Considerations for pharmacoepidemiology study design

Right now, an unorganized list of concepts I don’t want to forget about.

Pharmacoepidemiology, general

• Rapid changes in the natural history of a disease or changes in treatment decisions will mean strong effects in relation to calendar time.

  • Example: The management of COVID-19 improved throughout the pandemic independent of medication use. If you wanted to assess the comparative effectiveness of a drug, you would need to control for calendar time.

• Nonuser designs tend to suffer from confounding by indication, as patients receiving treatment will be different from those receiving no treatment.

• Immortal time bias - this bias results when there is a period of follow-up time when it is impossible for the outcome to occur in the exposed individual.

  • The classic example is when person-time before an exposure is classified as “exposed”. By design that person had to have survived up to the exposure point, so this is not a real comparator to the unexposed person. This time should be categorized as “unexposed”.
  • Using a risk-set sampling design, where all people are eligible to be “nonusers” regardless of whether they later become “users”, avoids this bias.

• New user study design

  • For drugs with intermittent use, lookback period will have a strong impact on classification.

Using Claims Data

• Including open claims presents a challenge of defining a relevant denominator. Main concern is the under-inclusion of healthy individuals who are not"active" in the healthcare system.

References

Franklin, J.M., et al. Real-World Evidence for Assessing Pharmaceutical Treatments in the Context of COVID-19.