Case-Crossover Study Design

Overview

This page briefly describes case-crossover designs as an approach to investigating acute triggers that are potentially causing disease. An annotated resource list is provided.

Description

Triggers

A “trigger” can be thought of as the final step in leading from pathophysiology to disease, or the final component cause leading a susceptible person to experience a disease. Triggers thus may be important for our understanding of etiology. In addition, a new understanding of disease triggers can help us to prevent disease through trigger reduction, reduction of baseline risk, or a targeted intervention to reduce risk at a time when disease is more likely to occur.

For a potential hypothesis about a trigger to be tested using a case-crossover design, we would look for the following defining characteristics:

• short-term changes in exposure

• transient changes in disease risk

• acute-onset disease

Related study designs

Study designs used to examine exposure outcome association include cohort and case-control studies. Whereas cohort studies can be limited in power for rare disease outcomes, and case-control studies can be biased due to retrospective exposure assessment, case-crossover designs compare individuals to themselves at different times. This parallels the randomized crossover trial approach that compares individuals to themselves as they are going on and off treatment.

Something that the case-crossover design has in common with a case-control approach is the need to find representative controls. However, while case-control designs select control individuals, case-crossover designs select control time windows. This brings our focus to the plausible temporal relationship between exposure to a trigger and disease onset.

Other time-focused designs include ecological time-series data, and interrupted time-series data. Disease counts over time can be modeled across time in a generalized linear modeling framework, often using Poisson regression. For Poisson models the beta coefficient provides information about rate comparisons on a relative scale because they use a log link.

Setting up a case-crossover analysis

A key decision point in setting up a case-crossover study is to decide for what length of time before disease onset would exposure be compatible with triggering. That is your “case” window. For example, if you think physical activity could trigger a myocardial infarction in the subsequent 2 hours, you could identify a case window starting 2 hours before symptom onset, and ending at the time of symptom onset. Sensitivity analyses might be planned altering the length of this time window.

Selection of one or more control windows is then designed to identify whether exposure during the case window was atypical. By comparing exposures over time within the same person, you automatically condition on all stable characteristics of the individual. You might also want to match on potential time-varying confounders such as time of day. Usually, case and control windows are the same length. It may be efficient to select multiple control time windows for every case time window. Control windows should reflect the exposure distribution while at risk for the outcome, should be close enough in time that the baseline risk is similar, and should be far apart enough in time so that exposures are uncorrelated.

Once you have constructed your case and control time windows, compare the probability of exposure during case and control periods. This is usually done using conditional logistic regression (similar to a matched case-control study).

An opportunity to consider when working with a case-crossover design is that although you do not typically need to control for many individual characteristics, you can evaluate effect modification by individual characteristics. For example, while looking at physical activity and myocardial infarction, you might hypothesize that triggering would be most likely to occur for individuals with hypertension.

Readings

Methodological Articles

Maclure M. The case-crossover design: a method for studying transient effects on the risk of acute events. Am J Epidemiol 1991; 133(2):144-53.

Maclure M, Mittleman MA. Should we use a case-crossover design? Annu Rev Public Health 2000;21:193-221.

Lu Y, Zeger SL. On the equivalence of case-crossover and time series methods in environmental epidemiology. Biostatistics 2007;8(2):337-344.

Lu Y, Symons JM, Geyh AS, Zeger SL. An approach to checking case-crossover analyses based on equivalence with time-series methods. Epidemiology 2008; 19(2):169-75

Maclure M, Mittleman MA. Case-crossover designs compared with dynamic follow-up designs. Epidemiology 2008; 19(2):176-8.

Janes H, Sheppard L, Lumley T. Case-crossover analyses of air pollution exposure data: referent selection strategies and their implications for bias. Epidemiology 2005; 16(6)717-26.

Application Articles

Basu R, Dominici F, Samet JM. Temperature and Mortality Among the Elderly in the United States: A Comparison of Epidemiologic Methods. Epidemiology 2005;16(1):58-66

Hebert C, Delaney JA, Hemmelgarn B, Levesque LE, Suissa S (2007) Benzodiazepines and elderly drivers: a comparison of pharmacoepidemiological study designs. Pharmacoepidemiol Drug Saf 16: 845–849