What LOD method does your lab use — and why?

[Sachin]

Limit of Detection (LOD) is one of those validation parameters that looks straightforward on paper but generates a surprising amount of debate in practice.
There are three main approaches most of us work with:

Signal-to-noise (S/N) ratio — typically S/N ≥ 3 for LOD, ≥ 10 for LOQ. Simple and fast, but heavily dependent on how you define the noise region.
Blank standard deviation method — LOD = 3.3 × (SD of blank / slope). More statistically rigorous, but sensitive to how well your blank represents your matrix.
Calibration curve residuals — uses the standard error of the regression. Often preferred in regulated environments (ICH Q2(R1)) but can be overly optimistic with well-behaved curves.

In my experience, S/N is the default go-to for routine work, but the SD method tends to hold up better during audits.
A few questions for the community:
— Which method does your lab default to, and is it driven by your SOP, your instrument software, or personal preference?
— Have you ever had a method pass LOD on one approach and fail on another?
— For those working in GLP/GCP environments — has your LOD method ever been challenged during an inspection?
Would love to hear how others handle this, especially across different matrices and regulatory contexts.

[Author]

Posts