Comparative analyses of Drug-Adverse Event Associations
in Various European Databases
Raymond G. Schlienger, PhD, MPH. Global Clinical Epidemiology, Novartis Pharma AG, Basel, Switzerland
Mark de Groot, PhD. Division of Pharmacoepidemiology & Clinical
Pharmacology, Utrecht Institute for Pharmaceutical Sciences, Utrecht University, The Netherlands
First results from PROTECT WP2-WG1 ISPE Mid-Year Meeting, Munich, 12-Apr-2013
Disclosures
• RGS: full-time employee of Novartis Pharma AG, Basel Switzerland. The views expressed in this presentation are those of the author in his role of industry co-lead for IMI Protect WP2/WG1
• MdG: employee of Utrecht University, The Netherlands, and partly funded by TI Pharma Mondriaan grant T6.101
2
Acknowledgements • The research leading to these results was conducted as part
of the PROTECT consortium (Pharmacoepidemiological Research on Outcomes of Therapeutics by a European ConsorTium, www.imi-protect.eu) which is a public-private partnership coordinated by the European Medicines Agency
• PROTECT work in this presentation is work by WP2/WG1 colleagues
• The views expressed are those of the authors only
Contents • Background Working Group 1 (WP2/WG1)
– Framework of WP2/WG1 within Protect
• Objectives of WP2/WG1
• Methods
• Results
– Descriptive studies
• Current status and next steps
• Conclusions
4
5
WG1 Databases
WG2 Confounding
WG3 Drug utilization
Number of participants
n=46 33 public, 13 private
n=14 10 public, 4 private
n=9 5 public, 4 private
Public partners EMA, LMU-München, Witten
University4, AEMPS, CEIFE, CPRD, DKMA and UU
UU FIFC, LMU and Witten University4
Private partners
Amgen, AstraZeneca, Genzyme, GlaxoSmithKline, La-Ser, Merck,
Novartis, Roche and Pfizer
Amgen, Novartis, Roche and Pfizer
Amgen, Novartis and Roche
WG Coordinators
Raymond Schlienger1 (Novartis) Mark de Groot2 (UU)
Nicolle Gatto (Pfizer) Rolf Groenwold (UU)
Joan Fortuny3 (Novartis)
Luisa Ibanez (FIFC)
WP2 coleaders Olaf Klungel (UU) - Robert Reynolds (Pfizer)
WP2 coleaders alternates
Tjeerd van Staa (CPRD) - Jamie Robinson (Roche)
WP2 Project Manager Ines Teixidor (UU)
1 from October 2010 replacing John Weil (GSK), 2 from 1 February 2011 replacing Frank de Vries (UU), 3 from 15 March 2012 replacing Hans Petri (Roche), 4 New partner, accession approved by SC in January 2013
WP2: Participants and their roles
WP2: Framework for pharmacoepidemiological studies
6
To: • develop • test • disseminate
of pharmacoepidemiological studies applicable to:
• different safety issues • using different data sources
methodological standards for the: • design • conduct • analysis
Objectives:
Objective WP2 – WG1
• Explain differences in drug-adverse event (AE) associations due to choices in methodology and databases – Reduce variation due to methodological choice of individual
researchers
– Explain variation due to characteristics of country/database
– More consistency in drug-AE studies to improve B/R assessment of medicines
7
Methods
• Conduct of drug-AE studies in different EU healthcare databases, using different study designs
– Selection of 6 key drug-AE pairs AEs that caused regulatory decisions Public health impact (seriousness of the event, prevalence
of drug exposure, etiologic fraction) Feasibility Range of relevant methodological issues
– Development of study protocols for all drug-AE pairs
– Compare results of studies
– Identify sources of discrepancies
8
Methods: Drug - AE pair selection
• Selection of 6 key AEs and drug classes
– Initial list of 55 AEs and >55 drugs
– Finalisation based on literature review and consensus meeting
9
Drug – AE pair Antidepressants - Hip fracture Benzodiazepines - Hip fracture Antibiotics - Acute liver injury
Beta2 Agonists - Myocardial infarction Antiepileptics - Suicide
Calcium Channel Blockers - Cancer
Methods: Characteristics of individual databases
10
Data
base
Coun
try
Cum
ulat
ive
popu
latio
n (2
008)
Activ
e po
pulat
ion
(200
8)
Data
so
urce
Codi
ng
diag
nose
s
Codi
ng
drug
s
Reco
rdin
g of
dru
g us
e
BIFAP ES 3.2 Mio 1.6 Mio GP ICPC ATC Prescribing CPRD UK 11.0 Mio 3.6 Mio GP READ BNF Prescribing THIN UK 7.8 Mio 3.1 Mio GP READ BNF Prescribing Mondriaan
NPCRD AHC
NL 0.7 Mio
0.26 Mio
0.34 Mio 0.17 Mio
GP
GP/Pharmacy
ICPC ICPC
ATC ATC
Prescribing
Prescribing + dispensing
The Danish national registries
DK 5.2 Mio 5.2 Mio GP + specialist doctors
ICD-9 ATC Prescribing + dispensing
Bavarian Claims Database
DE 10.5 Mio 9.5 Mio Claims health insurance
ICD-10 ATC Claims
Methods: Designs
• Descriptive studies for drug-AE pairs in all DBs
– Prevalence of exposure of interest
– Prevalence/incidence of outcome of interest
• Association studies: Different study designs in selected DBs
– Cohort studies
– Nested case-control studies
– Case crossover studies
– Self-controlled case series 11
Methods: Overview of planned studies
Drug - AE pair
Descriptive Cohort Nested case control
Case crossover
Self-Controlled case series
AB - ALI All Databases
CPRD BIFAP
CPRD BIFAP
CPRD BIFAP
CPRD BIFAP
AED - Suicide All Databases
CPRD DKMA
n/a n/a n/a
AD - Hip All Databases
THIN Mondriaan BIFAP
THIN Mondriaan BIFAP
THIN Mondriaan
THIN Mondriaan
BZD - Hip All Databases
CPRD BIFAP Mondriaan
CPRD BIFAP Mondriaan
CPRD BIFAP
CPRD BIFAP
B2A - AMI All Databases
CPRD Mondriaan
n/a n/a n/a
CCB - Cancer All Databases
CPRD DKMA
n/a n/a n/a
12
Methods: Standardization of methods
• Common protocol for each drug-AE pair
• Common standards, templates, procedures
– Detailed data specifications (statistical analysis plan) including definitions, codes for diseases, drugs etc.
– Age-/sex-standardization to European reference population
• Blinding of results of individual DB analyses
• Submission of protocols to ENCePP study registry
13
Methods: Standardisation to European reference population • age stratification
14
Results: Benzodiazepine exposure prevalence 2001-2009
15
Crude Standardised
Results: Benzodiazepine exposure prevalence by age & sex (2008)
16
Females Males
Results: Benzodiazepine exposure prevalence – methodological issues • Marked differences of BZD exposure prevalence
across countries/DBs
– Unlikely to be primarily explained by DB characteristics (e.g. different drug coding systems) Prescribing vs dispensing information
– Rather explained by different prescribing habits, e.g. driven by country guidelines/policies, marketing, reimbursement, ...
• Age-/sex-standardization had relevant impact specifically on Mondriaan data
17
Results: Hip fracture incidence by sex
18
Adjusted Incidence of Hip Fractures in Males over 50 years old
0
5
10
15
20
25
30
35
40
2003 2004 2005 2006 2007 2008 2009
Years
Inci
d *1
0,00
0py
DKMA
BAVARIAN
AHC
BIFAP
CPRD
THIN
NPCRD
Adjusted Incidence of Hip Fractures in Females over 50 years old
0
10
20
30
40
50
60
70
80
2003 2004 2005 2006 2007 2008 2009
YearsIn
cid
*10,
000p
y
DKMA
BAVARIAN
AHC
BIFAP
CPRD
THIN
NPCRD
Results: Hip fracture incidence - methodological issues • Hip fracture: defined as a fracture of the proximal
femur in the cervix or in the trochanteric region
• Operational definition for this study: “any femur fracture”
– Some coding systems (International Classification of Primary Care ([ICPC-2]) don’t have a specific code for hip fracture, but only a broader code for “femur” fracture
19
Results: Hip fracture incidence - methodological issues (2) • Definition and coding of hip fracture/femur fracture
– ICPC-2: BIFAP and Mondriaan - 1 code
– ICD-10: Danish Registries and Bavarian Claims DB - 9 codes
– READ: THIN and CPRD - 64 codes
20
Results: Antidepressants and indications (2008)
21
Results: Antidepressants and indications - methodological issues • Major differences across DBs regarding underlying
‘indications’ for ADs
• Most DBs do not capture specific information on indication
• Time window defined ± 90 days around AD prescribing date to identify disorder which may correlate to AD prescribing
22
Results: ALI incidence rates by age and sex – definite cases (2008)
23
Results: Standardised ALI incidence rates – definite cases (2008)
24
Results: Incidence of acute liver injury - methodological issues • Major challenge to define idiopathic ALI in DB which
use different coding systems
– Codes specific of liver disease or symptoms (e.g. hepatitis , acute hepatic failure, icterus, ...)
– Non-specific codes (e.g. liver function tests abnormal, increased transaminases)
• Manual review of ‘free-text’ (in BIFAP and CPRD)
• Classification in ‘definite’, ‘probable’, ‘non-cases’ based on available DB information
25
Results: Incidence of acute liver injury - methodological issues (2)
26
Results: Incidence of acute liver injury - methodological issues (3)
27
Results: Antiepileptic drug prevalence
28
Results: AED exposure prevalence - methodological issues
• Definition of AED - literature provides different definitions of drugs belonging to that drug class
• Broad range of neurological and psychiatric indications for AEDs in addition to epilepsy
29
Results - Cohort studies
• Due to blinding of results policy we cannot show any results at this point
30
WG1: Progress of studies
31
Drug - AE pair
Descriptive Cohort Nested case control
Case crossover
Self-controlled case series
AB - ALI Completed Completed March 2013 May 2013 March 2013
AED - Suicide Completed March 2013 n/a n/a
n/a
AD - Hip Completed Completed Aug 2013
Dec 2013 n/a
BZD - Hip Completed Completed Sept 2013
Sept 2013 Sept 2013
B2A - AMI Completed March 2013 n/a
n/a
n/a
CCB - Cancer Completed April 2013 n/a
n/a
n/a
Conclusions
• WP2/WG1 provides unique framework for studying and explaining potential differences in drug-AE associations due to choices in methodology and DBs
• Descriptive studies on exposure and outcomes to better characterize the individual DBs have been finalized
• Association studies:
– Cohort studies on all outcomes across all DBs being finalized
– Other designs within same DB ongoing or starting soon
32
Conclusions (2) • Challenge to dissect identified differences (both of
exposure and outcome data)
– Due to different prescribing habits
– Due to true underlying differences in individual populations Life-style factors, genetics Different co-morbidities, risk factor distribution Latitude Other
33
Conclusions (3) – Due to differences in DB characteristics/structure Information on certain life-style factors (alcohol, smoking),
BMI Prescribing vs dispensing Primary care EMR db vs health claims DB vs population
registries Underlying coding systems Other
– Due to different interpretation of protocol/data specifications
– Differences because of different statistical software
– Impact of different study designs
34
Questions
35