Dr. Mahyar Etminan Scientist, Therapeutic Evaluation Unit...
Transcript of Dr. Mahyar Etminan Scientist, Therapeutic Evaluation Unit...
Pharmacoepidemiology
Dr. Mahyar Etminan Scientist, Therapeutic Evaluation Unit
Department of Pediatrics, UBC
Contacts Email: [email protected]
What is pharmacoepidemiology? study of drug efficacy, toxicity and patterns of
use in large populations uses epidemiologic methods to answer drug
related questions in large populations mainly examines drug safety but can also look at
comparative effectiveness
Pharmacoepidemiology
Biostatistics
Computer programming
EpidemiologyMedicine
Administrative dataPharmacology
Pharmacoepidemiology-collection of several disciplines
History of pharmacoepidemiology: the thalidomide tragedy
introduced in the 1960s marketed as thalidimide (Contergan®) used to control nausea in pregnancy reports of limb abnormalities need for pharmacovigilance
Dr. Frances Oldham Kelsey
Study designs in pharmacoepidemiology
Cross-sectional studies Case-control studies case-only designs
Case-cross over, self-controlled case series cohort studies
Limitations of RCTs underpowered to detect rare adverse events usually short follow up difficult to apply results to a real clinical setting difficult to execute; logistics, costs
Cross-sectional studies
Questionnaires or surveys Can only examine association at a
certain point in time Cant go back to look at exposure
lack of temporality
Case-control studies Ideal for rare adverse events First step: Identify all cases Second step: Select controls (none cases) Compare exposure distribution between
cases and controls Calculate an odds ratio
Calculating the OR
A BC D
drug
no drug
Case Control
OR=AD/BC
Pn-
Pn+
Cases
Controls
Cases Controls
drug+ A B
drug- C D
Pn+ Pn- OR = odds case exposed = ad/bc
odds control exposes
Case-Control StudyCompare drug
exposure in Cases and Controls
Group of patients
(RA Patients in BC Jan 1996-Dec 2004)drug No drug
Where is pharmacoepidemiology most effective?
effective tool in in identifying rare ADEs pharmacovigilance
requires large sample size to detect some examples of rare ADEs
bisphosphonates/jaw necrosis NSAIDs/tendon rupture dopamine agonist/heart valve abnormalities
Confounding Bias:Major threat to validity of pharmacoepid studies
People taking certain medications may have more comorbid conditions may be more adherent to medications may be more inclined to receive medications
from their physicians All of these reasons can change exposure
prevalence between cases and controls May lead to bias
What is a confounder? a variable that distorts the results a confounder is a variable that:
Is associated with the outcome Is associated with the exposure Is not an effect of the exposure
basically it causes a ‘mixing’ effectin the results of the study
Controlling for confounding How do we know the cases and controls
are not different with respect to other variables?
Example: Looking at the association between rofecoxib and myocardial infarction What if the rofecoxib users are sicker?
Bias in pharmacoepidemiology
bias types of bias
recall bias selection bias channeling bias misclassification bias
Systematic error in research
difficulty recalling drug
one group different than another
drug prescribed to sicker patients
Either disease or drug misclassified
Controlling for confounding
Restriction Stratification Matching statistical adjustment
Adverse drug reaction behavior-Hazard Function
ADRs onset may differ based on a drug’s pharmacokinetics
long half life-SSRIs Short half life-benzodiazepines
pharmacodynamics proton pump inhibitors
drug-drug interaction Ace-Is and K+diuretics
Duration of Drug use (Months)
Haz
ard
func
tion
0 3 6 9 12
NSAID use and Risk of GI bleed
Cohort studies by definition, group of people share similar
characteristics followed forward some exposed, some unexposed at the end of follow up, rate of exposed
compared to rate of unexposed using statistics, potential confounders is
controlled for
Cohort Entry
Time
Dynamic cohort
20152012
drug+ A B
drug- C D
Pn+ Pn- RR = a/(a+b)/c/(c+d)
Retrospective Cohort StudyCohort
(RA Patients in BC Jan 1996-Dec 2004)drug No drug Pn
+ Pn+
Pn+
Pn+
Pn+
Pn+
Pn+
Compare pneumonia incidence in exposed
and unexposed subjects
Relative Risk=probability of an event with the drug compared to the event in those without taking the drug
Link between basic, clinical pharmacology & pharmacoepidemiology
example with Nifedipine (Adelat XL) basic pharmacology
Ca+2 blockers block calcium influx clinical pharmacology
nifedipine’s starting dose in a 50 year old is 30mg daily
pharmacoepidemiology the relative risk of MI in nifedipine users
compared to thiazide users among cohort of antihypertensive users is 2.1
Canadian administrative databases originally used for admin purposes Saskatchewan data used first for
research 1980’s: Î number of pharmacoepi
research late 1990s:data from Quebec, BC,
Ontario used for research 2016-All Canadian provinces have
health admin databases data access can take many months
Population Based Registries
Advantages of BC Health Admin Data Low drop out rates Large Sample ~ 5 million Includes children, adults, older adults Will have lab data soon Captures procedures, surgeries, drugs, diagnoses, cancer, mental health
Examples of administrative databases administrative data in Canada
BCLHD Ontario Linked Databases RAMQ Saskatchewan Linked Databases
clinical databases General practice research database (GPRD.com) THIN database, Scotland
Disadvantages of administative data validity questioned for some outcomes may take years to access data costs-up to hundreds of thousands of $ lack of information for some confounding
variables data access-privacy
Clinical databases clinical databases-intended for clinical research include variables including
smoking status alcohol intake BMI cholesterol Blood pressure
usually cost more than admin data
General Practice Research Database (GPRD)
largest clinical database available based in the UK coding based on diagnosis captures information on
BMI, smoking, cholesterol, BP
Impact of pharmacoepidemiology population based drug related studies
rare drug adverse events drug efficacy drug interactions patterns of use
The Vioxx® controversy first reports of MI in VIGOR
trial confirmed by
pharmacoepi/cohort study published in the Lancet 400K US patients Î in risk of MI
Studying drug interactions in pharmacoepidemiology
drug interaction studies usually small human studies in vitro studies soft end-points
usually look at soft outcomes clinical significance of interaction-unknown
cimetidine P-450 inhibition not all interactions clinically significant
Drug interactions in pharmacoepidemiology case-control study by Juurlink (JAMA;2003) looked at interaction between
cotrimoxazole+glyburide hypoglycemia clarithryomcin+digoxin digoxin toxicity K+ sparing diuretics+Ace-inhibitors hyperkalemia
used the Ontario admin data using 1.5 million older adults
7 years time span matched for age, sex, medication, renal disease, comorbidity
Juurlink et al (JAMA, 2003) Results
Cotrimoxazole+glyburide OR=6.6 (95% CI, 4.5-9.7) Clarithryomcin+digoxin OR=11.7 (95% CI, 7.5-18.2) K+ sparing diuretics+Ace-I OR=20.3 (95% CI, 13.14-30.7)
Pharmacoepidemiology valuable tool in quantifying drug interactions
Statins and COPD mortality (Mancini JACC 2006)
COPD disease with high morbidity and mortality inflammation-important role in the pathology of COPD
statins-popular cholesterol lowering drugs ↓ death secondary to CAD may have anti-inflammatory properties antioxidative properties
study question: can statins lower the risk of COPD hospitalization/death ?
Inflammatory and antioxidative effects of statins
Statins and COPD mortality (Mancini et al. JACC 2006)
case-control study using Quebec data nested within cohort of subjects with CAD odds ratios statins
COPD hospitalization 0.88 (0.69-1.12) total mortality 0.62 (0.48-0.79) MI 0.44 (0.36-0.54)
Fen-Phen®
1995 1998 2002 2004 2006
Impact of pharmacoepidemiologic studies on drug policy
JAMA New Eng J Med BMJ BMJ Lancet New Eng J Med
Pitfalls of pharmacoepidemiology poly-pharmacy publication of poorly conducted studies misinterpretation of results by consumers
Pharmacoepidemiology programs in the US and Canada Boston Collaborative Drug Surveillance Program Center for Clinical Epidemiology and Biostatistics
university of Pennsylvania Brigham and Women Hospital
affiliated with Harvard McGill University
Dept of Clinical Epidemiology and Biostatistics
New Canadian Initiative in Drug Safety Sponsored by Federal government Answer drug safety questions from Health
Canada 30 million budget Have linked all Provincial admin data
Case-only Study Designs in Drug Safety Research: The self controlled Case-
Series
Case only designs
One major limitation of observational studies is the potential imbalance between the two group (drug vs no drug) with respect to other variables (age, medical conditions, other drugs etc).
Statistical analysis do not always correct for this imbalance which may lead to biased results
Case only design
Case only designs are a selected group of study designs in epidemiology that only use cases
Since only cases are used, between subject confounding that may occur in cohort studies or case-control studies is eliminated
The self-controlled case-series is a type of a ‘case only’ design that is used in drug safety
Self-controlled case-series (SCCS) A type of cohort study where
exposed/unexposed periods are examined in the same subject with respect to the outcome
time
The rate of exposed periods are compared to the baseline rate
drug periodNo drug
Event
When can SCCS be used Events are independent of drug use Event can not predict use of future drug use This design works best when risk periods are
short (acute events)
Bisphosphonates and macular degeneration
Bisphosphonates are a popular class of osteoporosis medications eg. Fosamax, Zometa
They are potent inflammatory drugs have been linked to uveitis, scleritis
Do Bisphosphonates increase risk of macular degeneration?
Self Controlled Case series
AMD
Exposed Rate
58
Time
Anti-VEGF injection
No Bisphosphonate use First Bisphosphonate Rx
Unexposed Rate
Rate Ratio=Rate exposed/Rate Unexposed
Exposure period
Unexposed Exposed Age-AdjustedRR (95 % CI)*
Patient Years
n AMD Patient Years
n/AMD
1 89.0 56 50.5 34 1.22 (0.76-1.95)
2 277.1 92 111.7 49 1.29 (0.86-1.94)
3 461.8 101 155.0 58 1.53 (1.05-2.25)
4 623.3 101 190.7 65 1.80 (1.25-2.62)
5 761.4 101 271.3 80 1.87 (1.32-2.67)
Overall 942.6 101 369.7 92 1.99 (1.41-2.79)
RRs for AMD with BPs stratified by years of use