Download Analyzing Adverse Drug Reactions with SRS Databases and Stats and more Study notes Genetics in PDF only on Docsity!

1

Data Mining in

Data Mining in

Pharmacovigilence

Pharmacovigilence

Aimin Feng, David Madigan, and Ivan Zorych

dmadigan@rutgers.edu

http://stat.rutgers.edu/~madigan

2

Overview

Overview

 

Intro. to Post-marketing Surveillance

Intro. to Post-marketing Surveillance





SRS Databases

SRS Databases





Existing Analysis Methods

Existing Analysis Methods





Our Approaches

Our Approaches



 Bayesian Logistic Regression

Bayesian Logistic Regression



 Propensity Score

Propensity Score





Conclusions

Conclusions

4

Why Post-marketing Surveillance

Why Post-marketing Surveillance





Limitations on pre-licensure trials

Limitations on pre-licensure trials

Size

Size

Duration

Duration

Patient population: age,

Patient population: age,

comorbidity

comorbidity

, severity

, severity





Fact

Fact

Several hundred drugs have been removed

Several hundred drugs have been removed

from market in the last 30 years due to safety

from market in the last 30 years due to safety

problems which became known after approval

problems which became known after approval

5

Databases of Spontaneous

Databases of Spontaneous ADRs

ADRs



 FDA Adverse Event Reporting System (AERS)

FDA Adverse Event Reporting System (AERS)

Online 1997

Online 1997

replace the SRS

replace the SRS

Over 250,

Over 250, ADRs

ADRs reports annually

reports annually

15,000 drugs - 16,

15,000 drugs - 16, ADRs

ADRs



 CDC/FDA Vaccine Adverse Events (VAERS)

CDC/FDA Vaccine Adverse Events (VAERS)

  Initiated in 1990Initiated in 1990

12,000 reports per year

12,000 reports per year

50 vaccines and 700 adverse events

50 vaccines and 700 adverse events

  Other SRS

Other SRS

WHO - international

WHO - international pharmacovigilance

pharmacovigilance program

program

7

Existing Methods

Existing Methods



 Multi-item Gamma Poisson

Multi-item Gamma Poisson Shrinker

Shrinker (MGPS)

(MGPS)

US Food and Drug Administration (FDA)

US Food and Drug Administration (FDA)



 Bayesian Confidence Propagation Neural Network

Bayesian Confidence Propagation Neural Network

  WHO Uppsala Monitoring Centre (UMC)WHO Uppsala Monitoring Centre (UMC)

  Proportional Reporting Ratio (PRR and

Proportional Reporting Ratio (PRR and aPRR

aPRR )

)

UK Medicines Control Agency (MCA)

UK Medicines Control Agency (MCA)



 Reporting Odds Ratios and Incidence Rate Ratios

Reporting Odds Ratios and Incidence Rate Ratios

Other national spontaneous reporting centers and drug

Other national spontaneous reporting centers and drug

safety research units

safety research units

8

Existing Methods (Cont

Existing Methods (Cont

d)

d)

Focus on 2X2 contingency table projections

Focus on 2X2 contingency table projections

15,000 drugs * 16,

15,000 drugs * 16,

AEs

AEs

= 240 million tables

= 240 million tables

Most

Most

N

N

ijij

0, even though

0, even though

N

N

.. very large

.. very large

AE j =

Yes

AE j =

No

Total

Drug i = Yes a =20 b =100 120

Drug i = No c =100 d =980 1080

Total 120 1080 1200

10

10

These Measures not

These Measures not

Robust

Robust

Reverend

Reverend

Bayes

Bayes

to the rescue!

to the rescue!

11

Bayesian Statistics

Bayesian Statistics

The Bayesian approach has deep historical roots but required

the algorithmic developments of the late 1980’s before it was

of any use

The old sterile Bayesian-Frequentist debates are a thing of the

past

Most data analysts take a pragmatic point of view and use

whatever is most useful

13

13

Hospital Example (0/27)

Hospital Example (0/27)

!

f ( " | data ) =

f ( data | ") f ( ")

f ( data )

f ( data | ") f ( ")

posterior distribution

prior distribution

likelihood

!

27

0

"

$

%

&

'

(

0

( 1 ) ()

27

!

c "

a

( 1 # ")

b

!

" #

a + 0

( 1 $ #)

b + 27

16

16

What to report? Mode? Mean? Median?

Posterior probability that theta exceeds 0.2?

theta* such that Pr(theta > theta*) = 0.

theta* such that Pr(theta > theta*) = 0.

Posterior probability that theta is in (0.002,0.095) is 90%

17

More formal treatment

More formal treatment

Denote by θ

i

the probability that the next operation in Hospital i

results in a death

Assume θ

i

~ beta( a , b )

Compute joint posterior distribution for all the θ

i

simultaneously

!

19

Relative Reporting Ratio

Relative Reporting Ratio



 If the Drug and the AE were independent, what would

If the Drug and the AE were independent, what would

you expect

you expect a

a to be?

to be?

Overall (

Overall ( a

a

c

c )/(

a

a

b

b

c

c

d

d )=120/1200=10% have the AE

)=120/1200=10% have the AE

So, 10% of the

So, 10% of the “

Drug

Drug ”

reports should have the AE

reports should have the AE

  That is (That is ( aa ++ bb )(()(( aa ++ cc )/()/( aa ++ bb ++ cc ++ dd ))=12010%))=12010%=12==12= EE

ijij

Note

Note N

N

ij

ij

/E

/E

ij

ij

=a/

=a/ (

a

a

b

b )*((

a

a

c

c )/(

a

a

b

b

c

c

d

d ))=RR

))=RR

RR = 20/12 = 1.67 =

RR = 20/12 = 1.67 =

N

N

E

E

= Pr(AE|Drug)/Pr(AE)

= Pr(AE|Drug)/Pr(AE)

d=

d= c=

c=

Not

Not

Drug

Drug

ii

b=

b= a=

a= DrugDrug

ii

Not

Not AE

AE

j

j

AE

AE

j

j

N

N

ijij

20

Relative Reporting Ratio

Relative Reporting Ratio

RR

RR

ijij

=N

=N

ijij

/E

/E

ijij

Advantages

Advantages

  SimpleSimple



 Easy to interpret

Easy to interpret

Disadvantages

Disadvantages



 Extreme sampling variability when baseline and

Extreme sampling variability when baseline and

observed frequencies are small

observed frequencies are small

(

( N

N =1,

=1, E

E =0.01 v.s.

=0.01 v.s. N

N =100,

=100, E

E =1)

=1)



 GPS provides a shrinkage estimate of RR that

GPS provides a shrinkage estimate of RR that

addresses this concern.

addresses this concern.

E

ij

=N

ij

*N../N

i

.N.

j

N..

N.. N.

N.

j

j

Not

Not

Drug

Drug

i

i

N N

ii

NN ..

ijij

Drug

Drug

i

i

Not

Not AE

AE

j

j

AE

AE

j

j