Neil Ferguson

1
Antiviral use in a pandemicpredicting impact
and the risk of resistance
Neil Ferguson Dept. of Infectious Disease
EpidemiologyFaculty of MedicineImperial College
2
Introduction

• Modelling antiviral use in a pandemic
• Effect of treatment on transmission
• Post-exposure prophylaxis
• Use in containment at source
• Uncertainties
• Antiviral resistance
• Seeded resistance
• Evolution of resistance

3
Modelling approach

• State-of-the-art large scale simulation (up to
  300 million pop.)
• Individuals reside in households, but go to
  school or a workplace during the day.
• Transmission probabilities are specified
  separately for households and different place
  types.
• Local movement/travel random contact between
  strangers, at a rate which depends on distance.
• Air travel incorporated.

4
Influenza natural history

• New analysis of best available data on pandemic
  and inter-pandemic flu.
• Short incubation period 1-2 days.
• People most infectious very soon after symptoms.

Frequency
Days
5
Assumptions about antiviral effect

• Values initially used estimated by Longini et al
  from analysis of Roche data.
• Treatment (or PEP) assumed to reduce
  infectiousness by 60, from time treatment
  starts.
• Uninfected individual on prophylaxis has 30 drop
  in susceptibility (risk of infection per
  exposure event).
• Prophylaxis reduces chance of becoming a case
  by 65.
• Now using updated values, but results v. similar.

6
Clinical influenza

• Previous work assumed 50 of infections become
  clinical cases i.e. have ILI, independent of
  age.
• Have also looked at 67 (value used by Longini
  and others).
• More important quantity is proportion of
  infections seeking healthcare here Longini and
  Ferguson assumptions more similar (Ferguson
  assumed 90 cases sought healthcare, Germann
  assumed 60).
• Cases assumed to be 2-fold more infectious than
  non-ILI-generating infections (assumption based
  on data from Hayden et al. Cauchemez et al.).
• Aetiology of disease complex and variable, even
  for pandemics. No clear basis to predict
  age-specific clinical attack rates.

7
A US pandemic

• Large urban centres affected first, followed by
  spread to less densely populated areas. Epidemic
  only a little slower than GB.

R02.0/1.7
Up to 12 absenteeism at peak
8
Mitigation case treatment

• Main effect is to reduce severity of cases, but
  treatment within 24h of onset can also reduce
  transmission (reduction the proportion ill from
  34 to 28).
• 25 stockpile is then just enough, assuming 90
  of cases receive drug but demand may be
  higher.
• Effect relies on very early treatment within
  24h since infectiousness peaks soon after
  symptoms start.
• 48h delay gives no reduction in transmission and
  much poorer clinical benefit.
• So 25 stockpile is bare minimum could well
  lead to rationing.

No treatment 2 day delay 1 day delay 0 day delay
9
Household prophylaxis (PEP)

• Household prophylaxis treatment of everyone in
  house of case, not just case herself.
• 2006 Nature paper results Combined with school
  closure and rapid case treatment, PEP can reduce
  clinical case numbers by 1/3 for R02 but needs
  antiviral stockpile of 50 of population.
• UK now increasing stockpile to gt50, considering
  role for household PEP.

10
Varying timing and coverage in PEP

• Table shows cumulative clinical attack rate over
  pandemic.
• Results assume case treatment and prophylaxis of
  households of treated cases.
• No NPIs.
• Even with only 75 coverage and a 2 day delay,
  PEP can reduce attack rates by 25.
• But effect v limited for gt2 day delay.

11
Stockpile sizes required for PEP

• As previous slide, but showing antiviral courses
  used, as of population size.
• No allowance for wastage made here (e.g. due to
  treating non-flu ILI).
• Conservatively, need drug for 75 of population
  to cover all these scenarios and allow for some
  wastage.

12
Use of AV incontainment at source

• Need to add geographically targeted mass
  prophylaxis to treatment and close contact PEP to
  block transmission enough to achieve control.
• Still also need NPIs (and vaccine also helps).
• Need a maximum of 3m courses of drug if you
  need more then outbreak is too large to be
  contained.
• Need to detect outbreak at lt50 cases, react to
  new cases in 2 days.
• Too intensive to be used except in containment at
  source.

13
Uncertainties

• Nature of virus modelling assumes next
  pandemic virus will look like past pandemic
  viruses but H5N1 might be different, and the
  duration and dose of NAI required for treatment
  may differ.
• Transmission rates in different settings.
• (Real-world) effectiveness of drug.
• Adherence.
• Behavioural responses to epidemic and other
  controls.
• Antiviral resistance.
• ..

14
Antiviral resistance

• Resistance only a major issue during a pandemic
  if a resistant strain emerges with close to the
  transmission fitness of wild-type.
• Current spread of oseltamivir-resistant H1N1
  strains demonstrates this is a possibility.
• But we have no idea of the probability (per
  treated /or infected person) of such a strain
  emerging during a pandemic.
• So can only look at plausible illustrative
  scenarios.
• Two possibilities
• Resistance emerges elsewhere and a mixture of
  sensitive and resistant strains are seeded into
  your country.
• Resistance emerges for the first time in your
  country.

15
Selection of resistance theoretical worst case
Worst-case 100 of cases get instantaneous
treatment or treatment household PEP from day 1.
No NPI.

• Amplification of resistance depends on level and
  promptness of treatment prophylaxis.
• Reduction in attack rate from antivirals also
  quantifies selection pressure for resistance.
• If all cases were treated instantaneously, attack
  rate would be reduced to 24. Adding 100
  prophylaxis would give 16.
• But if 1 of infections entering country at start
  of epidemic are resistant, antiviral effect
  substantially reduced.
• Resistance substantially ampilfied, esp. by PEP.

16
Real-world selection pressurefor resistance

• Delays in real-world treatment mean weaker
  selection, so much less amplification of
  resistance.
• Large-scale use of household prophylaxis would
  amplify resistance more, but effect still v.
  limited.
• Final level of resistance can be less than seeded
  proportion, due to head start of sensitive
  epidemic.

Treatment of 60 of cases within 1 day of onset.
Treatment starts after 1000 cases in US.
17
de novo evolution of resistance

• Assume risk per infected person per day of
  generating a transmission fit resistant virus of
  10-4 and10-5 - pessimistic values.
• In reality evolution of transmissible resistant
  strain probably requires multiple changes, so
  this is worst case.
• Treatment of clinical cases never results in
  substantial resistance overall.
• For very pessimistic assumptions, household PEP
  can strongly select for resistance.

Instantaneous treatment of all cases from 1st
case in US, 10-4 mutation rate.
60 of cases treated in 24h from 1000th case,
10-5 mutation rate.
18
Conclusions

• Treatment needs to be delivered rapidly to have
  best direct and indirect effect.
• Household prophylaxistreatment on its own can
  reduce attack rates by 1/3, if delivered
  rapidly to gt75 of households of cases.
• Large scale prophylaxis (i.e. community rather
  than household), can achieve near-control, but
  delivery equally challenging.
• A combination of interventions gives more
  failsafe policy (e.g. NPIs slow spread of
  resistance).
• Antiviral resistance only likely to be a larger
  problem in the first wave if it emerges very
  early in the pandemic, with virus being fully
  fit.
• If transmissible resistant strains do emerge
  early, prophylaxis should be used with caution.

19
Collaborators
Christophe Fraser Simon CauchemezAronrag
Meeyai Don BurkeDerek Cummings Steven Riley
Sopon IamsirithawornRTI IncNCSA NIH MIDAS
programme
20
Private stockpiles(bought in advance by
households)

• US Govt. initiative to encourage households to
  stockpile antiviral medkits.
• Modelling predicts impact of 25 private
  stockpile on attack rates negligible.
• Some reduction in demand for public stockpile
  (25 public stocks might then be enough).
• Could be huge geographic (income-related)
  disparities in uptake.
• The same money far better invested in public
  stocks if distribution efficient.