Cost-effectiveness of models for prevention of vertical HIV transmission – voluntary counseling and testing and choices of drug regimen

Study model

We use a hypothetical cohort of 100,000 pregnancies as the study model. The decision tree in Figure 1 presents a flow of the programme options. Cost and outcome parameters are based on Thai settings. The combination of one and two VCT sessions and four antiretroviral therapy (ART) regimens leads to eight case options being considered within the model. For simplicity's sake we assign codes 1 and 2 for single and double VCT strategy, and A-D for four drug regimens.

Cost analysis is conducted from the perspective of the Ministry of Public Health (MOPH) as the Thai government pays all costs for VCT, ART and substitute feeding. We measure the programme outcomes as the net cost to the public-sector payer, total number of cases of paediatric HIV infection averted, and cost per paediatric HIV infection averted.

The cost-effectiveness of each intervention is calculated as (IC+AC-HC)/IA, where IC are the programme intervention costs, AC additional healthcare cost due to NVP resistance, HC the lifetime health care cost of an HIV infected infant (or cost offset), and IA the number of HIV infections averted by the intervention. We converted all cost and effectiveness at the present value (2003) with discounting of 5%.

Input parameters (table 2)

For each parameter we determine base-case values. A maternal HIV infection rate of 1.5% was reported by national sentinel surveillance 2001 [19]. A prospective mother-to-child study from 1992 to 1994 in Bangkok [20] found that with a second VCT round during pregnancy an additional 4.7% (95% CI 2.7% and 7.5%) HIV positive mothers were detected who had become infected during pregnancy. In other word, we assume the second VCT picks up 4.7% of the incidence cases.

In Thailand, all pregnant women have at least one ANC visit during pregnancy [21]. As it takes two weeks to know the maternal HIV status the 7.4% of pregnant women arriving late for ANC are likely to receive ART later than 36 weeks compromising the optimal period for effective AZT treatment of four weeks [8].

The risk of perinatal transmission without treatment is 18.9% [3] and with breast-feeding is assumed to be an additional 12% [22]. Lallemant et al [6] report that the odds ratio of transmitting the virus by regimen D versus control (no prevention) is 0.23 (95% CI 0.05 to 0.41). The effectiveness of other drug options is derived from a Cochrane systematic review [23] indicating an odds ratio compared to placebo of 0.46 (95% CI 0.35 to 0.60) for regimen A and 0.51 (95%CI 0.33 to 0.79) for regimen B.

To estimate the risk of transmission among late arrivals for ANC who get AZT treatment less than 4 weeks in regimen A, we apply an odds ratio of 1.40 (95% CI 0.82 to 2.38) found in a Thai study [24] comparing the risk of transmission between a short (from 36 weeks onwards) and long (from 28 weeks) maternal course of AZT. In the absence of evidence of the efficacy of regimen D started after 28 weeks but before 36 weeks of gestation, we assume the same efficacy if treatment is started before 34 weeks, and the lower efficacy of regimen A for those starting after 34 weeks.

In 13 provinces in the North of Thailand [8] 75% of infected pregnant women accepted AZT treatment after knowing their HIV status. A recent study found a higher proportion (84%) of pregnant women accepted regimen D [17]. We use the former figure as the base case scenario for infected women who knew their HIV status before 36 week of gestation and accepted AZT treatment in regimens A and C, and the latter for regimen D.

As we know that women make their first antenatal visit late in pregnancy tend to report low education and socioeconomic status [25]. We, therefore, assume a lower proportion of 65% of infected pregnant women who know their HIV status after 36 weeks accept AZT in programmes A and C, and 75% of those accept AZT and NVP in programme D. For those who refuse AZT treatment for whatever reason, we assume 50% would accept the simpler regimen of NVP in programme C.

For the regimen of single NVP, we propose 85% of infected pregnant women who know their serologic status before 36 weeks and 75% thereafter enroll in the programme B. This enrollment rate is slightly higher than with AZT as drug administration is simpler.

The study presents all prices in US$ (USD) at 2003 price units. Costs of providing VCT, maternal and infant antiretroviral treatment are derived from 160 MOPH hospitals [8]. Briefly, data on the units of each category of resources used and valued were gathered by mean of a questionnaire sent to each hospital participating in the study. Intervention unit costs include recurrent labour and non-labour expenditures but exclude capital depreciation.

A study in Thailand found that 17.4% (95% CI 12.0% to 22.7%) of infected mothers who were treated with a single dose NVP developed a strain of HIV resistant to the drug [17]. Several studies report that the resistant virus spontaneously reverts to a wide-type genotype by one year [14, 26], meaning that the resistance would not have altered effectiveness of prevention in further pregnancies. Nevertheless, the concern is that those who were resistant and were then treated for AIDS following delivery would have a much more difficult time controlling the virus [13, 18]. AIDS experts therefore recommend a more expensive treatment regimen based on protease inhibitors (PI) as a first line treatment for mothers with the NVP resistant virus instead of the common and cheaper regimen, based on Non-Nucleoside Reverse Transcriptase Inhibitors (NNRTI) [27]. A report from MOPH [28] reveals 33.9% of infected mothers need treatment for AIDS by a year after delivery and the incremental cost of switching from the cheaper to the more expensive treatment regimen is 497 USD (95% 147 USD to 847 USD) assuming the incremental cost occurs only for the first three years of the treatment [29].

Assuming 80% of lifetime paediatric HIV/AIDS treatment cost is shouldered by the public sector, the lifetime medical care cost was estimated at 3,300 USD in 1997 (25 Baht per 1 USD) [30]. Converting the 1997 figures to 2003 values using the general consumer price index, the public sector health expenditure for a case of paediatric AIDS is 1,680 USD (40 Baht per 1 USD). A range of 1,340–2,015 USD is proposed for sensitivity analysis. A clinical trial of morbidity and mortality among breast fed and formula fed infants of HIV-1-infected women in Kenya found a similar overall mortality rate, incidence of diarrhea, pneumonia and other serious complications among the two groups [31]. Therefore, we do not add costs of treatments for excess diarrhoea and respiratory tract infections among formula-fed infants.

Uncertainty analysis

To handle uncertainty in the model input parameters of interest are ascribed a distribution that reflects the uncertainty associated with their true value (table 2) and entered in a probabilistic uncertainty analysis. For example, the beta-distribution was the choice of distribution for probability parameters which were bounded zero-one and the gamma distribution was modelled for unit cost parameters [32].

The results from 1,000 calculations are presented in a CE acceptability curve based on the concept of net-benefit approach suggested by Stinnett and Mullahy [33] and Briggs et al [34]. It shows the probability in the 1,000 iterations of the model that any of the eight interventions is the most cost-effective option given different willingness to pay thresholds.

Cost-effectiveness

Compared with no routine prophylactic ART, programme 1D is the most cost-effective regimen costing 556 USD to avert one paediatric HIV infection. Option 2B is the least cost-effective regimen at more than three times the cost per additional case prevented.

With relative little benefit gained from the second VCT, all regimens with the single VCT strategy are more cost-effective than those with 2 VCT sessions.

Uncertainty analysis

We plotted acceptability curves using net-monetary benefit approach for the choice of prevention strategy in figure 3. To explain this, we consider table 3 where the most effective drug option is programme D. The incremental cost of moving from 1D to 2D is 260,000 (450,000-190,000) USD for 16 (353-337) additional infections averted. In other words, the incremental cost-effectiveness ratio of giving up the 1D regimen and adopting 2D is 16,000 USD per additional infection averted. In Figure 3 this incremental cost-effectiveness value is represented by showing the line of 1D crossed the line of 2D at the ceiling ratio of 15,000USD (they were not exactly the same value since the first is a deterministic and the latter is a probabilistic value). If the decision maker would prefer a confidence level greater than 95%, the threshold is 35,000 USD per infection averted. Further studies would be needed to improve the accuracy of the cost-effectiveness results. In particular, better information on the proportion of infected pregnancy detected by second VCT and the cost of VCT would improve the model because these parameters have the greatest bearing on uncertainty.

Having mentioned above that the rate of developing NVP resistant virus among Thais was considerably lower than of Sub-Saharan setting, we therefore tested the results assuming a rate of developing HIV resistant to NVP as high as in a Sub-Saharan setting (40%). Figure 4 illustrates that the regimens that contains no NVP (A) or contains less NVP (C) are dominant. Programme 2D is still to be the preferable choice if the willingness to pay threshold greater than 40,000 USD per HIV infection averted with the statistical confidence level 95%. It is interesting to note that 1A is a significantly dominant regimen at the lowest threshold of zero willingness to pay, that is, where a decision has been made that no further resources would be attributed to healthcare.