Transcript Case-Control Studies: An Example

1
CASE-CONTROL
STUDIES:
AN EXAMPLE
EP 711
November 8, 2011
2
Issues in Study Design and Interpretation
• Type of study
• Case definition
• Control definition
• Sources of exposure information
• Exposure definition
• Potential confounders
• Potential sources of bias
3
Louik C et al: N Engl J Med 2007;356:2675-83
4
Background
• ≥10% of pregnant women experience depressive
symptoms in pregnancy
• Medications represent one treatment option
• SSRIs first marketed in 1988 and gained widespread
acceptance
• RDD survey in 2005 indicated that among women aged 18-44,
~8% used an SSRI in the week preceding interview
5
Background
• Advantages of SSRIs
• Established efficacy
• Favorable side effect profile
• Serum monitoring not required
• Toxicity/overdose not a major problem
6
Background
• SSRIs include
• Fluoxetine (Prozac)
• Sertraline (Zoloft)
• Paroxetine (Paxil)
• Citalopram (Celexa)
• Fluvoxamine (Luvox)
• Escitalopram (Lexapro)
7
Background
• Congenital malformations
• Major defects affect about 2-3% of livebirths
• Encompasses a wide variety of defects
• Some are extremely rare, e.g. limb reduction defects (2-4/10,000)
• Others are more common, e.g. cleft lip (1/700), neural tube defects
(1/1000)
• Heart defects occur about 1/200 births
8
Background
• Initial studies
• Small cohorts from pregnancy registries
• Results reassuring with respect to birth defects overall
9
Background
• Later studies
• Larger
• Found no increased risk for birth defects overall, but
• Increased risk for some specific defects
• 3 independent studies reported increased risk for heart defects
10
Objectives
• Examine 1st trimester use of specific SSRIs in relation to
specific birth defects
• Test existing hypotheses
• Craniosynostosis
• Omphalocele
• Heart defects
• Explore other defects not yet reported
• Focus on specificity to help reduce misclassification
12
Choice of Study Design
• Birth defects occur in 2-3% of livebirths (not uncommon)
• Short latent period (9 months)
• So why not cohort??
13
Choice of Study Design
• “Birth Defects” is NOT a single outcome, and individually,
THEY ARE RARE
• Major birth defects affect 2-3% of newborns
• However, specific defects typically affect only ~2/1000 to 1/10,000
• Known teratogens typically increase the risk of a specific
defect or a group of related defects (syndrome or
sequence)
14
Choice of Study Design
• Therefore, a case-control approach is better suited to our
research question looking at specific defects.
Types of Case-Control Studies
Specific
Semi-Specific
Non-Specific
16
Methods
• Study subjects
• Infants with any of a wide range of malformations (isolated
minor defects excluded)
• Sample of non-malformed infants from same birth hospitals
17
Methods
• Identified in 5 study centers
• Study staff review clinic/surgical logs,
admission/discharge lists, contact newborn nursery
and labor/delivery rooms
• In 2 study centers, use statewide birth defects
registries
• Nonmalformed infants identified at study hospitals
• In Mass., population-based random sample of
newborns
18
Methods
• Mothers interviewed within 6 months of birth
• Trained study nurses
• By telephone
• Questions address
• Demographics
• Reproductive history
• Medical history
• Lifestyle habits
• Detailed medication history (includes prescription, over-the-counter,
and herbal products)
19
Case definition--Theory
• Specific as possible
• Strict criteria
• Should be reasonable to think that they have a common
etiology
20
Case Definition--Applied
• Cases are infants with heart defects
• Concerns
• No rigorous definition
• Not homogeneous
• Solution
• Blind review of all infants with a heart-related defect
• Sub-classify into embryologically meaningful groups to increase
homogeneity
21
Case Definition--Applied
• Case subgroups
• Looping, laterality, and single ventricle defects
• Conotruncal defects
• Atrioventricular canal defects
• Right ventricular outflow tract obstruction defects
• Left ventricular outflow tract obstruction defects
• Septal defects
• Anomalous pulmonary venous return
Normal Heart
23
Pulmonary Valve Stenosis
24
Ventricular Septal Defect
25
Control Definition--Theory
• Sample of the population that produced the cases
• Sampled independently of exposure status
26
Control Definition--Applied
• Infants with other malformations
• Advantages
• From same population as cases
• Should remember pregnancy events similarly to cases
• Disadvantages
• May not reflect exposure status in population that gave rise to cases, i.e.
exposure may be associated with increased risk of these defects too
27
Control Definition--Applied
• Nonmalformed infants
• Advantages
• From same population as cases
• Should reflect exposure status in the population that gave rise to cases
• Disadvantages
• May remember pregnancy events differently than cases (recall bias)
28
Control Definition--Applied
• We will use nonmalformed controls
• Address disadvantages by
• Asking focused exposure questions (evidence exists that this enhances
recall)
• Considering other antidepressants (recall bias, if present, should apply
to these too)
29
Exposure Assessment--Theory
• Need sufficient detail of nature, duration, and timing of
exposure
• Accuracy is critically important
• Sources of Information
• Biomarkers
• Records
• Interviews
30
Exposure Assessment--Applied
• Biomarkers—not applicable
• Records
• Medical records
• Multiple providers
• Prescription ≠ Consumption
• Complete?
• Pharmacy
• No OTC meds, herbals
• Dispense ≠ Use
31
Exposure Assessment--Applied
• Interview
• Greater detail available
• Potentially includes all drugs
• Relies on mother’s recollection
• Because we chose to use nonmalformed controls, may differ between
cases and controls (recall bias)
Source of Selected Prescription Drugs 5,435
Mothers; Boston, Philadelphia, Toronto, Iowa, 1976-1984
Source (in %)
Total
Users
Drug
Bendectin
Valium
Fiorinal
Hydrodiuril
Clomid
Compazine
Darvon
Seconal
Diuril
859
135
105
76
66
59
42
42
27
Physician’s
Prescription
Other
98
82
96
97
100
91
78
95
85
1
18
4
3
0
9
22
5
11
Mitchell AA, Cottler LB, Shapiro S. Effect of questionnaire design on recall of drug exposure in pregnancy.
Am J Epidemiol 1986;123:670-6.
32
33
Exposure Assessment
J Womens Health (Larchmt). 2008 Sep;17(7):107380.
Prescription medication borrowing and sharing among
women of reproductive age.
Petersen EE, Rasmussen SA, Daniel KL, Yazdy MM,
Honein MA.
National Center on Birth Defects and Developmental
Disabilities, Centers for Disease Control and
Prevention, Atlanta, Georgia, USA.
33
34
Exposure Assessment
• Overall, 28.8% of women and 26.5% of men reported
ever borrowing or sharing prescription medications.
• Women of reproductive age were more likely to report
prescription medication borrowing or sharing (36.5%)
than women of nonreproductive age (>or=45 years)
(19.5%)
34
35
A quarter of new prescriptions go unfilled, study finds
February 18, 2010
35
36
Exposure Assessment
More than a quarter of new prescriptions are unfilled,
especially when the drugs are for symptomless
conditions, researchers from Boston's Brigham and
Women's Hospital have found. Physicians have long
been concerned that many patients fill new prescriptions
one time, then never get refills. But it has been
impossible in the past to determine adherence to new
prescriptions. The new study was made possible by the
implementation of an electronic-prescribing initiative by
two Massachusetts health plans.
36
37
Exposure Assessment—Evidence of
Completeness*
• False negatives
• Valium reports—1975
• Medical records—1.2%
• Maternal interview—5.8%
• 20% from sources other than physician
• Bendectin—late 1970’s, used exclusively in pregnancy
• Manufacturer’s data—25%
• Interview data—24%
*Mitchell AA, Cottler LB, Shapiro S. Effect of questionnaire design on recall of drug exposure in pregnancy.
Am J Epidemiol 1986;123:670-6.
38
Exposure Assessment—Evidence of
Completeness*
• False positives
• Elamar
*Mitchell AA, Cottler LB, Shapiro S. Effect of questionnaire design on recall of drug exposure in pregnancy.
Am J Epidemiol 1986;123:670-6.
39
Exposure Assessment—Addressing
Recall Bias
• If recall is complete, recall bias cannot exist
• Enhancement of recall will diminish recall bias
Reported Drug Use by Question Asked
532 Women; Boston - 1975, and Toronto - 1978
Drug
Total
Users
Percent Ascertained by
Open-ended
Indication
Name
(n)
Valium
21
55
26
(90)
Fiorinal
37
35
28
(43)
Darvon
23
45
32
(40)
Aspirin
1
79
20
(377)
Acetaminophen
2
55
43
(177)
Prescription Drugs
Non-Prescription Drugs
Am J Epidemiol 1986; 123:670-6.
40
41
Exposure Assessment
• Series of questions about medications
used any time from 2 months prior to
LMP through the pregnancy
• Illnesses they may have had, and medications used
42
Exposure Assessment
• List of indications for which medications might be used
• Anxiety, depression, other psychological conditions
• Use of specific medications
• Prozac, Zoloft, Paxil, Effexor, Elavil, Celexa, Luvox, Lexapro,
Wellbutrin
43
Exposure Assessment—Details
• Timing
• First trimester
• 28 days prior to LMP through the 4th lunar month
• Nature
• Any SSRI
• Specific SSRI
• Non-SSRI antidepressants
• Helps assess presence of recall bias
• Helps assess “confounding by indication”
44
Exposure Assessment
• Referent group
• Women who reported no exposure to any antidepressant at any
time (56 days prior to LMP through the pregnancy)
• Exclusions
• Women whose only exposure was either 28-56 days prior to LMP
or after the 4th lunar month
45
Potential confounders
• Factors related to exposure and to outcome that explain
results
• Maternal age
• Maternal race/ethnicity
• Maternal education
• Smoking history
• Alcohol consumption
• Family history of birth defects
• BMI
Parity
History of seizures,
diabetes, hypertension
Infertility
Folic acid use
LMP year
Study center
46
Results
• 5,860 nonmalformed infants
• Rate of exposure to any SSRI in controls—2.8%
47
Results
• 3,724 infants with congenital heart defects (100
exposed)
• 186 looping, laterality defects (2 exposed)
• 620 conotruncal defects (13 exposed)
• 164 atrioventricular defects (0 exposed)
• 363 RVOTO (15 exposed)
• 482 LVOTO defects (15 exposed)
• 1161 septal defects (32 exposed)
• 17 anomalous pulmonary venous return (0 exposed)
All cardiac defects
Cases
Controls
OR
(95% CI)
Any SSRI
2.7%
2.7%
1.2
(0.9, 1.6)
Non SSRI
Antidepressant
0.6%
0.8%
0.8
(0.5, 1.5)
48
49
Results
• Create more homogeneous classes
• Specific cardiac defects
• Septal defects
• RVOTO defects
• Specific SSRIs
• Fluoxetine
• Sertraline
• Paroxetine
Septal defects
Cases
Controls
OR
95% CI
Fluoxetine
0.9%
1.1%
Sertraline
1.1%
0.8%
1.0
(0.5, 2.2)
2.0
(1.2, 4.0)
Paroxetine
0.5%
0.5%
Non-SSRI
Anti-depressants
0.9%
0.9%
0.8
(0.3, 2.2)
1.1
(0.6, 2.4)
50
RVOTO Defects
Cases
Controls
OR
(95% CI)
Fluoxetine
1.1%
1.0%
Sertraline
0.8%
0.8%
1.0
(0.2, 3.4)
2.0
(0.6, 6.8)
Paroxetine
1.7%
0.5%
Non-SSRI
Anti-depressants
0.6%
0.8%
3.3
(1.3, 8.8)
0.9
(0.2, 3.8)
51
52
Interpretation
• Possible explanations for findings
• Selection bias
• Information bias
• Confounding
• Chance
• “Real”
53
Interpretation
• Selection bias
• Not population-based, so must be considered
• Is entry into study related to exposure
• “medicalization”?
• These are serious defects, unlikely to escape detection
• Differential effects among SSRIs and no effect for non-SSRI
antidepressants
54
Interpretation
• Information bias
• Interviewers “blind” to hypothesis
• Recall bias
• Structured interview
• Antidepressants used on a regular basis
• Antidepressants used for non-trivial indications
• Null effect for non-SSRI antidepressants
55
Interpretation
• Confounding
• Many demographic, medical, reproductive, and
administrative variables “controlled” for in analysis
• “confounding by indication”
• No increased risk for non-SSRI antidepressants
• Uncontrolled confounding by factors not considered
is always possible
56
Interpretation
• Chance
• Can never be ruled out
• In this study, many comparisons were made, so
possibility of chance findings are increased
• Place greatest reliance on those findings that are
consistent with other studies
RVOTO Defects (NBDPS data)
Fluoxetine
Sertraline
Paroxetine
No. of
infants
No.
Adjusted
OR
(95%CI)
No.
Adjusted
OR
(95%CI)
No.
Adjusted
OR
(95%CI)
RVOTO
669
4
0.9
(0.4-2.1)
4
0.8
(0.3-2.3)
7
2.5
(1.0-6.0)
Controls
4092
29
32
18
Alwan S, Reefhuis J, Rasmussen SA, Olney RS, Friedman JM.
Use of selective serotonin-reuptake inhibitors in pregnancy and the risk of
birth defects. N Engl J Med 2007; 356:2684-92
57
58
Conclusions
• This study does not support overall effect with cardiac
defects, but does find evidence of associations for specific
SSRIs with specific cardiac defects
• Paroxetine and RVOTO
• Sertraline and septal
59
Conclusions
• Specific SSRIs may increase the risk of specific birth
defects
• Absolute risks still small
• Baseline prevalence of RVOTO: about 5.5 cases per 10,000
livebirths
• 4-fold increase in risk means that the absolute risk is only 0.2%.
60
Conclusions
• Depression in pregnancy is not a good thing for the
mother or her infant.
• As with any therapeutic decision, one must weigh the
risk of possible adverse outcomes with the benefits
of treating depressed women with any SSRI and with
specific SSRIs.
62
Source of Selected Prescription Drugs 5,435
Mothers; Boston, Philadelphia, Toronto, Iowa, 1976-1984
Source (in %)
Total
Users
Drug
Bendectin
Valium
Fiorinal
Hydrodiuril
Clomid
Compazine
Darvon
Seconal
Diuril
Physician’s
Prescription
Other
98
82
96
97
100
91
78
95
85
1
18
4
3
0
9
22
5
11
859
135
105
76
66
59
42
42
27
Am J Epidemiol 1986; 123:670-6.
63
64
Fetal Circulation
65
Circulation after birth
67