Transcript Safety Data Mining Perspective, e

Safety Data Mining:
Background and Current Issues
Ramin Arani, PhD
Safety Data Mining
Global Biometric Science
Bristol-Myers Squibb Company
SAMSI: July, 2006
Outline

Rationale for Pharmacovigilance
 AERS Data Base
Data base issues

Methodologies
 BCNN (WHO)
MGPS (FDA)

Summary

Challenges and Opportunities
Pharmacovigilance - Rationale
Information obtained prior to first marketing is inadequate to cover all
aspects of drug safety:

tests in animals are insufficiently predictive of human safety,

in clinical trials patients are selected and limited in number,

conditions of use in trials differ from those in clinical practice,

duration of trials is limited
 information about rare but serious adverse reactions, chronic
toxicity, use in special groups or drug interactions is often not
available.
Pharmacovigilance - Rationale
Pre Approval Data
- Controlled
- Limited # Pts
- Safety data not mature
Post Approval Data
- Real life ; uncontrolled
- Off label use
-Generic
Population
Subjects for
approval
- Solicited Safety
Data
- Unsolicited Safety
Data
Spontaneous AE Reports
 Safety information from clinical trials is incomplete
° Few patients -- rare events likely to be missed
° Not necessarily ‘real world’
 Need info from post-marketing surveillance & spontaneous reports
 Pharmacovigilance by reg. agencies & mfrs carried out.
 Long history of research on issue
° Finney (MIMed1974, SM1982)
Royall (Bcs1971)
° Inman (BMedBull1970)
Napke (CanPhJ1970)
Issues
 Incomplete reports of events, not necessarily reactions
 How to compute effect magnitude
 Many events reported, many drugs reported
 Bias & noise in system
 Difficult to estimate incidence because no. of pats at risk, duration of
exposure seldom reliable
 Appropriate use of computerized methods, e.g., supplementing
standard pharmacovigilance to identify possible signals sooner -early warning signal
Pharmacovigilance - Definition
Phamacovigilance
Set of methods that aim at identifying and quantitatively
assess the risks related to the use of drugs in the entire
population, or in specific population subgroups
Adverse Drug Reaction
A response to a drug which is harmful and unintended, and which
occurs at doses normally used.
Safety Signal:
Reported information on a possible causal relationship between
an adverse event and a drug.
AERS Database

Database Origin 1969


SRS until 11/1/97; changed to AERS
3.0 million reports in database

All SRS data migrated into AERS

Contains Drug and "Therapeutic" Biologic Reports

exception = vaccines (VAERS)
Source of AERS Reports

Health Professionals, Consumers / Patients

Voluntary : Direct to FDA and/or to Manufacturer

Manufacturers: Regulations for Postmarketing Reporting
AERS Limitations

Different populations, Co-morbidities, Co-prescribing, Off-label
use, Rare events

Report volume for a drug is affected by, volume of use,
publicity, type and severity of the event and other factors,
therefore the reporting rate is not a true measure of the rate or
the risk

An observed event may be due to the indication for therapy
rather than the therapy itself; therefore observed associations
should be viewed as signal, and causal conclusions drawn
with caution
Examples
Claritin and arrhythmias (channeling and need for detailed
data not in data base)
Increased number of reports due to preexisting
condition. Selection of high risk patients for the drug
deemed safest for them.
Prozac and suicide (confounding by indication) Large
increase in reports following publicity and stimulated
reporting
The Pharmacovigilance Process
Traditional
Methods
Insight from
Outliers
Type A
(Mechanism-based)
Type B
(Idiosyncratic)
Data
Mining
Detect Signals
Generate Hypotheses
Refute/Verify
Public Health
Impact, Benefit/Risk
Estimate
Incidence
Act
Inform
Change Label
Restrict use/
withdraw
Methodologies
Finding “Interestingly Large” Cell Counts
in a Massive Frequency Table
No. Reports AE1
…
AEn
Total
Drug 1
N11
…
N1n
N1+
:
Nij
:
:
Nm
…
Nmn
Nm+
…
N+n
N++
:
Drug m
1
Total
N+1

Rows and Columns May Have Thousands of Categories

Most Cells Are Empty, even though N++ Is very Large

Only 386K out of 1331K Cells Have Nij > 0

174 Drug-Event Combinations Have Nij > 1000
Method - Basics

Endpoint: No of AEs

Most use variations of 2-way table statistics
No. Reports
Target
AE
Other
AE
Total
Target Drug
a
b
a+b
Other Drug
c
d
c+d
a+c
b+d
n
Total
Some possibilities
Reporting Ratio:
Proportional Reporting Ratio:
Odds Ratio:
OR > PRR > RR when a > E(a)
E(a) = (a+b)  (a+c)/n
E(a) = (a+b)  c / (c+d)
E(a) = b  c / d
Basic idea:
Flag when
R = a/E(a) is
“large”
Bayesian Approaches
 Two current approaches: DuMouchel & WHO
 Both use ratio nij / Eij where
nij = no. of reports mentioning both drug i & event j
Eij = expected no. of reports of drug i & event j
 Both report features of posterior dist’n of ‘information criterion’
ICij = log2 nij / Eij = PRRij
 Eij usually computed assuming drug i & event j are mentioned
independently
 Ratio > 1 (IC > 0)  combination mentioned more often than
expected if independent
WHO (Bate et al, EurJClPhrm1998)
 ‘Bayesian Confidence Neural Network’ (BCNN) Model:
 nij = no. reports mentioning both drug i & event j
 ni+ = no. reports mentioning drug i
 n+j = no. reports mentioning event j
Usual Bayesian inferential setup:
 Binomial likelihoods for nij, ni+ , n+j
 Beta priors for the rate parameters (rij, pi, qj)
WHO, cont’d
 Uses ‘delta method’ to approximate variance of
Qij = ln rij / piqj = ln 2  ICij
 However, can calculate exact mean and variance of Qij
 WHO measure of importance = E(ICij) - 2 SD(ICij)
 Test of signal detection predictive value by analysis of signals 19932000: Drug Safety 2000; 23:533-542
 84% Negative Pred Val, 44% Positive Pred Val
 Good filtering strategy for clinical assessment
WHO, cont’d
 WHO. (Orre et al 2000)
WHO, cont’d
Let A denote adverse events and D denote the drug.

 P di A
P  A D   P  A d 1 ,  d 2   P  A  
 P d 
i
i

 P d i1 , A 
 P  A  

1
i
 P d i P  A 
 log P  A  

i
 log P  A  


P d i , A 
k
1
di
 
P d
k
i
k

P d i , A 
log  

k
 k P d i P  A 

i
 
k
P d i , A 
k
di
k
log
P d i
k
P  A 

P  A 
k
di

k
di








if   0 ,1
Mutual information I(A,D) is a measure of association
I  A, D  

P  A , D log
P  A, D 
P  A P  D 
   

IC
DuMouchel (AmStat1999)
 Eij known, computed using stratification of database --
ni+(k) = no. reports of drug i in stratum k
n+j(k) = no. reports of event j in stratum k
N(k) = total reports in stratum k
Eij = k ni+(k)n+j(k) / N(k)
(E (nij) under independence)
 nij ~ Poisson(ij) -- interested in ij = ij/Eij
 Prior dist’n for  = mixture of gamma dist’ns:
f(; a1, b1, a2, b2, ) =  g(; a1, b1) + (1 – ) g(; a2, b2)
where
g(; a, b) = b (b)a – 1e-b/(a)
DuMouchel, cont’d
 Estimate , a1, b1, a2, b2 using Empirical Bayes -- marginal dist’n of
nij is mixture of negative binomials
 Posterior density of ij also is mixture of gammas
 ln2 ij = ICij
 Easy to get 5% lower bound (i.e. E(ICij) - 2 SD(ICij) )
The control group and the issue
of ‘compared to what?’

Signal strategies, compare

a drug with itself from prior time periods

with other drugs and events

with external data sources of relative drug usage and
exposure

Total frequency count for a drug is used as a relative surrogate for
external denominator of exposure; for ease of use, quick and
efficient;

Analogy to case-control design where cases are specific AE term,
controls are other terms, and outcomes are presence or absence of
exposure to a specific drug.
Other useful metrics and methods

Chi-square statistics

P-value type metric- overly influenced by sample size

Modeling association through directly Multivariate Poisson dist

Incorporation of a prior distribution on some drugs and/or
events for which previous information is available - e.g. Liver
events or pre-market signals
Interpreting the Signal Through
the Role of Visual Graphics

Four examples of spatial maps that reduce the scores to
patterns and user friendly graphs and help to interpret
many signals collectively
Example 1
A spatial map showing the “signal scores” for the most
frequently reported events (rows) and drugs (columns) in
the database by the intensity of the empirical Bayes signal
score (blue color is a stronger signal than purple)
Example 2
Spatial map showing ‘fingerprints’ of signal scores allowing one
to visually compare the complexity of patterns for different drugs
and events and to identify positive or negative co-occurrences
Example 3
Cumulative scores and numbers of reports according to the
year when the signal was first detected for selected drugs
Example 4
Differences in paired male-female signal scores for a specific
adverse event across drugs with events reported (red means
females greater, green means males greater)
Summary
1.
There is NO Golden Standard method for signal detection.
2.
The signals become more stable over time, however there is a
limited time window of opportunity for signal detection.
3.
Use Time-slice evolution of signal.
-Fluctuation might reveal external risk factors.
-Robustness can be assessed.
4.
Consider other endpoint such as time to onset, duration of
event, etc.
5.
For spontaneous case reports, the means to improve content is
to standardize and improve intake
6.
Data mining likely will generate many false positives and
affirmations of what was previously known
7.
Causality assessments should largely be reserved refining
important signals
Challenges in the future

More real time data analysis

More interactivity ( Visual Data mining, e.g. ggobi )

Linkage with other data bases to control the bias
inherent in data base

Quality control strategies (e.g. Identifying duplicates

Methods to reduce the false positive and negative?