Spotlight | What are mathematical modellers doing about AMR?

Last week, over 400 mathematical modellers of infectious disease dynamics met for the biennial conference, Epidemics. This year, for the first time, there were two sessions on antimicrobial resistance (AMR). There was also a plenary by Marc Bonten, which focused on whether there really is any evidence for the spread of AMR from livestock to humans. For his three examples, the answer was VanA may have had an animal origin in VRE, livestock associated MRSA is not clinically important and there’s no evidence for ESBL but its complex! This chimes well with trying to formulate a model of livestock driving clinical AMR – can we really expect to see many infections originating from meat that we cook?

LSHTM alumni, Esther van Kleef, kicked off the first AMR session with a hospital model that showed nicely that although we might expect “horizontal” hospital infection control methods to reduce the transmission of sensitive and resistant equally, if the resistant strains are hospital adapted and somehow not sustained by incoming rates then we might expect a bigger impact on infections with resistance than sensitive strains. Other talks in the sessions ranged from exploring whether we should model breadth vs. intensity of antibiotic use, modelling the lag time between seasonally varying antibiotic use and resistance to uncover a “stabilising” factor, quantification of potential bystander selection exposure and modelling individual cells and antibiotic concentration degradation to explore optimal dosing strategies.

A key focus of these sessions was co-existence: why don’t drug-resistant strains dominate? What is the link between prescribing and resistance? The correlation between outpatient prescribing and penicillin resistance in Streptococcus pneumoniae across Europe from Herman Goossen and the ESAC project was shown in multiple presentations. It all started with a great talk from our own Nick Davies showing how a new type of underlying model structure, allowing for small subpopulations of strains to exist alongside a dominant population, might explain the variation we see. However, several other mechanisms for this coexistence were also proposed from, among others, Christophe Fraser’s group at Oxford, including differential lengths of carriage for resistant vs. susceptible strains, competition and population structure.

There were no talks on drug resistance in TB, although there were a few interesting posters. Most presentations instead focused on data from S. pneumoniae – is this because there is already a lot of data for this “interesting” bug to model with the competing serotypes and vaccine impact implications? Or is it a “representative bug” for AMR? Can we make “generalisable” modelling assumptions that, for example, resistant strains are carried for less time than susceptible ones?

Looking across the rest of the conference, many talks described models for outbreak forecasting, a lot were on influenza and several new R packages were discussed. Compared to the complex, individual patient data driven models presented for the outbreak diseases, such as Zika, the underlying mechanics of AMR are still being debated. We have a way to go to get to model forecasting in AMR, but this conference highlighted a wave of exciting models and modellers making predictions that now need to be tested with detailed data and presented to the wider public health community.