Periodic Reporting for period 1 - EstAMR (Estimating the Prevalence of AntiMicrobial Resistance)
Période du rapport: 2020-04-01 au 2022-03-31
To prevent drug resistance, treatment needs to be effective and the full prescribed course of treatment needs to be completed. However, this can be difficult to monitor, especially in low to middle income countries where health care access may be limited. For example, perhaps a treatment centre is using low quality drugs, which is known to promote drug resistance occurring and spreading. This model allows these treatment centres to be identified, whilst accounting for uncertainty in the data collection, and varying disease prevalence over the region.
To demonstrate the main model of my project, consider a toy example in Square Land over 20 years (see figures). Within Square Land you have sampled infected people, at different locations and times, and identified whether they carry a resistant strain or not. In addition, suppose you know the location of five treatment access points, such as health care centres or hospitals (however knowledge of the exact locations is not a requirement of the model). There is data on infected people in Square Land for 20 years. Some patients are carrying a sensitive strain, and some are carrying a resistant. We know the location and time that people were tested, and that there is more disease prevalence at the bottom of Square Land.
The model created here identifies which hotspot is introducing more resistant infections into the population. In this example, the model identifies that the top left hotspot (hotspot 1) is contributing the most resistant infection into the population. Therefore, if hotspot 1 was a health care centre, it would be worth investigating the quality of the drugs administered here and/or the adherence of the patients. If this hotspot was a transport hub, it would inform us that drug resistance is entering the population from outside. The middle hotspot (hotspot 3) has more resistant infections in the region (due to higher prevalence of malaria cases here), however because the model explicitly accounts for the higher prevalence of malaria (not necessarily drug resistant malaria) at the bottom of Square Land, it stills successfully identifies hotspot 1. Analysis without a mechanistic component, as presented here, would incorrectly suggest investing hotspot 3, not hotspot 1.
To demonstrate the main model of my project, consider a toy example in Square Land over 20 years (see figures). Within Square Land you have sampled infected people, at different locations and times, and identified whether they carry a resistant strain or not. In addition, suppose you know the location of five treatment access points, such as health care centres or hospitals (however knowledge of the exact locations is not a requirement of the model). There is data on infected people in Square Land for 20 years. Some patients are carrying a sensitive strain, and some are carrying a resistant. We know the location and time that people were tested, and that there is more disease prevalence at the bottom of Square Land.
The model created here identifies which hotspot is introducing more resistant infections into the population. In this example, the model identifies that the top left hotspot (hotspot 1) is contributing the most resistant infection into the population. Therefore, if hotspot 1 was a health care centre, it would be worth investigating the quality of the drugs administered here and/or the adherence of the patients. If this hotspot was a transport hub, it would inform us that drug resistance is entering the population from outside. The middle hotspot (hotspot 3) has more resistant infections in the region (due to higher prevalence of malaria cases here), however because the model explicitly accounts for the higher prevalence of malaria (not necessarily drug resistant malaria) at the bottom of Square Land, it stills successfully identifies hotspot 1. Analysis without a mechanistic component, as presented here, would incorrectly suggest investing hotspot 3, not hotspot 1.
(i) Developed a mechanistic hierarchical Bayesian model that captures underlying spatiotemporal dynamics on the emergence and spread of drug resistant infections.
RESULTS: Using my model, resistance hotspots are ranked, thus enabling resources to be targeted - such as verifying the quality of drugs at a particular health care centre. I demonstrated the model using a simulated dataset, where even sophisticated regression based models incorrectly rank the hotspots. This paper promotes the need to revisit the sampling approach when performing molecular surveillance.
DISSEMINATION: Preprint on Biorxiv, and under review with Royal Society Open Science where I am the first author.
(ii) Predicted the prevalence of drug resistant malaria infections in Indian states where data is lacking.
OVERVIEW: In low to middle income countries, socioeconomic status effects the prevalence of drug resistance. In India, people in states with a low socioeconomic status are more likely to share their antimalarial medicine (and not complete the course themselves), or visit a health care centre that uses substandard drugs. Using data available online, my Masters student predicted the prevalence of drug resistant malaria in Indian states where data is absent. We showed that socioeconomic factors can be stronger influencers than epidemiological factors.
RESULTS: We showed that the best performing model used the (average) number of people per government hospital in each state, and the capita income for each state. This model outperformed models with a focus on epidemiological factors, such as the prevalence of malaria cases.
DISSEMINATION: A blog entry written by a Master student is available on my website.
(iii) Built a model that captures the rise and decrease of drug resistant malaria in Guyana.
OVERVIEW: A dataset gathered by my colleague (Pablo Martinez de Salazar) is unique because it captures that young men, with no immunity, moved from the capital to mining regions due to an increase in the price of gold. Therefore, this real life data allows us to quantify key influencers on the spread of drug resistant malaria. I developed a model that captures the rises and fall of the prevalence in drug resistant malaria, as seen in the data, such as whether treatment was sought in reputable health care centres, and the increasing immunity of the miners.
RESULTS: When a malaria region receives a large influx of hosts without immunity, malaria cases increase and drug resistance can occur when health facilities are poor. However, as the hosts immunity levels increase, and many hosts leave the region, drug resistant infections become less prevalent.
DISSEMINATION: A paper currently in draft where I am the last author on the paper. Furthermore, Pablo Martinex de Salazar has submitted an application for a five year grant to the Cloetta foundation, part of which will develop this work further.
RESULTS: Using my model, resistance hotspots are ranked, thus enabling resources to be targeted - such as verifying the quality of drugs at a particular health care centre. I demonstrated the model using a simulated dataset, where even sophisticated regression based models incorrectly rank the hotspots. This paper promotes the need to revisit the sampling approach when performing molecular surveillance.
DISSEMINATION: Preprint on Biorxiv, and under review with Royal Society Open Science where I am the first author.
(ii) Predicted the prevalence of drug resistant malaria infections in Indian states where data is lacking.
OVERVIEW: In low to middle income countries, socioeconomic status effects the prevalence of drug resistance. In India, people in states with a low socioeconomic status are more likely to share their antimalarial medicine (and not complete the course themselves), or visit a health care centre that uses substandard drugs. Using data available online, my Masters student predicted the prevalence of drug resistant malaria in Indian states where data is absent. We showed that socioeconomic factors can be stronger influencers than epidemiological factors.
RESULTS: We showed that the best performing model used the (average) number of people per government hospital in each state, and the capita income for each state. This model outperformed models with a focus on epidemiological factors, such as the prevalence of malaria cases.
DISSEMINATION: A blog entry written by a Master student is available on my website.
(iii) Built a model that captures the rise and decrease of drug resistant malaria in Guyana.
OVERVIEW: A dataset gathered by my colleague (Pablo Martinez de Salazar) is unique because it captures that young men, with no immunity, moved from the capital to mining regions due to an increase in the price of gold. Therefore, this real life data allows us to quantify key influencers on the spread of drug resistant malaria. I developed a model that captures the rises and fall of the prevalence in drug resistant malaria, as seen in the data, such as whether treatment was sought in reputable health care centres, and the increasing immunity of the miners.
RESULTS: When a malaria region receives a large influx of hosts without immunity, malaria cases increase and drug resistance can occur when health facilities are poor. However, as the hosts immunity levels increase, and many hosts leave the region, drug resistant infections become less prevalent.
DISSEMINATION: A paper currently in draft where I am the last author on the paper. Furthermore, Pablo Martinex de Salazar has submitted an application for a five year grant to the Cloetta foundation, part of which will develop this work further.
BEYOND STATE OF THE ART: The model itself is beyond state of the art because we develop a partial differential equation that is specifically aimed for capturing the effect of varying hotspots for drug resistant infections. As with previous mechanistic hierarchical models, the partial differential equation has diffusion and growth terms. To capture the importance of drug resistance hotspots, we added a new term which accounts for the distance from a resistance hotspot, and thus we can determine which hotspot has more drug resistance occurring in the surrounding region, whilst explicitly accounting for confounding factors such as disease prevalence.
INNOVATION IN DRUG RESISTANCE MODELLING: Before this project, mechanistic hierarchical models were not used to capture the spread of drug resistance because of biases in the data collection. The data is biased because currently, when drug resistance is suspected to be in a region, patients in the region are sampled and the suspicion is confirmed. This sampling approach means that strategies to lengthen the lifespan of drugs are
reactionary, not proactive. However, because testing for drug resistant infections is becoming easier and cheaper, allowing the possibility to sample more people, now is the right time to revisit the sampling approach and consider the public health questions that are currently difficult to answer because of a lack of insight into the underlying dynamics. My work demonstrates the possibilities if we evolve our questions about the emergence and spread of drug resistance alongside the technology evolution.
BENEFITS FOR GLOBAL HEALTH: Being able to rank hotspots means strategies to prolong the lifespan of a drug can be improved by efficiently targeting hotspots, such as health care centres.
SHAPING FUTURE RESEARCH: Bringing state of the art models to epidemiology, because in comparison to regression based models, such as the generalized additive model, the model more accurately predicts the spatiotemporal density of patients infected with a drug resistant genotype. This is particularly relevant for antimicrobial resistance in general, an action point of the European Union.
INNOVATION IN DRUG RESISTANCE MODELLING: Before this project, mechanistic hierarchical models were not used to capture the spread of drug resistance because of biases in the data collection. The data is biased because currently, when drug resistance is suspected to be in a region, patients in the region are sampled and the suspicion is confirmed. This sampling approach means that strategies to lengthen the lifespan of drugs are
reactionary, not proactive. However, because testing for drug resistant infections is becoming easier and cheaper, allowing the possibility to sample more people, now is the right time to revisit the sampling approach and consider the public health questions that are currently difficult to answer because of a lack of insight into the underlying dynamics. My work demonstrates the possibilities if we evolve our questions about the emergence and spread of drug resistance alongside the technology evolution.
BENEFITS FOR GLOBAL HEALTH: Being able to rank hotspots means strategies to prolong the lifespan of a drug can be improved by efficiently targeting hotspots, such as health care centres.
SHAPING FUTURE RESEARCH: Bringing state of the art models to epidemiology, because in comparison to regression based models, such as the generalized additive model, the model more accurately predicts the spatiotemporal density of patients infected with a drug resistant genotype. This is particularly relevant for antimicrobial resistance in general, an action point of the European Union.