Who needs monitoring? Modelling individual risk of HIV drug resistance from routine clinical data

Background

Almost all people living with HIV now start the same first-line regimen, but outcomes on that regimen vary. Some remain virally suppressed, some experience rebound and later resuppress, and a minority develop resistance mutations that reduce the number of effective treatment options available to them. Resistance genotyping can identify which mutations are present and which drugs are likely to remain active, and in well-resourced health systems a baseline genotype is often obtained before treatment begins. After treatment has started, however, genotyping is usually repeated only once viral rebound has been detected. It is therefore used mainly to guide management after failure, rather than to identify in advance who is at increased risk of developing resistance. 

As a result, decisions about who may need closer monitoring, earlier switching, or an alternative regimen are generally made without an individual estimate of resistance risk. This is particularly important because resistance testing is concentrated in high-income settings, whereas treatment failure is common in settings where genotyping is limited and viral load monitoring may be intermittent. A prediction strategy that depends on genotype may therefore be least accessible in the settings where it could be most useful. 

Aims

First, the student will characterise variation in resistance outcomes among people receiving WHO first line Dolutegravir-based regimens, describing which mutations are detected, in what combinations, and over what timescales.  

Second, they will build a stochastic simulation model of within-host viral rebound and mutation accumulation under interrupted drug exposure, with parameter values drawn from the published literature. The model will be used to generate synthetic cohorts in which each individual’s underlying trajectory is known. These trajectories can then be sampled according to realistic clinical monitoring schedules. The student will fit time-to-event models to predict rebound with resistance and accumulation of resistance to a second drug class from routinely recorded clinical features. They will compare standard parametric survival models with more flexible machine learning approaches, asking whether the additional flexibility improves prediction. 

Third, the student will ask how much additional predictive information is provided by additional information such as genotype over clinical history alone. Using the same simulation framework, they will test why prediction based on routine clinical data falls short: because relevant information is not recorded, or because it is recorded too sparsely to be useful. Varying the monitoring schedule in the simulated cohorts will allow these possibilities to be separated. This has a direct practical implication: the first would support collecting additional information, including genotype, whereas the second would point towards more frequent viral load monitoring. These imply different priorities in settings where resources are limited. 

Training outcomes 

The student will be trained in mathematical and statistical modelling for infectious disease. At Edinburgh, this will include stochastic simulation, survival analysis, likelihood-based and Bayesian inference, simulation study design, and machine learning methods for prediction. They will also gain experience working reproducibly within a secure data environment and handling sensitive clinical data. At Glasgow, this will include novel mathematical modelling for within-host viral evolution and dynamics as well as the protein-level underpinnings of HIV resistance. The project will also provide experience of working with a national clinical dataset.  The student will be equipped to lead interdisciplinary research that links molecular mechanism to clinical outcome, with a skillset transferrable to viral and bacterial AMR across pathogens.  The course will set them up for ongoing work in the fields of academia, industry or public health, whether in the UK or in health systems operating under greater resource constraint. 

References

• Feder AF, Harper KN, Brumme CJ, Pennings PS (2021). Understanding patterns of HIV multidrug resistance through models of temporal and spatial drug heterogeneity. eLife 10:e69032.  
• Blassel L, Zhukova A, Villabona-Arenas CJ, Atkins KE, Hué S, Gascuel O (2021). Drug resistance mutations in HIV: new bioinformatics approaches and challenges. Current Opinion in Virology 51:56–64. 
• World Health Organization (2024). HIV drug resistance: brief report 2024. Geneva: World Health Organization. ISBN 978-92-4-008631-9. Available at https://www.who.int/publications/i/item/9789240086319