Title: Active postmarketing surveillance for vaccine adverse events: The experience of the Vaccine Safety D
1Active post-marketing surveillance for vaccine
adverse events The experience of the Vaccine
Safety Datalink
Katherine Yih, PhD, MPH Harvard Medical School
and Harvard Pilgrim Health Care
2Why we need active post-marketing surveillance
- Rare adverse events may be impossible to detect
in pre-licensure studies - Drawbacks of passive surveillance systems (e.g.,
the Vaccine Adverse Event Reporting System) - Underreporting
- Reporting bias
- No denominators (so no calculation or comparison
of rates in vaccinated vs. unvaccinated possible) - Safety studies using traditional approaches can
take months to years
3Outline
- VSD background
- Active surveillance for vaccine safety in VSD
- Signals
- False signals
- True signal
- Monitoring influenza vaccine safety
- Conclusions
4Vaccine Safety Datalink (VSD)
- Collaboration between CDC and 8 managed care
organizations - Data annually captured from 8.8 million members
(2.9 of US population)
Group Health Cooperative
Northwest Kaiser Permanente
HealthPartners
Harvard Pilgrim
Marshfield Clinic
No. CA Kaiser Permanente
Kaiser Permanente Colorado
So. CA Kaiser Permanente
CDC
5VSD data
Health Outcomes (Hospital) (Emergency
Dept) (Outpatient)
Patient Characteristics (Birth Certificate) (Censu
s / Geocode)
Vaccination Records
Linked by Study IDs Data are linked and kept at
each site, not at CDC
6VSD features enabling active surveillance
- Population-basedsupplements VAERS, a passive
surveillance system, where underreporting and
lack of denominators are problems - Fastdata updated on weekly basis
- Good data qualitydata quality checked
frequently, feedback goes to data managers at
sites
7Vaccines for which VSD has conducted active
surveillance
- Adolescent and adult
- MCV4 (Menactra)
- Tdap (Adacel and Boostrix)
- HPV (Gardasil)
- Influenza (TIV and LAIV)
- Childhood
- MMRV (ProQuad)
- Rotavirus (RotaTeq)
- Influenza (TIV and LAIV)
8Basics of active vaccine safety surveillance in
VSD
- For each vaccine, choose specific outcomes to
monitor (5-10) - Hypothesis-testing, not data-mining
- Each week, evaluate the number of outcomes in
vaccinated persons - Compare it to the expected number of outcomes
based on a comparison group
9Choosing outcomes
- 1. Select outcomes based on plausibility
- Pre-licensure data
- Known biologic properties of the vaccine
- VAERS reports
- Literature on this or similar vaccines
- 2. Additional criteria
- Clinically well-defined
- e.g., Guillain-Barré syndrome rather than
neurologic problems - Acute-onset
- Serious
- Relatively uncommon
10Example Outcomes, HPV study
- Guillain-Barré syndrome
- Appendicitis
- Stroke
- Venous thromboembolism
- Seizures
- Syncope
- Allergic reactions
11Historical comparison method
- Uses incidence rates from historical data
- Advantage Allows earlier recognition that a
small number of excess cases is unusual - Limitation Background rates may be uncertain,
vary over time, or hold potential for confounding
12Concurrent comparison method
- Uses matched controls, e.g., visits for other
vaccines or just preventive care - Advantage Avoids false signaling or missed
signals due to error/trends/ confounding in
historical data - Limitations
- Need to define an appropriate control group not
simple! - Vaccines may be adopted rapidly, leaving few
controls
13Sequential analysis methods
- Each week, our analysis includes data from all
previous weeks - Problem Repeated testing increases the chance
of false-positive results - Need to adjust for this statistically
- Solution Sequential analysis
14Maximized Sequential Probability Ratio Testing
(maxSPRT)1
- Refinement of a classical statistical method2
- Null hypothesis no excess risk
- Alternative hypothesis increase in risk
- Test statistic log likelihood ratio (LLR)
- Depends on numbers of events in exposed vs.
unexposed observation time - If critical value of LLR exceeded, potential
association
- Kulldorff M et al. Working paper. Available at
www.dacp.org/faculty_Kulldorff.html. - 2. Wald A. Annals of Mathematical Statistics,
16117186, 1945.
15Sequential analysis methods
- Method to be used for comparison with
- historical data, large sample size
- - MaxSPRT, Poisson data
- historical data, small sample size
- - Conditional MaxSPRT, takes uncertainty in
estimated historical rates into account - concurrent matched controls
- - Flexible exact sequential analysis
- self-controls
- - MaxSPRT, binomial data
16Example Rotashield vaccine and intussusception
(retrospective analysis)
Vaccine suspended
Vaccine licensed Aug 98 By Jul 99, 15 reports to
VAERS
Withdrawn
1999
17Example Rotashield vaccine and intussusception
(retrospective analysis)
Vaccine suspended
Vaccine licensed Aug 98 15 VAERs reports through
Jul 99
Withdrawn
MaxSPRT analysis would have signaled in May 1999
Log likelihood ratio
Critical value 3.3
1999
18SIGNAL ? ASSOCIATION ? CAUSALITY
19Signals observed
- 11 signals observed in 4 studies
- Only 1 true association
- Example of false positive ?
20HPV and appendicitis(Age group 18-26 years)
- When surveillance started, HPV had been in use
for several months - 1 case in Week 2 of data produced retrospective,
transient signal - At time of actual look, relative risk (RR) and
LLR had decreased to null values - Signal ascribed to chance
21Data from 8/20/06 to 6/8/08 19 cases, 108,184
vaccines given (16.4 cases expected) RR1.2,
LLR0.1995.
22Reasons for false signals
- Error in estimated background rates
- Very few cases in historical data
- Changes in true incidence or coding over time
- Inappropriate comparison group
- Confounding
- Miscoding of outcomes
- Chance
23Signals sometimes do represent true increases in
risk
24MMRV study background
- Age 12-23 months
- Outcomes monitored
- Ataxia --Thrombocytopenia
- Seizures --Arthritis
- Meningitis encephalitis --Allergic reactions
- Post-vaccination window 42 days
- Comparison group MMR visits in 2000-2006
- MMRV usage began 2006
- MMRV doses given as of March 2009 84,000
25MMRV and seizures
- By 33rd week of data (Sept. 2006) 22 seizure
diagnoses had occurred among 9,000 one-year-old
MMRV recipients, vs. 10.7 expected from the
historical incidence rate after MMR - Relative risk 2.1
- LLR surpassed threshold ? signal
26MMRV and seizures Log likelihood ratio and
relative risk over time
Relative risk
Log likelihood ratio
Relative risk
Log likelihood ratio
Critical value of LLR
27Number of seizures among 12-23 mo. olds by day
after MMRV vaccination, 2006-2007
Number of Seizures
Days Post-MMRV Vaccine
(2/06-9/07, after 47,137 vaccine visits)
28Seizure rate among 12-23 month olds by day after
vaccination, 2000-2008
Temporal Scan
29MMRV and seizures in 1-year-olds
- 2-fold increased risk of seizure on Days 7-10
after MMRV compared to MMRV, adjusting for age,
VSD site, year, flu season - For every 2,000 doses of MMRV given instead of
MMRV, 1 additional seizure on Days 7-10 - Same result for febrile seizure after chart
review and reanalysis - Same result with (not yet chart-reviewed) data
updated through 10/08 and alternative analyses
30(No Transcript)
31Studying influenza vaccine safety
- Cannot use usual comparison groups to study flu
vaccine safety - Vaccinated vs. unvaccinated
- Misclassification of vaccinated
- Confounding by indication
- Vaccinated (current) vs. vaccinated (past)
- Population recommended to receive vaccines
changed over time - Secular trends in potential adverse events
32Influenza vaccine safety analytic approaches
- Self-controlled case series
- Compares risk of AEs in risk vs. control periods
of current season - Difference-in-difference
- Adjusts for risk vs. control periods of
cumulative previous seasons
33Status of adolescent and adult vaccine studies
34Conclusions
- Sequential evaluation of accumulating data
complements traditional, passive, adverse event
reporting - Makes best, early, use of routinely collected
data - Signals can provide early evidence of risk
- Lack of signals may provide rapid reassurance
when assessing concerns that arise from another
source
35Upcoming adolescent/adult vaccines to be
monitored for safety after FDA licensure
- MCV4 (Menveo)
- HPV (Cervarix)
- Influenza A/H1N1 vaccine
- 2009-2010 influenza vaccine
36Local VSD colleagues
- Rich Fox
- Sharon Greene
- Virginia Hinrichsen
- Martin Kulldorff
- Grace Lee
- Lingling Li
- Renny Li
- Tracy Lieu
- Rich Platt
- Ping Shi
- Irene Shui
- Ruihua Yin