Quantifying bias in epidemiologic studies evaluating the association between acetaminophen use and cancer

Area of Science:

• Epidemiology
• Pharmacovigilance
• Biostatistics

Background:

• Observational studies investigating the link between acetaminophen use and cancer are common but suffer from inherent biases.
• Limitations like channeling, protopathic bias, and uncontrolled confounding compromise the interpretability of these studies.
• Understanding the potential magnitude of bias is crucial for accurate risk assessment of acetaminophen.

Purpose of the Study:

• To quantify the potential bias introduced by various design choices in observational studies of acetaminophen and cancer.
• To evaluate the impact of different study design variants on effect estimates using negative controls and cancer outcomes.

Main Methods:

• Identified key design choices in observational studies and defined 10 distinct study design variants.
• Applied these variants to 37 negative control outcomes and 4 cancer outcomes within the Clinical Practice Research Datalink (CPRD) database.
• Assessed bias by examining confidence intervals for odds and hazards ratios in negative controls and comparing them to cancer outcomes.

Main Results:

• Evaluated design variants demonstrated substantial bias, with confidence intervals for negative controls frequently not including 1.
• Effect-size estimates for cancer outcomes were similar to those observed for negative controls, indicating potential bias.
• Significant baseline differences between exposed and unexposed groups were noted, often uncorrected in published studies.

Conclusions:

• Common design choices in observational studies investigating acetaminophen and cancer can introduce considerable bias.
• Results from published observational studies on acetaminophen and cancer should be interpreted with caution.
• Further research with robust designs is needed to clarify the true association between acetaminophen and cancer risk.