A researcher studies whether cell phone use immediately before a car crash increases crash risk. For each crash victim, they compare phone use in the 10 minutes before the crash to phone use during the same 10-minute window the previous day. What type of confounding does this design automatically eliminate?
ATime-varying confounders only, such as traffic density that changes by the hour
BAll forms of confounding, including secular trends in cell phone adoption
CTime-invariant personal characteristics like driving skill, risk tolerance, and visual acuity
DConfounders that affect only the control period, not the hazard window
Because the same person serves as their own control, all stable personal characteristics — driving experience, personality, vehicle type, usual routes — are held constant by design. They cannot differ between the hazard window and the reference window for the same person. However, time-trend bias (e.g., cell phone use rising steadily over time) is NOT automatically eliminated; it requires careful choice of reference periods.
Question 2 Multiple Choice
A researcher wants to study whether long-term low-level air pollution exposure increases the risk of developing type 2 diabetes, which accumulates over years. Is a case-crossover design appropriate?
AYes — air pollution is an environmental exposure, which is exactly what case-crossover designs study
BYes — as long as the hazard window is extended to cover the relevant exposure period
CNo — the exposure is not transient (it is chronic), and the outcome has a long latency, violating both design requirements
DNo — case-crossover designs can only be used for cardiovascular events, not metabolic diseases
Case-crossover design requires both a transient (intermittent) exposure and an acute outcome with a short, identifiable causal window. Chronic exposure like long-term air pollution cannot be captured by comparing a brief hazard window to a brief reference period — the exposure is essentially the same in both windows. Outcomes with long latency (years) also lack a meaningful 'just before the outcome' comparison point. The design is suited to acute triggers of acute events, not cumulative exposures and slowly developing diseases.
Question 3 True / False
In a case-crossover design, the comparison group consists of a separate set of healthy individuals matched to the cases on demographic characteristics.
TTrue
FFalse
Answer: False
This describes a conventional case-control design. In a case-crossover design, there is no separate comparison group — each case is compared to themselves at a different time. The hazard window (just before the event) is compared to one or more reference windows (the same person at other times when no event occurred). This within-person design is precisely what eliminates time-invariant confounders.
Question 4 True / False
A case-crossover study comparing exposures on the day of a heart attack to exposures one week earlier is susceptible to time-trend bias if the prevalence of the exposure is changing over calendar time.
TTrue
FFalse
Answer: True
If exposure prevalence is trending upward (or downward) over time, then comparing 'today' (hazard window) to 'last week' (reference window) will systematically differ in exposure rate for reasons unrelated to the event itself — the secular trend creates an artificial difference. Symmetric reference periods (comparing times equidistant before and after the event) or multiple reference windows from prior weeks are strategies used to mitigate this bias.
Question 5 Short Answer
Why does the case-crossover design automatically control for time-invariant confounders, and what is the essential condition the exposure must meet for this design to be valid?
Think about your answer, then reveal below.
Model answer: Time-invariant confounders (stable characteristics like genetics, personality, or occupational history) are held constant because the same person appears in both the hazard and control windows — they cannot differ between windows for the same individual. For validity, the exposure must be transient (it turns on and off), so there is meaningful variation between the hazard period and a reference period. Chronic or stable exposures cannot be studied this way because there is nothing to compare — exposure would be the same in both windows.
This is the design's defining strength and its defining limitation simultaneously. The within-person comparison is powerful precisely because it never needs to measure stable confounders — but it only works when the exposure is the kind of thing that varies day-to-day or hour-to-hour. Applications like air pollution spikes and asthma attacks, physical exertion and cardiac events, or drug initiation and adverse events all share this transient structure.