"Renaissance in Signal Management" highlights recently published studies and their insights regarding opportunities to refine signal management processes and leverage different data sources including electronic health records for signal detection. The talk discusses the published work of many teams including CIOMS, EMA, IMI UMC, FDA and others and what the results of those studies offer regarding the relative strengths of various data sources and statistical methods for signal detection. This talk was presented at the 2016 World Drug Safety Congress by Commonwealth Informatics.
5. v
CIOMS-VIII Signal Management Process
Copyright 2016 Commonwealth InformaCcs Inc. Slide 5
Impact
assessment
and
prioriCzaCon
NEED
FURTHER
INVESTIGATION?
Monitor via rouCne
pharmacovigilance
(if signal is indeterminate)
OR
Close out
(if signal is refuted)
INDIVIDUAL CASE ADVERSE EVENT REPORTS
• Clinical trials (serious adverse events)
• Post-markeCng sources (serious and non-
serious adverse events)
• Literature report
SIGNAL DETECTION IN ADVERSE EVENT
REPORTING SYSTEM
• Health authority / monitoring center
systems
• Company databases
TRADITIONAL PHARMACOVIGILANCE
METHODS
• Review of individual cases
• Aggregate analyses of case report data,
using case counts, crude or adjusted
reporCng rates, etc.
DATA MINING ALGORITHMS
• ProporConal ReporCng RaCo (PRR)
• MulC-item Gamma-Poisson Shrinker
(MGPS)
• Bayesian Confidence PropagaCon Neural
Network (BCPNN)
TRIAGE OF OUTPUTS
• Interpret within the context of all other
relevant sources of safety data, disease
knowledge, biological plausibility,
alternaCve eCologies for suspected adverse
drug reacCons, etc.
SIGNAL EVALUATION
• Case series analysis
• Analysis of exisCng clinical trial data
• Literature search and review
• Pharmacoepidemiologic studies
• MechanisCc studies
• AddiConal clinical trials
• Other types of studies
OTHER SAFETY DATA TO BE MONITORED
• Non-clinical / pharmacology studies
• Non-intervenConal studies
• Published literature (study reports,
mechanism of acCon, etc.)
• Periodic safety reports
• InformaCon on drugs in the same class
• Other relevant informaCon
From Figure 1, page 22, Report
of CIOMS Working Group VIII
6. v
ICSRs – EMA Signaling Study
§ Study period Sep-2003 to Mar-2007 using
Eudravigilance data
§ 191 chemically disCnct products
§ 532 adverse reacCons added to the SPC in the period
§ Searched EMA records to see when the agency first
became aware of each of the adverse reacCons
§ Looked retrospecCvely to find the earliest point when
the signal of disproporConate reporCng (SDR) criteria
would have occurred
§ Compared Cme of SDR with Cme of EMA awareness
Reference: Alvarez, Y., Hidalgo, A., Maignen, F., & SlaOery, J. (2010). ValidaCon of staCsCcal signal
detecCon procedures in EudraVigilance post-authorizaCon data. Drug safety, 33(6), 475-487.
Slide 6
7. v
ICSRs – EMA Signaling Study
§ Used PRR 025 > 1 to define a signal of
disproporConate reporCng (SDR)
§ 6356 SDRs
§ 1561 SDRs survived first triage and required further
invesCgaCon
§ 217 SDRs represented real issues found earlier by SDR
that were eventually added to SPC
§ 1561 / 217 = 7.2 invesCgaCons per signal
§ 405 IME ADRs added to the SPCs
§ 217 (54%) found earlier by SDR than exisCng methods
§ 79 found later by SDR
§ 109 not signalled by SDR in the study period
Reference: Alvarez, Y., Hidalgo, A., Maignen, F., & SlaOery, J. (2010). ValidaCon of staCsCcal signal
detecCon procedures in EudraVigilance post-authorizaCon data. Drug safety, 33(6), 475-487.
Slide 7
9. v
Electronic Health Records (EHR)
§ IMI-PROTECT
§ Work Package 3 Sub-package 10, "Signal detecCon in
Electronic Health Records"
§ Key points
§ Developed and tested a quesConnaire for structured
assessment of temporally associated drug-event
combinaCons
§ Demonstrated that exploratory analysis of electronic
medical records can detect important potenCal safety
signals
§ An expert triage step is required to achieve an acceptable
false posiCve rate (as with signal detecCon from ICSRs)
Copyright 2016 Commonwealth InformaCcs Inc. Slide 9
Reference: Cederholm, S., Hill, G., Asiimwe, A., Bate, A., Bhayat, F., Brobert, G. P., ... & Norén, G. N.
(2014). Structured Assessment for ProspecCve IdenCficaCon of Safety Signals in Electronic Medical
Records: EvaluaCon in the Health Improvement Network. Drug safety, 38(1), 87-100.
10. v
Signal detecCon from EHR - Methods
§ 42 drugs randomly selected
§ Up to 20 events per drug selected from those
with a temporal associaCon
§ AssociaCon determined by calibrated
self-controlled cohort analysis
§ Data from UK primary care electronic medical
records
§ The Health Improvement Network (THIN)
§ 6 assessors evaluated 7 drugs each
§ up to 20 medical events per drug
§ using a pre-specified quesConnaire
Copyright 2016 Commonwealth InformaCcs Inc. Slide 10
Reference: Cederholm, S., Hill, G., Asiimwe, A., Bate, A., Bhayat, F., Brobert, G. P., ... & Norén, G. N.
(2014). Structured Assessment for ProspecCve IdenCficaCon of Safety Signals in Electronic Medical
Records: EvaluaCon in the Health Improvement Network. Drug safety, 38(1), 87-100.
13. v
CIOMS-VIII Signal Management Process
Copyright 2016 Commonwealth InformaCcs Inc. Slide 13
Impact
assessment
and
prioriCzaCon
NEED
FURTHER
INVESTIGATION?
Monitor via rouCne
pharmacovigilance
(if signal is indeterminate)
OR
Close out
(if signal is refuted)
INDIVIDUAL CASE ADVERSE EVENT REPORTS
• Clinical trials (serious adverse events)
• Post-markeCng sources (serious and non-
serious adverse events)
• Literature report
SIGNAL DETECTION IN ADVERSE EVENT
REPORTING SYSTEM
• Health authority / monitoring center
systems
• Company databases
TRADITIONAL PHARMACOVIGILANCE
METHODS
• Review of individual cases
• Aggregate analyses of case report data,
using case counts, crude or adjusted
reporCng rates, etc.
DATA MINING ALGORITHMS
• ProporConal ReporCng RaCo (PRR)
• MulC-item Gamma-Poisson Shrinker
(MGPS)
• Bayesian Confidence PropagaCon Neural
Network (BCPNN)
TRIAGE OF OUTPUTS
• Interpret within the context of all other
relevant sources of safety data, disease
knowledge, biological plausibility,
alternaCve eCologies for suspected adverse
drug reacCons, etc.
SIGNAL EVALUATION
• Case series analysis
• Analysis of exisCng clinical trial data
• Literature search and review
• Pharmacoepidemiologic studies
• MechanisCc studies
• AddiConal clinical trials
• Other types of studies
OTHER SAFETY DATA TO BE MONITORED
• Non-clinical / pharmacology studies
• Non-intervenConal studies
• Published literature (study reports,
mechanism of acCon, etc.)
• Periodic safety reports
• InformaCon on drugs in the same class
• Other relevant informaCon
From Figure 1, page 22, Report
of CIOMS Working Group VIII