1. Example – anti-psychotic drugs and venous
thromboembolism
• Some research suggests antipsychotic drugs
might be associated with an increased risk of
venous thromboembolism
– Drugs also widely prescribed for nausea, vomiting,
and vertigo
– Venous thromboembolism
• Deep vein thrombosis
• Pulmonary embolism
2. Example – anti-psychotic drugs and VT
• Nested case-control study examining whether
antipsychotic drugs are associated with an
increased rate of venous thromboembolism
– Examined by type of antipsychotic, potency, and dose
– (BMJ 2010;341:c4245)
3. Example – anti-psychotic drugs and VT
• Cohort was an open cohort of people registered
with UK general practices - QResearch
– Data have been collected over 16 years
– Includes the anonymised records of over 7 million
patients who have been registered with over 500
practices spread throughout the UK
– UK has national insurance
4. Example – anti-psychotic drugs and VT
• Nested case-control included time period 1996-
2007
• Cases
– All patients aged 16-100 yrs with first ever record of
venous thromboembolism
– Identified based on diagnostic codes in cohort
database
5. Example – anti-psychotic drugs and VT
• Controls
– Used risk-set sampling to identify controls - matched
on calendar time of the incident case
– Additionally matched on age, sex and practice
– Defined as members of the practices with no
diagnosis of venous thromboembolism up to the date
of the incident case
– Selected up to 4 controls per case (will discuss in
matching)
6. Example – anti-psychotic drugs and VT
• Exposure to drugs assessed with prescriptions
on or before the index date
– For anti-psychotics collected information on drug
name, formulation, dose, instructions and date
– For other drugs of concern as confounders collected
only whether they were prescribed in the 24 months
before
7. Example – anti-psychotic drugs and VT
• Not eligible for analysis if part of the database for
less than 24 months
– Due to inadequate exposure data
• 25,532 venous thromboembolism cases
• 89,491 controls
11. Example – anti-psychotic drugs and VT
• Notes on the methods in this example
– Primary study base – population defined as those
attending a set of general practices, all cases within
that population identified
– Risk-set sampling
• What does OR estimate?
12. Example – anti-psychotic drugs and VT
• Notes on the methods in this example
– Case-control design made collection of data on full
prescription details for all antipsychotics (drug name,
formulation, dose instructions, dates) a manageable
task
• Allowed detailed examination of anti-psychotic drugs (types,
doses, timing)
• Allowed comparison of risk between new users vs longer
term users
13. Example – anti-psychotic drugs and VT
• Notes on the methods in this example
– Access to the prescription data due to the larger
cohort also critical to success of the exposure
assessment
– Extensive control for confounding, however…
– For most participants could not identify the reason for
the prescription
• Underlying condition and not the drug prescribed to treat it
might be the cause of VT (confounding by indication)
• Analyses removing those with schizophrenia and bipolar
disorder did not change the results
14. Example – anti-psychotic drugs and VT
• Authors suggested a case-crossover design to
strengthen the evidence on anti-psychotic drugs
and VT
– What do you think of this suggestion?
15. Case-control studies outline
• Strengths and challenges
• Example: anti-psychotic drugs and venous thromboembolism –
nested case-control
• Example: diarrhea outbreak in India – cumulative case-control xx
• Example: hip fracture – comparison of hospital and community
controls
• Summary
16. Example – diarrhea outbreak in India
• On 26 September 2007, the Gayeshpur
municipality reported cluster of diarrhea cases
• Investigation conducted to identify the agent as
well as the source of infection
• Cumulative case-control design
• Saha et al. 2009 (Natl Med J India. 2009 Sep-Oct;22(5):237-9.)
17. Example – diarrhea outbreak in India
• Cases
– Reviewed data on diarrheal diseases available from
the municipal health office
– Defined a possible case-patient as a resident of the
Gayeshpur municipality who had diarrhea (>3 loose
stools during a 24-hour period) between September
and October 2007
– Asked healthcare workers in all facilities to report
similar new cases
– Cases for the case-control analyses met the above
definition and resided in one of 4 wards (ward
numbers 2, 5, 6 and 8)
18. Example – diarrhea outbreak in India
• Controls
– Were matched on age-, sex- and neighborhood to
each case
• Exposure assessment
– Municipal health workers collected information among
participants regarding demographic characteristics,
date of onset, signs and symptoms, outcome, food-
handling practices, water intake and sanitation
23. Example – diarrhea outbreak in India
• Notes on the methods in this example
– Primary study base – population defined as those
residing in one of 4 wards in Gayeshpur municipality,
all cases within that population identified
– Case-control design facilitated a rapid assessment of
differences in exposures among those with diarrheal
disease and those without
– Self-report of exposures so accuracy of recall may be
different between cases and controls
24. Example – diarrhea outbreak in India
• Notes on the methods in this example
– Confounding handled by matching on age, sex and
neighborhood
• To be discussed in matching module
– Concerns about other confounders?
– Problematic that OR in study estimates OR in
population and not CIR or IDR?
25. Example – hip fracture
• Case-control study of risk factors for hip fracture
– Comparison of results using hospital controls vs
community controls
– N Engl J Med. 1991 May 9;324(19):1326-31
26. Example – hip fracture
• Cases
– White and black women aged 45 years or more with
radiologically confirmed diagnosis of first hip fracture
– Selected from 30 participating hospitals in NYC and
Philadelphia 1987-1989
– Primary or secondary study base?
27. Example – hip fracture
• Hospital controls
– Hospitalized women from a surgical ward or an
orthopedic ward with no previous hip fracture or hip
replacement
– Frequency matched to cases by age, hospital and
race
– Admissions diagnoses included cardiovascular
disease or peripheral vascular disease, digestive
disorders, cancers, osteoarthritis and other
musculoskeletal disorders, infections
– Why might you use hospital controls?
28. Example – hip fracture
• Community controls
– For cases aged 65 or older, community controls were
randomly selected from the Health Care Financing
Administration lists of Medicare recipients – frequency
matched to cases on age, race and zip code
– For cases under 65 years of age, community controls
selected by random digit dialing and matched by age,
race and telephone prefix
– Why might you use community controls?
29. Example – hip fracture
• Exposure assessment
– In-person interviews to assess lower-extremity
function, vision, medical and surgical history, use of
medications before hospitalization, dietary and
reproductive history, height, weight, smoking, alcohol
consumption, symptoms related to balance and gait,
and sociodemographic information
– Cases and hospital controls were generally
interviewed in the hospital
– Community controls were generally interviewed in
their homes
30.
31. Example – hip fracture
• Associations observed were different depending
on the control group selected
– Hospital controls likely less healthy than study base
– Community controls perhaps healthier than study
base
• Comparison of community controls with
representative sample of elderly in urban
northeastern areas
– Very similar
• Conclude community controls more appropriate
32. Example – hip fracture
• Notes on the methods in this example
– Secondary study base – highlights challenge of
finding controls to represent secondary study base
– Case-control design provided a less resource
intensive approach to identifying risk factors, but
challenges with control group identification probably
mean a trade off with bias
– Self-report of exposures so accuracy of recall may be
different between cases and controls
• Probably less of an issue for the hospital controls than for the
community controls
33. Example – hip fracture
• Notes on the methods in this example
– What does OR estimate?
• Cumulative case-control so OR estimates OR in
population
– What could have been changed about the design to
have the OR estimate a different MoA?
• Matching controls on time of case occurrence
could have produced a density-sampling approach
in which OR = IDR
• However, the control groups already bringing in bias, so
perhaps perceived not to be worth the additional effort
34. Example – hip fracture
• Notes on the methods in this example
– With the cumulative case-control design, when does
OR approximate IDR?
• Rare disease