individual refernce change values based on individual variability of lab values
20 views
Skip to first unread message
Thomas Keller
Jun 28, 2022, 10:08:53 AM6/28/22
to MedStats
Hi,
in clinical chemistry, the concept of RCV exists (RCV - reference change values). Given a patient's or proband's specific steady state value of a measured value a change of the value of sqrt(2) Z x sqrt(CVa**2 + CVi**2) refers to clinical relevant change, whereby CVa iand CVi are the analytical and biological variation expressed as CVs. They are taken from databases (CVi) and information of the lab or manufacturer (CVa). This concept was introduced by Fraser more then 20 years ago and is generally accepted (although some assumptions have to be met to be valid).
There are new (already published) concepts which estimate (CVa**2 + CVi**2) from individual patient's data, whereby minimal 5 datapoints are recommented. In fact the prediction interval of the mean (based on individual CV) is used as the RCV.
(The values are measured in periods of several weeks/months, so they can be regarded as independend).
This is then called personalized RCV based on personal biological variance.
Given, that the RCV concept described in the 1st paragraph is correct, I have criticized this concept of personal RCV based on at least 5 measurements as not helpful because of the huge uncertainty of the estimated SDs (CVs). I was not able to convince the community by presenting CI of SD, presented simulations or real data. It was argued by using outlier detection and trend analysis this huge uncertinty would not occur. May be you can help me to find arguments (or I'm not correct with my critics). Thank you,
kind regards
Thomas
John Whittington
Jun 28, 2022, 10:41:18 PM6/28/22
to meds...@googlegroups.com
Hi Thomas,
Interesting. In the hope that I have understood you correctly, just a few thoughts/comments, for what they are worth! .
I have not come across the concept of a 'personalised RCV' as such. The nearest I can recall, a while ago, was a lab which, in additional to publishing the usual 'normal reference ranges' also published figures relating to within-subject variability, but they were derived from a group of 'healthy' subjects - and I assumed (perhaps wrongly!) that the figures presented were based on a reasonable sample size,
Whilst the concept of a 'personalised RCV' is theoretically attractive, I certainly see some limitations. One of my greatest concerns is that there would seem to be potential for a 'Catch 22' situation. One is most likely to be interested in the relevance of observed changes in a patient with a disease or pathological process which might result in ('real') changes in the quantity being measured - so if one estimated (CVa**2 + CVi**2) within such a subject by serial measurements "over a period of weeks/months", then one could well be looking partially at real changes (that one was interested in), rather than just variabilities. I would have thought that it would be more sensible (albeit not 'personalised') to do as I describe above, and use estimates of within-patient variability based on a group of 'healthy' subjects (i.e. not expected to show any real changes in the quantity over time.
For that reason, I can't really see how one can estimate a 'personnalised RCV' derived from measurements over time in a patient who might well have 'real' changes in the quantity concerned during that time period.
Particularly if I put on my physician's hat (and really also whilst wearing my Statistician's one!) one thing I would take issue with is the notion that any such technique would/could result in an indicator of a "clinically relevant change". It is surely the case that the approach is seeking to determine whether (given analytical and biological variation) an observed within-patient change is statistically relevant ('significant') - i.e.is likely to be a real change, rather than just a consequence of the variabilities. A change could be highly relevant/'significant' in that (statistical) sense, yet far too small to be of any clinical relevance/importance - only clinicians can make that judgement.
As I said, just my few thoughts!
Kindest Regards,
John
Interesting. In the hope that I have understood you correctly, just a few thoughts/comments, for what they are worth! .
I have not come across the concept of a 'personalised RCV' as such. The nearest I can recall, a while ago, was a lab which, in additional to publishing the usual 'normal reference ranges' also published figures relating to within-subject variability, but they were derived from a group of 'healthy' subjects - and I assumed (perhaps wrongly!) that the figures presented were based on a reasonable sample size,
Whilst the concept of a 'personalised RCV' is theoretically attractive, I certainly see some limitations. One of my greatest concerns is that there would seem to be potential for a 'Catch 22' situation. One is most likely to be interested in the relevance of observed changes in a patient with a disease or pathological process which might result in ('real') changes in the quantity being measured - so if one estimated (CVa**2 + CVi**2) within such a subject by serial measurements "over a period of weeks/months", then one could well be looking partially at real changes (that one was interested in), rather than just variabilities. I would have thought that it would be more sensible (albeit not 'personalised') to do as I describe above, and use estimates of within-patient variability based on a group of 'healthy' subjects (i.e. not expected to show any real changes in the quantity over time.
For that reason, I can't really see how one can estimate a 'personnalised RCV' derived from measurements over time in a patient who might well have 'real' changes in the quantity concerned during that time period.
Particularly if I put on my physician's hat (and really also whilst wearing my Statistician's one!) one thing I would take issue with is the notion that any such technique would/could result in an indicator of a "clinically relevant change". It is surely the case that the approach is seeking to determine whether (given analytical and biological variation) an observed within-patient change is statistically relevant ('significant') - i.e.is likely to be a real change, rather than just a consequence of the variabilities. A change could be highly relevant/'significant' in that (statistical) sense, yet far too small to be of any clinical relevance/importance - only clinicians can make that judgement.
As I said, just my few thoughts!
Kindest Regards,
John
--
--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .
MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules
---
You received this message because you are subscribed to the Google Groups "MedStats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to medstats+u...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/bc74aced-0bf4-4ad2-a7c3-cff4f340fa59n%40googlegroups.com .
John
----------------------------------------------------------------
Dr John Whittington,
Voice: +44 (0) 1296 730225
Mediscience Services
Fax: +44 (0) 1296 738893
Twyford Manor, Twyford, E-mail:
Joh...@mediscience.co.uk
Buckingham MK18 4EL,
UK
----------------------------------------------------------------
Reply all
Reply to author
Forward
0 new messages