Simulation modeling for the design and evaluation of HIV drug resistance interventions: a systematic review of literature

Abstract

Background

The utility of antiretroviral therapy (ART) in the management of human immunodeficiency virus (HIV) is being challenged by growing HIV drug resistance (HIVDR). Although simulation modelling is useful for understanding complex problems, the extent to which it is used in HIVDR is unknown. This review aimed to determine how modelling has been used to inform HIVDR interventions and how its use can align with the World Health Organization (WHO) Global Action Plan (GAP) for HIVDR 2017-2021.

Methods

This review involved a literature search across PubMed, Scopus, Web of Science, and Embase databases. Articles published after the introduction of ART, 1997 to 28th November 2025, were considered. Relevant information, including metadata, model descriptions, interventions, and their outcomes, was extracted. Findings from included papers were categorized according to their area of focus within the five strategic objectives of the WHO GAP for HIVDR 2017–2021.

Results

A total of 2346 articles were screened, and 17 articles were included in the final analysis. Most studies modelled HIVDR in sub-Saharan Africa (n = 13). Acquired resistance (n = 15) was assessed in most of the studies, followed by transmitted resistance (n = 7) and pretreatment resistance (n = 3). Most of the models were stochastic models(n = 11), with about one third of them analyzing cost effectiveness(n = 6). Ten models focused on the WHO GAP HIVDR strategic objective of prevention and response, four aligned with the objective of monitoring and surveillance, while the remaining three assessed a combination of the two objectives. None of the models assessed the remaining three objectives: research and innovation, laboratory capacity, or governance and enabling mechanisms.

Conclusion

This review identifies a need for more cross-cutting analyses of multiple strategic objectives for HIVDR, including system-wide models to provide holistic insights into the complexity surrounding HIVDR. Employing patient and public involvement (PPI) in model development and intervention design would further strengthen model validity and transition to real-world settings (PROSPERO ID: CRD42024553557).

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-27723-4.

Background

Antiretroviral therapy (ART) and pre-exposure prophylaxis (PrEP) are cornerstones of HIV transmission prevention and management and essential components of achieving the UNAIDS fast-track goals by 2030 [1], including attaining 95% viral suppression in persons receiving ART, thereby disrupting HIV transmission to sexual partners [2]. Over 30 million people globally were receiving ART in 2023, and approximately 73% of them were virally suppressed [3]. However, the emergence and persistence of HIV drug resistance (HIVDR) endangers the sustained success of treatment programs, eventually leading to a resurgence in HIV incidence and AIDS-related mortality [4].

ART typically consists of three different drugs chosen from one of the four main classes of anti-HIV drugs: nucleoside reverse transcriptase inhibitors (NRTIs), non-nucleoside reverse transcriptase inhibitors (NNRTIs), protease inhibitors (PIs), and integrase strand transfer inhibitors (INSTIs) [5]. In 2022, five million people living with HIV-1 (PLHIV) were estimated to have experienced failure to one or more ART regimens [6]. Furthermore, more than 10% of people initiating first-line ART experience pretreatment drug resistance to NNRTIs; the proportion climbs to more than 50% in treatment-experienced people [7]. In response to rising resistance, in 2018, the World Health Organization (WHO) recommended a change in the first-line regimen by replacing NNRTIs with the INSTI, dolutegravir (DTG). However, recent reports suggest DTG resistance levels are already as high as 19.6% in long-term ART patients who switched to DTG-based ART [8].

In 2017, the WHO released the Global Action Plan(GAP) on HIV Drug Resistance to steer synergistic efforts towards managing and preventing HIVDR as a potential threat to achieving global HIV targets. The global plan includes five strategic objectives (SO): (1) prevention and response; (2) monitoring and surveillance; (3) research and innovation; (4) laboratory capacity; and (5) governance and enabling mechanisms [9].While many global guidelines have been developed for the management and prevention of HIV/AIDS through the various stages of the HIV care continuum, some including sections on HIVDR, this global action plan stands out as a guideline that is specific to HIVDR. The extent to which these strategic objectives can be met, intervention design, and the most appropriate places for intervention, remain a challenge for countries seeking to adopt plans to tackle HIVDR.

Chronic illnesses like HIV are often determined by a range of contributing risk factors, making it difficult for decision makers to frame the entire problem while targeting effective solutions [10]. Simulation modelling can be a valuable decision support tool to aid in understanding complex problems, as well as test scenarios for the outcomes of proposed health interventions and associated costs, thereby avoiding risks to patients and supplementing time-consuming and costly traditional randomized controlled trials (RCTs) in evidence-based medicine [11, 12]. In the context of HIVDR, simulation models can capture the complex dynamics of resistance emergence and transmission, evaluate interventions under different adherence and coverage scenarios, and project long-term population-level outcomes that cannot be directly measured in empirical studies [13–15].

Furthermore, the decision-making process itself can be complicated by external factors that challenge the adoption of research evidence, including resource constraints and changing political priorities; modelling can provide a low-bias perspective of the benefits and costs of interventions and protect against ineffective decisions [10]. Modelling can be used to simulate disease progression in populations over time, analyze medium- and long-term outcomes of various policies, and provide benefits in understanding complex systems with competing interests such as HIVDR [16, 17]. In addition, modelling removes barriers such as time limitations, costs, and safety risks associated with empirical research. Different aspects of the healthcare system can be modelled, including patient flows in health facilities, resource allocation, disease transmission, and cost-utility [18]. By enabling counterfactual comparisons and mechanistic evaluation of interventions, simulation models can directly inform strategies to mitigate HIVDR and guide resource allocation [13–15].

The alarming propagation of HIVDR demands the development and rollout of effective interventions [19]. To our knowledge, no systematic reviews of the literature have been performed examining the use of simulation modelling for a better understanding of interventions to reduce the impact of HIVDR [20–22]. This review aims to examine how population-level simulation models have been used to design and evaluate interventions for mitigating HIVDR, and in doing so, synthesize existing literature on modelling approaches. The review assesses modelling recommendations for intervention strategies and evaluates the extent to which these recommendations meet global strategic objectives for HIVDR.

Materials and methods

A systematic search of the literature was carried out in PubMed, Scopus, Web of Science, and Embase databases. The search was guided by the defined Population, Intervention, Comparison, Outcomes and Study (PICOS) framework and used MeSH/Emtree terms adjusted to databases for collections of terms covering HIVDR interventions, and modelling. Since the focus of the review was HIVDR, articles published before the introduction of ART in 1997 were excluded.

We restricted inclusion to English-language publications because mathematical modeling research in infectious diseases is overwhelmingly published in English. A bibliometric analysis of infectious disease modeling studies found that approximately 97.2% of publications were in English [23]. Additionally, our preliminary scoping search did not identify relevant non-English modeling studies, suggesting a low risk of language-related selection bias.

The searches were conducted on the 28th of November 2025. Duplicates were removed, and the selection of relevant publications continued according to the inclusion/exclusion criteria listed below. The full search strategy is available in Additional File 1. A two-stage screening process was applied in accordance with the pre-defined inclusion/exclusion criteria (Table 1). Two reviewers (BA and GB) first screened the titles and abstracts. This was followed by three reviewers (BA, DM, and LG) screening full texts. When ambiguity regarding full text inclusion arose, a vote was conducted among the three reviewers, with a decision made by a 2-to-1 majority ruling. Reasons for article exclusion at the full-text stage were recorded. A summary of the process is outlined in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) diagram in Fig. 1. A protocol summary is registered on PROSPERO (ID: CRD42024553557). Traditional risk-of-bias assessment frameworks are not directly applicable to mathematical modeling studies, as the nature of potential biases differs from those typically assessed in empirical research.

Table 1.

Summary of inclusion and exclusion criteria

Inclusion Criteria

Exclusion Criteria

1. Any mathematical or computational models simulating HIVDR interventions and considering transmission at the population level 2. Studies involving any population affected by HIV, regardless of geographical location or demographic characteristics 3. Reported outcomes focus on the epidemiological impact of HIV drug resistance (e.g., prevalence of resistant strains, transmitted resistance, and treatment failure rates) 4. English language

1. Modelling studies focusing on general HIV epidemiology without a specific focus on drug resistance interventions 2. Within-host models of disease progression, immune response, or treatment outcomes 3. Pharmacokinetic-pharmacodynamic models 4. Molecular modelling studies 5. Descriptive statistical analyses without a simulation modelling component (e.g., regression) 6. Reviews, opinion pieces, commentaries, letters to the editor, and any other publication types that do not present original work 7. Literature published before the introduction of ART in 1996

Fig. 1.

PRISMA flow diagram displaying detailed database searches and screening process

Data extraction

Relevant data were extracted and categorized from selected articles into a Microsoft Excel spreadsheet by the first reviewer (BA) and checked by a second reviewer (LG). The following information was retrieved from the papers: metadata (authors, institutional affiliations, funding, publication dates), location (country), setting (school/family, hospital, community, farm, etc.); data sources (cohort, literature, expert interviews); model type (stochastic, deterministic model); model classification (individual-based vs. compartmental, ); time horizon (years, months, days), intervention type (pharmacological, behavioral), type of resistance addressed (acquire resistance vs. transmitted resistance), measures of outcomes (mutation rate, prevalence of resistance); intervention description and; intervention outcome (success, failure, inconclusive). In this context, acquired resistance refers to resistance that develops in an individual during or after treatment, whereas transmitted resistance refers to resistance acquired directly from another person at infection [24]. We also did contextual mapping of the level the interventions were modelled, whether micro (institutional level) or macro (regional/national level).

Domain analysis

The studies were classified according to the strategic objectives of the WHO Global Action Plan (GAP) on HIV drug resistance 2017–2021 [9]. The plan defines five strategic objectives for combating HIV drug resistance: prevention and response, monitoring and surveillance, research and innovation, laboratory capacity, and governance and enabling mechanisms (Table 2). We added a category for “other” interventions that did not specifically fit the five categories. We also explored the potential overlap of interventions between categories and their interconnectedness.

Table 2.

Summary strategic objectives from the WHO Global Action Plan (GAP) on HIVDR

Gap strategic objective		Description
1	Prevention and response	Strengthening ART service delivery in the treatment cascade including: HIV testing, HIV treatment and care services. This involves ensuring the availability and appropriate use of antiretroviral (ARV) drugs, adequate adherence support and efforts towards maximizing retention in care.
2	Monitoring and surveillance	Implementing systems to regularly monitor HIVDR trends through surveillance, conducting periodic HIVDR surveys, encouraging routine viral load and HIVDR testing as well as monitoring quality of routine patient care delivery.
3	Research and innovation	Urges investment into relevant research to fill knowledge gaps in understanding HIVDR dynamics, create novel diagnostics and design innovative interventions to increase viral load suppression and contain HIVDR.
4	Laboratory Capacity	Support enhancement of laboratory infrastructure for the expansion of viral load and HIVDR monitoring to allow accurate detection and quantification of HIVDR as well as prompt reporting and use of results.
5	Governance and enabling mechanisms	Establish adequate governance and enabling mechanisms (country ownership, advocacy and sustainable funding) to facilitate action against HIVDR.

Results

A total of 2346 scientific articles were identified through the database searches. After removing duplicates, 1701 articles were screened by title and abstract, which led to the removal of a further 1593 articles. Full-text article screening began with 108 articles.

Upon reviewing these, 91 of the articles did not meet the inclusion criteria, leading to a final of 17 articles included in this study (Fig. 1) below.

Most of the studies (n = 12) were done before the WHO introduced the “treat all” HIV treatment policy in 2015, which began the movement toward immediate initiation of ART for people with confirmed HIV infection, as opposed to the previous immunity-based initiation, where treatment was guided by CD4 counts or clinical symptoms [25]. With regard to HIVDR, eight of the studies addressed acquired resistance, two addressed pretreatment resistance, two addressed transmitted resistance, four addressed both acquired and transmitted resistance and one addressed acquired, transmitted as well as pretreatment drug resistance. All the studies used secondary data from cohort studies, databases, or scientific literature for parameterization and validation, with no primary data collection involved in the modelling process. The simulation time for the models ranged from one year to a lifetime.

Over half (n = 10/17) of the models identified by this review were stochastic models, while the rest were deterministic models with both compartmental and individual population distributions. Various analyses were applied to the models; however, over a third of the studies (n = 6/17) were cost-effectiveness analyses. The models that did not focus on cost-effectiveness analyzed the effects of various interventions on disease epidemiology and progression. The majority (n = 13/17) of the studies were in sub-Saharan Africa, with studies of the whole region (n = 1), South Africa (n = 9), Kenya (n = 2), Uganda (n = 1), and Zimbabwe (n = 1). Four (n = 4) studies were set outside Africa, one in the USA exclusively, one in Southeast Asia, a multinational study in Europe, and one multinational study set in Europe, the Americas, and Australia.

The studies included a range of scenarios with impacts on HIVDR in the care continuum. Over three quarters (n = 11/17) of the studies focused on determining the best choice of regimen and/or scheduling of pharmacological therapy with ART. Four (n = 4) of the studies were treatment monitoring interventions that involve pretreatment diagnostics and/or in-treatment diagnostics, including CD4 counts, viral load tests, or drug resistance tests, for the aim of informing ART initiation time or choice and switch to second-line ART. One study assessed a target group drug allocation intervention, and one assessed an adherence intervention. These interventions have been categorized according to the strategic objectives of the WHO GAP on HIVDR in detail below.

SO: prevention and response

Of the 17 studies included in the review, 10 assessed aspects that aligned with the WHO GAP HIVDR for prevention and response. Of these, four studies modelled ART initiation strategies, three studies did cost-effectiveness analysis for different ART, one study modelled the utility of PrEP, one study modelled adherence interventions, and one study modelled drug allocation strategies. Within this category, studies addressed HIV prevention strategies, ART initiation strategies, treatment optimization, regimen switching, and non-pharmacological interventions.

Prevention strategies

Combined prevention strategies involving ART and PrEP reduced HIV incidence but had complex implications for resistance. While combined ART and PrEP rollout prevented more infections than either strategy alone, it was associated with higher prevalence of drug resistance, particularly when PrEP was initiated in individuals with undiagnosed HIV infection [26]. Use of non-overlapping drug regimens between ART and PrEP reduced resistance only modestly, as ART remained the primary driver of resistance [26].

ART initiation strategies

Findings on ART initiation strategies were mixed. ART initiation strategies focused on the choice and timing of first-line regimens. A multi-country model in sub-Saharan Africa demonstrated that first-line DLT initiation without prior NNRTI resistance testing was cost-effective across different groups of new ART initiators [27].

Comparisons between “treat all” and CD4-based ART initiation strategies showed mixed effects on HIVDR. Some models found that targeted CD4-based ART initiation performed similarly or better than universal ART in limiting resistance [28, 29]. However, when universal ART was combined with additional interventions such as voluntary medical male circumcision (VMMC) and adherence support, lower overall resistance prevalence was observed compared to CD4-based strategies [28].

In contrast, other models found no significant difference in NNRTI resistance between “treat all” and CD4-based initiation, particularly among unsuppressed individuals (> 500 copies/ml), and highlighted that improved diagnosis and retention could allow CD4-based initiation (< 350 cells/ml) to achieve similar incidence reductions [29].

Expanding CD4 thresholds for ART initiation was also associated with reduced new infections and transmitted resistance, but required enhanced virological monitoring and access to second-line therapies to prevent HIVDR in the expanded treatment population [30].

Treatment optimization

Treatment optimization studies focused on individuals with existing drug resistance. Two studies conducted in Europe, the USA, Latin America, and Australia evaluated the cost-effectiveness of darunavir (DRV) and lopinavir (LPV) based protease inhibitor regimens among treatment-experienced PLHIV with PI resistance and found that DRV-based regimens were more cost-effective in this population [31, 32].These findings were influenced by population characteristics, as one study assumed LPV-naïve individuals [31] while the other included participants with prior LPV exposure [32].

ART regimen switch strategies

Switching strategies from first-line to second-line ART were identified as critical determinants of resistance. Models showed that switching to second-line therapy following first-line failure reduced NNRTI resistance, particularly when implemented promptly [33]. This analysis also identified the magnitude of ART rollout and low frequency of monitoring of first-line treatment failure as key drivers of resistance, and highlighted that continued use of NNRTI-based first-line regimens limits the ability to fully prevent resistance emergence to this ARV drug class [33].

Non-pharmacological interventions

Non-pharmacological interventions demonstrated variable effects on HIVDR. Modified directly observed therapy (MDOT), an adherence intervention modelled in a USA population, improved viral suppression and reduced mortality among individuals achieving high adherence (~ 90%), but did not prevent and may increase HIVDR [34]. Similarly, drug allocation strategies influenced resistance patterns, with urban-only allocation associated with the lowest levels of transmitted resistance compared to population-based and HIV prevalence–based urban–rural distribution strategies [35].

SO: monitoring and surveillance

Four studies modelled scenarios related to monitoring and surveillance of HIV progression and treatment response as a method of controlling the emergence and propagation of resistance.

Overall, these studies highlight that the type, frequency, and use of monitoring tools influence both treatment outcomes and resistance patterns.

A study from Zimbabwe demonstrated that viral load (VL) monitoring outperforms other monitoring strategies, with the lowest levels of transmitted drug resistance observed among individuals initiating ART. The addition of resistance testing as a confirmatory tool did not significantly reduce resistance further [36].

A model from Kenya assessed pretreatment drug resistance (PDR) testing among women not initiating DTG-based first-line regimens due to safety concerns. The study found that PDR testing improved viral suppression rates among women with PDR, decreased PDR prevalence more rapidly, and increased quality-adjusted life years (QALYs) compared to ART strategies without PDR testing. Despite these benefits, neither of the two methods modeled was found to be cost-effective due to the higher cost of PI-based regimens used in women identified with PDR [37].

A study in Southeast Asia examining monitoring frequency found that more frequent viral load testing (e.g., quarterly) resulted in the lowest transmitted resistance, although less frequent testing, even once every two years, still provided benefit. Notably, the model emphasized that availability of second-line and subsequent therapies for individuals failing first-line ART is essential to reducing resistance among treatment-naïve individuals [38].

The final study evaluated switching from first-line dolutegravir-based regimens to protease inhibitor-based second-line regimens, demonstrating that earlier switching (within 6 months of viraemia) reduced resistance more effectively than delayed switching (6–12 months). Genotypic resistance–informed switching provided only modest additional benefit compared with immediate switching without testing, highlighting the importance of timely action within routine monitoring systems [39].

SO: research and innovation, laboratory capacity, and governance and enabling mechanisms

No studies in this review were found to examine research and innovation, laboratory capacity, or governance and enabling strategies.

Transversal studies

Three studies examined scenarios spanning multiple WHO GAP HIVDR strategic objectives, integrating prevention and response as well as monitoring and surveillance strategies in South Africa. These studies demonstrated that combinations of ART regimen choice, diagnostic testing, and monitoring strategies influence both cost-effectiveness and resistance outcomes. For example, viral load monitoring at 6 months became more cost-effective over time as NNRTI resistance increased, while PI-based regimens were only cost-effective in settings where higher willingness-to-pay thresholds are acceptable [40].

Rosen et al. showed that confirmatory resistance testing reduced unnecessary switching to second-line therapy in patients with virologic failure, although overall costs across strategies remained similar [41]. In the final study, long-acting injectable cabotegravir and rilpivirine were evaluated as alternatives to oral dolutegravir-based regimens for both initiation and second-line therapy. Results indicated higher INSTI and rilpivirine resistance with injectable use compared to oral DTG-based ART first line, although genotype-informed eligibility could mitigate this risk. The injectable regimen showed modest benefits in reducing resistance among patients with adherence challenges [42] (Table 3).

Table 3.

Characteristics of modelling studies identified by the systematic review

No	Author (Year)	Model description	Location & setting	SO	Simulated time	Type of resistance	Intervention description	Intervention outcome
1	Phillips et al. (2018) [27]	HIV Synthesis Model - Stochastic microsimulation model using synthetic population Cost effectiveness	Sub Saharan Africa Macro (health sector perspective)	1	20 years (2018 start)	PDR	Selection of DLT-based VS non-DLT-based ART regimen in prior ART-exposed vs. new initiators, with or without being informed by resistance testing results	1st line DLT in all ART initiators is the lowest cost policy with least percentage of occurrence of NNRTI resistance and highest number of suppressed people
2	Phillips et al. (2014)[40]	HIV Synthesis Model - Stochastic microsimulation model using synthetic population Cost effectiveness	South Africa Macro (health sector perspective)	1&2	15 years (2017 start)	ADR	Comparison of NNRTI-based VS PI-based first-line ART in new initiators, with or without resistance testing, and monitoring strategies using CD4, VL testing, or both	VL testing at 6 months found to be cost effective with increasing NNRTI resistance. PI-based regimens only favorable as first line regimens at the highest cost-effectiveness threshold.
3	Kagay et al. (2004) [34]	Markov Chain Monte Carlo Simulation model	USA Meso	1	1 year	ADR	Effects of an MDOT adherence intervention on adherence, virus load and drug resistance.	MDOT programs achieving 90% adherence can enhance virological suppression. However, they may not prevent drug resistance in the population.
4	Brogan et al. (2010) [32]	Markov Model adapted from Mauskopf et al., 2010 Cost effectiveness	Multinational study: Europe, America, Australia Macro (US societal perspective)	1	Lifetime	ADR	Treatment with DRV/r 600/100 mg bid VS LPV/r 400/100 mg bid in treatment- experienced, PI-resistant, HIV-infected adults	DRV/r 600/100 mg bid is cost-effective compared to LPV/r 400/100 mg bid
5	Moeremans et al. (2010) [31]	Markov Model adapted from Mauskopf et al., 2010 Cost effectiveness	Multinational European study: Belgium, Italy, Sweden and UK Macro(Public payers’ perspective)	1	Lifetime	ADR	Treatment with DRV/r 600/100 mg bid VS LPV/r 400/100 mg bid in treatment experienced, LPV naïve, PI-resistant, HIV infected adults	DRV/r 600/100 mg bid is cost-effective compared to LPV/r 400/100 mg bid
6	Nichols et al. (2014) [30]	Compartmental deterministic model	Kampala, Uganda, and Mombasa, Kenya Micro	1	10 years (2012 start)	TDR	ART initiation at less than 200 CD4 microcells per ml VS less than 350 microcells per ml VS less than 500 microcells per ml.	Averted HIV infections due to expanded treatment eligibility are predicted to offset the risk of increased transmitted resistance. Simultaneous efforts to increase VL testing and 2nd line drug accessibility should be done.
7	Abbas et al. (2019) [28]	Stochastic microsimulation Bayesian model extended from Glaubius et al., 2016	South Africa Micro	1	52 years (1978 start)	ADR	CD4-based ART initiation VS Universal ART initiation scenario with adherence-support intervention	Total resistance from the implementation of universal ART initiation, is less compared to that from CD4-based ART
8	Hauser et al. (2019) [33]	Extended from MARISA - deterministic compartmental model	South Africa Macro	1	11 years (2005 start)	ADR TDR	Effect of time of adoption of “Treat all” policy, drug resistance testing and immediate second line treatment in individuals harboring a resistant strain	Emergence and spread of NNRTI resistance is inevitable with NNRTI-based first line regimens. Two main resistance drivers are magnitude of the ART roll-out and low frequency of monitoring of first-line treatment failure.
9	Phillips et al. (2014) [36]	HIV Synthesis Model - Stochastic microsimulation model using synthetic population	Zimbabwe Macro	2	10 years (2015 start)	ADR TDR	Monitoring with VL alone VS VL and resistance testing, with either no regimen switch or a switch to second-line therapy after failure	Viral load monitoring without confirmation provided highest DALYs averted and lowest rate of transmitted drug resistance
10	Hoare et al. (2010) [38]	Compartmental deterministic model	Southeast Asia Macro	2	10 years	ADR TDR	Effect of frequency of VL testing on drug resistance(DR): once every 2 years, every year, twice yearly, or quarterly.	VL monitoring quarterly in ART users is associated with the lowest prevalence of DR although modest strategies are also beneficial. Availability of subsequent lines of therapy for those who have failed first line regimens is required to reduce prevalence of DR among treatment naive individuals
11	Abbas et al. (2013) [26]	Compartmental deterministic model	South Africa Macro	1	10 years (2003 start)	ADR	Impact of orally administered overlapping and nonoverlapping PrEP and ART on HIV transmission and drug resistance	Combined ART + PrEP prevented HIV infections than either strategy individually but with higher resistance risk. This risk can be reduced modestly with non-overlapping ARV although ARTs remain main drivers of resistance.
12	Rosen et al. (2011) [41]	Markov model	South Africa Meso	1&2	6 years (2004 start)	ADR	Costs of different monitoring strategies and regimen switch decisions: routine VL monitoring, resistance testing and limited testing	Confirmatory resistance testing led to fewer individuals with virological failure without mutations switching to second line regimen. All three strategies had similar net costs
13	Duarte et al. (2020) [37]	Stochastic microsimulation model adapted from the Standford Health Policy model Bendavid et al., 2008 Cost effectiveness	Kenya Macro (health sector perspective)	2	15 years (2019 start)	PDR	Cost effectiveness of No pre-treatment drug resistance (PDR) testing Vs. PDR testing with oligonucleotide ligation assay (OLA) VS. PDR testing with consensus sequencing (CS)	None of the tests are cost-effective. PDR testing improves rates of viral suppression among women with PDR and leads to more rapid decrease in PDR prevalence and leads to increases in QALY in Kenya compared to ART strategies without PDR testing.
14	Cambiano et al. (2014) [29]	HIV Synthesis Model - Stochastic microsimulation model using synthetic population	South Africa Macro	1	20 years (2012 start)	ADR	Compare projected levels of resistance in the context of expanded eligibility criteria for ART, and enhancements in diagnosis and retention	Prevalence of NNRTI resistance in unsuppressed individuals not significantly different with the “treat all” policy.
15	Wilson et al. (2006) [35]	Compartmental deterministic spatial model	South Africa Meso	1	4 years (Start 2004)	TDR	Drug allocation Strategy (DAS): Population-Based Allocation of ART Vs. Prevalence-Based allocation Vs. Urban only allocation	Urban-only DAS generates the lowest levels of transmitted resistance
16	Loosli et al. (2025) [39]	Extended from MARISA - deterministic compartmental model	South Africa Macro	2	30 years (Start 2005)	ADR TDR	Comparing projected levels of resistance with immediate vs. genotypic resistance testing informed switch from dolutegravir based regimen to PI based regimen while considering length of viraemia	Both immediate and genotypic resistance informed switching reduced resistance, with slightly greater benefit from genotypic guidance. Additional benefit is seen with shorter time with viraemia.
17	Han et al. (2025) [42]	Extended from MARISA - deterministic compartmental model	South Africa Macro	1&2	40 years (Start 2005)	ADR PDR TDR	Comparing projected levels of resistance between dolutegravir based regimen and long-acting injection (LAI) cabotegravir and rilpivirine for initiation and switch with or without genotype guidance	LAI cabotegravir and rilpivirine is predicted to lead to higher INSTI and rilpivirine resistance compared with continued use of oral dolutegravir-based ART. Genotype-dependent eligibility further reduces the increases in resistance levels, especially for rilpivirine

Discussion

Use of modelling for informing policies and interventions on HIVDR remains modest and mostly confined to single-strategy objectives within the WHO GAP HIVDR. This review found that of the seventeen studies assessing HIVDR, only three included cross-cutting, transversal analyses that addressed multiple strategic objectives. Given the complexity and expense of conducting randomized controlled trials, the relatively low use of modelling to inform HIVDR interventions represents a missed opportunity to complement empirical evidence and explore potential outcomes beyond trial time horizons. As a positive outcome, six studies are adaptations of previously developed models showing that when robust, well-documented models are available, these can be used and reused to continue to inform policy decisions. Leveraging existing models can further save time and resources, even where capacity for developing models may be limited. Notably, the studies that are included in this review explore several scenarios relevant to reducing HIVDR and achieving the strategic objectives, including: ARV choices in first- and second-line therapies, ART initiation policies, PrEP, adherence interventions, drug stock allocation strategies, as well as ART monitoring strategies and their effect on acquired, transmitted, or pretreatment resistance.

Generally, expanded ART initiation, DLT-based first-line regimens, PI-based second-line regimens, use of PrEP in at-risk populations, regular monitoring of viral load, and timely switch from first-line regimen to second-line regimen when resistance is detected were recommended to reduce the occurrence as well as propagation of HIVDR. Decision-makers looking to use more modelling are not obliged to start from scratch. Six of the studies in this analysis were done with models that were adapted to different settings or updated with new input data [43–45].

The model outcomes across several different areas relevant to HIVDR have been consistently supported by findings from observational studies including timing of ART initiation [46], use of DLT as a first line ART [47–49], use of PI based second line ART [50–52], use of combined PrEP and ART [53], pretreatment resistance testing with long acting injectables cabotegravir [54] and routine viral load testing and HIVDR surveillance [49, 55, 56]. These studies validate the findings and prolonged relevance of the modelling studies included here and indirectly support the utility of modelling as a flexible, fast, low-risk, and relatively inexpensive way of evaluating policies for HIVDR [46–53, 55, 56].

The utilization of inferences from modelling studies observed aligns with recommendations of the WHO GAP to use quantitative modelling and cost-effectiveness analysis as tools to inform decision-making on the impact of interventions to prevent and respond to HIVDR [9]. It is encouraging that researchers have modelled HIVDR interventions from as far back as 2004, and some have even made projections beyond the 2030 FastTrack timepoint. On the other hand, none of the models include broad model boundaries encompassing biological, behavioral, socioeconomic, cultural factors, clinical care and delivery or governance and policy mechanisms associated with wicked problems [57]. Furthermore, some of the areas of leverage identified by researchers for developing interventions against HIVDR, such as stigmatization and poverty, remain unexplored in these models [58].To add on, the conclusions of cost-effectiveness studies become less applicable as costs change over time. This demonstrates the necessity of incorporating systems thinking to adequately and holistically represent the complexities of HIVDR while also designing and testing interventions that align with identified leverage points and reflect the complexity of the problem.

The models themselves are instrumental to developing policies, as evidenced by the fact that three modelling studies in this review were referenced in the WHO GAP HIVDR [4, 29, 59]. Furthermore, the updated action plan released in 2021 [60] includes the same modelling studies, together with literature reviews that included modelling papers [61–63]. Modelling provides a powerful tool not only to inform how interventions may be planned, but also how targets and strategies are developed and operationalized. It allows for the simulation and refinement of innovations such as new long-acting ART and differentiated service delivery strategies. Additionally, modelling can be utilized to explore the effects of funding disruptions on ART availability and the emergence of HIVDR.

Nevertheless, the use and uptake of modelling can be difficult to ascertain. Although these studies were referenced in the action plan, it was not obvious from their titles that they were modelling studies and required one-by-one searching and screening of the abstracts of the papers to determine the study types. This approach may be difficult to replicate with a guideline with more references, and it is very time-consuming. The FAIR (findability, accessibility, interoperability, and reusability) principles for research data stewardships emphasize the need for adoption of standardized methods of reporting research to promote wider access and reuse of this data, as well as reduce research gatekeeping [64]. Any adaptation, extension, or future use of modelling requires adherence to the FAIR guidelines and a minimum standard of reporting on the methods, parameters, and data used to be truly reproducible.

There remain many areas of the WHO GAP HIVDR that have not been incorporated into modelling or evaluated in modelling studies. Of the five strategic objectives, only two have been explored by the models in this review; prevention and response were the overwhelming focus of modelling studies, and to a lesser degree, monitoring and surveillance. Unfortunately, this means that no modelled scenario testing of HIVDR interventions for any of the other three strategic objectives of laboratory capacity, research and innovation, as well as governance and enabling mechanisms has been done. Paradoxically, the objectives not modelled could act as critical roadblocks in the success of the interventions that have been prioritized in HIVDR modelling research. For example, low laboratory capacity could hinder the success of any intervention tailored towards monitoring of resistance through routine viral load and/or CD4 cell counts. This narrow view on the modelling boundaries means that critical insights are missed, and conclusions that are drawn would be erroneous. Additionally, governance and enabling mechanisms such as monitoring of health policy implementation and resource management, although vital in improving health systems, were not modelled [65]. This observation underscores a lack of a system-wide perspective and system-based modelling encompassing all five objectives, which could facilitate more robust recommendations.

The limited inclusion of the strategic objectives of research and innovation, laboratory capacity, and governance and enabling mechanisms in modelling studies likely reflects methodological challenges rather than omission [23, 66]. Governance processes are inherently complex, involving coordination across multiple levels of decision-making with differing priorities, timeframes, and contexts, and are further constrained by imprecise and difficult-to-measure indicators, making causal relationships hard to establish [67, 68]. Laboratory capacity is similarly difficult to represent in models, as improvements often stem from changes in work processes rather than new tools or infrastructure, making it appear as a less impactful component. It is further shaped by operational uncertainties such as equipment failures and limited access to technical expertise for maintenance, particularly in LMICs, rendering system performance difficult to predict [69–71]. Research and innovation are also poorly suited to traditional modelling approaches due to their non-linear, iterative, and often unpredictable nature, and reliance on external funding streams in many settings [72, 73]. However, mathematical modelling may still contribute to this domain through approaches such as Value of Information (VOI) analysis, which can support prioritization of future research investments, particularly in resource-limited settings [74, 75]. Across all three strategic objectives, limitations in the availability and quality of data restrict the ability to accurately capture these areas [66]. These challenges may help explain their limited representation in modelling studies, although this remains an area for further investigation.

Another critical observation to mention is that over 80% (n = 14/17) of the studies were set in sub-Saharan Africa. This is closely consistent with the infection epidemiology in the region, which estimates that more than two-thirds (67%) of the HIV burden is in the region [76]. However, the studies in the region are disproportionately distributed, with South Africa accounting for 10 of the 14 models. Although South Africa has the highest number of PLHIV, the HIV prevalence is greatly exceeded by other countries in the region, such as Eswatini and Lesotho, where models were absent [77]. Outside of the region, high burden areas like Southeast Asia [38] and South America (no studies) were also underrepresented. This uneven concentration underscores regional and global underutilization of modelling as policy-informing tools that can lead to advancements as we draw closer to the fast-track targets in 2030 [78].

Finally, none of the models developed included “patient and public involvement (PPI)” in the design of the interventions, nor the development of the model. While including these perspectives can be challenging, there are established and validated methods for doing so that benefit both researcher and stakeholders [79]. Major deficiencies in clinical care delivery have spurred the idea of partnering with patients and the public as important stakeholders that can influence the success of interventions [80]; this is no different for the development of models. In practice, implementation issues are known to arise due to contextual issues that researchers, funders and decision makers may not have considered during intervention design [81]. This likely also influences the assumptions that are made by researchers when designing and simulating interventions in mathematical models. The limited PPI in HIVDR intervention modelling represents a missed opportunity for intervention and model co-creation that can contribute to model validity and lead to a more seamless transition to implementation [82].

Conclusions

Simulation studies have been used to design and evaluate interventions for HIVDR, identifying key strategies to reduce resistance, including expanded ART initiation, DTG-based first-line regimens, PI-based second-line therapy, PrEP for at-risk populations, regular viral load monitoring, and timely switching following detection of resistance. These findings both reflect and, in some cases, have informed current clinical practice, highlighting the value of simulation modelling in supporting and shaping HIVDR intervention strategies while enabling evaluation of their population-level impact.

Despite these advances, gaps remain in the extent to which modelling addresses the strategic objectives of the WHO Global Action Plan (GAP) on HIVDR. Future modelling efforts could benefit from cross-cutting approaches that enable system-wide evaluation of HIVDR strategies, inherently capturing their complexity and supporting broader geographic coverage. In parallel, incorporating patient and public involvement (PPI) in the development of models and intervention prototypes may enhance contextual relevance, feasibility, and stakeholder buy-in. In addition, improving model accessibility and transferability may support researchers in building on existing work and strengthening the cumulative evidence base. Together, these advances could enhance the utility of modelling for informing HIVDR policy, research, and clinical decision-making.

Abbreviations

ADR

Acquired Drug Resistance

ARV

Antiretrovirals

ART

Antiretroviral Therapy

DRV

Darunavir

DTG

Dolutegravir

FAIR

Findability, Accessibility, Interoperability, and Reusability

GAP

Global Action Plan

HIVDR

HIV drug resistance

INSTI

Integrase Strand Transfer Inhibitors

LPV

Lopinavir

MDOT

Modified directly observed therapy

NNRTI

Non-Nucleoside Reverse Transcriptase Inhibitors

NRTI

Nucleoside Reverse Transcriptase Inhibitors

PLHIV

People Living With HIV-1

PICOS

Population, Intervention, Comparison, Outcomes and Study

PPI

Patient and Public Involvement

PrEP

Pre-Exposure Prophylaxis

PDR

Pretreatment Drug Resistance

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-analyses

PI

Protease Inhibitors

QALY

Quality-Adjusted Life Years

RCT

Randomized Controlled Trials

SO

Strategic Objective

TDR

Transmitted Drug Resistance

VMMC

Voluntary Medical Male Circumcision

WHO

World Health Organization

Authors’ contributions

BA led the conceptualization of the review, designed and conducted the literature searches, performed data extraction and thematic analysis, and drafted the first version of the manuscript. LG contributed to study screening, data extraction, and thematic analysis, and critically reviewed and edited the manuscript. DM and GB participated in study screening and contributed to manuscript review and editing. GN contributed to the conceptualization of the review, supported the search strategy, and provided critical input during manuscript revision. RS, NV, and CD provided overall supervision of the work and contributed to critical review and revision of the manuscript. JK provided strategic oversight and guidance on the study scope and interpretation of findings. All authors read and approved the final manuscript.

Funding

Funder: Flemish Interuniversity Council – University Development Cooperation (VLIROUS) Grant No: TZ2022TEA530A101 PI: NV URL: https://www.vliruos.be/.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.