Interrupted time-series analysis
An interrupted times series (ITS) analysis is a quantitative, statistical method in which multiple (sometimes as many as 40 to 50) repeated observations are made at regular intervals before and after an intervention (the “interruption” in the time series). Statistical analysis can be used to determine whether there is a change in the scores or trends in scores of the observations after the intervention. There are multiple variations, including: using a non-intervention control group; removing the intervention, then taking more measurements; using two groups with differing times for the intervention; and using two groups of subjects, altering which is the intervention and which is the control.
This type of design has a long history in the hard sciences, and arose out of simple pretest-posttest observation designs. It has been used as well in behavioral sciences. It was further characterized and formalized in the 1970s with the development of variations and statistical methods for analyzing the data.
ITS is useful in complex situations in which the observed variable changes over time, either before or after the intervention. It is also useful when there is only small change in the observed variable, especially if there are multiple factors other than the intervention which may affect the variable.
ITS can detect changes that are delayed or intermittent. It can also determine if the change is permanent or temporary. In addition, it allows evaluation of variables which are changing before the intervention, for instance, by comparing slopes of trend lines before and after the intervention. The design may be simple, without need for randomization. In some cases, historical data can be used. Finally, ITS makes it easier to control for confounding variables and regression to the mean.
One of the biggest problems is in determining whether a change noted is due to the intervention or to other factors, such as another event occurring at a similar time to the intervention. Cyclical changes, such as effects of seasons, may also be overlooked if the number of observations or interval between observations is not great enough. These may be overcome through some of the variations, such as a non-intervention control group, removal of the intervention, or applying the intervention to two or more groups at different times. These steps, however, increase the cost and time required for making the large number of observations, which may already be problematic.
Examples in Informatics
- Park WS, Kim JS, Chae YM, Yu SH, Kim CY, Kim SA, et. al. Does the physician order-entry(POE) system increase the revenue of a general hospital? Int J Med Inf. 2003;71:25-32.
- Kolde D, Ohe K, Ross-Degnan D, Kalhara S. Computerized reminders to monitor liver function to improve use of etritinate. Int J Med Inf. 2000;57:11-19.
- Rosenbloom ST, Chiu K, Byrne DW, Talbert DA, Neilson EG, Miller RA. Interventions to regulate ordering of serum magnesium levels: report of an unintended consequence of decision support. J Am Med Inform Assoc. 2005;12:546-553.
- Harris AD, McGregor JC, Perencevich EN, Furuno JP, Zhu J, Peterson DE, Finkelstein J. The use and interpretation of quasi-experimental studies in medical informatics. J Am Med Inform Assoc. 2006; 12:16-23.
- England E. How interrupted time series analysis can evaluate guideline implementation. The Pharmaceutical Journal. 2005;275:344-347.
- Hartmann DP, Gottman JM, Jones RR, Garnder W Kazdin AE, Vaught RS. Interrupted time series analysis and its application to behavioral data. J Applied Behavioral Analysis. 1980; 13:543-559.