Calibration Interval (Frequency) Adjustment using Hilliard type 2 algorithm

[A C Diggs]

We have a new manager for our pharmaceutical calibration lab, and he has asked me for help deciphering our SOP on cal interval adjustments. Currently, the SOP states that we use the Hilliard type 2 calibration interval assignment algorithm. I have run some senerios through the formula and cannot come up with a situation that would qualify to extend the interval (not even zero OOTs worked). Perhaps, I'm not applying the formula correctly.
I work in a fair size shop with 3000 instruments to cal (approx 400 per month). We do not have a validated program to crunch these numbers for interval adjustment according to the Hilliard method.
We have kicked around the idea of rewriting the SOP and using a simple look-up table based on the previous 4 calibrations, but then we ran into the question of how do we test this table for its ability to meet our relaibility target?

[Marc]

Can any stats experts help out on this one?

I know it's dated, but I'm trying to catch up on some zero reply threads.

[Hershal]

I confess to ignorance of the specific algorithm (until I read up on it). Graeme might be the best source for a good answer here.

Hershal

[Marc]

I tried a search for "Hilliard type 2 algorithm" (with and without the quotes) on just about every search engine and haven't found anything. I did find references to Cahn-Hilliard type equations but they are beyond me.

[rmccallum]

If you are willing to spend a few $$ there is software out there that will optimize your cal intervals for you. I would have to assume it is statistically significant. Correct me if I am wrong.

[Marc]

I've never seen any company go so far as to use something like a Hilliard type 2 algorithm (whatever that is). It's typically factors such as use frequency and environment, and results of previous calibrations.

Let's see - If we're going to analyze data to set calibration intervals, I'd look at 'as received' vs. 'as returned' if adjusted.

Anyone do data analysis and have a formula, or do most of you folks just look at the calibration results and 'gut feel' it?

This thread may be of interest: http://Elsmar.com/Forums/showthread.php?t=997

[rmccallum]

Currently we don't statistically analize to establish calibration frequency. We use the "gut feeling" method. Curious if this is acceptable for a ISO 17025 accredited automotive lab. The standard does call out uncertainty analysis. I just assumed this would be applied to calibration optimization. I did find the calculations on a google search to optimize calibration using fitted modelling and reliability but I don't think I have the time or willpower to program an excel spread sheet so I found a company that supplies the software to perform this function.

I am just starting out in pursuit of the accreditation. I have a pretty good foundation but a long ways to go. You will most likely see me hanging around over the next while. One question that may sound stupid but can I reference ISO 17025 requirements such as internal auditing or corrective action to our ISO 9000: 2001 system or do I need to rewrite all these?

[Hershal]

If we get away from the algorithm, then I recommend bouncing over to NCSLI RP (Recommended Practices) 1.

http://www.ncsli.org

It is well respected, established, and sound.

Some of the calibration software in the newer versions include cycle adjustment. Check with Cybermetrics and Blue Mountain.

Hershal