Optimization Models for HIV/AIDS Resource Allocation: A Systematic Review

Models for malaria control optimization-a systematic review

BACKGROUND

Despite advances made in curbing the global malaria burden since the 2000s, progress has stalled, in part due to a plateauing of the financing available to implement needed interventions. In 2020, approximately 3.3 billion USD was invested globally for malaria interventions, falling short of the targeted 6.8 billion USD set by the GTS, increasing the financial gap between desirable and actual investment. Models for malaria control optimization are used to disentangle the most efficient interventions or packages of interventions for inherently constrained budgets. This systematic review aimed to identify and characterise models for malaria control optimization for resource allocation in limited resource settings and assess their strengths and limitations.

METHODS

Following the Prospective Register of Systematic Reviews and Preferred reporting Items for Systematic Reviews and Meta-Analysis guidelines, a comprehensive search across PubMed and Embase databases was performed of peer-reviewed literature published from inception until June 2024. The following keywords were used: optimization model; malaria; control interventions; elimination interventions. Editorials, commentaries, opinion papers, conference abstracts, media reports, letters, bulletins, pre-prints, grey literature, non-English language studies, systematic reviews and meta-analyses were excluded from the search.

RESULTS

The search yielded 2950 records, of which 15 met the inclusion criteria. The studies were carried out mainly in countries in Africa (53.3%), such as Ghana, Nigeria, Tanzania, Uganda, and countries in Asia (26.7%), such as Thailand and Myanmar. The most used interventions for analyses were insecticide-treated bed nets (93.3%), IRS (80.0%), Seasonal Malaria Chemoprevention (33.3%) and Case management (33.3%). The methods used for estimating health benefits were compartmental models (40.0%), individual-based models (40.0%), static models (13.0%) and linear regression model (7%). Data used in the analysis were validated country-specific data (60.0%) or non-country-specific data (40.0%) and were analysed at national only (40.0%), national and subnational levels (46.7%), or subnational only levels (13.3%).

CONCLUSION

This review identified available optimization models for malaria resource allocation. The findings highlighted the need for country-specific analysis for malaria control optimization, the use of country-specific epidemiological and cost data in performing modelling analyses, performing cost sensitivity analyses and defining the perspective for the analysis, with an emphasis on subnational tailoring for data collection and analysis for more accurate and good quality results. It is critical that the future modelling efforts account for fairness and target at risk malaria populations that are hard-to-reach to maximize impact.

TRIAL REGISTRATION

PROSPERO Registration number: CRD42023436966.

Evaluation of the use of modelling in resource allocation decisions for HIV and TB

INTRODUCTION

Globally, resources for health spending, including HIV and tuberculosis (TB), are constrained, and a substantial gap exists between spending and estimated needs. Optima is an allocative efficiency modelling tool that has been used since 2010 in over 50 settings to generate evidence for country-level HIV and TB resource allocation decisions. This evaluation assessed the utilisation of modelling to inform financing priorities from the perspective of country stakeholders and their international partners.

METHODS

In October to December 2021, the World Bank and Burnet Institute led 16 semi-structured small-group virtual interviews with 54 representatives from national governments and international health and funding organisations. Interviews probed participants' roles and satisfaction with Optima analyses and how model findings have had been used and impacted resource allocation. Interviewed stakeholders represented nine countries and 11 different disease programme-country contexts with prior Optima modelling analyses. Interview notes were thematically analysed to assess factors influencing the utilisation of modelling evidence in health policy and outcomes.

RESULTS

Common influences on utilisation of Optima findings encompassed the perceived validity of findings, health system financing mechanisms, the extent of stakeholder participation in the modelling process-including engagement of funding organisations, sociopolitical context and timeliness of the analysis. Using workshops can facilitate effective stakeholder engagement and collaboration. Model findings were often used conceptually to localise global evidence and facilitate discussion. Secondary outputs included informing strategic and financial planning, funding advocacy, grant proposals and influencing investment shifts.

CONCLUSION

Allocative efficiency modelling has supported evidence-informed decision-making in numerous contexts and enhanced the conceptual and practical understanding of allocative efficiency. Most immediately, greater involvement of country stakeholders in modelling studies and timing studies to key strategic and financial planning decisions may increase the impact on decision-making. Better consideration for integrated disease modelling, equity goals and financing constraints may improve relevance and utilisation of modelling findings.

A quantitative assessment of the consistency of projections from five mathematical models of the HIV epidemic in South Africa: a model comparison study

BACKGROUND

Mathematical models are increasingly used to inform HIV policy and planning. Comparing estimates obtained using different mathematical models can test the robustness of estimates and highlight research gaps. As part of a larger project aiming to determine the optimal allocation of funding for HIV services, in this study we compare projections from five mathematical models of the HIV epidemic in South Africa: EMOD-HIV, Goals, HIV-Synthesis, Optima, and Thembisa.

METHODS

The five modelling groups produced estimates of the total population, HIV incidence, HIV prevalence, proportion of people living with HIV who are diagnosed, ART coverage, proportion of those on ART who are virally suppressed, AIDS-related deaths, total deaths, and the proportion of adult males who are circumcised. Estimates were made under a "status quo" scenario for the period 1990 to 2040. For each output variable we assessed the consistency of model estimates by calculating the coefficient of variation and examining the trend over time.

RESULTS

For most outputs there was significant inter-model variability between 1990 and 2005, when limited data was available for calibration, good consistency from 2005 to 2025, and increasing variability towards the end of the projection period. Estimates of HIV incidence, deaths in people living with HIV, and total deaths displayed the largest long-term variability, with standard deviations between 35 and 65% of the cross-model means. Despite this variability, all models predicted a gradual decline in HIV incidence in the long-term. Projections related to the UNAIDS 95-95-95 targets were more consistent, with the coefficients of variation below 0.1 for all groups except children.

CONCLUSIONS

While models produced consistent estimates for several outputs, there are areas of variability that should be investigated. This is important if projections are to be used in subsequent cost-effectiveness studies.

Modeling the impact of different PrEP targeting strategies combined with a clinic-based HIV-1 nucleic acid testing intervention in Kenya

BACKGROUND

Up to 69% of adults who acquire HIV in Kenya seek care for acute retroviral symptoms, providing an important opportunity for early diagnosis and HIV care engagement. The Tambua Mapema Plus (TMP) trial tested a combined HIV-1 nucleic acid testing, linkage, treatment, and partner notification intervention for adults with symptoms of acute HIV infection presenting to health facilities in coastal Kenya. We estimated the potential impact on the Kenyan HIV epidemic of providing PrEP to individuals testing negative in TMP, if scaled up.

METHODS

We developed an agent-based simulation of HIV-1 transmission using TMP data and current Kenyan statistics. PrEP interventions were layered onto a model of TMP as standard of care, to estimate additional potential population-level impact of enrolling HIV-negative individuals identified through TMP on PrEP over 10 years. Four scenarios were modeled: PrEP for uninfected individuals in disclosed serodiscordant couples; PrEP for individuals with concurrent partnerships; PrEP for all uninfected individuals identified through TMP; and PrEP integrated into the enhanced partner services component of TMP.

FINDINGS

Providing PrEP to both individuals with concurrent partnerships and uninfected partners identified through enhanced partner services reduced new HIV infections and was efficient based on numbers needed to treat (NNT). The mean percent of infections averted was 2.79 (95%SI:-10.83, 15.24) and 4.62 (95%SI:-9.5, 16.82) when PrEP uptake was 50% and 100%, respectively, and median NNT was 22.54 (95%SI:not defined, 6.45) and 27.55 (95%SI:not defined, 11.0), respectively. Providing PrEP for all uninfected individuals identified through TMP averted up to 12.68% (95%SI:0.17, 25.19) of new infections but was not efficient based on the NNT: 200.24 (95%SI:523.81, 123.23).

CONCLUSIONS

Providing PrEP to individuals testing negative for HIV-1 nucleic acid after presenting to a health facility with symptoms compatible with acute HIV adds value to the TMP intervention, provided PrEP is targeted effectively and efficiently.

FUNDING

National Institutes of Health, Sub-Saharan African Network for TB/HIV Research Excellence.

Modeling the Impact of HIV-1 Nucleic Acid Testing Among Symptomatic Adult Outpatients in Kenya

BACKGROUND

Up to 69% of adults who acquire HIV in Kenya seek care before seroconversion, providing an important opportunity for early diagnosis and treatment. The Tambua Mapema Plus (TMP) trial tested a combined HIV-1 nucleic acid testing, linkage, treatment, and partner notification intervention for adults aged 18-39 years with symptoms of acute HIV infection presenting to health facilities in coastal Kenya. We estimated the potential impact of TMP on the Kenyan HIV epidemic.

METHODS

We developed an agent-based network model of HIV-1 transmission using TMP data and Kenyan statistics to estimate potential population-level impact of targeted facility-based testing over 10 years. Three scenarios were modeled: standard care [current use of provider-initiated testing and counseling (PITC)], standard HIV rapid testing scaled to higher coverage obtained in TMP (scaled-up PITC), and the TMP intervention.

RESULTS

Standard care resulted in 90.7% of persons living with HIV (PLWH) knowing their status, with 67.5% of those diagnosed on treatment. Scaled-up PITC resulted in 94.4% of PLWH knowing their status and 70.4% of those diagnosed on treatment. The TMP intervention achieved 97.5% of PLWH knowing their status and 80.6% of those diagnosed on treatment. The percentage of infections averted was 1.0% (95% simulation intervals: -19.2% to 19.9%) for scaled-up PITC and 9.4% (95% simulation intervals: -8.1% to 24.5%) for TMP.

CONCLUSION

Our study suggests that leveraging new technologies to identify acute HIV infection among symptomatic outpatients is superior to scaled-up PITC in this population, resulting in >95% knowledge of HIV status, and would reduce new HIV infections in Kenya.

A reinforcement learning model to inform optimal decision paths for HIV elimination

The 'Ending the HIV Epidemic (EHE)' national plan aims to reduce annual HIV incidence in the United States from 38,000 in 2015 to 9300 by 2025 and 3300 by 2030. Diagnosis and treatment are two most effective interventions, and thus, identifying corresponding optimal combinations of testing and retention-in-care rates would help inform implementation of relevant programs. Considering the dynamic and stochastic complexity of the disease and the time dynamics of decision-making, solving for optimal combinations using commonly used methods of parametric optimization or exhaustive evaluation of pre-selected options are infeasible. Reinforcement learning (RL), an artificial intelligence method, is ideal; however, training RL algorithms and ensuring convergence to optimality are computationally challenging for large-scale stochastic problems. We evaluate its feasibility in the context of the EHE goal. We trained an RL algorithm to identify a 'sequence' of combinations of HIV-testing and retention-in-care rates at 5-year intervals over 2015-2070 that optimally leads towards HIV elimination. We defined optimality as a sequence that maximizes quality-adjusted-life-years lived and minimizes HIV-testing and care-and-treatment costs. We show that solving for testing and retention-in-care rates through appropriate reformulation using proxy decision-metrics overcomes the computational challenges of RL. We used a stochastic agent-based simulation to train the RL algorithm. As there is variability in support-programs needed to address barriers to care-access, we evaluated the sensitivity of optimal decisions to three cost-functions. The model suggests to scale-up retention-in-care programs to achieve and maintain high annual retention-rates while initiating with a high testing-frequency but relaxing it over a 10-year period as incidence decreases. Results were mainly robust to the uncertainty in costs. However, testing and retention-in-care alone did not achieve the 2030 EHE targets, suggesting the need for additional interventions. The results from the model demonstrated convergence. RL is suitable for evaluating phased public health decisions for infectious disease control.