Post on 27-Mar-2015
Choosing a Reference Group
Adriana PistolEPIET introductory Course, Lazareto
September 2011
Objectives of the presentation
• Define source population
• Understand importance of representativeness
• Understand the importance of a reference group in analytical studies
• Understand the advantages and disadvantages of different types of controls
• Incidence in exposed and unexposed (cohort studies)
• Exposure in cases and controls
• Incidence over time (surveillance)
Making comparisons
fundamental to epidemiology
Purpose of the comparison group (concept of counterfactuality)
Cohort study:
“The comparison group serves to provide an
estimate of the expected disease incidence in
the exposed group if the exposure had been
absent.”
Case control study:
“The comparison group serves to provide an
estimate of the exposure distribution in the
source population from which the cases
originate.”Rothman KJ, 1986
• Can be difficult to select reference
group
• Especially in case control studies
• Constraints of time and resource
Field epidemiology
Case control study
Outbreak• 24 cases of Salmonella Typhimurium • Cornwall (population 500,000)• onset through May 2004• age range 16 – 56 years• 9 male, 6 female • no recent travel abroad
Who are the right controls?
Controls
Controls should be representative of population from which cases arise (source population)
Control characteristics (1)
If controls represent source population:
• be representative of exposures in source population(the same chance that they were exposed as cases)
• be identified as cases if they had disease under study
• have same exclusion and restriction criteria as cases
Control characteristics (2)
• Controls should ideally be similar to cases with respect to:– Demographic characteristics – Cultural background– Socio-economic group – Employment
• Failure to recruit cases and controls from the same population lead to a bias
CasesExposed
Unexposed
Source population
Controls:• Sample of the denominator• Representative with regard to exposure
Controls
Sample
Considerations when choosing controls
• Validity – Selection bias – Information bias– Confounding
• Efficiency (most information for least cost) – Resources– Time scale – Sample size (precision)
Who is source population?
Start with your case definition
Case definition
Resident of Cornwall aged above 15 years with isolate of Salmonella Typhimurium in faecal sample during May 2004
Exclusion: Travel abroad in week before illness
What is source population?
Source population
Residents of Cornwall aged above 15 yearsduring May 2004 who have not recently travelled abroad
Controls should then be representative of this population
How to select controls?
• Aim for random sample of source
population
• Not always feasible
Selecting controls (examples)
1.Population– random from register/list/directory – stratified (age/sex/general practice)
2.Friends 3.Family4.Neighbourhood 5.Hospital
1) Population controls
• Is there a list or register of source population?
• Such a list should
– be complete
– contain all cases
– be readily accessible
– identify specified characteristics e.g. age
Take random sample
1) Population controlse.g. random digit dialling
• using residential directories or mobile numbers
• quick and easy • but may be bias in selection
– telephone ownership– availability– geographical area– participation
1) Population controlse.g. matching
• “Matching” = stratification in study designe.g. same age, same sex, same doctor
• Matching useful if – do not have full list of source population – only specific group affected – effect of matching variables is of no
interest
Advantages
• good matching for social factors
• can be quick and efficient
• validity in food poisoning investigations
2) Friend controls
2) Friend controls
Disadvantages
• Co-operation may be limited (concern about giving out names)
• if exposure same as in cases, may not detect causal association =
Overmatching
3) Family controls
• are rarely used in field epidemiology as exposures in family controls
• are often so similar to those of the cases that the association of interest may not be shown at all.
4) Neighbourhood controls
Advantages• no need for population register• similar socio- economic status
Disadvantages
• low co-operation
• may be time consuming and expensive
• might be too similar to cases
5) Hospital controls
Advantages • useful if all cases identified from hospital
register• easily identified • cost and time efficient
Disadvantages• different catchments for different diseases • overmatching on exposures for other
diseases
Example of a set of controls belonging to the same study base• Hospital-based study examining risk
factors for cholera during an outbreak• Controls:
– Patients admitted with meningitis into the same hospital
• The two diseases are of equivalent severity
• The population bases can be expected to be identical
Example of a set of controls not belonging to the same study base
• Hospital-based study examining risk factors for cholera during an outbreak
• Controls: – Patients admitted with minor complaints at
the outpatient clinic in the same hospital• The two diseases are not of equivalent severity• The population bases can be expected to be
different– Catchment areas will be larger for more
severe diseases
Sample size
• Often limited by number of cases available
• Unusual to select more than 2-3 controls/case– Little extra power beyond this number
Controls may not be easy to find
Source population
Residents of Cornwall aged above 15 yearsduring May 2004 who have not recently travelled abroad
Some common questions
A. Non-cases as controls
B. Asymptomatic cases
C. Immune populations
D. 100% exposure
A. Non-cases as controls
If attack rate high • high risk that non-cases do not represent
exposures in source population
If attack rate low • low risk that non-cases do not represent
exposures in source population • can use as controls
A. Non-cases as controls
High attack rate
Cases
Non- cases
start end
Sourcepopn
Low attack rate
B. Asymptomatic cases
• Does it matter if we fail to identify mild cases?
• Analogous to non-response• Example: 40 cases, 40 controls
Cases Controls
Exposed 20 10
Not exposed 20 30
OR = 20.30/20.10 = 3.0
B. Asymptomatic cases
• If we only identify half the cases and % exposure is the same
Cases Controls
Exposed 10 10
Not exposed 10 30
OR = 30.10/10.10= 3.0
No bias
C. Immune subjects
• Not eligible as cases • So not in source population • Difficult to identify • May have been cases in past • May have similar level of exposure to risk
factor as current cases in study • Bias in OR towards 1 (null value)
D. 100% exposure
• What if close to 100% of population exposed? – e.g. foodborne disease outbreaks where little
choice in menu
• Try to measure dose response • Reference group lowest level of exposure• What if 99% of population are cases?
– Try to conduct a cohort study!
Characteristics of good controls
Come from the same population as cases
May be exposed like casesCan develop the disease Could be recruited as cases if diseasedHave exposure windows identical to
casesAre adapted to the study objectives
Dealing with imperfect control groups
• Examine the limitations of your control group with respect to each criteria
• Assess in which way the limitation will affect the odds ratio– Towards one– Away from one
• Interpret your results in light of this review of limitations
• Recruiting two control groups is an option
Be prepared to defend your choice…
…and
do the study!
References
• Rothmann KJ, Greenland S. Modern epidemiology. Lippincott-Raven 1998.
• Hennekens CH, Epidemiology in Medicine. Lippincott-Williams and Wilkins 1987.
• Gregg MB. Field epidemiology. Oxford University Press 1996.
• Wacholder S, McLaughlin JK, Silverman DT, Mandel JS. Selection of controls in case control studies I-III. Am J Epidemiol 1992; 135: 1019-50.