2.1 Overview

A budget impact model was developed in Microsoft Excel. The model was adapted from the PEPFAR PVT Model [24], which has previously been validated against other commonly used PVT costing models, including those created by the Clinton Health Access Initiative, the National Center for Global Health and Medicine (Japan), and Avenir Health’s SPECTRUM [25]. This model, a mathematical representation of the real world, utilized country-level programme data and produced robust cost and other outcome outputs [25]. The model cost projections accounted for new programmatic inputs, such as annual LTFU, mortality rate with or without ART, vertical transmission rates, breastfeeding period, coverage of PVT services, and number of pregnant WLHIV; and epidemiological inputs, such as disease progression, differences in survival rates based on treatment eligibility, fertility rate, and the possibility of subsequent pregnancies.

2.2 Perspective

Following standard published guidelines for BIA studies [21], the current study was conducted from the payer’s perspective. Although funders like PEPFAR, through the US government, play a major role in HIV prevention and treatment, the Ugandan government, through the Ministry of Health, contributes 15% of the budget allocated to supporting health facilities in providing these services, making it a de facto payer. As a result, the Ugandan public payer perspective, which includes only direct healthcare costs and excludes all patient-related costs (out-of-pocket, informal care, and lost productivity), was considered for this study.

2.3 Target Population

The eligible population included all HIV-positive pregnant women currently receiving PVT services in Uganda, adjusted for data prior to pregnancy and before initiation of ART. The model follows all WLHIV and receiving PVT services over a period of 5 years (2025–2029). This target group is broken up into quarterly cohorts, and the model then projects the costs incurred and health outcomes for each cohort over the 5-year period. Notably, the continuous addition of WLHIV would overestimate the number of patient-years; we assumed that quarterly additions would allow for more accurate estimates of patient years on ART and allow WLHIV to be LTFU or have changes in the fertility odds based on breastfeeding status. Therefore, new patients (newly diagnosed with HIV or newly pregnant and engaging with HIV prevention services for the first time) were added to the model every quarter, while others were removed as they were lost to follow-up or died. These adjustments were essential to make plausible estimates of the BIA outcomes because, pragmatically, there is always an exit and entrance of in-system (women already receiving PVT services) or old (LTFU) and new patients, respectively.

2.4 Intervention Mix and Time Horizon

The new intervention in the current study was the ERS, which was compared to the SOC (Ministry of Health strategy) [20]. Table 1 explicitly shows the differences between the ERS and the SOC. Because this strategy set of interventions would compete favourably, we conservatively assumed a starting-static market share of 50: 50. Under the scenario analysis, the market share is projected to increase by 5% from 2026 onward, assuming that factors, such as changes in clinical guidelines and improved clinical efficiency, will support the adoption of the new intervention (ERS). Other strategies, such as the sexual and reproductive health strategy [26], were considered unfavourable and were excluded from the current analysis.

Table 1 The similarities and differences between the ERS and the SOC

This study adopted a 5-year time horizon, which was considered plausible for capturing all BIA outcomes as the country looks forward to achieving the 2030 targets of HIV elimination [27, 28]. No discounting was conducted because the budget holder’s interest is the expected budget impact at each point in time [21].

2.5 Analytic Framework Description

A static cohort of all women accessing PVT services entered the model. Each quarter of the year, a subset of this cohort was adjusted according to each arm’s specific parameter characteristics and the general HIV epidemiology in Uganda, such as disease progression, differences in survival rates based on treatment eligibility, and the possibility of subsequent pregnancies. In addition, our model considered only direct medical and non-medical costs associated with the testing and treatment of HIV. Adjusting for quarterly data is crucial because it resonates with the national budget framework, in which funds are released quarterly.

The main outcomes of the BIA were the net budget impact and new infant infections averted over 5 years.

2.6 Input Data

The model required inputs covering a range of programmatic, epidemiological, and cost information [25]. All input data on the projected number of HIV-infected pregnant women needing services, as well as other inputs, were obtained from the published literature, Ministry of Health reports, and the DolPHIN clinical trial [16]. Where necessary, assumptions were made on other cost-related inputs, as seen in Table 2.

Table 2 Budget impact analysis model inputs2.7 Costs

This BIA study estimated the costs of treating, supporting, and caring for a person living with HIV. These costs, particularly laboratory and service delivery, were obtained from recent data by Guthrie and colleagues [37] and inflated to 2023 US dollars [39]. Costs related to the use of the main ART regimens in Uganda—dolutegravir and efavirenz—during pregnancy were obtained from a recently published cost-effectiveness study [16]. The costs included in the model were divided into three categories: (1) drug costs, which include both adult ART (DTG [dolutegravir] + TDF [tenofovir disoproxil fumarate] + 3TC [lamivudine]) and antiretroviral (ARV) prophylaxis for women receiving PVT services; (2) service delivery (non-drug) costs, including both annual recurrent expenditures and annualized capital costs, and (3) supply chain management and central support costs.

2.8 Sensitivity Analysis and Uncertainty

We used the one-way sensitivity analysis to estimate the effect of adjusting individual model parameters on BIA estimates. Due to the unavailability of the 95% confidence intervals, we used ±20% on both cost and numeric parameters, and these results are presented as a tornado diagram.

2.9 Ethics Statement

No institutional review board approval was required because the current BIA model was parameterized using publicly available data and published literature with no protected patient or health information. Moreover, this study was an extension of the original DolPHIN clinical trial [36], which the respective institutional review boards had approved [14,15,16].