Example: Incidence - Time-Varying Exposure

Lead Author(s): Jeff Martin, MD

Analysis of Changing Exposure and Disease Incidence

Ray (2002) analyzed Medicaid recipients in Tennessee over a ten-year period to determine if there is a relationship between use of steroidal anti-inflammatory drugs (NSAIDs) and coronary heart disease (CHD).

Data: Tennessee Medicaid data base, 1987-1998

Use of NSAIDs could change over 11 years of study: same person could be in both using and non-using group at different times

Option 1: Cumulative Incidence

One option was to construct some fixed classification of persons as never, sometime, and frequent users and do cumulative incidence in each group.

In a prospective cohort with an exposure that can vary over time, subjects cannot be classified in groups by their amount of exposure at baseline because their future medication use cannot be known in advance.

In a cohort analysis done retrospectively (such as this example) or at the end of follow-up in a prospective cohort, it is possible to classify individuals by some measure of total use.

Problem with Cumulative Incidence at Follow-up

  1. It would give a rather crude categorization (say, 2 or 3 groups).
  2. It does not do a good job of distinguishing length of time of use, which may vary from months to years.

Option 2: Use Incidence Rates

BETTER: Incidence rates during times of use and non-use that account for time of exposure.

Analysis of Changing Exposure

In the Ray (2002) paper, the following data were given:

Person-time totaled for using and not using NSAIDs; MI or CAD death outcome The following were censoring variables: 181,441 person-years of use (persons who were new users of NSAIDS)

181,441 person-years of non-use (persons, matched by age, sex, and calendar date)

Subjects with no history of prior NSAIDS use who began use after having been enrolled in the database for 365 days (in order to have data on prior illnesses) were compared with subjects not using NSAIDS.


A person can contribute to the denominator both for use and non-use A subject who stopped using was eligible to be in the non-using group later, but to avoid any carryover effect, only after 365 days of non-use had elapsed.

Analysis of Person-Time Rates

Rate for NSAID use = 12.02 per 1000 pers-yrs

Rate for non use = 11.86 per 1000 pers-yrs

Rate ratio = 1.01

Concluded no evidence that NSAIDS reduced risk of CHD events.


Ideally, this is a question that should be resolved by a clinical trial, but a clinical trial of this question may never be done. In the absence of a randomized trial, an observational cohort study is the second best choice.

Again, a prospective cohort with better measurements of all the potential confounders, in particular aspirin use, would be preferable, but to get the numbers required would mean a very large cohort followed for a number of years. Possible but very expensive.

Analyzing existing data is less desirable, but it does provide an opportunity to assemble a cohort analysis on a large number over many years at minimal expense.

The question that remains is whether they were able to get adequate control of confounders.


Ray, W. A., Stein, C. M., Daugherty, J. R., Hall, K., Arbogast, P. G., & Griffin, M. R. (2002). COX-2 selective non-steroidal anti-inflammatory drugs and risk of serious coronary heart disease. Lancet, 360(9339), 1071-1073.