Should We Use a Case-Crossover Design?

Abstract

The first decade of experience with case-crossover studies has shown that the design applies best if the exposure is intermittent, the effect on risk is immediate and transient, and the outcome is abrupt. However, this design has been used to study single changes in exposure level, gradual effects on risk, and outcomes with insidious onsets. To estimate relative risk, the exposure frequency during a window just before outcome onset is compared with exposure frequencies during control times rather than in control persons. One or more control times are supplied by each of the cases themselves, to control for confounding by constant characteristics and self-confounding between the trigger's acute and chronic effects. This review of published case-crossover studies is designed to help the reader prepare a better research proposal by understanding triggers and deterrents, target person times, alternative study bases, crossover cohorts, induction times, effect and hazard periods, exposure windows, the exposure opportunity fallacy, a general likelihood formula, and control crossover analysis. INTRODUCTIONA case-crossover study (18, 26) is a scientific method to answer the question, "Was this event triggered by something unusual that happened just before?" The simple part of the question is "What happened just before?" The challenging part is to quantify "How unusual was that?" The latter question must be answered for all people in the study who experience the outcome event (disease onset, injury, or behavioral change), not just the people who had the hypothesized triggers just before.We started developing the design in 1988 to avoid control selection bias in the Myocardial Infarction (MI) Onset Study (29). The original study aim was to use a case-control approach to investigate why the incidence of MI peaks in the morning. Who should be the controls? We were concerned that noncases drawn from the general population would be a biased sample not only because of healthyvolunteer bias, but also healthy-day bias. These subjects would be more likely to decline to be interviewed on stressful days. We also ruled out patients hospitalized for other emergencies, because they would be biased by whatever triggered their Next we realized that we could make the opposite comparison by also asking morning cases about their afternoon exposures on the previous day. Finally, it dawned on us that we could compare each patient's experience on their MI day with their experience on the day before. This would give us a perfectly matching control for each patient-the same patient at the same time on the day before.Eventually we saw how we could have multiple control days for each patient by asking about exposure frequency during the entire week, month, or year before. When we applied epidemiologic theory to this approach, we realized it was like a nonexperimental crossover experiment observed retrospectively. We are often asked whether a case-crossover design would be appropriate to study a given hypothesis. This review is intended to...

Emotional triggering and low socio-economic status as determinants of depression following acute coronary syndrome

“…Triggering by emotional stress and physical exertion was analysed using the case-crossover approach as detailed by Maclure & Mittleman (2000). Onset of ACS was defined as the time at which the symptoms that led to admission began.…”

Adolescent alcohol involvement and suicide attempts: Toward the development of a conceptual framework

“…Proximal effects-Short-term, event-based, longitudinal designs, such as case-crossover designs (Maclure & Mittleman, 2000) and daily diary studies (including ecological momentary assessment; EMA; Stone & Shiffman, 1994), may be useful for studying directionality and proximal mechanisms underlying the proximal alcohol use-attempt relation. Case-crossover studies assess individuals at the time of a target event (e.g., presenting to the ER after a suicide attempt; e.g., Borges et al, 2004) and also at a within-subject control event (e.g., a week or month earlier).…”

Healthcare Utilization and Statin Re‐Initiation Among Medicare Beneficiaries With a History of Myocardial Infarction

“…Because exposure information for each beneficiary at the time of statin re‐initiation is compared with his/her own prior exposure experience, each person is their own control. This self‐matching eliminates confounding by risk factors that do not change over the study period within individuals but differ between individuals (eg, sex, obesity, socioeconomic status) 17. In the primary analysis, the case and control periods included the 14 days and 30 to 44 days before statin re‐initiation, respectively.…”

“…Conditional logistic regression was used to calculate the odds ratio (OR) and 95% confidence interval (CI) for statin re‐initiation associated with the occurrence of each exposure, including lipid panel testing, outpatient primary care visits, outpatient cardiologist care visits, emergency department visits, non‐CHD hospitalization, and CHD hospitalization, during the 14 days (case period) compared with 30 to 44 days (control period) before statin re‐initiation. This is the standard approach for matched case–control studies, except instead of case participants matched to control participants, the case and control periods were paired within the same Medicare beneficiary 17. ORs were calculated for the overall population and within subgroups defined by history of statin use before the index MI hospitalization, race/ethnicity, age (<75 and ≥75 years old), sex, and statin intensity discontinued.…”