The dynamics of drug resistance within hospital populations of Gram-negative bacteria

When bacteria become resistant to antibiotics the infections they cause are harder to treat. In the UK, and globally, we are seeing an increase in the number of infections being caused by antibiotic resistant bacteria. This could lead us back into the pre-antibiotic era before the early 1900s when infections of simple cuts may become life threatening and cancer treatments which suppress the immune system, and rely on antibiotics to prevent infections, will be unusable. Antibiotic resistance arises through the use and misuse of antibiotics and so, even if we develop new antibiotics, it is likely that we will always be faced with the problem of resistant strains.

We need to consider how best to preserve our existing antibiotics without compromising patient care. One of the ways to do this is to look at why some places have less resistance than others and to try to work out what they are doing right. For example, some hospitals have fewer infections with resistant bacteria than others. In this project I will explore what the reasons for these differences might be and translate these findings into ways to better prevent resistance from spreading.

To do this I will build mathematical models of antibiotic resistance spread. Mathematical models are frameworks in which different subpopulations are separated out and the rates at which these sub-groups increase or decrease are calculated. For example, a model would split a hospital population into those with and without infections with resistant or susceptible bacteria. It would then consider at what rate and by what mechanism people become infected, and then explore what the difference is between those with and without resistance. By writing this framework down mathematically, we get a better understanding of the processes underlying the spread of resistance and can identify the key targets - for example overuse of a certain type of antibiotic - that need to be tackled. From this understanding of the resistance spread, the model can be used to predict what will happen in the future without interventions and then compare this to what would happen if certain interventions were introduced.

I will build mathematical models that capture what is happening in different hospitals and determine why some have lower rates of resistance than others. In particular, I will look at the development of resistance within a group of bacteria called the Gram-negatives. These bacteria are often found living in the gut, but they can travel to other parts of the body and, in the UK, are the most common cause of serious hospital-associated infections such as bacteraemia (infection of the blood). Increasingly we are seeing strains of these bacteria becoming resistant to common, powerful antibiotics and so they are a key contributor to antibiotic resistance.

Groups of bacteria can have very different characteristics and can grow extremely rapidly. Their genetic make-up is very flexible which means that new genetic changes can occur or new pieces of genetic material can jump between bacteria creating and spreading resistance. In a bacterial population there will then be many different strains that may have many different resistances. This diversity has rarely been considered in mathematical models before, and so we may be missing a key part of resistance evolution. In this project I will develop mathematical models to incorporate this diversity and to determine how much of an impact it is having on resistance spread.

Mathematical models must be grounded in data in order to be relevant to clinicians and public health. In this project I will use the newly collected hospital level data on antibiotic usage and resistance to both gain parameters for my models and to determine what patterns the models should capture. My results will then be immediately useful for clinicians and the NHS, and will directly influence the interventions used to control the appearance and spread of antibiotic resistance.

Mathematical models of the spread of antibiotic resistant bacteria (ARB) in a hospital usually do not take into account individual hospital level data on antibiotic usage or resistance. Nor do they consider the full complexity of resistance mechanisms, usually considering only a binary state. In this project I will construct new frameworks for ARB spread in hospitals.
Firstly, I will extend an existing deterministic mathematical model to take into account not only diversity in fitness (or transmissibility) between bacteria, but also diversity in ability to survive at different antibiotic concentrations (i.e. resistance level). Diversity will be incorporated through a weighted mean function and the model will be analysed using a new diversity metric function. This will provide a theoretical framework to determine how resistance within a bacterial population changes under constant and then changing selective pressure. This framework will be adapted to account for mobile genetic element movement within Gram-negative bacteria and to include a stochastic rate at which resistances appear.
The second part of the project will utilise hierarchical statistical methods that I wish to gain knowledge of and will have training in during this fellowship. These will allow me to determine trends in antibiotic usage and resistance. I will use time series analysis and multivariate regression to link these two data. Multi-level modelling will then test the links between hospital and patient level factors in driving differences in resistance and antibiotic usage.
The final part will bring together the information on the bacterial population and the input the data on clinical trends. An initial deterministic homogeneously mixing compartmental model will be extended to include further population stratification and stochastic transmission, to compare intervention impact. This will be fitted to clinical data using Bayesian and Markov Chain Monte Carlo methods.

My choice of Professor Alison Holmes, Director of the Health Protection Research Unit (HPRU) on Hospital Associated Infections and Antimicrobial Resistance and Director of Infection Prevention and Control (DIPC) for Imperial College Healthcare NHS Trust, as my primary sponsor, means that I am basing my mathematical modelling in an interdisciplinary environment of clinicians, epidemiologists, statisticians and public health implementers with direct links to the NHS. This decision was made in order to maximise the clinical impact and relevance of the modelling tools developed during this project. Moreover by using data from Imperial College Healthcare NHS Trust and other Trusts, I will increase the credibility, applicability and impact of my modelling results.

This HPRU builds on a long-standing relationship between Imperial College London and Public Health England (PHE), which I have formalised here by including Dr Susan Hopkins and PHE as my second sponsor and Project Partner. I will therefore be able to not only have a direct impact on policy within the Imperial College NHS Trust, but also on national level decision making through PHE. Similarly, by engaging and directly modelling specific hospitals across England I aim to have an impact both on their individual methods of antibiotic resistance control, but also on the ways in which national guidelines are designed, implemented and their impact assessed. An extension to impact at the international stage will be made through Professor Holmes' international network, through international workshops and through continuing relationships with researchers at the London School of Hygiene and Tropical Medicine.

The other aspect of the impact will be on other modellers working in the field of antibiotic resistance by developing new tools and frameworks for the inclusion of antibiotic usage. Moreover, by analysing and publishing analysis of hospital level data in a format useful to modellers, I hope to improve the accessibility of data from a range of diverse sources, aiding in the development of a larger community of modellers engaging directly with antibiotic resistance and other problems within clinical settings. By developing such tools, I hope to make such modelling more accessible to those in developing countries where we have very little handle on resistance development.

By focusing this fellowship on a clear public health issue (antibiotic resistance spread), one aspect of the impact will be with clinicians and policy makers. This will be by providing a greater understanding of the drivers and differences in antibiotic resistance prevalence across England, which will inform better methods to control and prevent resistance spread. Moreover, I hope to improve the reporting and quantitative analysis of clinical data by demonstrating the power of such investigation. Ultimately, the impact will then be on improved patient care and a reduction in the number suffering from difficult to treat infections with resistant bacteria.