Antimicrobial risk assessment: The Macrolide example of a Deterministic Risk Assessment

1
Antimicrobial risk assessment The Macrolide
example of a Deterministic Risk Assessment
APPENDED Oct. 12. 2004

• H. Scott Hurd D.V.M, Ph.D.
• Iowa State University
• College of Veterinary Medicine
• Ames, Iowa
• Hurd-Health Consulting

J. Food Protection, May 2004 www.ifss.iastate.edu
shurd_at_iastate.edu
2
Additional comments

• A few additional slides are added to the end of
  this presentation in reply to Mr. Voses
  presentation at the same meeting.
• Mr. Vose stated that his presentation was under
  contract with FDA, although they did not have
  opportunity to clear his comments
• Therefore, we do not hold FDA liable for any
  defamatory comments made. We are concerned that
  Mr. Vose might be perceived as speaking on behalf
  of FDA, however. Finally, we have concern about
  the effect his remarks may have on open
  scientific exchange and advancement in this
  field.
• Science proceeds by conjecture and refutation
  (Karl Popper)

3
Objectives for this presentation

• Explain our methods
• Encourage you to use this same or a similar
  approach
• Highlight our lessons learned
• Generate discussion

4
Overview of lessons learned

• There is a system consisting of several
  measurable processes
• Must be multidisciplinary!
• Qualitative vs. quantitative
• Level of model and data resolution DEPENDS
• Risk Pr (hazard) consequence
• Estimation of risk is NOT the whole story

5
There is a system consisting of several
measurable processes
From Dickson, J.S., Hurd, H.S, and Rostagno,
M.H., Salmonella in the Pork Production Chain.
National Pork Board Review. 2003
6
Must be multidisciplinary!

• H. Scott Hurd, D.V.M., Ph.D., Hurd-Health
  Consulting, Roland, IA, USA
• Stephanie Doores, Ph.D., Pennsylvania State
  University
• Dermot Hayes, Ph.D., Iowa State University, USA
• Alan Mathew, Ph.D., University of Tennessee, USA
• John Maurer, Ph.D., University of Georgia, USA
• Peter Silley, Bs.C., Ph.D., MB Consult Limited,
  Bingley, West Yorkshire, UK
• Randall S. Singer, D.V.M., Ph.D., University of
  Minnesota, USA
• Ronald N. Jones, M.D., The JONES Group/JMI
  Laboratories, North Liberty, IA, USA

7
Objectives of the assessment

• Assess the potential public health risks relating
  to the use of the macrolides
• tylosin (Tylan) and
• tilmicosin (Pulmotil and Micotil) in cattle,
  swine and poultry.
• Provide input for regulatory decision making
• Stimulate (redirect) the scientific and public
  debate

8
Overall methods

• Define the real outcome of interest
• Identify the steps (events) leading to that
  outcome
• Express them in quantifiable terms
• Understandable by microbiologists
• Find the data
• Combine and understand

9
Qualitative Risk Definition (from Food Drug
Admin 152)

• Probability human illness is caused by a specific
  resistant bacteria, is attributable to specific
  animal-derived food commodity, and is treated
  with human antibiotic of interest.
• All events categorized as H, M, L
• Consequence considered after probability is
  estimated

10
Qualitative vs. quantitative

• It is relative
• Transparency
• Hard to explain
• Hard to defend
• Units of measure or breakpoints
• How do you combine multiple qualitative
  variables?
• L L L ?
• L L L ??

11
Hazard Definition

• Hazard defined as human illness that is
• Caused by macrolide-resistant Campylobacter spp.
  or Enterococcus faecium.
• Attributable to consumption of contaminated
  poultry, pork or beef.
• Treated with a human antibiotic of the macrolide
  class.
• Human salmonellosis not addressed because
  organism is neither routinely susceptible to nor
  treated by macrolides (Risk Ranking)

12
Hazard Definition (cont.)

• Included all U.S. label claim uses for two
  macrolide antibiotics, tylosin (Tylan) and
  tilmicosin (Micotil) in poultry, swine, and beef
  cattle.
• Administered via medicated feed or drinking water
  or by injection.
• Used for treatment, prevention and control of
  disease, and for increased feed efficiency and
  weight gain.
• Included two foodborne bacteria, Campylobacter
  spp. and E. faecium.

13
Risk Definition for this analysis

• Risk Pr (hazard) consequence
• The yearly probability that an average individual
  in the U.S. population would be affected by the
  defined hazard and would experience an adverse
  therapeutic event (treatment failure)
• Poorer efficacy than usual as manifested by
  longer duration of diarrhea, progression to more
  severe disease, or in worst case scenario
  mortality.
• OR The Pr that a random person in the U.S. will
  suffer an adverse therapeutic event in one year.

14
Methodology

• Event tree approach
• Some data are limited and thus are approximated
• When numbers uncertain, the most conservative
  (highest risk) estimates were used.
• Meets definition of semi-quantitative analysis
  or deterministically quantitative
• Describes all the necessary events that must
  occur to create the modeled risk.
• Provides greater transparency than qualitative
  analysis regarding calculations and assumptions
  at each point in the chain of events.
• Consistent with FDA guidance 152!!

15
Pathway of Events Leading to the Hazard
Release Assessment Describes the probability
that factors related to the antimicrobial use in
animals will result in the emergence of resistant
bacteria or resistance determinates
(RzD). Exposure Assessment Describes the
likelihood of human exposure to the RzD through
particular exposure pathways. Consequence
Assessment Describes the relationship between
specified exposures to the RzD (the hazardous
agent) and the consequences of those exposures
(CVM-defined hazard)
Risk
16
Node 1 Tylosin or Tilmicosin administered to
food animals

• Included all possible uses
• Medicated feed and injection counted as separate
  two exposure events
• Single dose was considered an exposure
• Estimated number of animals exposed/ treated with
  T-T in 2001 (Doanes, Rennier)
• Cattle 16.1 million
• Swine 49 million
• Broilers - 632 million

• Quan model V18c2v2 for PPT.xls

17
Node 2 Resistance selected above background

• This probability is a function of three factors
• Presence of ENT and/or CAMPY in treated animals.
• Intrinsic or background susceptibility of these
  bacteria.
• Mutation or RzD acquisition with survival of
  newly resistant strains.
• Used following data to estimate the parameters
• 1. Prevalence of CAMPY or ENT in livestock
• 2. Current resistance levels in livestock
• 3. Prevalence of resistance in human isolates

• Quan model V18c2v2 for PPT.xls

18
Node 3 RzD escapes from the farm

• Conservatively assumed as 100
• Carried in market animals
• No decreasing effect of early withdrawal
• Conservative (risk increasing) assumption
• Needs to be refined for length of treatment
  effects

• Quan model V18c2v2 for PPT.xls

19
Pathway of Events Leading to the Hazard
Release Assessment Describes the probability
that factors related to the antimicrobial use in
animals will result in the emergence of resistant
bacteria or resistance determinates
(RzD). Exposure Assessment Describes the
likelihood of human exposure to the RzD through
particular exposure pathways. Consequence
Assessment Describes the relationship between
specified exposures to the RzD (the hazardous
agent) and the consequences of those exposures
(CVM-defined hazard)
Risk
20
Node 4 Bacteria with RzD remain on carcass after
harvest

• Campylobacter spp. (FSIS data do not distinguish
  between C. jejuni and C. coli)
• 4 of beef carcasses
• 32 of swine carcasses
• 88 of poultry carcasses
• E. coli used as indicator of contamination from
  intestinal contents and/or fecal matter
• 8 of beef carcasses
• 31 of swine carcasses
• 100 of poultry carcasses
• No adjustment for imported meat
• All ground meat assumed contaminated

• Quan model V18c2v2 for PPT.xls

21
Node 5 Bacteria carrying resistant determinant
survives to retail meat case

• There is considerable evidence that CAMPY does
  not survive well under refrigeration, therefore,
  the risk should be reduced accordingly.
• But to what level?
• Due to the inconsistencies in retail studies and
  retail prevalence being only partly a function of
  wholesale carcass prevalence, retail data were
  not entered directly into the model.

• Quan model V18c2v2 for PPT.xls

22
Node 6 Mishandling and presentation of infective
dose for consumption

• To cause human illness or passage of RzD, it must
  be mishandled such that a sufficient dose of
  bacteria will survive for human consumption.
• Food borne risk FDA guidance
• National rates on the frequency of mishandling
  these commodities were not available, therefore,
  the effects of these situations were combined
  with the results from other nodes.
• Microbial growth models were not applied in this
  assessment.

• Quan model V18c2v2 for PPT.xls

23
Pathway of Events Leading to the Hazard
Release Assessment Describes the probability
that factors related to the antimicrobial use in
animals will result in the emergence of resistant
bacteria or resistance determinates
(RzD). Exposure Assessment Describes the
likelihood of human exposure to the RzD through
particular exposure pathways. Consequence
Assessment Describes the relationship between
specified exposures to the RzD (the hazardous
agent) and the consequences of those exposures
(CVM-defined hazard)
Risk
24
Node 7 Consumer becomes ill due to consumption
of organism containing RzD

• The output from this node is the number of
  illnesses associated with RzD attributed to
  meat-borne CAMPY and ENT.
• Limited data are available on the dose-response
  of humans to oral consumption of CAMPY and ENT.
• Human illness from meat-borne ENT is extremely
  rare and it is not considered a food-borne
  disease.
• Could not calculate the number of illnesses due
  to that exposure dose, therefore, calculations
  for this node were combined with other nodes.

• Quan model V18c2v2 for PPT.xls

25
Level of model and data resolution DEPENDS

• Available data
• Go from known to known
• Use of the model
• Could this process be changed as a risk
  management tool

26
Nodes 5-7 Ratio Method

• FSIS carcass swab data on contamination
• CDC data on human illness from CAMPY
• Combined effect of nodes 5,6,7
• Equals the Probability of human
  campylobacteriosis from a meal that originated
  from contaminated carcass regardless of RzD.

• Quan model V18c2v2 for PPT.xls

27
Nodes 5-7 Ratio Method

• Total human cases /
• contaminated servings
• Tot. Human cases
• CAMPY 13.3/100,000 U.S. pop
• Adjusted for under reporting (38x)
• 90 of human cases were attributed meat
• RESULT 8 x 10-6
• Same constant used for Enterococci

• Quan model V18c2v2 for PPT.xls

28
Node 8 Ill patient is treated with macrolide
class antibiotic

• For hazard (treatment failure) to occur, illness
  must occur AND be treated with a macrolide
• Results for CAMPY (3) from the probabilities of
• Patient seeking medical care (23.5)
• Submission of a culture (17.7)
• Positive test for CAMPY (94.5)
• Accurate diagnosis (75)
• Use of a macrolide (100)
• Probability of changing therapy from empiric
  regimen to a macrolide after CAMPY diagnosis
  conservatively assumed as 100
• Result ENT 1 in 1 million chance of food borne
  illness

29
Node 8 Ill patient is treated with macrolide
class antibiotic

• Commonest drugs selected are fluoroquinlones
  (FQs) and trimethoprim/sulfamethoxazole, not
  macrolides
• FQ (ciprofloxacin) or erythromycin are
  recommended "first-line" therapies for CAMPY
  gastroenteritis
• Initial therapy (FQ) would only infrequently be
  changed in the absence of clinical failure
• Routine practice would likely not result in
  macrolide treatment for diarrhea
• Therapy for CAMPY, if known from diagnosis or
  culture, is erythromycin, but not usually
  prescribed unless a FQ-treated case worsens.

• Quan model V18c2v2 for PPT.xls

30
Risk Pr (hazard) consequence
31
Node 9 Infection with a resistant organism
results in clinical treatment failure

• Risk probability of the defined hazard times
  the consequence, defined as treatment failure
• Treatment failure can have numerous definitions
• Death attributable to the episode (fatalities are
  very rare.)
• Persistence of presenting symptoms and laboratory
  test abnormalities.
• Lack of bacteriological evidence of pathogen
  eradication at designated evaluation intervals.
• Probability of CAMPY treatment failure
  conservatively estimated at 50
• Probability of ENT treatment failure when treated
  with erythromycin set at 100

32
Results

• Quan model V18c2v2 for PPT.xls

33
Conservative Assumptions

• Conservative estimates increased the risk
  estimate.
• All ground beef and pork assumed to be
  contaminated with RzD.
• For poultry, assumed that the entire 4-5 lb.
  carcass contaminated with RzD.
• Assumed that all uses of tylosin or tilmicosin
  produced the same risk.
• Assumed that a single treatment was equivalent to
  long-term feeding for growth promotion,
• Assumed 100 probability RzD escape from the
  farm.
• Assumed 50 failure rate for treatment of a
  macrolide-resistant CAMPY infection with a
  macrolide.
• Assumed 90 of CAMPY cases attributable to meat
  consumption.

34
Key data needs

• Selection growth of RzD in livestock
  populations exposed to antibiotic (2)
• Effect of withdrawal on RzD prevalence (3)
• National estimates of the contamination at retail
  (5)
• Estimate of the frequency and type of consumer
  mishandling (6)
• Potential of RzD transfer from animal ENT to
  human gut flora

35
How much danger are we really in??
INTERPRETATION
36
What is acceptable risk?

• 1 in 1 million?
• 1 in 7000?
• Function of
• Personal control
• Choice
• Benefits

37
Risk is NOT the whole story

• Secondary impacts of management change
• Unintended consequences
• Externalities
• Relative increased risk
• Benefits
• Economic
• Public health

38
Alternative risks to decreased antimicrobial uses

• Increasing human salmonellosis Eurosurveillance
  weekly (June 2002)
• Decreased airsacculitis ? decreased Salmonella
  and Campy on carcass (Russell, 2003)
• Decreased condemnations in turkeys? decreased
  Salmonella in ground product
• Stopped avoparcin ? Increase necrotic enteritis
• Increased human illness ? increased Tx ?
  antibiotic resistance??

39
Future work

• Fully stochastic (uncertainty) model
• Benefits assessment
• Fill in critical research gaps,
• e.g. Pr (RzD develops)
• Revaluate E. faecium and RzD transfer
• Model at risk subpopulations

40
Following are slides taken from Mr. Voses
presentation with Dr. Hurds replies shown in
yellow italics font
41
It was (is) Dr. Hurds professional judgment as a
health care professional that a risk of 1 in 10
million could be deemed as safe (a risk
management opinion). Note, that nowhere in the
scientific (risk assessment) paper was that
terminology used.
42
 
Public Health Consequences of Macrolide Use in
Food Animals A Deterministic Risk Assessment.
Scott Hurd et al, (2004). J. Food Protection
67(5) p980-992

• Supported by Elanco Animal Health (acknowledged)
• Funded two meetings, not writing of paper
• Had antimicrobial use data thats great!
• There is often the suggestion that science funded
  by industry is biased and self-serving with
  possible insinuations against the integrity of
  the scientists. However, it should be noted that
  Eli Lilly Inc, parent of Elanco has much at stake
  by the loss human health effects and could be
  expected to have the highest interest in
  promoting human health over animal profit.
• Based on CVM-Vose FQ model mathematics
• Due credit was hopefully given. It was
  encouraging that basic methodologies could be
  similar

43
continued

• Vose comments on poster sent to Hurd
• Math is inappropriate for E.faecium as it is a
  reservoir not flow
• It became obvious to the authors, as we
  progressed that E.f. was an inappropriate
  candidate for inclusion in the foodborne route
  used in this model. However, we were attempting
  to follow FDA 152.
• However, that critique does not negate the
  information derived from the Campylobacter
  analysis
• Reduction of E.faecium risk by 10-6 not explained
  at all
• It was stated in the paper that Human illness
  from E.f. is extremely rare, and it is not
  considered a foodborne disease 1 in a million
  was a quantification of extremely rare (maybe
  it should have been lower)
• Sensitivity analysis not performed on important
  factors
• That is a matter of authors opinion based on the
  great amount of uncertainty in the parameters
  used. It represents an attempt at transparency.
• Published without any amendments
• The comments were on the poster only (a very
  short version of the whole). It was the authors
  hope that the issues would have been addressed in
  the publication and to press on with differing
  opinions for the advancement of science.

44
 

• Now saying The analysis presented in our paper
  was not meant to be the final answer about the
  risk of food animal macrolide use
• But results have already been widely publicised
  as if it was
• The authors realize that FDA is not swayed by
  media publicity, but we accept that the objective
  of this project as noted in this presentation was
  to Generate discussion. Failure to do so would
  not lead to scientific advancement.
• Purpose of paper was to criticise Guidance 152
  via an example
• Used FDA-FQ model principles
• Used Guidance 152
• It is absolutely incorrect assume the purpose of
  this paper was to criticise FDA, the guidance ,or
  Mr Vose!!!
• At the beginning of this project we had every
  intention to follow 152, and in many respects we
  did. In fact one reviewer of the manuscript
  commented that it was not contrary to 152 and
  should not be construed as an advancement or
  improvement to 152
• It is inevitable that improvements to 152 could
  be suggested from this exercise. It would be
  most unfortunate if FDA or its contractors take
  an attitude that hinders scientific advancement.

45
continued

• Calculated a minimum human risk from macrolide
  resistance
• Based on the conservative estimates used for many
  parameters, we conclude the real risk is likely
  much lower than our estimate
• Conclude that quantitative analysis better than
  qualitative
• The conclusions about the value of quantitative
  analysis are methodological issues only
  indirectly related to FDA decisions about
  macrolide use. The greatest value of quantitative
  analysis, which is relative to FDA decisions is
  TRANSPARENCY.
• i.e. 152 would disallow macrolide use, but
  numbers show very small risk
• That may be true, from that it appears that 152
  is a risk MANAGEMENT document and not a guide for
  risk ASSESSMENT.
• Definitely has a good point, but better just to
  make 152 criticism
• Criticism of 152 were meant to be a learning
  experience.
• Dr Tollefson also noted that CVM would look at
  low risk calculation
• Indeed CVM has a challenge as to how information
  of all types (Quant/Qual) should be utilized in
  decision making