risk assessment models for plasma products and blood components
TRANSCRIPT
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
09-01-2012
Risk assessment models for plasma products and blood components
Jan Over
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Outline of presentation
• Viral risk ‘models’ for blood components (single donor)
• Viral risk assessment models for plasma products (large pools): - deterministic model - probabilistic model
• Aspects of the probabilistic model for discussion
• Future of the probabilistic model
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Virus transmission risk of a blood product …
…… is the probability that an infectious virus particle (HIV, HCV, HBV, HAV or parvo B19) is present in a final container of blood product
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Infection risks of the Dutch blood supply
Single-donor (or small-pool) products….
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Infection numbers per 100,000 donors
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Infection numbers per 100,000 NL population / donors
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Residual HIV, HCV, HBV and HTLV risk of Dutch repeat blood donors (year 2008)
---------------------------------------------------------------------------------------- HIV 1/2 HCV HBV * HTLV I/II (HAV)
----------------------------------------------------------------------------------------
mean incidence 1) 2.7 0.0 32.8 2.7 (0.6)
probability 2) 0.1 0.0 5.2 0.4 (0.1) ---------------------------------------------------------------------------------------- * corrected by a factor of 3 for not detecting HBsAg peak 1) mean incidence number per million donor-years 2) probability of collecting an infectious donation per million donations
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Risk modelling for single-donor (or small-pool) products ?
In the absence of the pooling of a large number of donations and of many processing steps there is not much to model: a product produced from 5 donations has about a 5 times higher risk….
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
European regulation on viral risk of plasma products (1) Since 2004: • Guideline on Assessing the Risk for Virus Transmission
(CPMP/BWP/5180/03*) (new chapter 6 to NfG on Plasma-derived Medicinal Products (CPMP/BWP/269/95) (effective April 2005):
“The risk assessment should, wherever possible, include a quantitative estimation of the probability of a virus contaminant being present in a defined dose of final product.”
(Given the uncertainties in much of the required information) “.. realistic worst case scenarios should be considered…”
* Rev. 4, draft of February 2009
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
European regulation on viral risk of plasma products (2)
• Main elements of the risk assessment: - quantitative estimation of probability of the risk - additional parameters to take into account (comprehensive approach; not just log reduction factors by process steps)
• ‘New’ factors to take into account: - virus epidemiology in the donor population - virus titer in undetected donations - donation frequency - inventory hold - batch size - product yield - etc.
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
European regulation on viral risk of plasma products (3) Some issues:
• what is a realistic worst case scenario? pool contaminated?, smallest pool?, highest undetected virus load possible?, lowest product yield?, largest possible donation volume?, others?, a combination of all those?
• no specification of acceptable contamination risk of finished product < 1 per 106 final containers is contaminated? , < 1 per 104 or 103?
• number of virus particles per infectious dose is it 1 in all cases?
• probabilistic analysis instead of the conventional deterministic analysis? probabilities to be used instead of point estimates?
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Present-day practice of viral risk assessment
Deterministic analysis, using point estimates for:
- prevalence of undetected (rounded) average number, infectious donation: but at least one *
- viral load: highest undetected titer *
- vol. of infectious donation: largest volume possible *
- plasma pool size: smallest size *
- process reduction factor: average (of all) individual factors
- process yield: average
- product filling size: largest filling *
* worst case
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Calculation of deterministic risk C1-esterase inhibitor (Cetor®) and HAV
- HAV incidence: 1 : 20,000
- Plasma batch size: 6,400 L (20,000 don.)
- Undetected load of HAV (1 x 320 ml x 27.0 IU/ml x 576): 5.0 x 106 IU
- Number of vials per batch: 3,900
- HAV-RNA per vial prior to virus reduction: 103.1 IU (1.3 x103 IU)
- Total virus reducing capacity of the process: > 10.5 log10
- Residual HAV-RNA in 500 U vial < 10-7.4 IU
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Deterministic or probabilistic analysis ?
Disadvantages of deterministic analysis:
- worst case for a number of parameters - (too) many simplifying assumptions - some parameters / interactions not taken into account - no insight in spread in risk estimate - no insight in sensitivity and interaction of parameters
A probabilistic analysis provides a more comprehensive answer …..
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
IPFA Expert Group on Risk Assessment Methodologies
Aims:
• comparing existing risk assessment models between IPFA members (completed)
• evaluation of all models (including worst-case approach) by calculating the risk of two hypothetical products for two viruses (completed)
• identifying the best elements of all models (completed) and (preferably) combining those into more harmonised model(s)
• discussion of outcomes and questions with other parties
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Development of probabilistic viral risk analysis (1) * Assumptions:
- viral incidence rate is constant (HIV, HCV, HBV, HAV) - viral replication modeled by doubling time - probability of detection is dependent on contamination level - trailing window has no impact on viral risk - virus particles distribute evenly over final product - first-in, first-out principle followed - no GMP failures and testing errors
Use of real production data: - viral incidence, donation intervals, donation volumes (recovered/source plasma), pool sizes, inventory hold, product yield
* M.P. Janssen et al., Transfusion 48: 153-162, 2008
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Development of probabilistic viral risk analysis (2)
Analysis in two stages:
1) viral risk of production pool (calculation of probability that the pool is contaminated and calculation of the load of viral contamination)
2) contamination of finished product (starting from a contaminated pool)
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Model simulation
Manufacturing pool (size and composition)
Derive plasma collection time
Simulate contamination of the production pool (titre and probability)
Manufacturing information
Donation frequency, size, type
Inventory hold
Virus reduction Calculate probability of
contamination of final container
Product yield
Pool fraction used
Virus characteristics
Manufacturing time Test characteristics
Monte Carlo simulation
Number of donors
capacity of process
Monte Carlo simulation
Monte Carlo simulation
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Risk of a hypothetical product (1): calculation of contamination risk
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Risk of a hypothetical product (2): sensitivity analysis
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Probabilistic (median) risks are lower than deterministic risks
• HIV 2.1 log lower risk
• HCV 2.0 log lower risk
• HBV 0.5 log lower risk
• HAV 1.6 – 2.0 log lower risk
• parvo 1.6 log lower risk
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Virus titre modeling: HIV
* Viral doubling time: 0.85 days * NAT pool size: 48 donations * Test characteristics: X95 = 25 gEq/ml
X99.9 = 75 gEq/ml
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Probability of parvovirus B19 contamination of final container of Cohn fractionated albumin
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Pros and cons of the probabilistic model (1)
Advantages:
- more comprehensive, fewer assumptions, scientifically more sound - avoids unrealistic outcomes caused by addition of worst-case assumptions - provides insight in level of uncertainty - provides a sensible estimate for the worst-case risk - reveals which parameters and interactions are the most influential - predicts the effect of potential additional measures (cost-effectiveness)
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Pros and cons of the probabilistic model (2)
Disadvantages:
- some assumptions are still necessary
- outcome is a probability density function of the product vial risk
- complex model: not easy to calculate, to understand and to communicate !
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
The full-blown probabilistic calculation procedure
Model parameters:
- Incidence rate - Donor type
- Donation frequency distribution
- Virus doubling time - Test sensitivity - Test pool size
- Inventory hold period - Production pool size
and composition - Product yield
- Process viral reduction capacity
Monte-Carlo simulation
Model outcome:
Sampling of variability in model parameters
to obtain model outcome
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Approximation of the probabilistic calculation procedure
Model parameters:
Incidence rate (A) Donor type (B)
Donation frequency distribution (C)
Virus doubling time (D) Test sensitivity (E) Test pool size (F)
Inventory hold period (G) Production pool size and composition (H)
Product yield (I) Process viral reduction
capacity (J)
Meta Model
Model outcome and
approximation:
ƒ(A,F,G,I,J)
Mathematical equation providing model outcome
as function of input parameters
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Remaining questions
• Should we calculate (and present) the absolute risk or the conditional risk?
• Criteria for product safety: what is “an adequate safety margin”? What level of inactivation/removal capacity is “clearly exceeding” the potential virus input?
• How many genome equivalents/virus particles constitute an infectious dose?
• Should the clinical dose be taken into account? • Which parameters of the risk analysis are identical for every
manufacturer? (virus characteristics, clinical dose, calculation of risk….) • Do we need a probabilistic model in all cases?
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Conclusions
• The probabilistic approach of assessing viral risk of plasma-derived products provides major advantages
• Process log reduction factor has the biggest impact on outcome
• Not all criteria are defined yet (e.g. acceptable safety margin)
• Harmonisation is highly desirable
• A (generic) simplification of the probabilistic model is feasible
Plasma Products division IPFA/PEI 17th Workshop, 26-27 May 2010
Acknowledgements
- Julius Center for Health Science and Primary Care, University Medical Centre Utrecht, NL:
Mart Janssen
Cees van der Poel
- Sanquin:
Hans Zaaijer
Theo Cuijpers
- IPFA members