Cost-Effectiveness Analyses of Adolescent- and Young-Adult-Focused HIV Interventions Incorporating HIV Transmission: a Scoping Review of Modeling Methods and Proposed Guidance

Abstract

Background

Adolescents and young adults (AYA) are disproportionately affected by human immunodeficiency virus (HIV) globally, with high-burden population subgroups differing across regions. Cost-effectiveness modeling analyses of AYA-focused HIV interventions have provided vital information to policy makers by projecting long-term health and economic impacts of interventions’ effects on reduced HIV transmission. To provide a broad overview of available modeling approaches and identify gaps in methods reporting, we evaluated modeling methodologies used in AYA-focused cost-effectiveness analyses incorporating HIV transmission.

Methods

We searched PubMed, Embase, and Web of Science for peer-reviewed articles that were published January 2006 to August 2023 and described models that considered HIV transmission in estimating cost-effectiveness of AYA-focused HIV prevention interventions. We extracted selected study characteristics, transmission model properties, and methods to estimate long-term health and economic outcomes. We assessed study quality using published guidelines.

Results

Among 42 studies, 38% included individuals assigned male or female sex at birth, 19% included females only, and 43% included males only; 24% focused on AYA only; 88% were set in Africa; and 7% were restricted to certain population subgroups. The most common population subgroups examined were women who have transactional sex (24%) and men who have sex with men (17%). Most (88%) studied primary prevention interventions for people at risk of HIV; 29% examined secondary prevention interventions including treatment and testing. Most (98%) assessed incremental cost-effectiveness ratios (ICERs) were defined as incremental cost per incremental life year (either quality-adjusted or disability-adjusted), or cost per infection averted, or both. Of 24 different transmission models identified with distinct structures, 59% were dynamic. Of 20 studies that translated averted infections into long-term health benefit, 55% used their transmission models directly through assigning health utilities to modeled states. A total of 30 studies converted averted infections into long-term cost savings, among which, 73% used their transmission models directly by assigning costs to modeled states. Fewer than half captured age-specific sexual and care-engagement behaviors. Important gaps in quality included incomplete reporting of model validation and calibration results.

Conclusions

We identified heterogeneous modeling approaches in cost-effectiveness analyses of AYA-focused HIV interventions incorporating transmission. Reporting of key elements could be improved. We propose additional criteria that could clarify choices around modeling approaches and strengthen the reporting of model validation and calibration results.

Supplementary Information

The online version contains supplementary material available at 10.1007/s41669-025-00583-1.

Key Points for Decision Makers

A wide range of modeling approaches and model features were identified in published cost-effectiveness analyses incorporating HIV transmission into economic assessments of HIV interventions for adolescents and young adults.

To benefit the economic modeling community, the reporting of these modeling approaches, transmission model properties, and study outcomes can be improved.

Standardized and easy-to-use guidance to facilitate reporting, understanding, and comparison of modeling approaches, features, and outcomes are proposed.

Introduction

Over 30% of new human immunodeficiency virus (HIV) infections globally occur among young people aged 15–24 years [1]. Demographics of young people affected by HIV vary by region: adolescent girls and young women (AGYW) account for two-thirds of new HIV infections globally, bearing a disproportionate burden of the epidemic, particularly in African countries [2, 3], whereas young men who have sex with men (MSM) experience the highest rates of new HIV infections in Latin America [4]. From 2021 to 2022, HIV-related deaths decreased by 2.3% among 15–24-year-olds and 4.7% among persons older than 24 years [1]. Fewer than 25% of 15–19-year-olds in Eastern and Southern Africa—the regions most affected by HIV—report HIV testing and receipt of results in the past 12 months [5]. Among those with HIV globally, 10–19-year-olds have poorer access to antiretroviral therapy (ART) compared with older people, with 65% of adolescents and young adults (AYAs) versus 72–82% of older people receiving ART [1, 5].

While new treatment and prevention interventions, including long-acting medications [6–8] and digital self-testing modalities [9–11], hold promise for AYA, questions remain around their long-term clinical value and cost-effectiveness. Simulation models can provide key information for decision-makers by identifying benchmarks for uptake, effectiveness, and cost that will make an intervention clinically beneficial and cost-effective [12–14]. Simulation models can also inform how best to deliver programs given resource constraints [15]. Meanwhile, the quality of evidence generated by simulation models hinges upon the appropriateness and validity of modeling choices. While guidelines exist for model selection [16] and reporting [17], they are not universally implemented, which may be due to the wide variation in study outcomes, model properties, and approaches for estimating long-term intervention impact [18, 19].

Systematic reviews have assessed the modeling methodology in economic evaluations of adult-focused HIV interventions in specific settings [18, 20]. However, few reviews have evaluated the approaches to two key modeling objectives: estimating infections averted by interventions and translating averted infections into long-term health and cost impacts [21]; none, to our knowledge, have focused on AYA. This descriptive scoping review aims to: (1) document and analyze modeling approaches used in cost-effectiveness analyses of HIV interventions that include AYA as a primary population of interest and explicitly consider intervention impact on HIV transmission and (2) evaluate analysis quality. We use the information from the scoping review to propose standardized and easy-to-use guidance that addresses revealed gaps in methods reporting and to facilitate understanding and choices of modeling approaches.

Methods

We used the Joanna Briggs Institute (JBI) Manual for Evidence [22] and the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) extension for scoping reviews [23]. Given our primary purpose to provide a broad, descriptive overview of the range of modeling approaches used in cost-effectiveness analyses of adolescent- and young-adult-focused HIV interventions, the descriptive rather than analytic nature of our review fits well within a scoping review framework [24, 25].

Search

We searched Medline/PubMed (National Library of Medicine, NCBI), Embase (Elsevier, embase.com), and Web of Science Core Collection (Clarivate) on 1 August 2023, using terms meant to capture three concepts: HIV, cost-effectiveness analysis, and models (Appendix Table A1). Controlled subject headings (i.e., MeSH and Emtree) were included when available and appropriate. No language limits were applied. A publication date range of 2006–2023 was applied. The first integrase-strand-inhibitor drug (raltegravir) was introduced in 2006, greatly improving combination ART [26]. Thus, we chose 2006 to ensure the inclusion of most recent and relevant evidence reflective of current best practices in HIV care. Searches were developed and carried out by a librarian (C.M.). Title and abstract screening were conducted in Covidence systematic review software [27] by two independent reviewers (W.C. and S.L.), followed by full text screening (W.C.).

Table 1.

Summary of study characteristics in a scoping review of AYA-focused HIV cost-effectiveness literature incorporating transmission

Characteristic	Number	%	References
(A1) Population age of focus
Adolescent and young adult was a subgroup	32	76	[13, 15, 29, 39–67]
Adolescent and young adult only, aged 9–25 years	10	24	[12, 14, 68–75]
(A2) Population group studied
Full population in setting	39	93	[12–15, 29, 39–59, 61–65, 67–73, 75]
Women having transactional sex (± their clients)	10	24	[13, 29, 45, 50, 53–55, 57, 58, 61]
Men who have sex with men	7	17	[29, 53, 54, 57, 61, 66, 74]
Pregnant girls/(young) women	2	5	[29, 53]
Serodiscordant couples (or women in serodiscordant relationships)	2	5	[50, 58]
Sexual minority males	1	2	[71]
People who inject drugs	1	2	[57]
(A3) Country setting	Number	%
Single country	Total: 36	Total: 85
South Africa	15	36	[13, 15, 39, 49, 52–54, 59, 60, 62, 68–70, 73, 75]
Kenya	3	7	[12, 46, 63]
USA	3	7	[14, 71, 74]
Eswatini	2	5	[43, 61]
Mozambique	2	5	[40, 57]
Rwanda	2	5	[51, 55]
Uganda	2	5	[64, 67]
Zimbabwe	2	5	[41, 47]
Australia	1	2	[66]
Namibia	1	2	[65]
Peru	1	2	[29]
Tanzania	1	2	[44]
Zambia	1	2	[48]
Multiple countries (African and Caribbean)	Total: 6	Total: 14	[42, 45, 50, 56, 58, 72]
(A4) Geographic scalea
National	23	55	[12, 29, 41, 43, 45, 47–50, 55–57, 59–61, 64–66, 68–70]
Subnational (city, provincial, or regional)	10	24	[13, 15, 40, 44, 46, 52, 62, 63, 67, 71]
Unspecified/individual	6	14	[14, 39, 72–75]
Multiple (multiple scales with different scenarios)	3	7	[42, 51, 58]
(A5) Intervention(s) studiedb
Primary prevention
Male circumcision (any type/form)	18	43	[15, 39–44, 46–48, 51, 57, 61, 63–67]
Oral PrEP	13	31	[13, 46, 52–56, 58–61, 71, 73]
Behavioral/social support for preventionc	6	14	[29, 46, 49, 57, 61, 75]
Condom distribution/promotion for people without HIV	4	10	[29, 46, 57, 61]
Vaccination	4	10	[62, 68–70]
Long-acting PrEP	3	7	[54, 60, 73]
Dual prevention pills (coformulated oral contraception and PrEP)	1	2	[50]
Prevention of mother-to-child transmissiond	3	7	[29, 57, 61]
Secondary prevention
Testing (as an intervention)	8	19	[29, 45, 46, 49, 57, 61, 72, 74]
Behavioral/social support for treatmente	5	12	[14, 29, 49, 57, 61]
Condom distribution/promotion for people with HIV	4	10	[29, 46, 57, 61]
Oral ART	3	7	[31, 56]
Treatment as prevention (not further specified)	2	5	[46, 49]
Long-acting ART	1	2	[12]
Both
Sexually transmitted infections treatment	1	2	29]
(A6) Maximum time horizon, years
> 20, including lifetime	27	64	[13–15, 39, 40, 43–52, 59, 63, 65–67, 69–74]
11–20	7	17	[41, 42, 53, 54, 58, 62, 64]
≤ 5	6	14	[29, 55–57, 61, 75]
6–10	2	5	[12, 68]
(A7) Economic perspectivef	Number	%
Unspecified	Total: 23	Total: 55	[12, 29, 39–44, 47–49, 52, 55–57, 59, 63–65, 67, 68, 72, 75]
Specified	Total: 19	Total: 45
Healthcare providers/payers	13	40	[13, 45, 46, 50, 51, 53, 54, 58, 61, 62, 69, 70]
Government as a third party payer	1	5	[66]
Health systems	1	2	[60]
Societal	1	2	[71]
Funder	1	2	[15]
Modified healthcare sector	1	2	[74]
HIV program perspective (excluding medical care cost for individuals without HIV)	1	2	[73]
(A8) Cost-effectiveness measure(s) associated with averted infections
Cost per infection averted or maximum cost to achieve the same cost per infection averted	25	60	[15, 39–44, 46–49, 51, 53–59, 63–65, 67, 68, 75]
ICER or maximum price premiumg	20	48	[12–14, 29, 45, 46, 49–52, 54, 60, 62, 66, 69–74]
Cost per coverage (e.g., per diagnosed or treated)	3	7	[72]
Optimal budget allocation that minimizes infections and/or deaths averted	1	2	[61]
(A9) Non-cost-effectiveness measure(s)
Reduction in primary/secondary infectionsh	37	88	[12, 14, 15, 29, 39–54, 56–68, 71–75]
Reduction in population-level incidence/prevalence over time	13	31	[13, 15, 41, 43, 44, 48, 55, 56, 59, 64, 65, 67, 70]
Future HIV-related care cost savings	20i	48	[12–15, 39, 43, 46, 49, 50, 52–54, 64–71, 73, 74]
Life years/QALYs gained, DALYs averted	16i	38	[12–14, 29, 46, 49, 50, 52, 54, 60, 69–74]
Intervention efficiencyj	14	33	[14, 15, 39–41, 43–45, 47, 48, 58, 62, 66, 67]
Deaths averted/reduction in mortality	11	26	[12, 14, 45, 50, 54, 57, 59, 61, 68, 70, 73]
HIV care continuum outcomesk	5	12	[14, 40, 42, 59, 74]
Budget impactl	4	10	[56, 62, 65, 73]
Prevention of mother-to-child transmissionm	1	2	[68]
Reduction in opportunistic infections	1	2	14]
Miscellaneous clinical outcomesn	2	5	[13, 74]
Unintended pregnancies averted	1	2	[50]

ART antiretroviral therapy, DALY disability-adjusted life year, ICER incremental cost per incremental effectiveness (life years saved or quality-adjusted life yearss gained or disability-adjusted life years averted) ratio, PrEP preexposure prophylaxis, QALY quality-adjusted life year

^aGeographic scale is reflected by the size of the primary cohort, policy context of the primary outcome(s), or the intended coverage of intervention(s) studied

^bA total of six studies evaluated more than one intervention; thus, fields do not add up to 42

^cIncludes mass media campaign for people without HIV, peer counseling/education for people without HIV, cash transfer for in-school girls without HIV, community outreach/behavior or social support with prevention services

^dThough included studies needed to have nonvertical transmissions as an outcome, some also looked at prevention of mother-to-child transmissions

^eIncludes mass media campaign for people with HIV, adherence counseling to improve viral suppression on ART, community outreach/behavior or social support (e.g., text messaging follow-up) with treatment services (e.g., ART scale-up), and pre-ART tracing

^fAs stated by the studies themselves

^gMaximum cost of an intervention to obtain the same ICER

^hPrimary infections refer to infections occurring among those receiving the prevention intervention. Secondary infections refer to infections occurring among those not receiving the treatment intervention

ⁱNumber different from the number of studies that translated averted infections into health/cost impacts owing to some not separately reporting them but incorporating them into ICER

^jIncludes number of circumcisions/vaccines/tests needed per infection/death averted, infections averted per circumcision, person-years of PrEP per infection averted, and number needed to treat to prevent one HIV-related death

^kIncludes increase in circumcision/treatment/diagnosis coverage, cross-sectional viral suppression, diagnosis, and retention in care

^lIncludes total cost of the intervention(s) studied, number of circumcisions needed given a fixed coverage level, and infections averted given fixed budget

^mThough included studies needed to have nonvertical transmission as an outcome, one study looked at prevention of mother-to-child transmission

ⁿIncludes increase in CD4 counts, decrease in prevalence of nucleoside reverse transcriptase inhibitors (NRTIs) resistance, decrease in time spent with undiagnosed HIV

Inclusion Criteria

Studies were included in the final data extraction if they were full-length and peer-reviewed; used a mathematical model to estimate reductions in nonvertical new HIV infections (acquisition or transmission, respectively, if specific to a population without or with HIV) from HIV prevention interventions, which included primary prevention (i.e., focused on people without HIV, including preexposure prophylaxis, or PrEP; male circumcision; condom distribution/promotion; vaccination; and behavioral/social support with prevention services) or secondary prevention (i.e., treatment and testing); included AYA as a population of interest (i.e., AYA received the intervention(s) studied and/or reported AYA-specific outcomes); and reported a cost-effectiveness measure (any outcome that combines cost and intervention effectiveness into a single measure) associated with averted HIV infections. To capture a comprehensive range of modeling papers, we defined AYA as persons aged 9–25 years, corresponding to the World Health Organization definition of 10–24 years [28] ± 1 year. If study eligibility for inclusion was unclear, two authors (W.C. and A.M.N.) reached a consensus through ensuring the alignment between study eligibility and independently predefined key review objectives.

Data Extraction

One reviewer (W.C.) extracted study characteristics, transmission model properties, and approaches for estimating long-term intervention benefits from articles and supplemental materials (where available) (Appendix Table A2). Study characteristics include: (A0) publication details (title, author names, journal, and year of publication); (A1) population age; (A2) population group; (A3) country setting; (A4) geographic scale (national, subnational, multiple scales, or unspecified/individual (only individual-level outcomes reported)); (A5) intervention(s); (A6) maximum time horizon; (A7) economic perspective; (A8) cost-effectiveness measure(s) associated with averted infections; and (A9) non-cost-effectiveness measure(s). Cost-effectiveness outcomes included: incremental cost-effectiveness ratio (ICER), defined either as incremental cost per infection averted or as incremental cost per incremental life year (LY, quality- or disability-adjusted); maximum price premium, defined as the greatest price differential for an intervention to remain cost-effective compared with the next less costly intervention that is nondominated; and, optimal budget allocation, defined as the optimal distribution of funds across multiple interventions to maximize health outcomes within a budget. Non-cost-effectiveness outcomes included intervention efficiency (e.g., coverage needed per infection or death averted), clinical and epidemiological outcomes, and budget impact.

To assess transmission model properties, we extracted information on the: (B10) type of transmission model used in terms of four distinct aspects (analytical/simulation-based, stochastic/deterministic, dynamic/static, and compartmental/individual-based). In analytical models, outputs can be hand-derived from inputs using closed-form mathematical equations expressing the functional relationship between the two (e.g., a simple calculation of infections averted as a constant incidence multiplied by the cohort size and PrEP efficacy). Simulation-based models utilize computer simulations to estimate outputs from inputs, often in contexts where underlying functional relationships are analytically intractable. A model is stochastic if randomness is introduced in any model component (e.g., using an outcome probability that varies across individuals or time steps according to some distribution). In a dynamic model, the force of infection among susceptible persons is a time-varying function of HIV prevalence. We also evaluated whether the model included: (B11) age-specific HIV care-engagement behaviors (with respect to, e.g., PrEP uptake, linkage to HIV care, or adherence to care or ART); (B12) age-specific risk of HIV acquisition (e.g., by using age-specific incidence rate) or risk of HIV exposure/transmission in an exposure (e.g., by modeling age-specific frequency of coital acts in a partnership); (B13) any nonrandom mixing between different age or population subgroups; and (B14) if sexual behaviors explicitly modeled characteristics of partnership dynamics (e.g., multiple partners overlapping in time (i.e., concurrency) and partner change rate) or partner- or act-level behaviors (e.g., condom use and role preference).

Additionally, we evaluated the approaches used to estimate long-term (C15) health and (C16) cost impacts associated with averted infections. Health impacts refer to (quality- or disability-adjusted) LYs saved (years). Cost impacts correspond to future HIV-related care costs saved. We grouped the approaches into four categories: (1) whether outcomes (health and/or cost impacts) were directly linked to modeled health states within the transmission model (approach 1) and if not, then the ways in which outcomes were attached to model projected reduction in infections through an intervention (approaches 2, 3, and 4). We labelled these approaches: (1) “using the transmission model directly” (e.g., assigning evidence-based utility weights and costs to modeled health states and tallying total life expectancy and costs over simulated lifetime trajectories); (2) “using external published estimates” for health/cost impacts that accrue as a result of an infection averted (published estimates can be based on any approach, e.g., by using an economic model distinct from the transmission model in the study); (3) “using the transmission model indirectly” (only in studies evaluating secondary prevention interventions on the basis of a static transmission model, e.g., quality-adjusted life years (QALYs) gained per secondary infections averted were estimated as the delta QALYs between two cohorts—one without HIV and another with acute infection, both simulated using the transmission model and discounted by the time when the transmission was averted); and (4) “using analytical methods” (e.g., Fox–Rushby approach [29] that calculates disability-adjusted life years, or DALYs, as a function of standard life expectancy, age of death, and disability weight, which were independently projected using the transmission model).

Of particular relevance to AYA is the extent to which delaying infection may impact health and/or economic gains. Older age at infection may be associated with longer life expectancy and lower costs owing to living more years without HIV, a phenomenon we termed a “delaying effect”. For studies that evaluated long-term health benefit, we extracted information about the (C15.1) methods used for estimation and (C15.2) whether the delaying effect was incorporated into the health benefit (e.g., through assigning age-varying mortalities to modeled states). For studies that evaluated long-term cost impacts, we extracted information about (C16.1) methods used for estimation and (C16.2) whether the delaying effect was captured into cost impacts (e.g., tallying costs on the basis of number of years lived with HIV).

Quality Assessment

We assessed methodological quality by combining recommendations from the International Society for Pharmacoeconomics and Outcomes Research Society (ISPOR), the Society for Medical Decision Making (SMDM) Modeling Good Research Practices Task Force, and the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) checklist (Appendix Table A2) [16, 17, 30–32]. We evaluated whether studies satisfied these recommendations in four domains as categorized by ISPOR-SMDM: model conceptualization, transmission modeling, handling of uncertainty, and model transparency and validation (Appendix Table A2). For model conceptualization [33], we examined whether the study stated the: (D1) decision problem, modeling objective, population of interest, alternative interventions, and outcomes; (D2) economic perspective (e.g., healthcare providers); and (D3) time horizon. For transmission modeling, we evaluated whether the study (D4) directly addressed structural assumptions or described potentially important variations and limitations related to not varying these assumptions (e.g., not modeling sexual mixing between adolescent and other age groups might underestimate the intervention impact) and (D5) reported model-estimated HIV incidence or prevalence as reflections of intervention effects on the underlying transmission dynamics over time. For uncertainty in modeling, we evaluated whether the study: (D6) conducted sensitivity analyses around input parameters (deterministic or probabilistic); (D7) conducted model calibration; and, if calibration was performed, (D8) reported ranges or posterior distributions around calibrated input parameters and/or uncertainty intervals (e.g., credible/confidence intervals) around model outcomes. Lastly, for model transparency and validation, we assessed whether: (D9) the study documented the model and methods (e.g., supplemental materials containing model inputs, equations, and assumptions), and (D10) external validation was described (i.e., comparing model estimates to actual event data that were not included in model calibration) [32].

Some recommendations were adapted to permit definitive and/or objective answers. For example, ISPOR-SMDM guidelines [31] recommend assessing whether sensitivity analyses are relevant to the decision problem being addressed, which depends on the decision-makers’ perspective. If no rationale is reported for varying certain parameters, assessing the relevance between sensitivity analyses and the decision-making context may be unclear. Thus, rather than assessing relevance, we evaluated in (D6) whether a study conducted sensitivity analyses around input parameters at all (Y/N). In addition, we determined whether studies: (D11) defined any cost-effectiveness thresholds or benchmarks, which permit comparison across studies; (D12) stated the funding source; and (D13) addressed the implications of results for policies, programs, or clinical practice, which explicitly link model-projected outcomes to their practical relevance for decision-making in policy [17].

On the basis of our evaluation of the quality of modeling methods and reporting, we identified areas where execution of the existing guidelines was rare and suggested new, more actionable recommendations for these areas. In addition, we developed recommendations of aspects not mentioned in existing guidelines that could strengthen model design and reporting.

Results

We identified 4476 records by the literature searches and removed 2062 duplicates identified using EndNote software [34] (Clarivate, Philadelphia, PA, USA), 14 additional duplicates identified during import into Covidence, and 7 duplicates identified manually, resulting in 2393 unique references for screening (Fig. 1). After screening titles and abstracts, 506 studies proceeded to full-text review. After full text review, 42 studies were included in the detailed extraction. Overall, 24 unique models were identified. Notably, 38% (9) of the distinct models identified were used in 64% (27) of the analyses.

Fig. 1.

The PRISMA flow diagram. This flow diagram shows the results of the search and selection process, from the number of records identified in the search to the number of studies included in the review. PRISMA indicates the Preferred Reporting Items for Systematic Reviews and Meta-analyses.

Study Characteristics

Most studies (32, 76%) included AYA as an age subgroup of interest and the remaining (10, 24%) evaluated intervention(s) focused on AYA only, with varying age ranges defined for AYA (Table 1). Among all studies, 16 (38%) assessed individuals assigned male or female sex at birth, 8 (19%) assessed females only, and 18 (43%) assessed males only. The most commonly studied population subgroups were women having transactional sex (WTS) in ten (24%) and MSM in seven (17%) (Table 1). The majority (36, 85%) were set in a single country: 15 (36%) in South Africa, 16 (38%) in other sub-Saharan Africa countries, 3 (7%) in the USA, 1 (2%) in Australia, and 1 (2%) in Peru. Interventions were evaluated at the national and subnational scale in 23 (55%) and 10 (24%) studies, respectively. Most evaluated primary prevention interventions, including male circumcision in 18 (43%) and oral PrEP in 13 (31%). HIV testing as an intervention was considered in eight (19%). Treatment as prevention interventions such as oral or long-acting antiretroviral therapy (LA ART) appeared in three (7%) and one (2%), respectively.

Among eight (19%) that studied interventions specific to individuals assigned female at birth (including AGYW), six (75%) considered PrEP, one (11%) vaccination, and one (11%) cash transfer. In the 18 (43%) that studied interventions specific to individuals assigned male at birth (including MSM), male circumcision was considered in 15 (83%), PrEP in 2 (11%), and testing in 1 (6%) (Appendix Table A3). Among ten (24%) studies examining WTS, interventions considered included: PrEP in five (50%), behavioral/social support for prevention in two (20%), testing in two (20%), and dual prevention pill (coformulated oral contraception and PrEP) in one (10%). Among seven (17%) studies examining MSM, interventions considered included behavior/social support for prevention in two (29%), testing in two (29%), PrEP in two (29%), and male circumcision in one (13%) (Appendix Table A3).

Clinical and Cost-Effectiveness Analysis

Nearly two-thirds of studies (27, 64%) considered a time horizon beyond 20 years (Table 1). Of 19 (45%) studies that stated an economic perspective, most (13, 68%) adopted a healthcare provider perspective. The two most common cost-effectiveness measures were incremental cost per infection averted (25, 60%) and incremental cost per incremental (quality- or disability-adjusted) life year (20, 48%). The latter was only seen in studies evaluating a time horizon greater than 10 years (results not shown). Of the non-cost-effectiveness measures reported, the most common were reductions in new HIV infections among susceptible persons receiving an intervention and/or declines in population-level incidence (41, 98%).

Model Structure and Approaches

There were substantial differences in transmission model types across the 24 unique models identified: 10 (42%) were deterministic, dynamic, compartmental models; 6 (25%) were analytical, which are deterministic and static and have closed-form solutions for outputs given inputs; 4 (17%) were stochastic, dynamic, individual-based models (e.g., agent-based network models); 2 (8%) were stochastic, static, individual-based models (e.g., microsimulation models); and 2 (8%) were deterministic, static, compartmental models (e.g., Markov cohort models) (Table 2). In reviewing model features, 7 (29%) models captured age-specific care engagement behavior; 16 (67%) captured age-specific risk of HIV exposure and/or HIV acquisition/transmission in an exposure; and 13 (54%) captured some form of nonrandom sexual mixing. The most common characteristics representing partnership dynamics or partner- or act-level sexual behaviors were condom use (13, 54%), concurrency (12, 50%), and coital act frequency (12, 50%). Studies varied in their methods for translating averted infections into long-term health and/or cost impacts. Of 20 (48%) studies that converted averted infections into health impacts, 11 (55%) used transmission models directly; others used the transmission models indirectly (3, 15%) or analytical methods (3, 15%), 2 (10%) used external published estimates, and 1 (5%) used a mixture of approaches. More studies (30, 71%) converted averted infections into cost impacts than into health impacts; the majority (22, 73%) used their transmission models directly, three (10%) used published estimates, and the remainder used their transmission models indirectly or analytical methods. All studies that converted averted infections into health impacts also converted averted infections into cost impacts and reported incremental cost per incremental life-year (quality- or disability-adjusted). The “delaying effect” of later age at infection was captured in 80% and 83% of studies that evaluated long-term health and cost impacts, respectively.

Table 2.

Summary of model characteristics and modeling approaches in included studies

Characteristics	Number (out of 24 models)	% (out of 24 models)	References
(B10) Types of transmission model
Deterministic, dynamica, compartmental model	10	42	[12, 29, 39, 46–48, 52–54, 57–59, 61, 66]
Analytical model (deterministic, static)	6	25	[15, 40–44, 51, 55, 56, 60, 64, 65, 67, 75]
Stochastic, dynamic individual-based (e.g., agent-based network model)	4	17	[13, 45, 49, 50, 62, 63, 71]
Stochastic, static individual-based (e.g., microsimulation model)	2	8	[14, 68, 73, 74]
Deterministic, static, compartmental (e.g., (semi-)Markov cohort model)	2	8	[69, 70, 72]
(B11) The model captured age-specific care engagement behaviorb	7	29	[12–14, 45, 53–56, 60, 74]
(B12) The model captured age-specific HIV exposure risk or acquisition/transmission risk in an exposurec	16	67	[12–15, 39–56, 59, 60, 64, 65, 67, 71, 73, 74]
(B13) The model captured nonrandom mixingd	13	54	[12, 13, 39, 45–50, 52–56, 62, 63, 66, 71]
(B14) If sexual behaviors explicitly modeled, characteristics of partnership dynamics and/or partner- or act-level sexual behaviors
Condom use	13	54	[29, 39, 46, 49, 50, 52–63, 66]
Number of concurrent partners (i.e., concurrency), number of partners per year	12	50	[12, 13, 29, 39, 45, 47–50, 52–58, 62, 63, 71]
Frequency of coital acts (within a partnership)	12	50	[12, 29, 39, 47–50, 52–58, 61, 66, 71]
Not modeled	7	29	[14, 15, 40–44, 51, 64, 65, 67–70, 72–75]
Types of relationship/relationship duration/tendency to form various type of relationships	6	25	[13, 45, 47–50, 53, 54, 61–63]
Partner change rate	5	21	[46–48, 52, 59, 62, 63]
Role preference (insertive/receptive/both)	1	4	[66]

Characteristic	Number (out of 20 studies)	% (out of 20 studies)	References
(C15) Approach to long-term health benefit
(C15.1) Methods of translating averted infections into LYs/QALYs gained or DALYs averted
Using the transmission model directly	11	55	[12, 13, 45, 46, 49, 50, 52, 54, 62, 69, 73]
Using the transmission model indirectlyd	3	15	[14, 72, 74]
Calculated independent of the transmission model using an analytical approache	3	15	[29, 51, 60]
Using external published estimates for LYs/QALYs/DALYs per averted infection	2	10	[66, 71]
Mixed	1	5	[70]
(C15.2) Effect of delaying age at infection incorporated into LYs/QALYs gained or DALYs averted
Yesf	16	80	[12–14, 45, 46, 49–52, 54, 60, 69, 70, 72–74]
No	2	10	[66, 71]
Unspecified (insufficient details about the model provided)	2	10	[29, 62]

Characteristic	Number (out of 30 studies)	% (out of 30 studies)	References
(C16) Approach to long-term cost impacts
(C16.1) Methods of translating averted infections into future cost savings on HIV-related care
Using the transmission model directly	22	73	[12, 13, 15, 43, 45, 46, 49, 50, 52–54, 58, 59, 61, 62, 65, 67–70, 72, 73]
Using external published estimates	3	10	[51, 66, 71]
Using the transmission model indirectlye	3	10	[14, 39, 74]
Calculated independent of the transmission model using an analytical approache	1	3	[60]
Unspecified	1	3	[64]
(C16.2) Effect of delaying age at infection incorporated into future cost-savings on HIV-related care
Yesg	25	83	[12–15, 39, 43, 45, 46, 49, 50, 52–54, 58–60, 64, 65, 67–70, 72–74]
No	3	10	[51, 66, 71]
Unspecified	2	7	[61, 62]

DALY disability-adjusted life year, LY life year, QALY quality-adjusted life year

^aA dynamic model means that force of infection is a function of the size or proportion of the population infected (i.e., prevalence), which changes over time in one simulation run [16]

^bIncludes PrEP uptake, linkage to HIV care, and adherence to care or ART, not including studies without sufficient details to infer this feature

^cExcluding the effect of intervention, not including models without sufficient details to infer this feature

^dIncludes mixing between sexual activity groups and/or different age groups, not including models without sufficient details to infer this feature

^eSee Appendix Table A2 for detailed descriptions of the approach used in each study

^fAs an example, if a transmission model was used to tabulate life-year measures, and the model uses age-dependent mortality, then the effect of delaying age at infection is captured because an infection incurred at age 25 years had a different impact on mortality than that incurred at age 60 years; a counterexample would be to use a published estimate that assumes constant lifespan regardless of age at infection

^gAs an example, if a transmission model was used to tabulate HIV-related care every cycle, then the effect of delaying age at infection is captured because the later the infection happens, the less time costs will be incurred (though costs incurred in later years could be higher); a counterexample is when a published estimate was used which assumed a fixed lifetime HIV cost. In this case, regardless of when the infection happens, it is assumed to incur a constant lifetime HIV cost

Quality Assessment

Most studies (40, 95%) stated the decision problem, modeling objective, target population(s), intervention(s), and outcomes (Table 3). In the remaining two (5%), the decision problem seemed inferable (Appendix Table A3). Economic perspective was stated in 20 (48%) studies. All studies stated time horizon(s). Most studies (30, 71%) acknowledged the importance of varying certain structural assumptions and either conducted additional analyses or noted limitations of not conducting those analyses. Incidence and/or prevalence of HIV over time were reported in 25 (59%). Sensitivity analyses were conducted in 37 (88%). In total, two (5%) studies met all quality assessment criteria (Appendix Table A3).

Table 3.

Summary of quality assessment of included studies in a scoping review of AYA-focused HIV transmission modeling literature

Quality assessment criteria	Number	%	References
(D1) Clear statement of the decision problem, modeling objective, population of focus, interventions, and outcomes	40	95	[12–15, 29, 40–74]
(D2) Statement of the economic perspective	20	48	[13–15, 41, 42, 45, 46, 48, 50, 51, 53, 54, 60, 63, 66, 69–71, 73, 74]
(D3) Statement of the time horizon	42	100	[12–15, 29, 39–75]
(D4) Considerations for variations on or limitations of structural assumptions/parameters	30	71	[12, 13, 15, 29, 39, 40, 45–49, 51–54, 56–61, 65, 66, 68–74]
(D5) Showed incidence and/or prevalence over time	25	60	[12, 15, 39–41, 43, 44, 46–49, 51–54, 57, 59, 61–67, 71]
(D6) Sensitivity analyses performed	37	88	[12–15, 29, 39–48, 50–56, 58–60, 62, 63, 65–74]
(D7) Conducted model calibration	23	55	[12, 13, 29, 39, 45–50, 52–58, 61–63, 66, 71, 74]
(D8) When calibration is used to derive model parameters, uncertainty around calibrated input values reported or reflected in sensitivity analyses	9	21	[13, 29, 48, 53–56, 61, 71]
(D9) Nontechnical model documentation exists and easily accessible	42	100	[12–15, 29, 39–75]
(D10) Evidence of external validation of the model exists	11	26	13,14,29,52–54,59,62,63,73,74]
(D11) One or more cost-effectiveness threshold or benchmark defined	35	83	[12–15, 29, 41, 43–48, 50, 51, 53–63, 65–74]
(D12) Statement of the funding source	37	88	[12–15, 29, 39–54, 56–61, 63, 65–70, 72–74]
(D13) Considerations for policy implication of results	38	90	[12–15, 29, 40–48, 50–54, 56, 58–75]

Uncertainty in results related to calibration and model validation had the least adequate reporting. Of 23 (55%) studies that contained evidence of model calibration (Table 3), uncertainty ranges around calibrated parameters were reported or reflected as uncertainty intervals around model outcomes in 9 (39%, Appendix Table A3). Evidence of external validation of the model was found in 11 (26%) studies. The majority (35, 83%) defined one or more cost-effectiveness thresholds or benchmarks. Funding source was identified for 37 (88%). Most studies (38, 91%) discussed the implications of results for policy implementation or broader program and practice decisions.

For three studies, model documentation was hard to retrieve, as no direct reference was mentioned in the publication (Appendix Table A3). Of 27 studies that were based on (or used variations of) one of the nine commonly used models, 4 (15%) did not specify the model version (Appendix Table A3).

Discussion

This scoping review documents and assesses the modeling methods used in cost-effectiveness analyses of HIV prevention interventions that explicitly considered transmission and pertained to adolescents and young adults. A total of 42 studies were included, spanning the years 2006–2023. Transmission modeling approaches fell into five categories: (1) analytical; (2) deterministic, dynamic, and compartmental; (3) stochastic, dynamic, and individual-based; (4) stochastic, static, and individual-based; and (5) deterministic, static, and compartmental. Of 20 studies that translated averted infections into long-term health benefit, 55% used their transmission models directly through assigning health utilities to modeled states. A total of 30 studies converted averted infections into long-term cost-savings, among which, 73% used their transmission models directly by assigning costs to modeled states. Except for economic perspective, reporting of model conceptualization was high-quality, with most studies (95%) stating the decision problem, modeling objective, population of interest, interventions, outcomes, and time horizon. Economic perspective, by contrast, was noted in only 49%. Model validation and reporting of the impact of uncertainty in parameters used for calibration were the areas with the most opportunities for improvement. In Table 4, we suggest additional user-friendly guidance that can supplement existing guidelines to facilitate the understanding and comparison of modeling approaches and outcomes across cost-effectiveness analyses of HIV interventions in general.

Table 4.

Proposed additional guidance for model-based analysis design and reporting in a scoping review of AYA-focused HIV transmission modeling literature

Guidance for reporting
1. Is an outcome reported which can be compared with other studies? (Y/N)
Yes	List outcomes	Example^: ICER, cost/infection averted.
No	List rationale	Example: preference-based utilities not available for population of interest.
2a. Are long-term health benefitsa associated with averted infections estimated? (Y/N)
Yes	Describe the methodology used: - Using the transmission model? (Y/N) - Using published estimates? (Y/N) - Using an analytic approach outside the transmission model? (Y/N) - Using a simulation-based approach outside the transmission model? (Y/N)	Example: difference in life years gained was by simulating two cohorts separately and taking the difference of the projected average life years, one with HIV infection and one without, both beginning at the time the infection was projected to be averted by the transmission model.
2b. Are long-term cost impactsb associated with averted infections estimated? (Y/N)
Yes	Describe the methodology used: - Using the transmission model? (Y/N) - Using published estimates? (Y/N) - Using an analytic approach outside the transmission model? (Y/N) - Using a simulation-based approach outside the transmission model? (Y/N)	Example: estimated HIV-related care cost-savings per averted infection used published estimates.
3. Is a simulation-basedc transmission model used to estimate HIV acquisition/transmission? (Y/N)
Yes	3a. Is the model: - Compartmental or individual-based? - Stochastic or deterministic? - Dynamic or static? 3b. If static, discuss the impact of ignoring indirect community benefitd on results	Example: an individual-based stochastic dynamic model is used. Example: since the size of susceptible persons outside of the primary cohort modeled is much larger than the size of the primary cohort, reductions in incident infections in the primary cohort due to indirect community benefit are negligible.
No	Identify the analytical approach usedc	Example: using literature-based incidence rates and life expectancy with or without HIV to estimate infections averted.
4. Is there a specific situation or realistic scenario that corresponds to the range of parameter values varied in a sensitivity analysis? (Y/N)
Yes	Describe the specific situation or realistic scenario	Example: 50% screening uptake and $20 extra dollar reward represents a hard-to-reach scenario.
No	Describe the framing of the sensitivity analysis	Example: extreme values, “even if” or threshold analysis.
5. Is external model validation conducted? (Y/N)
Yes	Provide comparisons between modeled outcomes with real-world evidence through text/tables/figures	Example: the model of the natural history of HIV and the effect of antiretroviral therapy has been derived previously and compared with a range of observed data (not shown here).
No	List challenges to model validation	Example: lack of high-quality data sources with sufficient descriptions to allow replication of design/the model does not include all elements needed to accurately simulate the population or setting from which data were derived.

^{^}See Appendix, Detailed Examples for Table 4 for more detailed examples

^aLong-term health benefits refer to life years saved, or quality-adjusted life years gained, or disability-adjusted life years averted

^bLong-term cost impacts estimate future HIV-related care cost savings

^cA simulation-based approach/model contrasts with an analytical approach/model, where given inputs/outputs can be derived using closed-form mathematical formulas

^dIndirect community benefit refers to the benefit of an intervention on nonusers of the intervention due to having lower virus (source of infection) in the community

There was notable variation in stated economic perspective: 13 studies adopted a healthcare provider perspective; the remaining studies with stated perspectives used terminology including health systems (1), government as the third-party payer (1), societal (1), funder (1), modified healthcare sector (1), and HIV program perspective (1). Since standard definitions of economic perspective are challenging, reporting costs included and their associated components (e.g., direct medical cost) would facilitate the understanding of the study context [17].

In total, two cost-effectiveness measures were frequently used in conjunction with averted infections: incremental cost per incremental (quality- or disability-adjusted) life year (48%) and incremental cost per infection averted (60%). Other outcomes not combining cost with intervention effectiveness spanned clinical, epidemiological, and intervention efficiency measures. Often, the reason for choosing specific outcomes over others, along with their relevance to the study context, could only be inferred. Building on published guidance [17], it would be helpful to explicitly state whether or not a study includes an outcome that permits comparison—for example, to interventions for the same or other conditions (e.g., cost per QALYs gained or cost per infection averted), as well as the rationale for the choice of the specific outcomes and their relevance to the decision-making context (Table 4 and Appendix, Detailed Examples for Table 4, no. 1).

A major benefit of modeling is its ability to translate short-term data about clinical and economic outcomes into long-term, value-for-money measures that can help to inform decisions about healthcare resource allocations. The approaches used for such projections in the reviewed studies were heterogenous, which may reflect the availability of data to inform employment of different model structures, intervention types assessed, and research questions of interest. Few authors discussed the rationale behind their chosen approaches, the corresponding impact on their cost-effectiveness estimates, or whether there was any potential limitation or alternative to the chosen approaches. For example, choosing to use a fixed, published estimate for future HIV care-related costs saved per infection averted may result in underestimation of costs if the estimate was based on a population older than the modeled AYA population, who would need more years of care if infected. Improved reporting, including through formal journal guidance or requirements of the rationale behind chosen approaches (Table 4, no. 2a and no. 2b), could facilitate comparability and enhance confidence in policy recommendations.

We used four dimensions as defined in Methods to classify the fundamental features of a model: 1) simulation-based or analytical; 2) dynamic or static; 3) deterministic or stochastic; and 4) compartmental or individual-based. Defining models in these terms, or presenting model equations [16] to make model type easily inferable, can facilitate identifying essential features. Among 24 unique transmission models, 50% were compartmental. Compared with individual-based models, compartmental models allow more ease of model development in general (owing to the lower requirement of disaggregated data) and yet may be limited in modeling heterogeneity across subgroups [35, 36]. This limitation may be overcome by an individual-based model that tracks individuals’ behavior [35, 36]. A total of 41% of unique models were static. A potential disadvantage of a static model is that a constant force of infection disregards any indirect benefit that an HIV intervention may have on disease transmission in the population beyond the specific cohort that received the intervention [16]. Still, static models can be sufficient if, for example, the time horizon of interest is short or if intervention effects of interest are expected to be mostly direct (e.g., oral PrEP among people without HIV who are in long-term, monogamous partnerships with people with HIV) [16]. Regardless of time horizon, geographic scale, or population subgroup, static models were used in studies that evaluated male circumcision and vaccinations. Few studies stated the rationale for choosing static over dynamic models or whether excluding the indirect benefit of the interventions for nonusers might meaningfully impact population-level transmission or policy recommendations.

On the basis of these observations, we propose the following guidance for reporting model-based analyses evaluating the cost-effectiveness of HIV prevention interventions (Table 4, no. 3): (1) if simulation-based, describe three properties of the model used: dynamic versus static, deterministic versus stochastic, and compartmental versus individual-based; otherwise, identify the analytical approach to modeling; (2) if a static model is used, discuss the impact of ignoring indirect intervention effects. Concise and clear communication of model features and the rationale for these choices could both help readers identify key model features more easily and improve understanding of their appropriateness for the decision-making context.

Adolescence and young adulthood is a distinct developmental period, during which risk-taking and impulsive decision-making are normative behaviors [37, 38]. These factors impact the health behaviors of AYA and their subsequent health outcomes. We identified gaps in existing model structures to capture AYA-specific features: only 28% of models permitted age-specific care engagement behaviors, and 67% captured age-specific risks of HIV exposure, acquisition, and/or transmission. Additionally, in studies that estimated long-term intervention health and/or cost impacts, ~ 20% did not consider whether the delaying effect might alter cost-effectiveness owing to using external published estimates that were invariant of age at which those infections were averted [16]. This phenomenon is another feature of particular relevance to AYA, for whom the effects of averting infection at a young age may not manifest for decades. Inclusion of these features—or a statement of why their inclusion would not change policy conclusions—will allow more robust economic modeling of AYA.

Most studies conducted deterministic or probabilistic sensitivity analyses. In studies that conducted calibration to derive parameter values, there might be multiple parameter values (or combinations) with similar fits to calibration target(s). However, uncertainty around calibrated values and the potential impact on outcomes were seldom reported (e.g., if a Bayesian framework was used, reporting of posterior distributions around calibrated parameters and credible intervals around model outcomes). Understanding uncertainty in calibrated parameters and other forms of uncertainty (e.g., structural uncertainty) is important for full consideration of the insights a model seeks to illuminate or actions it seeks to inform. Reporting the rationale for sensitivity analyses and the impact of results on the robustness of policy conclusions can strengthen the transparency and utility of cost-effectiveness analyses (Table 4, no. 4).

External model validation (reported in 26% of studies) also presents an opportunity for more in-depth reporting and better methods development. Not assessing external validity can undermine decision-makers’ confidence in results [32]. However, establishing a formalized validation process can be difficult for several reasons, such as when data useful for validation targets are insufficient or nonexistent. Our proposed criteria provide a more actionable format by asking authors to report (1) whether external validation was conducted and (2) challenges that arose in the process of considering external validation, if not done (Table 4, no. 5).

This scoping review has several limitations. First, the exclusion of gray literature or conference abstracts could potentially result in selection bias. However, more extensive details than provided by these formats were necessary for this review. Second, study quality assessment may be subjective, and people familiar with a specific model or approach may readily infer details that were not directly stated. Third, studies often used their limited word space to report projected outcomes in lieu of detailed descriptions of modeling methods, and supplemental methods descriptions were not universally provided. For complex models where only prior publications detailed certain portions of a model, our assessment of a given analysis may be limited. This limitation highlights the potential benefit of having standardized reporting elements. Fourth, comprehensively outlining practical solutions for challenges that could arise in the process of external validation, as well as approaches to sensitivity analyses and calibration, would be useful to modelers, though are beyond the scope of this review.

Conclusions

In this scoping review of studies incorporating HIV transmission in cost-effectiveness analyses of HIV interventions among AYA, we identified a wide range of modeling approaches to extrapolating long-term health and cost impacts from reduced HIV transmission. Reporting of modeling methodology could be strengthened. Improved reporting on specific model features and the reasons behind them would facilitate assessment and comparison of models’ strengths, limitations, results, and implications.

Supplementary Information

Below is the link to the electronic supplementary material.

Declarations

Funding

This work was supported in part by the National Institute of Health through the Eunice Kenney Shriver National Institute of Child Health and Human Development Adolescent Medicine Trials Network for HIV/AIDS Interventions Modeling Core (1UM2HD111102, terminated 21 March 2025, to A.M.N. and R01HD111355, to A.M.N.), the MGH Department of Medicine Transformative Scholars Award and the Executive Committee on Research Claflin Scholar award (both to A.M.N.), and the James and Audrey Foster MGH Research Scholar Award (to A.C.). The funders had no control over the design or conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication. The National Institutes of Health had no role in the design and conduct of this study.

Conflict of interest

None of the authors report any potential conflicts of interest.

Availability of data and material

All data used in this study are presented in the text of the paper or the supplemental materials.

Ethics approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

Not applicable.

Code availability

Not applicable.

Author contributions

All authors contributed substantively to this manuscript in the following ways: study design (W.C. and A.N.), search strategy (W.C., C.M., and A.N.), article screening (W.C. and S.L.), data extraction (W.C.), synthesis of findings (W.C. and A.N.), interpretation of results (all authors), drafting the manuscript (W.C. and A.N.), critical revision of the manuscript (all authors), and final approval of submitted version (all authors).