Background:England experienced an unusually early and rapid increase in influenza A/H3N2 subclade K infections in 2025/26. Antigenic change and a fast selective sweep raised concerns over a potentially severe season. Building on analysis conducted as the subclade emerged, we aim to compare epidemic dynamics of the 2025/26 season to previous years and to model plausible epidemiological scenarios. Methods:We compared peak epidemic growth rates and reproduction numbers across influenza seasons from 2011/12 to 2025/26 using routine surveillance data in England. Weekly epidemic growth rates were estimated using a Gaussian random walk model, and time-varying reproduction numbers using EpiEstim. We also developed an age-stratified transmission model and interactive web tool to explore scenarios varying immune escape, transmissibility, and seed date, using 2022/23 as a baseline season. Results:Peak A/H3N2 growth rates and time-varying reproduction numbers for the 2025/26 season are of similar magnitude but earlier than previous severe seasons. Scenario analyses suggest early trends are compatible with moderate levels of immune escape, a 10% higher R 0 , or an earlier seed date, though it is not possible to distinguish the relative importance of these mechanisms from these data alone. Conclusions:The 2025/26 influenza season is characterised by early but not unusually rapid growth. Earlier growth does not systematically lead to especially large epidemics due to earlier susceptible depletion combined with a dampening effect from school holidays. Laboratory evidence for antibody escape does not directly translate to large reductions in population immunity, supporting the need for complementary real-time epidemiological analyses and modelling.

BACKGROUND:Data on the population-scale impact of dolutegravir (DTG)-based HIV regimens in sub-Saharan Africa are extremely limited. We used data from a surveillance cohort in southern Uganda to assess viral suppression and antiretroviral (ART) resistance over 10-years alongside DTG scale-up. METHODS:Consenting participants in the population-based Rakai Community Cohort Study between August 2011 and March 2023 aged 15-49 completed questionnaires and provided samples for HIV testing, viral load quantification, and viral deep-sequencing. We collected data on DTG utilization at HIV care clinics. We estimated the prevalence of HIV suppression and ART resistance using robust Poisson regression. Bayesian logistic regression quantified associations between resistance and individual-level suppression across surveys. RESULTS:Among 8781 people with HIV (PWH), suppression increased from 57.1% (2014, 95% confidence interval [CI], 55.4%-58.8%) to 90.3% (2022, 95% CI, 89.2%-91.4%). By 2020 84.4% (95% CI, 83.7%-85.2%) and 64.6% (95% CI, 63.9%-65.3%) of men and women on ART were on DTG. Among treatment-experienced viremic PWH, any intermediate/high resistance decreased from 51.1% (95% CI, 40.7%-64.2%, 2014) to 27.9% (95% CI, 21.3%-36.5%, 2022). Two of 258 (0.8%) 2022 participants harbored intermediate/high-level DTG resistance (inQ148R, inE138K, and inG140A). inS153Y (2-fold INSTI resistance) was observed in 23/306 (7.5%) of viremic individuals, with evidence of transmission. By 2022, NNRTI/NRTI resistance was not associated with a reduction in individual-level suppression (risk ratios: 1.15, 95% HPD: 0.93-1.39; 1.14, 0.86-1.42). CONCLUSIONS:Viral suppression increased during the DTG transition with minimal emerging intermediate/high-level resistance. Falling resistance among treatment-experienced PWH underscores the role of ART adherence in reducing viremia. The emergence of inS153Y justifies continued surveillance.

The phenotypic fitness landscape defines the action of natural selection on pathogens, linking changes in their phenotypes to transmission and evolution. The rapidly changing nature of epidemic spread and antigenic landscapes pushes viruses to evolve on fitness seascapes. As a result, evolution of viruses such as SARS-CoV-2 proceeds in a never ending series of waves, driven by epistatic interactions and by the arms race between viral adaptation and human immunity. Phenotypic characterisation of these rapidly changing fitness seascapes is an open challenge. Using a Phenotypic Selection Inference framework that links phylogenetic estimates of mutation fitness effects with deep mutational scanning data, we traced how selective pressures on viral phenotypes have shifted throughout the COVID-19 pandemic. Natural selection has favoured enhanced ACE2 binding since the emergence of SARS-COV-2, with relatively constant selective pressure even for the most recent variants. The strength of selection for antibody escape was comparable to ACE2 binding during early evolution, but as population immunity rose, escape from class 3 and then class 2 antibodies became dominant. For variants circulating in 2024, natural selection shifted toward class 3 antibody escape, while those circulating in 2025 have experienced dynamic, rapidly changing pressures for escape from all antibody classes. These transitions reflect an ongoing arms race between viral adaptation and human immunity. Our findings reveal that SARS-CoV-2 antigenic evolution is governed by dynamic, class-specific immune pressures, and that selection for replication capacity has been continuously present during the pandemic, presumably to compensate for the effects of antigenic escape on viral replication. Our approach for the inference of phenotypic selection provides a framework to understand and anticipate the evolution of future variants.

The prospective design of vaccine efficacy trials for deployment in outbreaks requires advance consideration of plausible outbreak scenarios, anticipated vaccine characteristics, and logistical and ethical constraints. As part of CEPI’s 100 Days Mission to accelerate vaccine development against a novel Disease X, we evaluated trial designs for a hypothetical Nipah-X outbreak. We assumed Nipah-X would share key features with Nipah, including high case fatality rates and substantial super-spreading, but with sustained human-to-human transmission. Using simulations based on infection models, including an extended chain-binomial model incorporating super-spreading, we compared ring-trials using cluster-randomisation with individual-randomisation within rings. High levels of super-spreading markedly reduced the power of cluster-randomised designs due to strong intra-cluster correlations in case numbers, whereas individual-randomisation retained power. These findings highlight that understanding and accounting for super-spreading is critical when designing ring-trials, as cluster-randomised designs may fail unless vaccine efficacy is nearly complete.

BackgroundHIV incidence among adolescent girls and young women (AGYW) in eastern and southern Africa has declined substantially over the past two decades. These declines are often attributed to biomedical HIV prevention strategies, though concurrent changes in sexual behavior may also contribute. We evaluated the contributions of biomedical and behavioral drivers to historical incidence decline in AGYW and projected their impact on incidence trajectories over the next 30 years.Methods and findingsWe conducted a mathematical modeling study using data from the Rakai Community Cohort Study (RCCS), an open, population-based cohort of adults aged 15-49 years in 30 communities in Rakai, Uganda. We used an agent-based HIV-1 transmission model calibrated to cohort data to estimate HIV incidence trends among AGYW, aged 15-24, and to quantify the independent and combined effects of antiretroviral therapy (ART), voluntary medical male circumcision (VMMC), and changes in age at first sex (AFS). HIV incidence among women aged 15-24 declined by 71% between 2000 and 2019, from 1.57 to 0.45 per 100 person-years, representing the largest decline across female age groups in the cohort. Increasing AFS over the study period (by approximately 3 years in women and 2 years in men) was the largest contributor to incidence declines among adolescent women aged 15-19, averting 17% of cumulative infections between 2000 and 2020 and 37% between 2000 and 2050. Among women aged 20-24, ART scale-up had the greatest impact, averting 13% of infections by 2020 and 43% by 2050. VMMC contributed modestly to historical declines but had larger projected effects over longer time horizons. ART, VMMC, and delays in AFS acted additively to reduce HIV incidence among AGYW. Study limitations include reliance on self-reported sexual behavior and the use of a mathematical model that cannot capture all real-world sexual network dynamics.ConclusionsBoth biomedical HIV interventions and broader behavioral changes contributed to declines in HIV incidence among AGYW. Sustaining continued incidence declines in young women will require maintaining both the protective changes in sexual behaviors and effective biomedical interventions.

The Coalition for Epidemic Preparedness Innovations’ (CEPI) 100 Days Mission, and broader global pandemic preparedness efforts, require overcoming challenges in evaluating vaccine efficacy during emerging epidemics and outbreaks. Key challenges, addressed in a workshop jointly hosted by CEPI and the University of Oxford’s Pandemic Sciences Institute, arise from the sporadic and heterogeneous nature of outbreaks, high fatality rates that may preclude traditional placebo-controlled trials and biases from small or real-world studies. Here, we review recent developments, potential solutions and areas for innovation emphasized in the workshop to address these challenges. These opportunities for research and methodological development encompass four main areas: serology, exposure measurements, statistical modelling and trial design.

BACKGROUND:Orthoebolavirus outbreaks place health-care workers (HCWs) at substantial risk, and HCW illness or death can weaken response capacity. The 2026 Bundibugyo virus outbreak in DR Congo highlights the need for deployable countermeasures when species-specific vaccines are unavailable. With candidate antivirals under evaluation, we aimed to estimate the impact of HCW-targeted antiviral post-exposure prophylaxis (PEP) across different readiness, disruption, and allocation scenarios. METHODS:We adapted a previously published stochastic branching-process model of orthoebolavirus transmission, representing health care, community, and funeral transmission; time-varying non-pharmaceutical interventions; and HCW-targeted PEP. The model was calibrated to two historical outbreaks using sequential approximate Bayesian computation: the 2013-16 west Africa epidemic, to define a high-burden, reasonable worst-case scenario archetype (west Africa-like archetype); and the 2018-20 North Kivu and Ituri outbreak in eastern DR Congo, to define an archetype with longer transmission under conflict-related response disruption (DR Congo-like archetype). The primary outcome was HCW deaths averted. For both archetypes, we simulated three antiviral deployment readiness scenarios (scenario 1: 100% coverage on day 0; scenario 2: scaled up to 80% coverage over 180 days; and scenario 3: scaled up to 50% coverage over 1 year) and compared their impact on HCW deaths with a scenario of no antiviral. For the DR Congo-like archetype only, we simulated four disruption scenarios: no antiviral PEP, ideal delivery (100% coverage and no dosing delay), delayed dosing with coverage preserved, and delayed dosing with delayed coverage. As a secondary outcome, we assessed number of PEP doses required per HCW death averted under different allocation scenarios. FINDINGS:At baseline (no antiviral PEP), cumulative HCW deaths reached a median of 553 (IQR 208-983) in the west Africa-like archetype by week 60, compared with 61 (19-125) in the DR Congo-like archetype by week 80. Assuming 80% efficacy and 80% coverage with antiviral PEP in the same timeframe in a central analysis, cumulative HCW deaths fell to 200 (68-349; equivalent reduction of 64% [63-66] relative to baseline) in the west Africa-like archetype and 22 (10-44; equivalent reduction of 64% [60-68]) in the DR Congo-like archetype. Under different scenarios of deployment readiness at 80% antiviral efficacy, the median reduction in HCW deaths compared with no PEP was 80% (95% CrI 79-81) in the west Africa-like archetype and 80% (76-84) in the DR Congo-like archetype for scenario 1; 60% (57-62) and 52% (41-58), respectively, for scenario 2; and 19% (16-22) and 22% (7-29), respectively, for scenario 3. In the DR Congo-like operational disruption analyses, an ideal scenario (PEP delivered at 100% coverage without a delay after exposure) averted 83% (79-87) of HCW deaths compared with no antiviral; maintaining 100% coverage but introducing delayed dosing (1-5 days post-exposure) reduced this finding to 50% (40-55) compared with no antiviral. Delayed coverage and dosing resulted in only 35% (25-47) of the ideal scenario impact. At 80% antiviral efficacy with same-day dosing, targeted allocation of recognised high-risk exposures (such as personal protective equipment breaches or direct body-fluid contact) required 44 doses (95% Crl 43-44) per HCW death averted versus 109 doses (85-161) with broad allocation. INTERPRETATION:HCW-targeted antiviral PEP could substantially reduce HCW deaths during orthoebolavirus outbreaks if efficacious antivirals can be delivered rapidly and high operational coverage is maintained. Comparisons of antiviral use cases and alternative response investments are needed to determine how resources can best support outbreak response. FUNDING:Gilead Sciences, UK National Institute for Health and Care Research, Oxford Martin School, Miller Institute, EU Global Health EDCTP3, and Coalition for Epidemic Preparedness Innovations. TRANSLATIONS:For the French and Swahili translations of the abstract see Supplementary Materials section.

Background:Novel HIV prevention interventions such as long-acting pre-exposure prophylaxis (PrEP) could substantially reduce HIV transmission in Africa. However, efficient implementation in high-prevalence settings where incidence has declined requires an understanding of the contemporary dynamics driving new infections. Methods:We identified incident HIV cases from a longitudinal, population-based cohort in Uganda. We individually matched cases to HIV-negative controls; traced and enrolled reported sexual partners; and enrolled female sex workers (FSWs) from reported venues. Conditional logistic regression, transmission modeling, and phylogenetics were used to characterize transmission networks. Findings:From 2021-2024, 38,899 HIV tests among 22,255 people identified 187 people with incident infections (47.6% male); 164 (88%) were enrolled and matched to 164 HIV-negative controls. Overall, 593 non-sex-worker partners (371 enrolled,62.6%), 146 FSW partners (21 enrolled,14.4%), and 28 venues (208 FSWs enrolled) were reported. Incident infection was most strongly predicted by partnership with a FSW (odds ratio:15.5; 95%CI:3.7-64.8), identified in 43.0% of male cases versus 6.3% of controls. Men with FSW partners had larger sexual networks than men without (median:6 vs 2 partners), and 91.2% of men with FSW partners also had non-sex-worker partners. Transmission modeling attributed 34.4% (95%CI:31.5-36.8%) of all male infections and 80.0% (95%CI:73.2-84.4%) of infections among male clients to sex with FSWs. Oral PrEP use among HIV-negative partners of incident cases was low (8.9% in women; 2.1% in men). Interpretation:Men with FSW partners accounted for a substantial share of incident HIV infections and had markedly higher odds of infection than men without such partnerships. Together with the high potential for onward transmission within male client networks, these findings suggest that inclusion of male clients in long-acting HIV prevention strategies could be highly efficient and impactful. Funding:National Institutes of Health, United States; Gates Foundation; National Health and Medical Research Council, Australia.

Introduction HIV viral load (VL) monitoring is essential for evaluating antiretroviral treatment effectiveness, but reliance on venepuncture in high-burden settings can delay sample collection and return of results. Emerging approaches suggest that lower blood volumes may be sufficient for VL and drug resistance testing using novel methods such as the next-generation sequencing platform, although these remain under evaluation. Self-sampling devices could enable pre-appointment collection, reduce clinic visits, and improve timeliness. We evaluated the usability of two devices, Tasso+ and Collect2Know v1.0 (C2K), among people living with HIV in Zambia. Methods An individually randomized crossover trial was conducted with 70 stable PLHIV attending routine ART visits (October–December 2024). Participants used each device in a randomized order, with supervision for the first use of each device and a second, unsupervised home sampling one week later. Success was defined as self-collection of ≥ 0.5 mL of blood during unsupervised use. A questionnaire captured perceptions of device usability, compared to standard venepuncture options, and was complemented by findings from qualitative research conducted during both the formative and during the trial. Results Of 63 participants completing the study (35.5% male, 64.5% female, aged 18–50+), 87.1% (54/63) successfully collected ≥ 0.5 mL with Tasso+, compared to 37.1% (24/63) with C2K (McNemar test, p <0.001; OR 11.45, 95% CI: 4.83–27.10). Among participants who successfully collected a sample using the Tasso+ device, 61.1% (33/54) failed to achieve successful collection with the C2K v1.0 device (Table 1) Healthcare workers highlighted reduced clinic congestion as a key benefit of self-sampling, while patients cited less pain, reduced fear, and improved privacy as significant advantages. Conclusion This crossover RCT demonstrated that device-assisted, minimally invasive sampling offers high usability for self-collection among PLHIV. The adoption of such patient-led technologies could streamline HIV monitoring, enhance patient experience, and enable timely detection of treatment failure. Further exploration of minimally invasive sampling devices, including optimisation of device design and cost, is warranted in broader real-world contexts. Trial Registration Pan African Clinical Trial Registry PACTR202605883649675, registered on 24 February 2024.

Nipah virus causes sporadic outbreaks characterized by high mortality, with transmission concentrated within households and healthcare settings. Clinical trials of vaccines for high-mortality pathogens during outbreaks face significant methodological and ethical challenges, including due to short epidemic durations and variable community acceptance of standard individual randomized controlled trials (iRCTs). This paper examines key ethical issues in Nipah vaccine outbreak trial design. First, outbreak trials must recruit individuals at high risk of infection to reach efficacy endpoints before epidemics end, yet placebo controls may be controversial due to risks of infection among participants. Failure to accrue sufficient infections during outbreaks risks prolonged delays to vaccine licensure. Second, alternatives to classical iRCTs involve ethical and methodological trade-offs: cluster randomization may delay efficacy endpoints, especially when high transmission occurs in only a minority of clusters; trials involving delayed vaccination may involve similar risks (for control participants who are infected during the delay to vaccination) as for those in standard placebo iRCTs; and single arm trials risk failure to clarify vaccine efficacy due to their lack of controls. Third, although high pre-trial probability of experimental vaccine efficacy (>50% for recent Phase III vaccines) challenges traditional concepts of clinical equipoise, rigorous vaccine trials remain ethically acceptable. Fourth, policies for post-trial access to efficacious vaccines should be revised to provide accelerated access to control arm participants. Finally, ethical trial design depends on appropriate community engagement, which should begin early in outbreaks and ideally continue in inter-epidemic periods.

To develop effective HIV prevention strategies to guide public health policy the main sources of infection in HIV prevention studies must be identified. Accordingly, we devised a statistical approach that estimates the relative contribution of different sources of infection in community-randomized trials of infectious disease prevention using deep- (or next generation) sequenced pathogen data. We applied this approach to the Botswana Combination Prevention Project (BCPP) and estimated that 90% [95% Confidence Interval (CI): 80-94] of new infections in communities that received combination prevention (including universal HIV test-and-treat) originated from individuals residing in communities outside the trial area. We estimate from our model that the relative benefit of providing the BCPP intervention to all communities nationwide would be a 59% [3-87] reduction in transmissions to recipients in trial communities, exceeding the 30% reduction observed when providing the BCPP intervention to trial communities only. Our results suggest that the impact of the BCPP trial intervention was curtailed by sources of transmission outside the trial area and could be considerably larger if applied nationally. We recommend that the impact of sources of transmission beyond the reach of the intervention be considered when designing and evaluating interventions to inform public health programs.

Background:As HIV incidence declines in African settings with high treatment coverage, it remains unclear how transmission is structured within populations and whether new infections arise from external introductions or local transmission. We characterized the molecular epidemiology of ongoing transmission in a mature multi-subtype epidemic in Uganda. Methods:We analyzed HIV genome sequences and survey data from the Rakai Community Cohort Study collected between 1994 and 2019. We identified phylogenetic clusters at 5·3% and 2·5% genetic distance thresholds and inferred long-horizon transmission chains with phylogeographic models. Newly diagnosed infections identified between 2016 and 2019 were mapped onto subtype- specific phylogenies to assess their origins and transmission context. A Bayesian negative binomial branching process model estimated undersampled chain sizes and case reproduction numbers. Findings:Among 4215 participants living with HIV between December 2016 and May 2019, 474 were newly diagnosed, of whom 269 had at least one pure-subtype sequence available. We identified 649 phylogenetic clusters at 5·3% genetic distance and 673 phylogeographic chains including ≥2 individuals. Most clusters and chains were small (median sizes 2 [IQR 2-3] and 3 [2-4], respectively), with new diagnoses rarely clustered together. Only 46/269 (17·1%) new diagnoses had phylogeographic external origins, while the remaining 82·9% were partially or fully linked to local chains. Mixed-subtypes/recombinant chains were larger and had higher case reproduction numbers (A1/D: 0.84 [95% CrI: 0.79-0.93]; mixed: 0.84 [0.73-0.97]) than single- subtype chains (A1: 0.56 [0.51-0.60]; D: 0.63 [0.59-0.66]; C: 0.55 [0.41-0.71]), yet all estimates were less than one. Interpretation:HIV transmission was fragmented across numerous, slowly propagating lineages, maintained by local clusters with occasional introduction. Continued transmission across many chains suggests that further reductions in HIV incidence will require maintaining high levels of population-wide treatment and prevention coverage. Funding:The National Institute of Allergy and Infectious Diseases, the Gates Foundation, and the HIV Prevention Trials Network Laboratory Center.

Mathematical modelling with agent-based models (ABMs) has gained popularity during the COVID-19 pandemic, but their complexity makes efficient and robust calibration to data challenging, particularly when applying Bayesian methods to quantify parameter uncertainty. We propose a method for calibrating ABMs that combines a Machine-Learning step with Approximate Bayesian Computation (ML-ABC). We showcase ML-ABC application with a proof-of-principle case study, in which we calibrate the Covasim -a stochastic ABM that has been used to model the English COVID-19 epidemic and inform policy at important junctions. Benchmarking against traditional Rejection-ABC (R-ABC), we illustrate the advantage of ML-ABC application in calibrating Covasim to data on hospitalisations and deaths from COVID-19 during the first and the second COVID-19 epidemic waves of 2020 and early 2021. Across scenarios, we demonstrate that using an ML screening step allows us to derive identical posterior distributions of the calibrated Covasim parameters as with the traditional R-ABC method, but faster. Specifically, we derive posterior distributions for input parameters around 52% faster when calibrating to the first epidemic wave and around 33% faster when calibrating parameters for the second epidemic wave, compared to the traditional R-ABC. Policy modelling requires calibration which is both efficient to adapt to fast-changing pandemic environments and robust to ensure confidence in policy decisions. However, existing ABM calibration often relies on myopic non-exhaustive searches in order to remain tractable, resulting in point parameter estimates. In this preliminary study, ML-ABC strictly improves upon existing ABC calibration approaches in all tested scenarios, indicating its potential to make ABC competitive with point-estimate calibration approaches. This novel approach offers a pathway to effectively calibrate ABMs in a way which is both efficient and quantifies parameter uncertainty, crucial for realising the potential of ABMs for timely and responsively modelling during an emerging epidemic.

A standard method in phylogenetic reconstruction for representing variation in substitution rates between sites in the genome is the discrete Gamma model (DGM). Relative rates are assumed to be distributed according to a discretised Gamma distribution, where the probabilities that a site is included in each class are equal. Here, we identify a serious bias in the branch lengths of reconstructed phylogenies when the DGM is used, with the magnitude of the effect varying with the number of sequences in the alignment. We demonstrate the existence of the bias, using both simulated datasets and real HIV-1 sequences; in both cases branch lengths are overestimated. The phenomenon is exacerbated by increasing the number of discrete rate categories, is only very slightly mitigated by the addition of an invariant sites category, and happens regardless of the software package used for reconstruction. We show that the alternative "FreeRate" model, which assumes no parametric distribution and allows the class probabilities to vary, is not subject to the issue. We further establish that the reason for the behaviour is the equal size of the class probabilities in the discretisation, not simply the fact that a continuous distribution has been discretised. We explore the mathematics of the phenomenon, showing how maximum likelihood branch lengths under the DGM may differ from the true ones used to generate the tree, and how the magnitude of this difference is equal to the departure of the mean maximum likelihood substitution rate across all sites in the genome from 1. We recommend that the DGM be retired from general use. While FreeRate is an immediately available replacement, it is known to be difficult to fit, and thus there is scope for innovation in rate heterogeneity models.

Abstract While much progress has been made in reducing the incidence of HIV-1 infection in sub-Saharan Africa in recent years, bringing the epidemic to an end will require identification of the demographic groups that continue to contribute to transmission. Pathogen phylogenetics and individual-based mathematical models (IBMs) of transmission are approaches that enable researchers to explore such questions. Here, we used both methods to characterise the ages and sexes of the individuals involved in heterosexual transmission in the context of the HPTN 071 (PopART) trial in Zambia. The results were concordant, and show that the male partner was on average older than the female by less than seven years, with larger age gaps in male-to-female than female-to-male transmissions. We found that the largest gaps for female recipients were amongst the youngest of those recipients. Conversely, the youngest male recipients saw the smallest gaps. We further used the IBM to demonstrate that transmission to new age cohorts first entering into sexual activity is driven predominantly by male-to-female transmission. We also simulated the PopART universal testing and treatment intervention into the future to show that effective treatment of under-35-year-olds accounts for 93.8% of the reduction in incidence by 2039, while effective treatment of under-35-year-old men accounts for 62.1%. Finally, we simulated a one-year cessation of ART treatment for the whole population, which resulted in an immediate increase in the average age at transmission of both sources and recipients. With it becoming ever more expensive and difficult to find treatment-naive individuals and link them to care, targeted interventions for demographic groups such as under-35 men may be the key to finally ending HIV.

BACKGROUND:With scale-up of antiretroviral therapy (ART) in sub-Saharan Africa, increasing pretreatment HIV drug resistance has been reported; however, the broader effect of ART expansion on population-level resistance patterns remains insufficiently quantified. We aimed to estimate the longitudinal prevalence of drug resistance and resistance-conferring mutations. METHODS:This study used data collected as part of the Rakai Community Cohort Study (RCCS), an open population-based census and cohort study conducted in southern Uganda. At each survey round, residents aged 15-49 years are invited to participate and receive a structured questionnaire that obtains sociodemographic, behavioural, and health information, including self-reported past and current ART use. Voluntary HIV testing is conducted using a rapid test algorithm and a venous blood sample. People with HIV provide samples for viral load quantification and deep sequencing. We analysed RCCS survey, HIV viral load, and deep sequencing (which was used to predict resistance) data from five survey rounds. The key outcomes were the population prevalence of viraemic people with HIV with non-nucleoside reverse transcriptase inhibitor (NNRTI), nucleoside reverse transcriptase inhibitor (NRTI), protease inhibitor, or multiclass resistance among all participants (regardless of HIV serostatus) in the 2015 and 2017 surveys. Prevalence of class-specific resistance and resistance-conferring substitutions were estimated using robust log-Poisson regression. FINDINGS:Between Aug 10, 2011, and Nov 4, 2020, there were 43 361 participants in the RCCS and 7923 (18·27%) people with HIV. Over five survey rounds, 93 622 participant visits occurred, among which 17 460 (18·65%) were from people with HIV. Over the analysis period, the median age of study participants remained similar (28 years [22-35] in 2012 and 29 years [21-38] in 2019). Sufficient data were available to reliably genotype 4072 (90·03%) of 4523 participant visits from 3407 people with HIV for at least one drug. Overall population prevalence of resistance contributed by viraemic pretreatment people with HIV decreased between 2012 and 2017 from 0·56% (95% CI 0·42-0·75) to 0·25% (0·18-0·33) for NNRTI and from 0·24% (0·15-0·37) to 0·05% (0·02-0·10) for NRTI (prevalence ratio 0·44 [0·29-0·68] for NNRTI and 0·21 [0·09-0·47] for NRTI). Between 2012 and 2017, NNRTI resistance among viraemic pretreatment people with HIV increased from 4·86% (3·69-6·42) to 9·61% (7·27-12·7; prevalence ratio 1·98 [1·34-2·91]). The prevalence of NNRTI and NRTI resistance was substantially higher among viraemic treatment-experienced people with HIV (51·49% [46·24-57·34] for NNRTI and 36·46% [30·06-44·22] for NRTI in 2017) than among pretreatment people with HIV. NNRTI and NRTI resistance was predominantly attributable to rtK103N and rtM184V. inT97A was observed at a similar prevalence among viraemic treatment-experienced (9·96% [6·41-15·48]) and viraemic pretreatment (10·56% [8·01-13·93]) people with HIV; no major dolutegravir resistance mutations were observed. INTERPRETATION:Despite rising NNRTI resistance among pretreatment people with HIV, overall population prevalence of pretreatment HIV drug-resistant viraemia decreased due to increasing ART uptake and viral suppression. This finding underscores the crucial role of achieving and maintaining high ART coverage in reducing transmission of drug-resistant HIV. The high prevalence of mutations conferring resistance to components of first-line ART regimens among viraemic people with HIV is potentially concerning. FUNDING:National Institutes of Health, Johns Hopkins University Center for AIDS Research, Bill & Melinda Gates Foundation, and the US Centers for Disease Control and Prevention.

Unlike classification, whose goal is to estimate the class of each data point in a dataset, prevalence estimation or quantification is a task that aims to estimate the distribution of classes in a dataset. The two main tasks in prevalence estimation are to adjust for bias, due to the prevalence in the training dataset, and to quantify the uncertainty in the estimate. The standard methods used to quantify uncertainty in prevalence estimates are bootstrapping and Bayesian quantification methods. It is not clear which approach is ideal in terms of precision (i.e. the width of confidence intervals) and coverage (i.e. the confidence intervals being well-calibrated). Here, we propose Precise Quantifier (PQ), a Bayesian quantifier that is more precise than existing quantifiers and with well-calibrated coverage. We discuss the theory behind PQ and present experiments based on simulated and real-world datasets. Through these experiments, we establish the factors which influence quantification precision: the discriminatory power of the underlying classifier; the size of the labeled dataset used to train the quantifier; and the size of the unlabeled dataset for which prevalence is estimated. Our analysis provides deep insights into uncertainty quantification for quantification learning.

Infectious disease threats to individual and public health are numerous, varied and frequently unexpected. Artificial intelligence (AI) and related technologies, which are already supporting human decision making in economics, medicine and social science, have the potential to transform the scope and power of infectious disease epidemiology. Here we consider the application to infectious disease modelling of AI systems that combine machine learning, computational statistics, information retrieval and data science. We first outline how recent advances in AI can accelerate breakthroughs in answering key epidemiological questions and we discuss specific AI methods that can be applied to routinely collected infectious disease surveillance data. Second, we elaborate on the social context of AI for infectious disease epidemiology, including issues such as explainability, safety, accountability and ethics. Finally, we summarize some limitations of AI applications in this field and provide recommendations for how infectious disease epidemiology can harness most effectively current and future developments in AI.

The spread of epidemics in populations is often inhomogeneous, consequently infection incidence varies between sub-populations. Age-structure is often particularly important in the dynamics of epidemics, due to the contact patterns between individuals of different ages. Public health interventions are often targeted at specific age-groups, therefore analysing the age-structure of transmission patterns is essential to evaluate the efficacy of these interventions. We develop a Bayesian model to estimate the contribution of different age-groups to the reproduction number (R) and to new infections for COVID-19 in England throughout 2021, using the ONS Infection Survey. We model a dynamic next-generation matrix in a novel way by splitting it into a static survey-derived social-contact matrix, multiplied by a low-rank dynamic matrix. We show that whilst R was typically highest for school-age children (5-11y and 12-17y) and lowest for the elderly (60y+), the former typically rose during term-time and fell during the school-holidays. The dynamics for young adults (18-29y) were particularly interesting, which increased relative to older adults in late-spring 2021 following the re-opening of entertainment venues. The R peaked for young adults in July 2021 coinciding with the period of the Euros football tournament, before rapidly dropping as the national vaccination program reached this group in August 2021. Our model is an important tool that can estimate R and attribute new infections by the infector's age, thus identifying core groups which sustain the epidemic and informing the design of targeted interventions.