I want to set an acceptance criteria for an assay result saying if the results of parallel samples are too far apart from each other the result is not valid.

Are you people doing this? How should I set it? For example if the repeatability of parallel samples was RSD = 4 % (n=6) in the validation.

1. Should I then say: the (mean) result is not valid if the percentual difference between individual results is more than x % (what would x be?)

or

2. Should I calculate percentual 95% confidence intervals for mean (from the validation) for a sample amount of 2 and say: (mean) result is not valid if either one of the individual results is outside the range?

I dont't want too much pain from this and the spec. should be wide enough allowing the acceptance of most of the results (no 2. seems quite tight).

Is there a rule of thumb for this? Any suggestions?

Thanks in advance.

what about this?

Routinely calculate mean from duplicates, lower acceptance limit is:

mean - 3 x RSD% in the validation

and upper acceptance limit is

mean + 3 x RSD% in the validation

check that both individuals are between limits.