In medical registry studies for rare conditions, a reporting error of less than 1% is enough to create fake 'statistically significant' survival differences between treatments.
April 1, 2026
Original Paper
Small Differential Outcome Ascertainment Error Can Generate Apparent Survival Differences in Low-Event Registry Studies: A Quantitative Bias Analysis with a Thyroid Cancer Example
SSRN · 6502241
The Takeaway
Researchers found that when outcome events are rare (like cancer death rates), the 'fragility' of the data is so high that a tiny 0.4% to 1.8% error rate in tracking patients completely invalidates the results. This suggests that massive medical databases might be showing us 'breakthroughs' that are actually just noise from minor clerical errors.
From the abstract
Background: Large registry-based studies often compare treatments in settings with rare outcome events. When event rates are low, even small differences in outcome ascertainment between groups can generate statistically significant but potentially non-causal survival differences. We quantified the minimum differential ascertainment error required to generate apparent survival gaps in historical thyroidectomy-versus-lobectomy studies. Methods: In this summary-level quantitative bias analysis, we