How hazard ratios can mislead and why it matters in practice

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Hazard ratios (HRs) are standard effect measures in clinical and observational research.
• Concerns exist regarding the causal interpretation of HRs in medical studies.
• Key issues include selection bias from susceptible depletion, non-collapsibility, and the proportional hazards assumption.

Purpose of the Study:

• To examine the relationship between selection bias, non-collapsibility, and the proportional hazards assumption.
• To illustrate these issues using an example of endocrine therapy for breast cancer.
• To highlight the advantages of survival curves and risk differences for causal inference.

Main Methods:

• Methodological review of hazard ratio interpretation.
• Case study analysis of breast cancer recurrence/mortality data.
• Comparative analysis of effect measures (HRs vs. survival curves/risk differences).

Main Results:

• Depletion of susceptibles can introduce selection bias, complicating HR interpretation.
• HRs are not collapsible, meaning they cannot be adjusted for covariates.
• The proportional hazards assumption is frequently violated in real-world medical data.
• Survival curves and risk differences avoid these interpretation pitfalls.

Conclusions:

• The causal interpretation of hazard ratios is problematic due to inherent methodological limitations.
• Selection bias, non-collapsibility, and assumption violations undermine HRs' causal validity.
• Survival curves and risk differences provide more reliable measures for causal effect estimation in survival analysis.