Title: Detecting Multi-Item Associations and Temporal Trends Using the WebVDME/MGPS Application
1Detecting Multi-Item Associations and Temporal
Trends Using the WebVDME/MGPS Application
- DIMACS Tutorial on Statistical and Other Analytic
Health Surveillance Methods - 18 June 2003
- Richard Ferris
2Pharmaceutical post-marketing surveillance
- Companies and regulatory agencies collect
databases of spontaneous adverse reaction reports - Relevant exposure data not readily available (the
denominator problem) - Can drug-event combinations of potential interest
be identified from internal evidence alone? - Approach
- Use an internally defined denominator
- Construct set of expected counts using a
stratified independence model
3Computation of Expected Counts
- The expected count for a given drug-event
combination is determined by the overall count
for the particular drug (across all events) and
the overall count of the particular event (across
all drugs) - For example, if 2 of all reports have PROZAC as
a drug, and 3 of all reports have RASH as an
event, then one would expect that 0.06
(0.020.03) of the reports will include this
combination (PROZAC in combination with RASH) - (MGPS carries out this computation separately for
each distinct stratum and sums the
strata-specific expected counts to obtain an
overall expected count)
4Comparing Observed and Expected CountsRelative
Reporting Rate
- Relative Report Rate (RR) RRij Nij / Eij
- Easy to interpret, easy to compute
- Statistically unstable if N is small or E is very
small - The following all have RR 100
- N 1000, E 10
- N 100, E 1
- N 10, E 0.1
- N 1, E 0.01
5Comparing Observed and Expected
CountsStatistical Significance
- What is the probability that Nij would be
observed by chance (sampling error) when
expected value is Eij ? (p-value for testing a
null hypothesis) - Harder to interpret (not expressed in same units
as RR) - Results in computation of absurdly small
probabilities that have no meaning - N100, E1 produces 10-158 !
- Small RR can be very significant (small p-value)
when sample size is very large - N 2000, E 1000, RR 2 is more
significant than - N 10, E 0.1, RR 100
6Comparing Observed and Expected CountsEmpirical
Bayes Multi-Item Gamma Poisson Shrinker
- Try for best of both previous approaches
- interpretability of relative rate
- adjust properly for sampling variation
- Focus on the distribution across the set of
drug-event combinations of the ratios - Estimate lij mij /Eij , where Nij
Poisson(mij ) - Fit a parameterized prior distribution function
(mixture of two gamma functions) to the empirical
distribution of the ls - Find posterior distribution of l after observing
N some value n - Use this to obtain posterior estimate of
expectation value of l given observation of Nij - This posterior estimate is what we call EBGM
(Empirical Bayes Geometric Mean) also get lower
and upper 95 confidence bounds (EB05, EB95). - EBGM is termed the shrinkage estimate for RR
7Multi-Item Associationsvs. Pairwise Associations
- Consider the case of an item triplet e.g. 2
drugs and an event - RRijk Nijk/Eijk where Eijk is based on
independence model - EBGMijk shrinkage estimate of RRijk
- Suppose a particular itemset (drug A, drug B,
event C kidney failure) is unusually frequent
(EBGM for the triplet is gtgt 2) - Important to ask
- Is this merely the result of one or more of the
pairs (AB, AC, BC) being unusually frequent? OR - Is this a drug-drug interaction
- Compare Empirical Bayes estimate of the frequency
count of the triplet to the prediction from the
all-2-factor log-linear model - EXCESS2 (EBGM E ) EAll2F
- E is the expected count from independence
- Computation of EAll2F uses shrinkage estimates of
pairwise counts - EXCESS2 is an estimate of how many extra cases
were observed over what was expected using the
all-2-factor model - Alternate approach Define Eijk from predictions
of all-2-factor model in which case resulting
EBGM directly measures divergence of observed
count from all-2-factor prediction
8Health Authority Adoption of Signal Detection
Technologies
- FDA
- CDER
- Experimented in Office of Biostatistics with GPS
for several years - Validated GPS
- Moving to production
- Have published data mining results on internal
web for almost all products - CBER
- initial GPS implementation (VAERS)
- CRADA between Lincoln and FDA to further develop
methodology and tools - CDC
- Collaborative GPS methodology development with
FDA - Includes simulation capability
- WHO Uppsala Monitoring Centre
- Production safety signal generation mechanism
using BCPNN
9FDA/GPS Validation Activities
- Positive controls
- Examine data mining results for drug-event
combinations corresponding to known labeled
adverse reactions - Negative controls
- Examine data mining results for several drugs
(with differing safety profiles) given for the
same indication - Roll back database in time to determine when
method would have provided first signal
10Databases of Spontaneous AE Reports
- FDA Spontaneous Report System (SRS)
- Post-Marketing Surveillance of all Drugs since
1969 - Dates from mid-60s thru 1997
- 1.5 Million Reports
- Encoded in COSTART
- FDA Adverse Event Reporting System (AERS)
- US cases, serious unlabeled events from all
manufacturers. - All products sold in the US 5000 Rxs
- Replaced SRS in 1997
- Reactions coded as MedDRA PTs
- Quarterly Updates, 4-6 month delay
- Drugs are Verbatim
- Includes initial and some follow-up reports
- Includes Demographics, Reactions, Drugs,
Outcomes, etc. - FDA/CDC Vaccine Adverse Events (VAERS)
- Stricter Laws for Vaccine Adverse Event Reporting
11Signal Detection DemonstrationUsing VAERS Data
12Significant EBGM and even extremely
conservative EB05 with small N
13Simple Rankings by Signal Strength
14Evolution of Signals Over Time
15Multi-Symptom Syndromes (Higher Order
Associations)
16The Serotonin Syndrome
- Could MGPS be used to identify unknown syndromes?
- Try mining the AERS data for significant event
triples using a known syndrome. - "The symptoms of the serotonin syndrome are
euphoria, drowsiness, sustained rapid eye
movement, overreaction of the reflexes, rapid
muscle contraction and relaxation in the ankle
causing abnormal movements of the foot,
clumsiness, restlessness, feeling drunk and
dizzy, muscle contraction and relaxation in the
jaw, sweating, intoxication, muscle twitching,
rigidity, high body temperature, mental status
changes were frequent (including confusion and
hypomania - a "happy drunk" state), shivering,
diarrhea, loss of consciousness and death. (The
Serotonin Syndrome, AM J PSYCHIATRY, June 1991)
17(No Transcript)
18Using Simulation to Testthe Signal Detection
Process
19Interpreting Simulation Parameters
Outcome
Yes
No
Yes
P-R
R
P
Exposure
1-P-QR
Q-R
No
1-P
Q
1-Q
1
- As R ? P and (Q-R) ? (1-P) gt No Signal
- As R ? P and (Q-R) ltlt (1-P) gt Strong Signal
- When R ltlt P and (Q-R)?(1-P) gt No Signal
- When R ltlt P and (Q-R) ltlt (1-P) gt Rare event
20Using Simulation to Create a Receiver Operating
Characteristic (ROC) Curve for EBGM
- An ROC curve displays the true-positive rate
(sensitivity) versus the false-positive rate(1
specificity) for a statistic - Ran a 20 iteration simulation using P 0.003Q
0.001 and R 0.00003 (RR 10) to check the
true-positive rate - Ran a 20 iteration simulation using P 0.003,Q
0.001 and R 0.0003 (RR 1) to check the
false-positive rate
21ROC Curve Based on Simulated Injection of Signals
22Simulating a Rare Event
- Sample 100,000 records from VAERS data
- Set P 0.003, Q 0.001, R 0.00003
- Iterate 20 Monte Carlo simulations
- Expect (on average)
- 0.003 x 100,000 300 Rare Exposures
- 0.001 x 100,000 100 Rare Outcomes
- 0.00003 x 100,000 3 Rare Exposure Rare
Outcome combinations - E (300 x 100) / 100,000 0.3
- RR 3/ 0.3 10
23Base Simulation on VAERS Data
24Sample Cases From VAERS
25Sample 100,000 Cases
26P 0.003 Q 0.001 R 0.00003
2720 Monte Carlo Iterations
28RareExposure Expected N 300
29RareOutcome Expected N 100
30RareExposure RareOutcome Expected N
3Expected RR 10
31Technical Details
- William DuMouchel. Bayesian Data Mining in Large
Frequency Tables (with Discussion). The American
Statistician (1999) pp 177-190. - William Dumouchel and Daryl Pregibon. Empirical
Bayes Screening for Multi-Item Associations.
Proceedings of KDD 2001.
32Methodology History and Key Contributors
- Stephan Evans
- MCA, UK
- Proportional reporting ratio (PRR) with Chi 2
analyses - Simple, highly intuitive, can be calculated by
hand - Bate, Lindquist, Edwards et. al.
- WHO Uppsala Monitoring Centre
- Bayesian neural network method for adverse drug
reaction signal generation - Ana Szarfman, FDA (CDER) and Bill DuMouchel (ATT)
- Empiric Bayes, more robust than PRR for small n
- MGPS method statistical parameter is EGBM
- William DuMouchel. Bayesian Data Mining in Large
Frequency Tables (with Discussion). The American
Statistician (1999) pp 177-190. - William Dumouchel and Daryl Pregibon. Empirical
Bayes Screening for Multi-Item Associations.
Proceedings of KDD 2001. - Multidimensional analyses possible
- Interactions, gender and other demographic
associates, syndrome identification - Can directly compare EBGM values of different
drugs, as well as for a specific drug
33Key Contributors (continued)
- WHO Collaborating Center for Internatl Drug
Monitoring M Lindquist, M Stahl, A. Bate, R.
Edwards, RH Meyboom. - Bayesian confidence propagation neural network
(BCPNN) . Information Component (IC) statistic is
the measure of the strength of DE relationship - Iterative approach
- L. Gould . Comparison and refinement of Bayesian
approaches for evaluating spontaneous reports of
ADRs. DIA Annual meeting, July 2001, (Denver) - EB vs BCPNN similar results
- Thakrar, BT, Blesch, KS, Sacks, ST, Wilcock, K
(2001) - (ISPE, Pharmacoepid. Drug Safety 10),
- PRR vs. EB similar sensitivity, EB better at
ranking events based on small N.