Rowan Martin-Hughes

Background: Antiretroviral therapy (ART) has greatly reduced HIV-related mortality and improved life expectancy in sub-Saharan Africa since the early 2000s. However, HIV-related mortality remains high. To guide intervention priorities, we used seven established HIV simulation models to identify where in the HIV care cascade deaths occur in Zimbabwe, Malawi, and South Africa. This multi-model approach provided a more comprehensive and robust assessment than could be achieved using any single modelling approach. Methods and findings: To assess alignment, each model generated key HIV metrics from 2000 to 2045, including total population, HIV prevalence, annual new infections, and proportions of people with HIV (PWH) who were diagnosed, on ART, and virally suppressed. HIV-related deaths were estimated annually and categorised by whether they occurred in individuals who were (1) undiagnosed, (2) diagnosed but had not yet started ART, (3) on ART, or (4) had interrupted ART. The models showed strong consistency with population estimates and HIV prevalence across the different settings. There was, however, variation in the absolute number of HIV-related deaths between models; in Zimbabwe in 2025, this ranged from 3,666 to 17,475; in Malawi from 4,580 to 17,835; and in South Africa from 39,470 to 114,968. In Zimbabwe, all models consistently indicated that PWH receiving ART made up the largest proportion of HIV-related deaths, though this proportion differed across models, ranging from 35% to 64%. This was followed by HIV-related deaths amongst those who had interrupted ART, ranging from 17% to 29%. Similar trends were seen in Malawi. In South Africa, a high proportion of HIV-related deaths occurred amongst people who had interrupted ART (23% to 74% in 2025) as well as those who were currently receiving ART (23% to 56%). Models are reliant on empirical data and limited data availability—in this instance, lack of national death registries—is a key constraint on modelling studies. Conclusion: Absolute numbers of HIV-related deaths vary substantially between models, highlighting wider issues around ascertainment of death in the region. However, our results do consistently show that most deaths are occurring amongst PWH on ART in Zimbabwe and Malawi, and amongst those who have interrupted treatment in South Africa, suggesting interventions around adherence counselling and retention in care need to be prioritised.

High rates of drug-resistant tuberculosis (TB) are a barrier to achieving End TB-strategy targets in Eastern Europe and Central Asia. This analysis collates results from five country-level modelling studies to identify priorities to reduce TB burden. Allocative efficiency studies were conducted in 2023 in Belarus, Kyrgyz Republic, Moldova, Tajikistan and Uzbekistan using the Optima TB model to determine the optimised distribution of funds to maximise health outcomes with given resources. A baseline scenario of continued 2022 spending was compared to scenarios with spending optimised across prevention, screening and treatment interventions to reduce TB incidence and deaths over 2024–2030. Modelled pulmonary TB incidence ranged from 25-119 per 100,000 population, and 14 − 43% of new/relapse TB cases were drug resistant. In all countries, optimizing current spending involved: expanding shorter treatment regimens (6–9 months) for drug-resistant-TB over standard regimens (18–20 months); reducing mass screening and mandatory testing and expanding community-based active case finding focused among populations at higher TB risk; and scaling-up TB preventive treatment. It was recommended to expand contact tracing in three countries and to improve cost-effectiveness in two countries by focusing on child household contacts first. With current spending optimised, pulmonary TB incidence was projected to decrease to 19 − 95 per 100,000 population by 2030, averting 1 − 13% of new/relapse TB cases and 1 − 18% of TB-related deaths from 2024-2030 compared to continued baseline spending. In three countries, optimised allocation of 150% of current spending had minimal additional epidemic impact. There are opportunities to reallocate TB funds more cost-effectively in Eastern Europe and Central Asia, but End TB targets may remain out of reach without new and prospective interventions.

Background: Characterising disparities in HIV infection across populations by gender, age, and HIV risk is key information to guide intervention priorities. We compared 9 models representing 15 different settings across Africa to assess how indicators measuring HIV acquisitions, transmissions, or potential long-term infections influence estimates of the contribution of different populations to new infections, including key populations (KPs, including female sex workers (FSW), their clients, men who have sex with men). Methods: We evaluated four indicators: I ₁ ) acquisition indicator measuring the annual fraction of all new infections acquired by a specific population, I ₂ ) direct transmission indicator measuring the annual fraction of all new infections directly transmitted by a specific population, I ₃ ) 1-year and I ₄ ) 10-year transmission population-attributable fractions ( tPAFs ). tPAFs measure the fraction of new infections averted if transmission involving a specific population was blocked over a specific time period. We compared estimates of the four indicators across 7 populations and 15 settings and assessed if the contribution of specific populations is ranked differently across indicators for 10 settings. Findings: Indicators identified distinct priority populations as the largest contributors: The acquisition indicator (I ₁ ) identified women aged 25+ years outside KPs as contributing the most to acquired infections in 8/10 settings in 2020, but to direct transmissions (I ₂ ) in only two settings. In 6/10 settings, the 10-year tPAFs (I ₄ ) identified non-KP men aged 25+ years and clients of FSW as the largest contributors to HIV transmission. Notably, non-KP women aged 15-24 years acquired (I ₁ ) more infections in 2020 (median of 1·7-fold across models) than they directly transmitted (I ₂ ), while non-KP men aged 25+ years and clients of FSWs transmitted more infections than they acquired in all but one model (median: 1·4 and 1·6-fold, respectively). Estimates of the 10-year tPAFs accounting for transmission in the long-term were substantially larger than the direct transmission indicator for all populations, especially for FSW (median: 2·0-fold). Interpretation: Indicators that reflect HIV acquisitions and transmissions over the short and long term can be utilised to capture the complexity of HIV epidemics across different populations and timeframes. The added nuance would improve the effectiveness of the HIV prevention response across all populations at risk. Funding: NIH, MRC. Research in context: Evidence before this study: Measures of the distribution of HIV acquisition across population groups are commonly used for assessing the contribution of populations to new HIV infections and prevention priorities. However, alternative indicators documented in the literature reflect transmissions or potential long-term effects. It is unclear how the choice of indicator affects the identification of populations that require additional prevention and treatment efforts to accelerate progress towards ending AIDS. We searched PubMed on March 08, 2025, with the terms (HIV) AND (Africa*) AND (acqui*) AND (transm*) AND (model*), with no language or publication date restriction, and identified no meta-assessment or mathematical model comparison studying differences in estimates of the fraction of all infections acquired and transmitted by a population when using different epidemiological indicators. Added value of this study: Using estimates from 9 models representing 15 different epidemic settings across Africa, we studied indicators of HIV epidemic contribution for 7 populations, including female sex workers, their clients, men who have sex with men, and non-key populations stratified by gender and age. We measured four commonly reported indicators of HIV contribution. One focused on acquired infections and the other three focused on transmissions. We found that estimates from these different indicators can differ greatly for the same model and population, to the extent that they identify different populations for prioritising interventions to accelerate HIV incidence declines. The acquisition-focused indicator (i.e. fraction of all infections acquired by a given population), the most used and communicated by UNAIDS, substantially underestimates the large contribution of men, and particularly male clients of female sex workers, to ongoing HIV transmission. Implications of all the available evidence: The choice of indicators measuring a population’s contribution to the HIV epidemic should be carefully considered and precisely defined. Modelling teams working in partnership with government, implementers, funders, and community members should systematically report both acquisition- and (long-term) transmission-focused indicators, instead of only measuring acquisitions in the short term as currently done. Multiple indicators will more comprehensively capture the potential impact of prevention efforts addressing acquisition and transmission risks of different vulnerable populations.

Introduction: Understanding how HIV epidemics are likely to behave in the future is key to informing HIV response strategies in low-income countries. Up-to-date HIV epidemiological estimates are important for policy decision- making, but surveillance data can be out of date. This study compared forecasts from HIV epidemiological models.Methods: Five independent modelling groups (EMOD-HIV, Goals, HIV Synthesis, Optima and PopART-IBM) calibrated their mathematical models to datapoints provided by the Ministry of Health and produced several indicators of the HIV epidemic in Zimbabwe for the period 1990 to 2040, under a status quo scenario in which it was assumed continuation of interventions at the current level.Results: All models predicted a continuous decline in HIV incidence and prevalence. However, there was variability in the estimated 2023 incidence rate (range: 2.0-3.3 per 1 000 person-years) and prevalence (range: 12.1%-14.3%). Variance was even larger in 2040 for incidence (range: 1.0-3.0 per 1 000 person-years), while this was not the case for prevalence (range: 3.9%-6.0%). All the models predicted that the country would reach a target of less than 7 800 new HIV infections per year by 2025.Conclusion: Five independent mathematical models fitted to the Zimbabwe Ministry of Health and Child Care's HIV surveillance data provided consistent predictions of continued decline in HIV incidence and prevalence in Zimbabwe if interventions continue to be implemented at the current levels, with prevalence predicted to be around a third of its level in 2000 by 2040.

The impact of outbreak response immunization (ORI) can be estimated by comparing observed outcomes to modelled counterfactual scenarios without ORI, but the most appropriate metrics depend on stakeholder needs and data availability. This study developed a framework for using mathematical models to assess the impact of ORI for vaccine-preventable diseases. Framework development involved (1) the assessment of impact metrics based on stakeholder interviews and literature reviews determining data availability and capacity to capture as model outcomes; (2) mapping investment in ORI elements to model parameters to define scenarios; (3) developing a system for engaging stakeholders and formulating model questions, performing analyses, and interpreting results; and (4) example applications for different settings and pathogens. The metrics identified as most useful were health impacts, economic impacts, and the risk of severe outbreaks. Scenario categories included investment in the response scale, response speed, and vaccine targeting. The framework defines four phases: (1) problem framing and data sourcing (identification of stakeholder needs, metrics, and scenarios); (2) model choice; (3) model implementation; and (4) interpretation and communication. The use of the framework is demonstrated by application to two outbreaks, measles in Papua New Guinea and Ebola in the Democratic Republic of the Congo. The framework is a systematic way to engage with stakeholders and ensure that an analysis is fit for purpose, makes the best use of available data, and uses suitable modelling methodology.

Background: Key populations (KPs), including female sex workers (FSWs), gay men and other men who have sex with men (MSM), people who inject drugs (PWID), and transgender women (TGW) experience disproportionate risks of HIV acquisition. The UNAIDS Global AIDS 2022 Update reported that one-quarter of all new HIV infections occurred among their non-KP sexual partners. However, this fraction relied on heuristics regarding the ratio of new infections that KPs transmitted to their non-KP partners to the new infections acquired among KPs (herein referred to as “infection ratios”). We recalculated these ratios using dynamic transmission models. Setting: One hundred seventy-eight settings (106 countries). Methods: Infection ratios for FSW, MSM, PWID, TGW, and clients of FSW were estimated from 12 models for 2020. Results: Median model estimates of infection ratios were 0.7 (interquartile range: 0.5–1.0; n = 172 estimates) and 1.2 (0.8–1.8; n = 127) for acquisitions from FSW clients and transmissions from FSW to all their non-KP partners, respectively, which were comparable with the previous UNAIDS assumptions (0.2–1.5 across regions). Model estimates for female partners of MSM were 0.5 (0.2–0.8; n = 20) and 0.3 (0.2–0.4; n = 10) for partners of PWID across settings in Eastern and Southern Africa, lower than the corresponding UNAIDS assumptions (0.9 and 0.8, respectively). The few available model estimates for TGW were higher [5.1 (1.2–7.0; n = 8)] than the UNAIDS assumptions (0.1–0.3). Model estimates for non-FSW partners of FSW clients in Western and Central Africa were high (1.7; 1.0–2.3; n = 29). Conclusions: Ratios of new infections among non-KP partners relative to KP were high, confirming the importance of better addressing prevention and treatment needs among KP as central to reducing overall HIV incidence.

Background: The distribution of new HIV infections among key populations, including female sex workers (FSWs), gay men and other men who have sex with men (MSM), and people who inject drugs (PWID) are essential information to guide an HIV response, but data are limited in sub-Saharan Africa (SSA). We analyzed empirically derived and mathematical model-based estimates of HIV incidence among key populations and compared with the Joint United Nations Programme on HIV/AIDS (UNAIDS) estimates. Methods: We estimated HIV incidence among FSW and MSM in SSA by combining meta-analyses of empirical key population HIV incidence relative to the total population incidence with key population size estimates (KPSE) and HIV prevalence. Dynamic HIV transmission model estimates of HIV incidence and percentage of new infections among key populations were extracted from 94 country applications of 9 mathematical models. We compared these with UNAIDS-reported distribution of new infections, implied key population HIV incidence and incidence-to-prevalence ratios. Results: Across SSA, empirical FSW HIV incidence was 8.6-fold (95% confidence interval: 5.7 to 12.9) higher than total population female 15–39 year incidence, and MSM HIV incidence was 41.8-fold (95% confidence interval: 21.9 to 79.6) male 15–29 year incidence. Combined with KPSE, these implied 12% of new HIV infections in 2021 were among FSW and MSM (5% and 7% respectively). In sensitivity analysis varying KPSE proportions within 95% uncertainty range, the proportion of new infections among FSW and MSM was between 9% and 19%. Insufficient data were available to estimate PWID incidence rate ratios. Across 94 models, median proportion of new infections among FSW, MSM, and PWID was 6.4% (interquartile range 3.2%–11.7%), both much lower than the 25% reported by UNAIDS. Conclusion: Empirically derived and model-based estimates of HIV incidence confirm dramatically higher HIV risk among key populations in SSA. Estimated proportions of new infections among key populations in 2021 were sensitive to population size assumptions and were substantially lower than estimates reported by UNAIDS.

Background: Previously, The Joint United Nations Programme on HIV/AIDS estimated proportions of adult new HIV infections among key populations (KPs) in the last calendar year, globally and in 8 regions. We refined and updated these, for 2010 and 2022, using country-level trend models informed by national data. Methods: Infections among 15–49 year olds were estimated for sex workers (SWs), male clients of female SW, men who have sex with men (MSM), people who inject drugs (PWID), transgender women (TGW), and non-KP sex partners of these groups. Transmission models used were Goals (71 countries), AIDS Epidemic Model (13 Asian countries), Optima (9 European and Central Asian countries), and Thembisa (South Africa). Statistical Estimation and Projection Package fits were used for 15 countries. For 40 countries, new infections in 1 or more KPs were approximated from first-time diagnoses by the mode of transmission. Infection proportions among nonclient partners came from Goals, Optima, AIDS Epidemic Model, and Thembisa. For remaining countries and groups not represented in models, median proportions by KP were extrapolated from countries modeled within the same region. Results: Across 172 countries, estimated proportions of new adult infections in 2010 and 2022 were both 7.7% for SW, 11% and 20% for MSM, 0.72% and 1.1% for TGW, 6.8% and 8.0% for PWID, 12% and 10% for clients, and 5.3% and 8.2% for nonclient partners. In sub-Saharan Africa, proportions of new HIV infections decreased among SW, clients, and non-KP partners but increased for PWID; elsewhere these groups' 2010-to-2022 differences were opposite. For MSM and TGW, the proportions increased across all regions. Conclusions: KPs continue to have disproportionately high HIV incidence.

Sciris aims to streamline the development of scientific software by making it easier to perform common tasks. Sciris provides classes and functions that simplify access to frequently used low-level functionality in the core libraries of the scientific Python ecosystem (such as numpy for math and matplotlib for plotting), as well as in libraries of broader scope (such as multiprocess for parallelization and pickle for saving and loading objects). While low-level functionality is valuable for developing robust software applications, it can divert focus from the scientific problems being solved. Some of Sciris' key features include: ensuring consistent dictionary, list, and array types (e.g., enabling users to provide inputs as either lists or arrays); enabling ordered dictionary elements to be referenced by index; simplifying datetime arithmetic by allowing date input in multiple formats, including strings; simplifying the saving and loading of files and complex objects; and simplifying the parallel execution of code. With Sciris, users can often achieve the same functionality with fewer lines of code, avoid reinventing the wheel, and spend less time looking up recipes on Stack Overflow. This can make writing scientific code in Python faster, more pleasant, and more accessible, especially for people without extensive training in software development.