09.08.2013 Views

Fundamentals of epidemiology - an evolving text - Are you looking ...

Fundamentals of epidemiology - an evolving text - Are you looking ...

Fundamentals of epidemiology - an evolving text - Are you looking ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Validity<br />

The validity <strong>of</strong> a case-control study requires that:<br />

Cases in the study adequately represent the relev<strong>an</strong>t cases (the population <strong>of</strong> cases about whom<br />

inferences are to be made) with respect to the variables <strong>of</strong> interest (notably, prevalence <strong>of</strong><br />

exposure). This depends upon whether the cases available do in fact reflect the rates <strong>of</strong> disease<br />

in exposed <strong>an</strong>d unexposed individuals undistorted by differential m<strong>an</strong>ifestation, detection, or<br />

short-term survival (e.g., selective survival, access to care, detection bias);<br />

Controls accurately reflect the exposure proportions in the study base (the source population for<br />

the cases). For example, hospitalized controls may overrepresent exposures associated with<br />

hospitalization for other conditions.<br />

Both <strong>of</strong> these requirements, especially the latter, c<strong>an</strong> be difficult to ensure. Therefore, case-control<br />

studies are regarded as highly susceptible to bias from problems with the:<br />

Identification <strong>of</strong> cases<br />

Reli<strong>an</strong>ce on medical care system<br />

Often miss subclinical cases (detection bias?)<br />

C<strong>an</strong> miss rapidly fatal cases (selectively?)<br />

Selection <strong>of</strong> controls<br />

Selection <strong>of</strong> controls c<strong>an</strong> determine the study results<br />

Which controls are appropriate is <strong>of</strong>ten not obvious<br />

Trade-<strong>of</strong>f between sampling <strong>an</strong>d data collection<br />

Hospitalized controls, community controls, dead controls<br />

Controls may be reluct<strong>an</strong>t to cooperate<br />

Measurement <strong>of</strong> exposure for cases <strong>an</strong>d controls<br />

Reli<strong>an</strong>ce on recall or records (differential?)<br />

Effect <strong>of</strong> disease on exposure assessment<br />

Effect <strong>of</strong> disease on exposure (confounding by indication)<br />

There is also the thorny problem <strong>of</strong> establishing temporality, i.e., did the exposure precede the<br />

disease?<br />

Interpretability <strong>of</strong> the odds ratio<br />

Why does the OR from the cases <strong>an</strong>d controls we have assembled estimate <strong>an</strong>ything in the<br />

population? Consider what the cells in the table below represent. Assume that the cases were<br />

selected as newly occurring cases <strong>of</strong> endometrial c<strong>an</strong>cer over a period <strong>of</strong> time in a defined<br />

_____________________________________________________________________________________________<br />

www.sph.unc.edu/courses/EPID168, © Victor J. Schoenbach 1999 8. Analytic study designs - 235<br />

rev. 9/6/1999, 10/7/1999, 12/17/1999

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!