When Zero Events Mislead: Evidential Fragility in Rare-Event Epidemiology

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Rare-event studies often rely on limited data, making conclusions sensitive to few observed outcomes.
• Evidential fragility is defined as the sensitivity of conclusions to minor data variations.

Purpose of the Study:

• To examine evidential fragility in rare-event studies.
• To illustrate this phenomenon using the hypothesis that early blindness protects against psychosis.

Main Methods:

• Utilized aggregate data from a Danish nationwide registry.
• Employed Bayesian analysis to compare protection versus no protection.
• Assessed the impact of prior Australian data using a discounted power prior approach for cross-study heterogeneity.

Main Results:

• Evidence for protection varied from anecdotal to strong when zero cases were observed.
• One observed case significantly weakened support; two cases reversed the evidence towards no protective effect.
• External prior information initially boosted support for zero cases but diminished rapidly with accumulating events.

Conclusions:

• Epidemiologic claims from rare-event studies can be inherently fragile.
• Minimal changes in observed case counts can reverse study conclusions.
• Bayesian stress-testing can evaluate the robustness of rare-event findings before clinical or policy application.