The National Academies of Science engineering

1
Combating Antimicrobial Resistance with Big Data
Outcomes in the DoD and Challenges for Large
Health Systems

• The National Academies of Science engineering
  medicine
• 
  forum on microbial
  threats
• Emil P. Lesho, DO, FACP, FIDSA, FSHEA
• Colonel, Medical Corps, U.S. Army

2
Disclaimer / Disclosures

• Solely the views of the author.
• Not official or to be construed as the official
  views of WRAIR, Army, Navy, or DOD
• No conflicts, nothing to disclose

3
Why care about what the DoD is doing?

• AMR borderless problem global threat to public
  health
• Travel, displacement, and conflict associated w/
  AMR
• Special population deserving of safe, high
  quality care
• Aligned with, and component of, the National
  Action Plan
• Transparency - taxpayer funded publically
  available
• Database, isolates / pathogen panels
• Portable / translatable approaches products
• Challenges and barriers If DoD can do it, any
  program should be able to
• (Although currently no match exists for DoDs
  magnitude and speed)
• 48 hr. TAT for SNP-based outbreak support,
  regardless of location
• Large HMOs could face same constraints and
  restraints.

4
(No Transcript)
5
Outline

• I. Size of the challenge, how the DoD
  generates, analyzes stores AMR data
• II. Outcomes
• Big data vs. Gram negatives
• Big data vs. acute resp. infection for antibiotic
  stewardship
• III. Challenges
• Big data vs. Gram positives more or bigger not
  always better
• .mil versus big data
• Information Management vs. big data

6
Potential Size of the Data Challenge
 
/ year
2/3 will stay healthy, not seek care or not have
infection related condition
Others gt356 day LOS gt 100 cultures
3 million -6 billion results
 
7
Aligned with One World/One Health - not limited
to human data
8
Creation and Analysis of Big Data in the DoD

• Created and analyzed by 1) leveraging public
  health and infection control surveillance
  mandates 2) using a high-throughput organism
  identification, susceptibility testing, and whole
  genome sequencing pipeline 3) establishing
  specialized datamarts.
• Datamarts
• Military Health System (MHS) Health Level 7 (HL7)
  repository of the U.S. Navy
• Pharmacovigilance Defense Application System
  (PVDAS) of the US Army.
• MHS HL7 receives data from the Composite Health
  Care Systems that contain laboratory, radiology
  and pharmacy data.
• MHS HL7 has gt2 billion records for beneficiaries.
  SQL algorithms identify ESKAPE pathogens and
  depict patterns that are run against the
  restructured data to generate specific
  surveillance products.
• The PVDAS receives data from a centralized data
  repository that captures, validates, integrates
  and stores data from medical claims, hospitals ,
  eligibility and enrollment, death files, and
  pharmacy transactions.
• The PVDAS incorporates data on 12 million
  patients including demographics, prescriptions,
  diagnoses, laboratory results. JAVA, SAS and SQL
  queries perform pharmacovigilance by monitoring
  prescribing and adverse drug reactions and
  safety alerts.

9
Creation of Big Data ARMoR
10
How the Central AMR Lab Creates and Stores Big
Data
11
How the Central AMR Lab Creates and Stores Big
Data
12
From ERIK SNESRUD
MRSN High Throughput Sequencing Data Analysis
Pipeline
Semi-Automated Analysis
Data Processing
PacBio Sequence
Illumina Sequence
Data Processing

• Outbreak Identification
• SNP and InDel Identification
• Gene Loss and Acquisition
• Rearrangement / Transposition

Merge Overlapping Reads
FLASH
Read Quality Trimming
Btrim

• Resistance Mechanism Detection
• SNP and InDel Identification
• Rearrangement / Transposition

De Novo Assembly
Newbler
Data Processing
Data Processing

• Assembly Quality Control
• Read Coverage
• Contig N50
• Contamination Detection

SMRT Analysis Software Package

• Resistance Transfer Mechanisms
• Transposon Structure and Function
• Plasmid Transfer Mechanisms

Automated Process
Species ID
ABR Genes
WG Annotation
WG Phylogeny
Outbreak ID
Automated Search of 16S Database
Antibiotic Resistance Gene Annotation
Whole Genome Annotation RAST/Prokka
Align Reads to Reference Genome
Complete Genome/Plasmid Comparison
Extract SNPs/Indels and Filter
Identify SNPs, InDels, and rearrangements
MLST
Plasmid Typing
VIR Genes
Bayesian Inference of Phylogeny
Multilocus Sequence Typing Database
Plasmid Replicon Typing Database
Virulence Gene Annotation
Genome Sequencing Database
13
Summary of Pipeline Functions

• 16S species confirmation
• Resistance gene content
• Virulence gene content
• MLST derivation
• Plasmid finder
• 4 methods for relatedness / phylogeny
• PFGE converter
• Mauve
• Nucleotide identity (BLAST- MUMmer)
• Read mapping to reference

14
Big Data vs. Gram Negatives

• Objective Assess correlations b/t ABX CRE at
  level of entire national heath system
• Data set 75 million person years 1.97 million
  cultures from 266 hospitals (globally, then by
  regional, facility, drug)
• Results Fluoroquinolones correlated w/ CR in E.
  coli at referral centers (P lt .001)

15
Big Data vs. Gram Negative Outcome CRE
Proportions in the DoD
16
Big Data vs. Gram Negatives Outcome Trends in
Carbapenemase producers

• The Challenges of Implementing Next Generation
  Sequencing Across a Large Healthcare System, and
  the Molecular Epidemiology and Antibiotic
  Susceptibilities of Carbapenemase-producing
  Bacteria in the Healthcare System of the U.S.
  Department of Defense
• PLOS One (at press)

17
Big Data vs Gram Negatives Fermenters vs.
Non-fermenters

• Objective Differential burden, relative risks,
  associations with antimicrobial consumption, and
  temporal trends of relevant taxa
• Data set 360,000 potentially carbapenem-resistant
  strains were identified from 14.7 million
  cultures
• Outcomes
• Isolation overseas or isolation from the
  bloodstream associated with a higher relative
  risks of carbapenem resistance (CR) (plt0.0001)
• Enterobacteriaceae isolated 11 times more
  frequently than P. aeruginosa and Acinetobacter
  spp.
• Compared to Enterobacteriaceae, CR was 73-fold
  and 210-fold higher in P. aeruginosa and
  Acinetobacter spp. respectively.
• Overall, CR rates increased for
  Enterobacteriaceae (p 0.03), and decreased for
  Acinetobacter spp. and P. aeruginosa (p lt0.0001).

Special Challenges for Trending Breakpoint
harmonization Breakpoint updates
De-duplication / identity management
Adjudication
18
Big Data for Better Stewardship

• There were gt13 million visits for acute
  respiratory infections. In 2006, 2011, and 2014,
  49, 54 and 48 respectively, of these visits
  had an antibiotic prescribed.
• Antibiotics were more likely to be prescribed to
  females, retirees and dependents of retirees, and
  persons 45 years and older
• 311 patients received potentially problematic
  prescriptions or formulations of an
  antimicrobial.
• Data from OTSG PVC (COL Trinka Coster)

19
Big Data Not Necessarily Better Data

A Example from Fighting Gram Positives

• Objective Methods
• S. aureus one of the most common and virulent
  pathogens
• Vancomycin MIC associated with outcomes in both
  methicillin susceptible and methicillin resistant
  S. aureus
• MIC 1gt lt4 susceptible (VSSA) but problematic
  PVSSA
• Manual broth dilution (MBD) gold standard, but
  laborious not feasible large scale
• STL (Seasonal Trend decomposition using Loess),
  ARIMA (AutoRegressive Integrated Moving Average)
  models.

• Data Set
• 230 million patient encounters
• 6.5 million bacterial cultures
• 81,018 unique cultures
• Isolates tested on
• VITEK (bioMérieux) 58
• Phoenix (BD) 14
• MicroScan (Siemens/BC) 20

20
Gram Positive Outcomes
21
Why only Phoenix and Vitek
Percent PVSSA Across the DOD
Percent PVSSA at Facility 123
22
Trends in Isolates Percentages
Combined Trends
23
(No Transcript)
24
Big Data vs. Gram Positive Outcomes

• S. aureus and PVSSA incidences are decreasing in
  this population / healthcare system
• Trends in the usage of most anti-staphylococcal
  drugs are decreasing or flat, and generally
  mirror trends of S. aureus incidence
  (Rx/infections is constant over time )
• PVSSA incidence is decreasing at a faster rate
  than the usages of some drugs (DAP, LIN, NAF)
• Downward trends are concurrent with maturation of
  MRSN-EDC (ARMoR Program) functionality
• Time, season/quarter, and usage of CEF, CEP, DOX
  and TRI were correlated with PVSSA percentages

25
Generating Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Lengthy approval processes and laborious
  acquisition requirements contract awards unable
  to keep pace with technologic advances

• Allow cooperative research agreements with
  operations and maintenance type of funds employ
  experienced acquisitions personnel within group
  to work closely with contracting agency vendors
  should notify contracting officer representatives
  or technical supervisors of impending major
  advancements or new releases allow clinical
  operations to be funded with research and
  development monies (not solely operations and
  maintenance monies)

26
Generating Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Balancing number of full time staff to workload

• 3-4 full time molecular laboratory technologists
  and one PhD level team lead for every 300-400
  isolates sequenced per month

27
Generating Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Limitations of shorter read platforms for certain
  types of bacterial antimicrobial resistance
  investigations (mobile genetic elements)

• Increase access to or funding for positioning of
  ultra or very long read sequencing platforms at
  surveillance or referral laboratories

28
Generating Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Limited availability of long read single
  molecule platforms)

• Wait for technologic advances to eliminate this
  constraint by making those platforms smaller and
  less expensive.

29
Generating Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Compared to research laboratories, clinical
  laboratories are more susceptible to higher
  staff turnover and may not have staff with
  specialized training needed for preparing high
  quality DNA libraries

• Increase and incentivize educational and training
  opportunities leverage automation or robotics
  for library preparation

30
Analyzing Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Balancing number of full time staff to workload

• 5-7 full time bioinformatacists and one PhD
  level team lead for every 300-400 isolates
  sequenced per month

31
Analyzing Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Limited access to open source and other state-
  of-the-art analytic software (primarily applies
  to government and military organizations) but may
  apply to healthcare systems ( Ransomware OPM
  Target breaches)

• Relax .mil restrictions on computer networks for
  facilities involved in biomedical research and
  clinical support allow use of .org or .net
  expedite process and shorten approval time for
  obtaining Certificates of .net Worthiness

32
Sharing and Storing Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Continuous sequencing of large volumes isolates
  (300-400 month) of creates extraordinary burdens
  for sharing and storage (Petabytes over the
  program lifecycle)

• Increase bandwidth or provide infrastructure to
  accommodate emailing of FASTQ / FASTA data files
  of 10s to 100s of isolates at once use tiered
  storage explore vendor or cloud-based solutions
  (but these can be prohibitively expensive for
  larger projects)

33
Sharing and Storing Data

• Challenge/Consideration

• Possible Mitigation/Solution

• Commercial 'off-the-shelf' database for managing
  isolate inventory and linking clinical and
  antibiotic susceptibility data to sequenced
  genomes does not yet exist
• Shortage of IT esp.w/ security certification

• Adopt the structure architecture of ARMoR-D
  which DOD can provide at no cost to nonprofit or
  other government agencies
• ??

34
(No Transcript)
35
Challenges Summary

• Generating - lengthy, burdensome and convoluted
  acquisition and awarding procedures long lags -
  obsolete highly skilled laboratory
  technologists in clinical labs access to robots
  long read platforms
• Storing internet restrictions / regulations
  shortages of qualified IT professionals (
  Business cases analysis required each time for
  more space requests)
• Analyzing Sharing access to high bandwidth
  LANs, sequencing pipelines, and commercial and
  open-source software

36
All credit belongs to the best big data team in
the business.
37
Acknowledgements

• Armed Forces Health Surveillance Center - Global
  Emerging Infections Surveillance and Response
  System
• P Waterman
• Navy and Marine Corps Public Health Center -
  EpiData Center
• U Chukwuma, M Kathryn
• US Army Pharmacovigilance Center
• T Coster, M LaCour, R Thelus, C Neumann
• Multidrug-resistant Organism Repository and
  Surveillance Network
• M Hinkle, Y Kwak, R Clifford, M Julius, P McGann,
  C Taylor, J Martinez, A Roth, A Ong, R Maybank, M
  Ly, G Flores, J Guzauskas, R Chavez, J Rosado, L
  Preston, N Litchfield, E Snesrud, L Appalla, F
  Onmus-Leone, J Stam, G Ward, LA Harden, J Padilla
  

38
Bibliography

• Lesho E, Clifford R, Onmus-Leone F, et al. The
  Challenges of Implementing Next Generation
  Sequencing Across a Large Healthcare System, and
  the Molecular Epidemiology and Antibiotic
  Susceptibilities of Carbapenemase-producing
  Bacteria in the Healthcare System of the U.S.
  Department of Defense. PLOS One. 2016. At press.
• Lesho EP, Clifford RJ, Chukwuma U, et al.
  Carbapenem-resistant Enterobacteriaceae and the
  correlation between carbapenem and
  fluoroquinolone usage and resistance in the US
  military health system. Diag Microbiol Infect
  Dis. 2014 doi.org/10.1016/j.diagmicrobio.2014.09.0
  17
• Lesho E, Lin X, Clifford R, Snesrud E, et al.
  From the battlefield to the bedside supporting
  warfighter and civilian health with the 'ART' of
  whole genome sequencing for antibiotic resistance
  and outbreak investigations. Mil Med. 2016. At
  press.
• Lesho E. How next generation sequencing might
  not transform infectious disease practice. Clin
  Infect Dis. 2016 doi10.1093/cid/ciw008.
• Lesho E, Craft D, Kirkup BC et al. Surveillance,
  characterization, and preservation of
  multidrug-resistant bacteria. Lancet Infect Dis
  2011118-10.
• Lesho E, Waterman P, Chukwuma U, et al. The
  Antimicrobial Resistance Monitoring and Research
  (ARMoR) Program the Department of Defenses
  Response to Escalating Antimicrobial Resistance.
  Clin Infect Dis. 201459390-7.
• Lesho E, Chuckwuma U, Sparks M, et al. Anatomic,
  Geographic, and Taxon-Specific Relative Risks of
  Carbapenem Resistance in the Health Care System
  of the U.S. Department of Defense. J Clin
  Microbiol. 2016. At press.
• McGann P, Bunin J, Snesrud E, et al. Real Time
  Application of Whole Genome Sequencing for
  Outbreak Investigation What is an achievable
  Turnaround Time? Diag Microbiol Infect Dis. 2016.
  At press.