Interrupted Time Series Design with Comparison Group
The ITS-CG method, like the related simple time series method, has multiple points of observation before and after an intervention or treatment. In addition, ITS-CG includes a non-equivalent (not randomly assigned) comparison group that did not undergo the intervention or treatment. The comparison group is chosen to be as similar as possible to the experimental group. Also, the repeated measurements should be equally spaced in time.
One of the best known historical examples of a quasi-experimental design is that of James Lind's experiments to cure scurvy among sailors on HM Bark Salisbury. The experiment, conducted in 1747, consisted of assigning sailors afflicted with scurvy non-randomly into one of six treatment groups. During the course of treatment, the group that received citrus fruits rapidly and dramatically improved, while the others did not. This episode fits the description of a "quasi-experiment" because there was no randomization and no formal control (except the other scurvitic sailors on board, whom Lind did not describe).
Donald Campbell and Julian Stanley first used the term "quasi-experiment" in 1963. Campbell went on to write one of the most widely cited texts on quasi-experimental methods with co-author Thomas Cook in 1979.
The earliest known use of the phrases "time series" and "time series analysis" are in the early 1900's, both in statistical journals of the period. (http://members.aol.com/jeff570/t.html)
The ITS-CG method should be used when true randomization of the treatment group and controls is impossible, but multiple measurements are still feasible. The method is well suited to clinical informatics research, where randomization may not be possible, but where an intervention can be targeted for one group, but not another (such as implementation of a new clinical decision support tool).
ITS-CG is a reasonable method to use where true randomization of cases and controls is not possible. Interrupted time series designs improve on the non-random control group pre-test / post-test design by introducing serial measurements before and after the intervention. This minimizes the weaknesses of single measurements such as regression to the mean and, to some extent, history as a threat to internal validity. By adding a comparison group, the effect of history is lessened, and the study is strengthened against other threats to internal validity, such as instrumentation (did the method of measurement change after the intervention?), maturation (did the change occur naturally over time), and ambiguous temporal precedence (did the change in outcome really occur after the intervention?).
Despite being one of the most robust quasi-experimental designs, ITS-CG is susceptible to some threats to internal validity, primarily selection. Since groups are not randomly assigned, the design cannot control for possible systematic differences in how the members of the groups were allocated.
Examples in Informatics
Goldberg HI, Neighbor WE, Cheadle AD, Ramsey SD, Diehr P, Gore E. A controlled time-series trial of clinical reminders: using computerized firm systems to make quality improvement research a routine part of mainstream practice. Health Serv Res. 2000 Mar;34(7):1519-34.
Ashton CM, Khan MM, Johnson ML, et al. A quasi-experimental test of an intervention to increase the use of thiazide-based treatment regimens for people with hypertension. Implement Sci. 2007 Feb 13;2:5.(not strictly an informatics intervention, but it included summarized data from an EHR as part of the intervention).
Desaib 20:18, 4 March 2007 (CST)