24 POPULATIONS AND SAMPLES
her further. These procedures will lessen the number of usable cases and may make
the sample less representative of patients with this disease. Most investigators try to
use the most recent patients, as their records may be better and they are easier to find.
Usually, a 100% sample of the most recent cases that are available to study are taken.
Hence, the sample of cases is a chunk sample taken in a given place and time.
The best sample of controls is more difficult to define. They will be used to contrast
their exposure history with that of the cases. The controls should be taken from the
same population that gave rise to the cases and the sample chosen in such a way
that the chance of being sampled is not dependent on the level of the major exposure
variables. They should be similar to the cases with regard to past potential exposure
during the critical risk period. Three types of controls have been commonly used and
each has its advocates: hospital or clinic controls, friend controls, and neighborhood
controls.
Typically, hospital controls are taken from the same hospital as the case, admitted
at a similar time, are of the same gender and close in age, but have a different disease
condition. One method of doing this is to individually find one or more matches for
each case. Here, you are trying to find a pseudo-twin. The investigator has to be
careful in what they choose to match, as it is not possible to analyze any factor that is
used for matching purposes since the cases and controls have been made artificially
equal in respect to that factor.
The matching controls from hospitals are sometimes found by sorting a list of
potential controls on gender, age, and time of admission (or whatever matching vari-
ables are used) with the computer, and then finding matches from this sorted list. This
procedure is sometimes called caliper matching because the investigator is trying to
find a control that is close to the case on the matching variables. Alternatively, if the
variables being matched fall into two groups, such as gender and age, and are grouped
into three age intervals, say, <30 years old, 3 1-50 years old, and >50 years old, then
what is called frequency matching is commonly done. The symbol “>” signifies
“greater than.” Here, the investigator makes sure that there are an equal number of
cases and controls in each gender and age category, or perhaps two controls for each
case.
Friend controls are chosen by asking the case (or a close relative if the case is
dead) for the name and address of a friend or friends who can be contacted and used
as a control. When obtaining friend controls is successful, it can be an inexpensive
way of getting a matched control; it is thought to result in a control who has a lifestyle
similar to that of the case. The three problems that have been noted with this method
are that (1) sometimes cases lack friends or are unwilling to name them; (2) cases
tend to name friends who are richer or have a better job or are in some way more
prestigious than themselves; and (3) if the critical exposure time was in the distant
past, then cases may have trouble recalling friends from that time period.
Sometimes neighborhood controls are used either in and of themselves or to sup-
plement unavailable friend controls. Neighborhood controls are thought of as suitable
matches on general lifestyle. These controls are found in the neighborhood of the
case by use of reverse directories, maps, or neighborhood searches. The general rule
for matched controls is to exclude the homes immediately adjacent to the case for