Residual sum of squares ICH

[ldcarthy]

The thing I’m struggling understanding is what the quoting of the residual sum of squares is for in ICH validation, obviously it needs to be as close to zero as possible to show no bias.

Other numbers show positive or negative bias but what would be an upper limit? I m guessing it's something to do with the degrees of freedom but I m not sure.

I don’t like the idea of quoting a number that does nt have a criteria so I won't know if it is a pass or fail result and hope someone can give me guidance on this.

[jeffspc88mx]

There is no criteria unless you establish one - based on your experience with the assay, it's characteristic precision/linearity - what you think is good vs. bad. Your limit has to be defensible - your reasoning for choosing that limit must be sound. No one will tell you what it should be, though an auditor may say it's unreasonable if you can't defend it.

Our RSS values are never near zero. We don't have a criteria (though we do have one for r^2).

[nonagon]

Try reverse engineering - Do remember that at the end of the day the RSS referes to an actual parameter be it area, height percent, amount,etc; with that in mind try this:
use a spreadsheet (excel or whatever) and create a realistic worst case table, calculate the RSS. This will give you an indication as to where you do not want to go ("the NoNo limit"). You can than play with the values to find your acceptable RSS upper limit by entering the y values you can live with.
you will probably find that the upper limit RSS is a few orders of magnitude below the NoNo limit. This may give you a general notion as to your own criteria and provide the rationale to defend your RSS limits by showing that they actualy translate to a deviation that is marginal (an RSS of 0.003 translates to a maximunm of 0.2% amount deviation etc.).
Warning: Do not fool the table into low RSS - keep it real. Do not try to correlate RSS with R^2.
All that without actual chromatography.

[Mattias]

I must admit that I have never really understood why this value needs to be reported. But since it is stated in the guideline, we always put it in the validation report. I have also never experienced that any authority has asked any questions about this.

The problem is that the RSS is stated in the units used by the detector. So sometimes 10 billions is a fair RSS, sometimes 0.4 is unacceptable.

[nonagon]

I usually convert areas back to the units of interest for RSS calculation for example:
plot of 70-130%, 5% increment (13 points)
slope=15000
intercept=-50000
RSq=0.996
areas in 1000000uAU/sec range
this will give area RSS of about 1E10 but extracting % back will show that 80% gives 79% so RSS accross the range will be around 15. with better RSq (0.9996) area RSS will be about 3E8 but 80% will give 79.85% with RSS around 1 or lower.

I have also never experienced that any authority has asked any questions about this.

...Neither have I. This would be an hyperanal auditor from hell!