Andy
Excellent!
Not 'embarrassingly dated' though - the term you want is 'fashionably retro'.
Paul Waller MSc CSci FIBMS FHEA
Associate Professor in Biomedical Science
Postgraduate Field Leader / Course Director MSc Biomedical Science
School of Life Sciences
T 020 8417 7783 (Direct Dial) / 67783 (Internal)
E [log in to unmask] / Web page Mr Paul Waller
Room PR MB 1019, Penrhyn Road, Kingston upon Thames, KT1 2EE
-----Original Message-----
From: Clinical biochemistry discussion list [mailto:[log in to unmask]] On Behalf Of Andy Minett
Sent: 08 August 2014 11:05
To: [log in to unmask]
Subject: Re: Using Patient Means as a QC measure
Hi,
This is something I have a little experience in from some work I did in a previous lab, which was part of a bigger QC software design project that I had set myself.
I also found there to be very little in the literature to go off, but managed to get hold of a copy of:
Cembrowski et al (1984): Assessment of "Average of Normals" quality control procedures and guidelines for implementation.
http://www.ncbi.nlm.nih.gov/pubmed/6702751
...which provided just enough explanation to properly implement AoN.
I'd have to dig the paper out to check, but going off memory alone the process of evaluating and setting up AoN goes something like this:
1) Collect a good history of patient data for a specific assay, during a period when the analyser is known to be functioning properly
2) Plot the data, and decide on upper and lower "hard limits" which would exclude extreme values and result in the remaining data having a normal distribution.
For some analytes (Na, Ca, K, etc) data will be normally distributed anyway, so just set the limits to exclude outliers/errors. For other tests (e.g. Creat, enzymes, etc) we would expect patient data to be positively skewed anyway, especially in an acute setting, and so limits should be set to exclude any non-normal data.
3) Calculate mean,sd for the new isolated patient data
4) Decide on +/- SD limits within which patient results will contribute to the AoN value. If the original patient data was normally distributed, use +/-3.0 SD, if it was skewed and you had to truncate a lot of data, use between 2.0 and 2.5 SD.
Note: the reason Cembrowski et al give for this is that if the data is skewed, there remains a lot of data outside of the limits set. In the event of the analyser suddenly developing a systematic error, the data within the skew may mask the shift. Hence a more severe limit is needed.
5) Find the SD of the analyser for this test, at the concentration of the patient data mean, or as close as you can get it from historic QC data
6) Calculate the AoN result groupings. This can be done via look-up tables provided in the referenced paper, alternatively I ran some curve-fitting algorithms over the look-up tables, and found that this closely approximates the data between n groupings of 5 - 1000:
n groupings = ((SDpatients/SDanalyser)/0.7029)^2
as an example for Sodium, if the "hard-limited" patient data SD was 3.313 and the analyser SD was 1.2:
n groupings = ((3.313/1.2)/0.7029)^2
n groupings = 15
Thus you would need to group together 15 consecutive patient results, and find their average. This value is now your AoN QC point. The average of the next 15 patient results is the next QC point, and so on.
7) This new AoN QC data can be evaluated in the usual way using power function analysis to generate upper and lower QC limits and QC rules, maximising error detection, and minimising false rejection.
After having set up a number of tests to monitor patient means, I found that there are only a handful of tests that really lend themselves to this sort of monitoring. For some tests where there is a inter-individual high biological variance, you would need to collect many hundreds of patient results before a suitable qc point can be generated. This prevents the use of AoN as a real-time QC monitoring process (you will probably run QC material on the analyser before the next AoN point is generated), but I found it was particularly useful in just generating a small number of points for the day/week - it cut a lot of the 'noise' out of the normal manual QC data.
We only ran AoN for a few tests, and only over a short period, but I remember on at least 4-5 occasions the AoN picked up ISE problems before the manual QC did.
The main problem with implementing AoN now is finding manufacturer software that actually supports AoN, and implements it properly on-the-fly applying the correct truncations etc.
Attached is a screenshot of the software I wrote showing Sodium evaluation - it looks embarrassingly dated now!
Best regards,
Andy Minett
Specialist Biomedical Scientist
Biochemistry
Pathology
Hull Royal Infirmary
------ACB discussion List Information--------
This is an open discussion list for the academic and clinical community working in clinical biochemistry.
Please note, archived messages are public and can be viewed via the internet. Views expressed are those of the individual and they are responsible for all message content.
ACB Web Site
http://www.acb.org.uk
Green Laboratories Work
http://www.laboratorymedicine.nhs.uk
List Archives
http://www.jiscmail.ac.uk/lists/ACB-CLIN-CHEM-GEN.html
List Instructions (How to leave etc.)
http://www.jiscmail.ac.uk/
This email has been scanned for all viruses by the MessageLabs Email
Security System.
This email has been scanned for all viruses by the MessageLabs Email
Security System.
------ACB discussion List Information--------
This is an open discussion list for the academic and clinical community working in clinical biochemistry.
Please note, archived messages are public and can be viewed via the internet. Views expressed are those of the individual and they are responsible for all message content.
ACB Web Site
http://www.acb.org.uk
Green Laboratories Work
http://www.laboratorymedicine.nhs.uk
List Archives
http://www.jiscmail.ac.uk/lists/ACB-CLIN-CHEM-GEN.html
List Instructions (How to leave etc.)
http://www.jiscmail.ac.uk/
|