GDRR Opening Workshop - PANEL SESSION: Risk Analysis: Antimicrobial Resistance - Jennifer Clarke, August 6, 2019
1. Antimicrobial Resistance
Challenges
Risk Analysis: Antimicrobial Resistance (AMR)
Jennifer Clarke
Department of Statistics, University of Nebraska-Lincoln
August 6, 2019
Opening Workshop, SAMSI GDRR
Jennifer Clarke Risk and AMR
2. Antimicrobial Resistance
Challenges
Antimicrobial Resistance (AMR)
First described in 1940 in E. Coli
Most antimicrobials in clinical use are naturally produced
by soil microorganisms; source of many resistance genes
AMR is ancient and natural part of genome of
environmental bacteria (self-preservation)
Intrinsic vs. acquired
Was rare in clinical isolates prior to antibiotics; Infections
caused by resistant bacteria associated with increased
morbidity, mortality and economic cost
Jennifer Clarke Risk and AMR
4. Antimicrobial Resistance
Challenges
Many Relevant Factors
Genomic factors:
Horizontal Gene Transfer (integrative and conjugative
elements (ICEs))
Mutations (hypermutators)
Selection pressure (Evolutionary advantage vs. ’Fitness
cost’)
Jennifer Clarke Risk and AMR
5. Antimicrobial Resistance
Challenges
Many Relevant Factors
Genomic factors:
Horizontal Gene Transfer (integrative and conjugative
elements (ICEs))
Mutations (hypermutators)
Selection pressure (Evolutionary advantage vs. ’Fitness
cost’)
Environmental factors:
Mechanism adoption (e.g., efflux pumps)
Impact of microbial ecology largely unknown
Jennifer Clarke Risk and AMR
6. Antimicrobial Resistance
Challenges
Many Relevant Factors
Genomic factors:
Horizontal Gene Transfer (integrative and conjugative
elements (ICEs))
Mutations (hypermutators)
Selection pressure (Evolutionary advantage vs. ’Fitness
cost’)
Environmental factors:
Mechanism adoption (e.g., efflux pumps)
Impact of microbial ecology largely unknown
Actor factors:
Medical use
Animal producer use and management practices
Public health economics
Jennifer Clarke Risk and AMR
8. Antimicrobial Resistance
Challenges
Statistical Issues
Statistical Issues
Feature Selection (think GWAS):
Many potential variables; correlated and multitype
Small but important effect sizes
Prevalence:
Events are (still) rare so data sparse
Importance of simulation
Jennifer Clarke Risk and AMR
9. Antimicrobial Resistance
Challenges
Statistical Issues
Statistical Issues
Feature Selection (think GWAS):
Many potential variables; correlated and multitype
Small but important effect sizes
Prevalence:
Events are (still) rare so data sparse
Importance of simulation
Risk Prediction (see Grunwald, Clarke, Bartlett):
Acquisition is a dynamic evolutionary process; prediction
may be sequential
If develop a learner that generates a posterior, what is
learning rate?? Rademacher complexity?
Adversarial loss? [Many machine learning models
vulnerable to adversarial attacks, e.g., adding adversarial
perturbations imperceptible to humans can make ML
models produce wrong predictions with high confidence.]
Jennifer Clarke Risk and AMR