Book contents
- Frontmatter
- Contents
- Preface
- 1 Introduction
- 2 Interventions
- 3 Evaluating an intervention
- 4 Randomized designs
- 5 Nonrandomized studies
- 6 Statistical analysis of intervention trials
- 7 Methods for adjusting for baseline differences between treatment groups
- 8 Time series analysis
- 9 Special topics
- 10 Research to action
- 11 Conclusion
- Index
- References
9 - Special topics
Published online by Cambridge University Press: 10 May 2010
- Frontmatter
- Contents
- Preface
- 1 Introduction
- 2 Interventions
- 3 Evaluating an intervention
- 4 Randomized designs
- 5 Nonrandomized studies
- 6 Statistical analysis of intervention trials
- 7 Methods for adjusting for baseline differences between treatment groups
- 8 Time series analysis
- 9 Special topics
- 10 Research to action
- 11 Conclusion
- Index
- References
Summary
What methods are available for evaluating interventions that do not occur to all subjects at the same time?
It is common in observational cohorts to have different subjects start an intervention at different times, with some subjects never receiving it. This is particularly common when new medications are introduced into practice.
In situations such as these, you may have no subjects on the medication (prior to approval) and then a gradual uptake in medication use. Let's assume you want to evaluate the impact of a new medication designed to prevent osteoporosis. Your outcome is occurrence of a pathologic fracture (yes/no). How would you do it?
At first blush you might think that you could compare the number of fractures in the 280 persons who took the drug to the number of fractures among the 720 persons who did not take the drug.
But this would be wrong from several points of view. First, for those people who dropped out we would only know what happened to them prior to the drop-out (e.g., whether they started taking the medicine, that they didn't develop a fracture). By year 5, some of them may have begun the medicine and some may have had a fracture. We could exclude dropouts from our analysis but then we would lose 15% of our total sample. And what about the deaths that occurred during the study prior to a fracture?
- Type
- Chapter
- Information
- Evaluating Clinical and Public Health InterventionsA Practical Guide to Study Design and Statistics, pp. 135 - 153Publisher: Cambridge University PressPrint publication year: 2010