The document discusses the case-crossover design, a method for studying intermittent exposures with immediate effects on abrupt outcomes by comparing cases to their own control periods. This design, highlighted in a 1997 analysis of car phone usage and collisions, is beneficial for capturing trends in time and involves using data retrospectively. An example study is presented, examining the link between sleep duration and unintentional childhood injuries, where control and case periods are matched for analysis.

1. Designs
• Case-crossover design
– When investigating an intermittent exposure with
immediate and transient effect on risk related to an
abrupt outcome
– All cases are incident occurrences of the disease of
interest
– Cases at another time (or other times) serve as their
own controls

2. Designs
• Case-crossover design
– Developed in 1988 due to concerns about ability to
find controls representing the study base for a study
of causes of MI peaks in the morning
– Realized could compare each patient’s experience on
their MI day with their experience the day before (or
multiple control days before)

3. Designs
• Case-crossover design
– Analogous to the experimental cross-over design
except that the data are:
• Non-experimental
• Observed retrospectively

4. Designs
• Case-crossover design
– Publicized in 1997 with media coverage of an analysis
showing that car phone calls were associated with
increased risk of collision

5. Designs

6. Designs
• Case-crossover design
– For each case, can sample:
• One control time
• Multiple control times
• Sample of the person-time over a defined period

7. Designs
• Case-crossover design
– Trends in time (e.g., seasonality) are main challenge
specific to this design

8. Case crossover example• Valent et al. A case-crossover study of sleep and childhood injury. Ped 2001;107:
E23
• Research question
• Does lack of sleep result in a greater number of unintentional childhood
injuries?
• Exposure = sleep <10 hours (EXP +) vs. >= 10 hours (EXP -)
• Outcome = Childhood unintentional injury
• Study population = 292 children with unintentional injuries
• Sleep time data collected for two periods (0-24 hours before injury) and (25-48
hours before injury)
• “Case period” = 24 hour period before injury; “Control period” = 25-48 hour period
before injury
• Note that each child has both one case and one control period...these two
periods together form a pair
• Analysis is done through a matched analysis in which the time periods (case and
control periods) are matched
• Four possible scenarios
• a. Case period = (EXP +) and Control period = (EXP +)...”concordant”
• b. Case period = (EXP +) and Control period = (EXP -)...”discordant”
• c. Case period = (EXP -) and Control period = (EXP +)...”discordant”
• d. Case period = (EXP -) and Control period = (EXP -)...”concordant”
• Concordant pairs offer no “information” about the effect of sleep < 10 hours (EXP +)
vs. sleep > 10 hours (EXP -)