Monte Carlo Process FDA

Embed Size (px)

Citation preview

  • 8/13/2019 Monte Carlo Process FDA

    1/17

    Monte Carlo Process

    Charles Yoe, Ph.D.

    College of Notre Dame of Maryland

  • 8/13/2019 Monte Carlo Process FDA

    2/17

  • 8/13/2019 Monte Carlo Process FDA

    3/17

    Simulation

    Numerical technique used to estimate analytical solutionsto a problem

    Not an optimization technique, answers what-if questionsResults are not analytical solutions

    Analytical solutions are preferred

  • 8/13/2019 Monte Carlo Process FDA

    4/17

  • 8/13/2019 Monte Carlo Process FDA

    5/17

    Suppose . . .

    You have a variable that varies between 10 and 50

    All you know is theoretical maximum and minimum, any

    number between is equally likely

  • 8/13/2019 Monte Carlo Process FDA

    6/17

    Monte Carlo Process

    Is a process that can generate numbers within that range

    According to the rules you specify

    In this case a min and a max

    Any number as likely as any other number

  • 8/13/2019 Monte Carlo Process FDA

    7/17

    Monte Carlo Process

    Two steps

    Generate a simple random number

    Transform it into a useful value using a specific probabilitydistribution

  • 8/13/2019 Monte Carlo Process FDA

    8/17

    Random Number Generation

    Pseudorandom Numbers [0,1]

    Seed = 6721 (any number)

    Mid-square Method (John von Neumann)

    (6721) 2 = 45171841; r1= 0.1718

    (1718) 2 = 29515240; r2= 0.5152

    (5152) 2 = 26543204; r3= 0.5432etc.

    More sophisticated methods now used

  • 8/13/2019 Monte Carlo Process FDA

    9/17

    Transformation (1)

    Assume Uniform Distribution, U(a,b) where a = 10 and b = 50

    To obtain a value, x, we use x = a + (b - a)uIn this case, x = 10 + 40u

  • 8/13/2019 Monte Carlo Process FDA

    10/17

    Transformation (2)

    Generate U~U(0,1), say u = 0.1718 then

    x = 10 +(50 - 10)0.1718 = 16.9

    x = 10 +(50 - 10)0.5152 = 30.6

    x = 10 +(50 - 10)0.5432 = 31.7, etc.

    Other distributions are similar but more complextransformations

  • 8/13/2019 Monte Carlo Process FDA

    11/17

  • 8/13/2019 Monte Carlo Process FDA

    12/17

    Some Language

    Iteration--one recalculation of the model during asimulation. Uncertain variables are sampled once duringeach iteration according to their probability distributions

    Simulation--technique for calculating a model output valuemany times with different input values. Purpose is to getcomplete range of all possible scenarios

  • 8/13/2019 Monte Carlo Process FDA

    13/17

    Monte Carlo Simulation

    Simulation model that uses the Monte Carlo process

    Deterministic values in models replaced by distributions

    Values randomly generated for each probabilistic variablein model and calculations are completed

    Process repeated desired # times

  • 8/13/2019 Monte Carlo Process FDA

    14/17

    Monte Carlo Simulation

    0.00

    0.08

    0 10 20 30 40 0.0

    0.4

    5.0 8.8 12.5 16.3 20.0

    X =

    0.00

    0.02

    0 100 200 300 400

  • 8/13/2019 Monte Carlo Process FDA

    15/17

    How Many Iterations?

    Means often stabilize quickly--few hundred

    Estimating probabilities of outcomes takes more

    Defining tails of output distribution takes many moreiterations

    If extreme events are important it make take many manymore

  • 8/13/2019 Monte Carlo Process FDA

    16/17

    Some Examples

    The Monte Carlo process is used for several riskassessments linked to the Clearinghouse

    Salmonellahttp://www.fsis.usda.gov/ophs/risk/index.htm

    Antimicrobial Resistant Campylobacterhttp://www.fda.gov/cvm/fda/mappgs/ra/risk.html

    http://www.fsis.usda.gov/ophs/risk/index.htmhttp://www.fda.gov/cvm/fda/mappgs/ra/risk.htmlhttp://www.fda.gov/cvm/fda/mappgs/ra/risk.htmlhttp://www.fsis.usda.gov/ophs/risk/index.htm
  • 8/13/2019 Monte Carlo Process FDA

    17/17

    The End