School of Public Health and Grantham Institute scholarships

Evaluating the Effectiveness of Large Language Models in Supporting Clinical Decisions Within Payor Settings

The PhD project will systematically examine the health insurance processes that are best suited to the application of large language models (LLMs) and assess the effectiveness of different LLMs in completing specific clinical decision support tasks in a health insurance context. 

It is expected that the results of the project will support the development of a framework for evaluating LLMs in health insurance. The project may also lead to the development of a commercial application to help insurance companies automate evidence-based claims adjudication for their members and thus reduce operational costs associated with the manual processing of health insurance claims.

The role of PFAS as immunosuppressors: risk of kidney cancer

There is consistent evidence that exposure to per-and polyfluoroalkyl substances (PFAS), a ubiquitous environmental exposure, can induce kidney cancer, among other health outcomes. In DISCERN, a large-scale multi-disciplinary study funded by the European Commission (https://discern.iarc.who.int/), we have analyzed mutational spectra from >1000 cases of kidney cancer. We will compare cancer cases showing specific somatic mutations (and mutational pathways) and cancers without. The comparison will be for selected environmental exposures – including PFAS - and omic patterns (e.g. immunological pathways based on proteomics). We will also compare cancer biopsies with their matched normal tissues (case-control approach) for exposures and omic patterns, to investigate clonal expansion. The overall aim is to understand whether exposure to PFAS can select mutated renal cells (for example in heavy smokers) thus increasing the risk of cancer. DISCERN measures the blood exposome by characterizing low-molecular-weight chemicals (<2000 Da) using untargeted high-resolution mass spectrometry (HRMS) combined with advanced data extraction and annotation algorithms to detect >10,000 chemical signals that include endogenous metabolites, diet, microbiome-derived metabolites, environmental chemicals (including PFAS, that are stable in blood), consumer chemical products, and pharmaceuticals. A suite of resources for enhanced annotation will be utilized. Proteomics (Somasca) data is also available for all cases. In addition to improving causal assessment of the relationship of PFAS with kidney cancer, we will contribute to the understanding of the mechanisms of carcinogenesis. PFAS are immunosuppressive, which is one of the “hallmarks of cancer”. The project builds upon the ENDOMIX consortium on PFAS exposure and health effects, funded by the European Commission, in which  mechanisms of action (in particular immunological) are investigated in humans, animals and cells in culture (funded by the European Commission). 

Health impacts of loss of biodiversity: built environment and dietary habits

Loss of biodiversity is a source of concern not only for the equilibrium of the biosphere, but also for human health. In this project we aim to use data from the global calculator (GC), a tool developed at Imperial College, and from a large European cohort study (EPIC) to estimate the impact of different human activities on the loss of biodiversity and on human health (metabolic health, gut microbiome). We have shown by using the GC that the food system and other forms of land use (related to urbanization) have a large impact on green-house gas emissions (1).  In EPIC we have demonstrated a relationship between food habits, GHG emissions, low food biodiversity and human mortality or the incidence of different types of cancer (2, 3). The aim of the current project is to extend the investigations on loss of biodiversity to the human metabolic landscape and the microbiota, whose variety is associated with exposure to natural environments (in contrast to the built enviornment), and exposure to bio-diverse diets. We will study the impact of loss of biodiversity on health by assessing (a) the impact of the built environment – measured with satellites analyses of people’s residences – on metabolic parameters (proteins associated with glucose metabolism, diabetes) and metabolites linked to the gut microbiota diversity, already measured in EPIC; (b) the impact of low food biodiversity on metabolic health and gut microbiota.  The outcome will be a model of the health outcomes associated with multiple sources of loss of biodiversity, and a contribution to the development of a policy focused on health co-benefits of biodiversity reintegration. 

Modelling strategies for the release of genetically-modified mosquitoes for arboviral control across different ecological settings

We will develop mathematical models to investigate the potential of novel strategies for controlling natural populations of insect pests and disease vectors that utilise cutting edge approaches in genome engineering. These strategies involve releasing insects carrying genes that spread through wild populations despite incurring fitness costs. These genes spread because they use CRISPR-based molecular mechanisms to bias their own inheritance above the normal Mendelian rate of 50%, a process known as “gene drive” (https://doi.org/10.1038/s41467-024-53065-z). 

We will develop an ecological modelling approach to inform the design of novel gene drive strategies to control Aedes aegypti mosquitoes, the major vector of several human arboviruses including the dengue, chikungunya and Zika viruses. Arbovirus transmission and prevalence is highly sensitive to climatic conditions, so it is critical to understand how our changing climate will influence their distribution and spread. Our modelling approach will represent key aspects of mosquito ecology, considering different environmental settings and projected climate change scenarios.

Our candidate will be part of an industrial collaboration with Biocentis (https://biocentis.com/), a biotechnology company building sustainable solutions for genetic control of insect pest and vector species. The candidate will benefit from a placement period with Biocentis, providing interdisciplinary training across academic and industrial settings.

Professor Mat Fisher (SPH)
Dr David Green (SPH)
Dr Nick Croucher (SPH)
Dr Andrew Singer (UK CEH)
Dr Liz Johnson (UKHSA)
matthew.fisher@imperial.ac.uk 

Will climate and landuse change drive antifungal resistance exposures?

Humans are increasingly exposed to microbes that are both pathogenic and are resistant to frontline clinical drugs, leading to a growing public health burden. The Environment Agency 2020 review of airborne antimicrobial resistance identified a major knowledge gap in the importance of airborne exposure to antimicrobial resistant microbes and the sources of this problem - the studentship will directly address this knowledge gap by seeking to understand the processes leading to AMR exposures in our air.

Trends in exposure to airborne AMR owe to a variety of changing environmental factors: One is that is that increasing ‘dual usage’ of chemically similar drugs in agriculture and the clinic is leading to burgeoning human infection by genotypes that have pre-acquired antimicrobial resistance in the environment. The student will explore the relationship between the characteristics of bioaerosols across a cross-section of landscapes and uses to measure the type and incidence of AMR bioaerosols that humans are exposed to. Methods will include citizen-science high-throughput low-yield bioaerosol surveillance coupled with low-thoughput high-yield bioaerosol sampling and experimentation. This project will be integrated into NERC-funded ongoing urban and landscape-scale surveillance of bioaerosols and directly addresses NERC’s strategic focus on the ‘Environment and Health; Sources, sinks and pathways of potentially harmful chemicals and organisms present in the natural environment that may have an effect on human health’, including ‘Sources, sinks and pathways of potentially harmful chemicals/bioaerosols in the atmosphere.’

Machine Learning for Reducing PM2.5 on the London Underground

Subway systems are essential for sustainable urban transport but are often associated with elevated PM2.5 levels, posing health risks to passengers and staff. The London Underground, known for its deep, poorly ventilated tunnels, often reports the highest PM2.5 concentrations globally. This presents challenges for Transport for London (TfL) in understanding concentration patterns, identifying influencing factors, and evaluating interventions. Machine learning can provide the tools to address these issues and reduce public health risks.

This project aims to develop a predictive model for PM2.5 concentrations using machine learning, analysing data from fixed-location sensors on platforms and portable sensor measurements. By understanding the transferability of results across lines and stations, the model will enable network-wide concentration assessments, even in unmonitored locations.

Outcomes include a robust predictive tool to assess PM2.5 concentrations, the exposure of staff and passengers and the efficacy of interventions such as air cleaning and enhanced ventilation. The collaboration with TfL ensures practical application and data access. This research will provide actionable insights to improve air quality on the London Underground, ultimately reducing health risks and enhancing the commuting experience.

Environmental Data Assimilation: Advanced integration of networks for air quality and climate studies

Exposure to poor air quality is one of the critical issues of our time. According to the WHO 99% of the global population is exposed to air above its guideline limits.  It is a climate and a social justice issue with low/middle income countries (LMICs) and economically disadvantaged groups likely more exposed and less resilient.

Emerging sensing technologies allow for investigation of our environment at previously unachievable fine scales. Transformative sensing networks in partnership with reference monitoring allows mapping of environmental composition at unprecedented resolutions. Posing fundamental questions about our complex environment. There has also been a huge increase in availability of environmental model and satellite data. There is a gap in knowledge around how to effectively link these products across spatiotemporal scales to best exploit their various advantages. These advancing capabilities have highlighted the need for optimisation to maximise benefits, maintain affordability and avoid redundancy in sensing. This is critically needed in developing economies where resources are limited and direct impacts on health and the environment are significant, clear and pressing.

This proposal is focused on interrogation of data in the big data challenges of deteriorating air quality and climate globally. It has the potential to drive the transition to advanced monitoring based on integration of new technologies/approaches into existing systems and the design of future variable capability networks globally. 

Convolutional Compartmental Modelling for Outbreak Analysis and Response
 
Convolution has long been recognized as a fundamental operation in infectious disease modeling, enabling the linkage of prevalence to incidence, the application of delay distributions for transitions between states (e.g., mild to severe symptoms), and the modeling of waning immunity. During the COVID-19 pandemic, convolution-based approaches proved invaluable, allowing accurate estimation of disease prevalence and fatality rates while also enabling rapid prototyping of new models.

However, despite its power and adaptability, convolution remains underutilized in infectious disease modelling. Instead, models are often expressed as systems of differential equations, which inherently assume a smoothness that rarely aligns with real-world data. Conversely, large agent-based models capture granular processes but are computationally intensive and difficult to fit. Convolution provides a compelling middle ground, balancing realism with computational efficiency.

This project will establish a new general-purpose framework that we refer to as Convolutional Compartmental Modelling (CCM). This framework will be applied to the problem of parameter estimation in an outbreak context, where speed and flexibility are key. Datasets will include high-quality UK COVID-19 data, and historical data from other disease outbreaks. The outputs of this project will have cross-cutting impact across multiple diseases and will enhance our capacity to respond to emerging threats.

Ecosystem disequilibria, droughts, and arboviral disease risk in Brazil

Droughts represent complex climate extremes operating across diverse spatio-temporal scales and severely impacting natural ecosystems. In Brazil, the intensification of droughts, driven by anthropogenic pressures such as Amazon deforestation and biodiversity loss, disrupts ecological balances and creates conditions that heighten the risk of arboviral diseases. This project aims to develop a composite drought index by integrating various metrics, ranging from short-term meteorological indices (e.g., SPI-1) to long-term hydrological indicators (e.g., SRI-6), to better capture the multifaceted nature of drought events and their environmental impacts.

The PhD project will analyse the cascading effects of droughts on ecosystems, focusing on how these dynamics influence the habitat of disease vectors and their interaction with climatic variability and socio-economic local conditions. By integrating these data within a novel spatio-temporal framework, this research will advance our understanding of the pathways linking drought dynamics, biodiversity loss, and disease transmission.

Ultimately, the findings will contribute to strategies for mitigating disease risks and address priorities in environmental resilience. 

Although the impact of these diseases is particularly high in tropical and subtropical regions, the increasing risk they pose in Europe and the UK highlights the potential for adapting this framework to other regional contexts.
This PhD project builds on the partnership between Imperial College, the University of São Paulo (Brazil), and the Paster Institute (Brazil) and will take advantage of a collaboration with the National Institute for Space Research (Brazil).

Somatic mutations in healthy tissue: the exposome, biological signatures and clinical outcomes

Somatic clones are groups of cells that share acquired DNA mutations, inherited from a progenitor cell. They occur naturally and increase with age, being a good biomarker of the ageing process. They also have been linked to cancers, although most clones don’t develop into tumours. Still, much is unknown about the origins and destiny of somatic mutations especially in healthy tissue. Here I propose an extensive study of the exposome’s influence on somatic clone formation and growth, as well as of biological signatures of such mutations when present. The main datasets available are (i) UK Biobank, a large and deeply characterised cohort of over 500k healthy participants and (ii) REACT-LC, a collection of over 10k COVID cases with persistent symptoms and controls. Both cohorts have produced a rich collection of exposures and molecular data in blood, including questionnaires, genomics, transcriptomics, proteomics, metabolomics and healthcare linkage data. The project can be further divided in the following objectives: 1) Identify environmental and lifestyle factors (i.e. the exposome) associated with somatic clonality in healthy populations, 2) Identify biological and molecular signatures of somatic clonality, 3) investigate clinical outcomes associated with somatic clonality. The project builds on the ongoing success of machine learning on large healthcare datasets and combines epidemiology and genetics to study somatic mutations at an unprecedented scale. Results from the project will significantly impact our knowledge of the underlying mechanisms involved in ageing and cancer, both open challenges for our healthcare.

Impacts of fuel duty changes on health and the environment

The transport sector is a key source of air pollution. Fiscal measures exert a strong influence transport behaviour and the UK government has considerable powers in this area, most prominently their ability to change fuel duty levied on petrol and diesel. Fuel duty raises significant revenue for the government, although rising fuel efficiency over time has meant that revenues from fuel duty have been decreasing as a percentage of GDP, raising concerns over the viability of the current system. Any changes to fuel duty may change levels of air pollution through changed transport behaviours in addition to changes in human health. There is however, a lack of evidence on whether this is the case. 

This project will test the hypothesis is that increases in fuel duty will reduce air pollution through changes in car use, and also impact other health outcomes. It will first synthesise evidence from the existing literature on the impacts of fuel duty changes on health, and second it will use routine data to examine the historical impacts of changes to the fuel duty regime in England on health and environmental outcomes. The final part of this project will use transport and air pollution modelling to provide estimates of potential impacts of changes to fuel duty on air pollution, car use and road traffic incidents under a range of different scenarios. 

Developing novel generalizable dengue forecasting tools for endemic settings

The heterogeneous patterns of dengue transmission pose substantial challenges to surveillance systems globally and especially in LMICs where limited resources require the optimal implementation of interventions. Reproducing dengue transmission dynamics is complex due to its epidemiology and changes in population demography and the climate. Building on recent spatiotemporal regression and machine learning models developed at small spatial resolutions for variable forecasting windows, we propose investigating the development of new models and software which can shift the current paradigm in dengue forecasting. In this project, we will assess the performance of nested forecasting models, e.g. the use of the output of a primary forecasting model as input to a secondary forecasting model. We also aim to develop generalizable software to allow the implementation of the forecasting tool by surveillance systems around the world. This will consist in the development of pipelines to (i) download and prepare climate and environmental variables from global gridded products, (ii) run individual and nested forecasting models with variable spatial and temporal resolution and lead times determined by the user and (iii) generate probabilistic forecasts including central and uncertainty estimates. The models developed during the PhD will be applied to global historical time series dengue data. 

Micro-Plastic Measurements

Micro- and nanoplastic pollution is a global environmental and health challenge. Originating from the degradation of plastic materials, including synthetic tyres and some PFAS, their ubiquity in the environment and food systems highlights the likelihood for population exposure. However, an understanding of environmental concentrations and exposure estimates, underpinned by robust and sensitive analytical techniques, is lacking but necessary to determine the level of risk that micro- and nanoplastics present. This project will aim to fill this gap, with opportunity for both field and laboratory-based work, and focus on air, but also scope to include freshwater environments. The project will be based within the Microplastics Team in the Environmental Research Group, where there is state of the art infrastructure for microplastics research, including pyrolysis gas chromatography mass spectrometry, a multimodal imaging platform with Raman capacity, and a dedicated clean environment. There is also critical infrastructure in the field, including access to air quality monitoring supersites, particle samplers and online instruments, and freshwater sampling sites. We have collaborators around the world, including Indonesia, Canada, and Australia, providing unique international opportunities for fieldwork, networking, and training.

Dr Dimitris Evangelopoulos (ERG)
Garyfallos Konstantinoudis (Grantham)
David Green (ERG)
d.evangelopoulos@imperial.ac.uk 

Assessing people's exposure and the associated health impacts of wildfires and extreme heat

Extreme heat and wildfires have become a public health issue worldwide, and they are projected to increase in frequency and intensity across Europe. Recent findings have shown that both short- and long-term exposure to wildfires lead to significant mortality and morbidity increases. However, there are still methodological challenges in epidemiological analyses, such as the air pollution exposure assessment that can be directly attributed to wildfires at fine spatial scale or the estimation of wildfire pollution-specific health impacts. In this project, the student will combine a large set of data sources, including air pollution measurements and models, satellite imagery and weather predictions within a machine learning multi-stage approach to estimate fine particulate matter (PM2.5) specifically from wildfires and its impact on health. The area of study will be the UK where more than 125 thousand hectares have been burned from 2009 onwards, but also other European countries, mainly from Southern Europe where wildfires are more often and more intense. The wildfire PM2.5 concentrations will be linked with cardiorespiratory mortality and morbidity data at fine spatial scale and their association will be quantified. Combined effects of wildfire exposure and extreme heat events will also be assessed. The aim of this study is to quantify the impacts of an emerging climate change hazard and inform policies on prevention strategies.