Examkin

Question

Difficulty: MediumBias and Confounding in Research

A clinical investigator is reviewing data from a large observational registry of patients with moderate-to-severe plaque psoriasis to assess the safety of a recently introduced interleukin-23 (IL-23) inhibitor. The primary outcome is the rate of severe systemic infections requiring hospitalization. When comparing patients receiving the IL-23 inhibitor to those receiving traditional systemic therapy, the unadjusted relative risk (RR) for severe infection is 2.82.8 (p<0.05p < 0.05).

Recognizing that patient populations may differ, the investigator separates the subjects into two subgroups based on concurrent methotrexate use at baseline. Among patients not taking methotrexate, the RR for severe infection is 1.01.0 (p=0.85p = 0.85). Among patients actively taking methotrexate, the RR is 1.11.1 (p=0.62p = 0.62).

Which of the following strategies, if implemented during the study design phase, would have best prevented the discrepancy between the unadjusted and subgroup risk estimates?

  1. Matching participants in both treatment arms based on their methotrexate statusAnswer
  2. B
    Adjusting the initial risk estimate using multivariable logistic regression
  3. C
    Blinding the assessing clinicians to the patients' treatment allocations
  4. D
    Expanding the registry to include a larger number of total participants
  5. E
    Implementing strict, standardized clinical criteria for defining severe infections

Answer

Matching participants in both treatment arms based on their methotrexate status
The data demonstrates confounding, as the unadjusted relative risk (2.82.8) is distorted by an unequal distribution of a third variable (methotrexate use). When the data is stratified by this variable, the risk estimates within each stratum (1.01.0 and 1.11.1) are similar to each other but significantly different from the crude estimate. To control for confounding at the study design phase, researchers can use randomization, restriction, or matching. Matching participants based on their methotrexate status ensures that this confounding variable is equally distributed between the treatment and control groups before data collection begins.

Step-by-Step Solution

1
Analyze the unadjusted versus the stratified relative risk (RR) data.
The unadjusted RR is 2.8, while the stratified RRs are 1.0 and 1.1.
Comparing crude and stratified data is the primary method to detect epidemiological phenomena like confounding or effect modification.
2
Determine the epidemiological principle at play.
Because the stratified RRs are similar to each other but significantly different from the unadjusted RR, the discrepancy is caused by a confounding variable (methotrexate use).
If it were effect modification, the stratified RRs would be significantly different from each other.
3
Identify methods to control for confounding based on the timeline specified in the stem.
The stem explicitly asks for a strategy implemented during the 'study design phase'. Matching is a design-phase control, whereas regression is an analysis-phase control.
Confounding must be controlled either before data collection (design phase: randomization, restriction, matching) or after data collection (analysis phase: stratification, multivariable regression).

Key Concept

Identifying confounding from stratified data and distinguishing between design-phase versus analysis-phase methods to control it.
Rate this question

Topics

Bias and Confounding in Research

More questions from this topic