Examkin

Question

Difficulty: Very hardBias, Confounding, and Validity

A retrospective cohort study using a national electronic health record database evaluates the association between a newly approved oral GLP-1 receptor agonist and the incidence of acute pancreatitis in patients with type 2 diabetes. The comparator group consists of patients prescribed DPP-4 inhibitors. Initial analysis yields a crude relative risk of 1.8 (95% CI: 1.4 - 2.3) for acute pancreatitis in the GLP-1 agonist group.

Upon further review, the investigators note that the GLP-1 agonist was disproportionately prescribed to patients with a higher baseline body mass index (BMI), a longer duration of diabetes, and a prior history of hypertriglyceridemia. All of these factors are known independent risk factors for acute pancreatitis. The researchers recognize that this unequal distribution of baseline characteristics constitutes confounding by indication and threatens the internal validity of the study.

Which of the following strategies represent valid methods to control for this specific methodological threat during the analysis phase of this study? (Select all that apply)

  1. Multivariable regression analysis incorporating BMI, diabetes duration, and hypertriglyceridemia as covariatesAnswer
  2. Propensity score matching of patients in the GLP-1 agonist and DPP-4 inhibitor groups based on baseline characteristicsAnswer
  3. C
    Reporting stratum-specific relative risks for demographic variables to determine if the drug's risk profile differs across subgroups
  4. D
    Retroactively applying blocked randomization to the existing electronic health record data to distribute covariates evenly
  5. E
    Calculating the positive and negative predictive values of the GLP-1 agonist for predicting acute pancreatitis

Answer

The correct methods for controlling this threat during the analysis phase are multivariable regression analysis incorporating the baseline disparities as covariates, and propensity score matching of patients based on their baseline characteristics.
Confounding by indication occurs when a provider's decision to prescribe a medication is influenced by clinical factors that also impact the patient's risk for the outcome. In retrospective observational studies, researchers cannot randomize patients. Instead, they must handle known confounders during the analysis phase. Multivariable regression analysis achieves this by mathematically modeling the independent contribution of the drug while holding the confounding covariates constant. Propensity score matching achieves this by calculating each patient's probability of receiving the drug based on their baseline characteristics, and then matching treated and untreated patients with similar scores, mimicking a randomized design in the analysis phase.

Step-by-Step Solution

1
Identify the methodological threat described in the scenario.
The unequal distribution of baseline risk factors (BMI, diabetes duration, hypertriglyceridemia) between the two treatment groups leads to confounding by indication.
Recognizing the specific type of bias is necessary to select the appropriate corrective strategy.
2
Determine the phase of the study in which the correction must be applied.
The question specifically asks for strategies applied during the 'analysis phase' of this retrospective cohort study.
Some strategies (like randomization or restriction) are applied during the design phase, while others (like regression or stratification) are applied during analysis.
3
Evaluate the options for valid analysis-phase methods to control confounding.
Multivariable regression and propensity score matching are both robust analytical techniques used to adjust for known, measured confounders in observational data.
These methods statistically balance the groups or adjust the risk estimates to account for the identified baseline differences.

Key Concept

Confounding by indication in observational studies can be controlled during the analysis phase primarily through multivariable regression analysis, propensity score matching, or stratified analysis.
Rate this question

Topics

Bias, Confounding, and Validity

More questions from this topic