addition of shipping and other ancillary costs raises the average loaded regimen costs for first and second line ART to $125 and$1119 a year, respectively. Furthermore, the cost of first line treatment rises over time (2-4% per month) because of changes in antiretroviral drugs within this regimen.

Table 1 also shows the costs per person year of use of health services in the trial. Differences across monitoring strategies are highest for laboratory monitoring, reflecting quarterly testing: $18.74 for CD4 and$137.30 for CD4 and viral load. The cost of ART varies noticeably by monitoring strategy because of large differences in regimen costs combined with modest differences in change rates. Observed use of diagnostic tests and treatment for opportunistic illnesses varied little (despite sharply different morbidity rates; see sensitivity analyses). Though inpatient costs were higher in the clinical strategy, with daily costs of less than $5 (as is common in the WHO CHOICE database for sub-Saharan Africa) the absolute values are low.16 17 20 Finally, differences in total costs of the arms during the trial reflect differential mortality and thus different total number of person years of receipt of ART. We estimated the cost of future care of trial survivors based on our simulations out to 15 years, incorporating the arm specific mortality and rates of change of antiretroviral regimens observed during the trial. The future cost per person alive after the trial is$4976 (discounted) for clinical, $4892 for clinical/CD4, and$6043 for clinical/CD4/viral load. Total costs for each arm represent the product of cost per person and the number alive of the original 100 in the arm.

### Comparison with cotrimoxazole

We examined the cost effectiveness of ART overall and of the least expensive ART arm (clinical monitoring) compared with treatment with cotrimoxazole alone. This extends an earlier analysis that compared the entire ART cohort at two years with cotrimoxazole.17 The cotrimoxazole comparison results are from a different trial in the same population. We included this comparison because some readers might be interested in the cost effectiveness comparison of ART versus cotrimoxazole alone, given that HIV care resources are increasingly constrained in Africa and a decreasing proportion of people with HIV will be able to access ART. Thus, the analysis provides comparisons in a cohort of people with HIV across the broad continuum of care from cotrimoxazole to ART with clinical monitoring, to ART with clinical and laboratory monitoring. We have included only the clinical benefits to the patient (not prevention of HIV transmission and other outcomes), so the results might not be precisely comparable.

### Sensitivity analyses

We examined the effect of uncertainty in inputs and modelling assumptions on the ICERs. We varied individual parameter values (such as the relative mortality across monitoring arms and costs of monitoring tests), as well as broader decisions (such as using intention to treat versus per protocol trial results, and limiting the analysis to the trial period). We used Monte Carlo simulations to portray the impact of simultaneous variation in key inputs.

## Results

### Base case

During the three year trial, the clinical strategy cost the least ($190 570 per 100 individuals starting the arm), followed by clinical/CD4 ($194 844), and clinical/CD4/viral load ($229 521) (table 2). The most important difference in cost across arms was monitoring, with a smaller difference for antiretroviral drugs. Total costs over 15 years follow a similar pattern, at$606 260, $626 718, and$769 177, respectively. The difference in total costs was $20 458 for clinical/CD4 versus clinical and$142 458 for clinical/CD4/viral load versus clinical/CD4 (table 3).

Table 2

Cost and health outcomes by monitoring strategy (per 100 people starting each arm), for cost effectiveness analysis, antiretroviral monitoring study, Tororo and Busia Districts, Uganda, 2003-7

Costs ($) Total during trial (three years)229 521194 844190 570 Laboratory monitoring CD4 and viral load*38 14251640 Antiretroviral drugs*44 40242 89746 370 Total after trial539 656431 874415 690 Total769 177626 718606 260 Health outcomes DALYs during trial27.130.544.4 Deaths11.112.217.5 DALYs from deaths25.828.340.5 DALYs from opportunistic infections1.362.173.90 Future DALYs294.5318.6422.0 Total DALYs321.6349.1466.4 *Selected cost detail presented to show major variation across strategy. Table 2 Cost and health outcomes by monitoring strategy (per 100 people starting each arm), for cost effectiveness analysis, antiretroviral monitoring study, Tororo and Busia Districts, Uganda, 2003-7 Table 3 Incremental costs, DALYs, and cost effectiveness ratios for antiretroviral monitoring study, Tororo and Busia Districts, Uganda, 2003-7 Cost ($)Incremental cost ($)DALYs incurredIncremental DALYs avertedICER:$ per DALY averted
Clinical606 260466.4
Clinical/CD4626 71820 458349.1117.3174
Table 3

Incremental costs, DALYs, and cost effectiveness ratios for antiretroviral monitoring study, Tororo and Busia Districts, Uganda, 2003-7

DALYs were determined almost entirely by mortality. During the trial, the most deaths (17.5 per 100 individuals starting the arm) and the most DALYs (44.4 per 100 individuals starting the arm, including 40.5 deaths) occurred in the clinical strategy arm. The figures for DALYs were 30.5 (28.3 deaths) in the clinical/CD4 arm and 27.1 (25.8 deaths) in the clinical/CD4/viral load arm. After the trial, DALYs continued to reflect differential mortality by arm and were about 10 times higher than DALYs incurred during the trial: 422.0, 318.6, and 294.5 for the three arms, respectively. The DALY total after the trial is high because each death, whether during or after the trial, incurs DALYs in each subsequent year. Total DALYs were 466.4 for clinical, 349.1 for clinical/CD4, and 321.6 for clinical/CD4/viral load. The difference in estimated total DALYs was 117.3 for clinical versus clinical/CD4 and 27.5 for clinical/CD4 versus clinical/CD4/viral load (table 3).

Table 3 shows incremental costs, health outcomes, and cost effectiveness. The incremental comparisons of interest, guided by the order of increasing cost, are clinical/CD4 versus clinical and clinical/CD4/viral load versus clinical/CD4. For the base case, each increase in cost is associated with increased health benefits (that is, fewer DALYs incurred), so no strategy is “dominated.”

The cost per DALY averted was $174 for clinical/CD4 versus clinical and$5195 for clinical/CD4/viral load versus clinical/CD4. The 30-fold higher ICER for the second comparison is because of sevenfold higher incremental costs and three quarters lower difference in DALYs.

### Comparison with cotrimoxazole

The cost of a cotrimoxazole strategy was $9299 for 100 individuals over 15 years.16 This reflects$50 per person year and reduced survival (and thus person years of care) compared with ART. DALYs over 15 years total to 1160. Thus, compared with cotrimoxazole, ART with clinical monitoring adds $596 961 and averts 693.7 DALYs for an ICER of$861 per DALY averted. ART overall (all three arms combined) adds $658 086 and averts 781.1 DALYs for an ICER of$843 per DALY averted. The second ICER is lower (by 2.1%) because of the attractive ICER of going from clinical to clinical/CD4. As noted above, these ICERs consider only the clinical value to the individual receiving treatment, omitting the health and cost effects associated with prevention and other benefits of ART, and thus are higher than the more inclusive ICERs below.

We estimate $518 per DALY averted for ART with the clinical/CD4 strategy versus cotrimoxazole, based on the previous finding of$573 per DALY averted for ART (all arms) versus cotrimoxazole17 and the lower ICER when CD4 is added to clinical monitoring. We estimated $637 per DALY averted for ART with clinical/CD4/viral load versus cotrimoxazole because of higher monitoring and costs of antiretrovirals. If a national programme has$100 million to spend for ART, a mean cost of $518 per DALY averted implies that 193 145 DALYs could be averted. If instead, the same money was spent at$637 per DALY averted, then 156 935 DALYs could be averted. The difference of 36 210 DALYs represents the potential health loss of allocating programme funds to viral load testing instead of starting more people on ART with clinical and CD4 monitoring. From the cost perspective, an additional $23 million would be needed to include viral load testing with no change in DALYs. ### Sensitivity analyses Table 4 reports the effect on ICERs of variation in key inputs and modelling assumptions. We examined uncertainty in the mortality rate as suggested by the 95% confidence intervals for observed adjusted hazard ratios in the intention to treat analysis. For clinical/CD4/viral load, with the most favourable hazard ratio and mortality rate, the ICER drops below$1600. With the worst mortality, the clinical/CD4/viral load strategy is dominated by clinical/CD4 (more expensive and less effective). For clinical/CD4, lower mortality worsens the ICER because the ICER approaches the cost effectiveness of keeping individuals alive on ART (that is, cost and a year’s DALYs v death and no cost). When mortality is higher, the clinical/CD4 arm becomes less expensive and effective than clinical, so the ICER is calculated in the reverse direction.

Table 4

Sensitivity analyses for key inputs and modelling assumptions for cost effectiveness analysis, antiretroviral monitoring study, Tororo and Busia Districts, Uganda, 2003-7

Input or assumptionValues/methods usedMost/least favourable ICER ($/DALY averted) Base caseMost/least favourable*Clinical/CD4 (v clinical) Clinical/CD4/viral load (v clinical/CD4) Deaths per 100 person years (intention to treat): Clinical/CD4/viral load, 95% CI†, other arms kept at base case3.72.22/6.11174/1741550/dominated‡ Clinical/CD4, 95% CI†, other arms kept at base case4.12.55/6.48368/1792§Dominated‡/1263 Clinical, 95% CI†, other arms kept at base case5.89.10/3.70454/2832§5181/5181 Deaths per 100 person years (per protocol): Viral load arm, CD4 arm, clinical armIntention to treat (above)NA/2.2, 2.0, 3.5NA/88NA/dominated‡ Annual rate of regimen change: Viral load arm, CD4 arm, clinical arm0.7%, 0.4%, 1.8%50-150% of base case values401/dominant‡4882/5466 Future DALYs averted and costs added per death: Inclusion or exclusion of projections beyond trialIncludedNA/excludedNA/307NA/10 257 Unit costs of monitoring test kits: CD43.801.90/5.70233/1155187/5175 Viral load27.2013.60/54.40176/1733322/7040 Costs of antiretroviral drugs (annual; per person): First line regimen12563/18873/2755169/5193 Second line regimen1119557/1679451/dominant‡4825/5538 Costs of opportunistic infection treatment: Observed (base case) v imputed from rates of opportunistic infection122, 123, 12797, 123, 194/NA1.06/NA4906/NA NA=not applicable. *For cost effectiveness of laboratory testing. †95% CI from clinical trial.11 ‡Dominated=higher cost and fewer DALYs than comparator; dominant=cheaper and better. §ICER for high end of mortality range represents reverse comparison among arms. Table 4 Sensitivity analyses for key inputs and modelling assumptions for cost effectiveness analysis, antiretroviral monitoring study, Tororo and Busia Districts, Uganda, 2003-7 The trial’s per protocol (as treated) analysis examined only the period after monitoring began (at 90 days), removing the early period of high mortality unrelated to ART monitoring. This analysis found adjusted mortality rates of 2.2, 2.0, and 3.5 per 100 patient years for the clinical/CD4/viral load, clinical/CD4, and clinical arms, respectively. The clinical/CD4 arm had an ICER of$88 per DALY averted compared with clinical alone, and the viral load option was dominated compared with the clinical/CD4 arm.

Varying the rate of change in regimen between 50% and 150% of base case values (for all arms simultaneously) had little effect on the ICER for viral load ($4882-5466). It has a proportionately larger effect on the ICER for CD4 versus clinical ($401-“dominant”). This is because of the small and similar switch rates in clinical/CD4/viral load and clinical/CD4 arms and the larger base case value of regimen switch rates in the clinical arm.

### Limitations

There are important limitations to our study. We treated mortality comparisons, which dominate our model’s health and cost effects, as meaningful regardless of statistical significance. Thus, the base case results reflect observed differences in mortality that were small and not quite significant for clinical/CD4 versus clinical monitoring (combined mortality and AIDS defined events were significantly different) and far from significant for clinical/CD4/viral load versus clinical/CD4 monitoring . As a result, they could be considered overly optimistic. Despite this lenient approach (counting as real a difference for which there is little evidence), viral load had a highly unfavourable ICER. Our sensitivity analyses highlight how the uncertainty about these comparisons contributes to a substantial likelihood that viral load monitoring is dominated. Furthermore, in the per protocol analysis, which is arguably appropriate for an intervention that starts three months after the initiation of treatment, the clinical/CD4/viral load arm had a mortality point estimate slightly higher than the clinical/CD4 arm.

We lacked data on mortality rates beyond the trial. To calculate long range ICERs, we assumed ongoing arm specific mortality rates as observed in the trial. The assumption of higher or lower rates after the trial, however, did not change our findings qualitatively—that is, clinical/CD4 monitoring retained its superior cost effectiveness. Specifically, lower long term mortality, as observed late in the trial, led to a more attractive ICER for clinical/CD4 versus clinical monitoring, and higher mortality increases the ICER just 50%. Within trial ICERs generate similar results. Thus, our findings seem robust to uncertainties in mortality.

The study had a home based component and was thus structurally atypical. The relative cost effectiveness among groups, however, is likely to be similar to facility based programmes as the overall cost effectiveness of ART was similar to facility based analyses in the Côte d’Ivoire and South Africa.25 26 In addition, all three strategies used quarterly monitoring. In some countries, governments promote biannual monitoring. Results for less frequent monitoring might differ from ours.

We did not examine the benefits of ART in preventing HIV transmission. In settings where viral load monitoring improves rates of full viral suppression, there would probably be benefits. These might not be as great as initially assumed as the effect of reduced viral load on HIV transmission is not dichotomous; even a plasma viral load of 5000-10 000 copies/mL would reduce the risk of transmission considerably compared with 500 000 copies/mL. Thus, the effect of identifying relatively slight increases in viral load a few weeks or months earlier could be less important than efforts to increase access to ART and retention in care.

We did not assess all plausible configurations of laboratory monitoring. For example, a strategy including clinical and virological but not CD4 monitoring could potentially reduce costs while detecting virological failure early. Viral load might be a better indicator than CD4 cell counts of treatment success and thus clinically advantageous. Yet our study did not indicate improved efficacy from adding viral load to CD4 monitoring. Furthermore, logistical challenges to viral load monitoring are substantial in Africa. CD4 cell counts are more widely available in peripheral health centres because of lower costs, ease of operation and maintenance, and use for determining eligibility for ART. Reliance on viral load could potentially work with efficient referral to a central laboratory or new technology, but currently in Uganda, routine viral load monitoring is unlikely to be feasible outside a research setting or major cities. As less expensive and logistically simpler viral load testing becomes available, a study comparing routine CD4 cell count monitoring with routine viral load monitoring alone would be valuable.

### Implications

Standards in clinical practice well attuned to resource rich settings might not maximise the opportunities available in resource constrained settings. Monitoring CD4 cell counts seems desirable clinically and economically in these settings. In contrast, viral load monitoring might be a relatively poor investment compared with offering ART to another person or monitoring CD4 cell counts for those currently taking ART. This approach supports current WHO guidelines.27

According to WHO, the attractiveness of the ICER can be determined by comparison with the country’s annual gross domestic product per capita. An intervention with an ICER below the annual gross domestic product per capita is considered “very cost effective.” An ICER below three times the annual gross domestic product per capita is considered “cost effective.”28 The annual gross domestic product per capita in Uganda in 2008 was an estimated 1200 international dollars ($485 using currency exchange rates).13 Thus by the gross domestic product per capita standard, CD4 count monitoring is “very cost effective” and viral load monitoring is not cost effective. In closing, we want to highlight another path to increased access to high quality ART in resource poor settings. We strongly support efforts to reduce the price of second line antiretroviral regimens and laboratory tests. If such efforts could replicate the impressive past decreases in the price of first line regimens and the also important reductions for CD4 counts, the types of analyses reported here would be less necessary. #### What is already known on this topic • ART in sub-Saharan Africa costs$500 to $1000 per disability adjusted life year (DALY) averted • Previous estimates of the cost effectiveness of laboratory monitoring relied on computer simulation models and varied widely #### What this study adds • Adding routine monitoring of CD4 count costs$174 per DALY averted, less than for ART. CD4 monitoring is cost effective by improving clinical outcomes and reducing changes to more expensive antiretroviral drugs

• Adding routine monitoring of viral load costs \$5181 per DALY averted, far more than for ART or CD4 monitoring and in a subset of analyses is both less effective and more costly than clinical or CD4 monitoring. Viral load monitoring has unattractive cost effectiveness because of low clinical benefit and high test costs

## Notes

Cite this as: BMJ 2011;343:d6884

## Footnotes

• Contributors: JGK and EM designed the analysis, constructed the cost effectiveness model, and wrote the manuscript. DM, RD, and JM provided data and interpretation for the clinical trial results. RD and PE provided healthcare use and cost data from the trial. DM, RB, WW, JWT, PE, FK, and JM reviewed and edited the manuscript to assure accurate interpretation and representation of the trial context, clinical practices, and economic findings. JGK is guarantor.

• Funding: This study was funded by US Centers for Disease Control and Prevention, Kenya, and US National Institute of Drug Abuse (R01 DA15612). The findings and conclusions in this manuscript are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention.

• Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.

• Ethical approval: This study was approved by the science and ethics committee of the Uganda Virus Research Institute, the Uganda National Council of Science and Technology; CDC Institutional Review Board; UCSF CHR 04024782.

• Data sharing: Detailed costing data available on request.

This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the license. See: http://creativecommons.org/licenses/by-nc/2.0/ and http://creativecommons.org/licenses/by-nc/2.0/legalcode.