Case-Control studies
COHORT DESIGN LIMITATIONS
• rare outcomes
• long latency periods
Szklo 2014
Motulsky 2018
CASE-CONTROL DESIGN
• Selection of subjects according to outcome
Cummings, Newman, Hulley. Designing Clinical Research. 2007
Rev. Henry Whitehead
Borgan et al. 2017
CASE-CONTROL STUDIES
STRENGHTS LIMITATIONS
• Less expensive and time • Exposure measurements
consuming than prospective taken after disease
cohorts occurrence
• Efficient for studying rare • Disease status can influence
diseases selection of subjects
• Can't estimate incidence
Target Population Identification
SUBJECT SELECTION
CASE SELECTION
1. Ideal SOURCES:
– Disease Registries with good population coverage
– group of regional hospitals (not just 1
hospital/center)
– include all subjects or a random sample
2. Ideal SELECTION: independent of exposure
status
AVAILABLE CASES & EXPOSURE REPRESENTATIVENESS
CONTROL SELECTION
CASES CONTROLS
EXPOSED EXPOSED EXPOSED
CASES CONTROLS
UNEXPOSED UNEXPOSED
UNEXPOSED CASES CONTROLS
• Include few cases available • What exposure distribution would I want
• Exposure distribution given, but... in my Control sample?
• Risk of selection bias (by not • That of the General Population
including undiagnosed/
misdiagnosed/ dead cases)
CONTROL SELECTION
1. SOURCES:
– Ideal: listed reference population (census)
– Alternatives: Neighbourhood, social networks, hospital,
friends, brothers
2. If using 1 hospital for case selection, the reference
population for control selection includes people that
-had they developed the disease- would have used
the hospital
3. Beware of hospital controls with other diseases (and
their exposure distribution)
• Controls shouldn't resemble cases
• They should resemble the population that originated the
cases
STUDY 1 (1981) STUDY 2 (1986)
Cancer Other GI dis. Cancer Other (not GI)
(n=367) (n=643) (n=172) (n=267)
Coffee 347 555 Coffee 143 204
No Coffee 20 88 No Coffee 29 63
OR = Odds in Exposed = 2,7 OR = Odds in Exposed = 1,5
Odds in Unexposed Odds in Unexposed
Controls without coffe: 14% (88/643) Controls without coffe : 24%
MacMahon 1981
Hsieh 1986
Odds Ratio and RARE DISEASE ASSUMPTION
• OR resembling RR with prevalences less than
<20% in both groups
OR
RR
INFORMATION BIAS
• + freq: Recall bias
– CASES remember more -> artificially more
exposure
• Try using data gathered before outcome
occurrence
• Blinding interviewer from outcome status
knowledge
REFERENCES
• Borgan O, Brelow N, Chatterjee N, Gail MH, Scott A, Wild CJ. Handbook of Statistical
Methods for Case-Control Studies. CRC Press, 2017
• Breslow N. Design and Analysis of Case-control Studies. Ann Rep. Public Health. 1982.
3:29-54
• Cook TD. Advanced Statistics: Up with Odds Ratios! A Case for Odds Ratios When
Outcomes Are Common. Acad Emerg Med 2002; 9:1430-1434
• Hulley, Cummings. Designing clinical research. 3° ed.
• Mansournia MA, Jewell NP, Greenland S. Case-control matching: effects, misconceptions,
and recommendations. Eur J Epidemiol. 2018 Jan;33(1):5-14
• Mansournia MA, Hernán MA, Greenland S. Matched designs and causal diagrams. Int J
Epidemiol. 2013 Jun;42(3):860-9
• Norman & Streiner. Epidemiology PDQ. 3ra ed. 2003
• Rothman K. Invited Commentary: When Case-Control Studies Came of Age. Am J
Epidemiol. 2017;185(11):1012–1014
• Rothman K, Lash T. Epidemiological study design with validity and efficiency
considerations. En: Modern Epidemiology, 4th ed. 2021
• Sackett DL, Deeks JJ, Altman DG. Down with odds ratios! Evidence Based Med 1996;1:164-
6.
• Szklo. Epidemiology beyond the basics. 3° ed.
• Vandenbroucke JP, Pearce N. Case-control studies: basic concepts. Int J Epidemiol. 2012
Oct;41(5):1480-9
