Re: [Gnumed-devel] measurement tooltip

[Jim Busser]

On 21-Jun-09, at 3:11 AM, Elizabeth Dodd wrote:

On Sun, 21 Jun 2009, Karsten Hilbert wrote:

BTW, I chanced upon this idea when I did the very same calculation

by hand two days ago on a certain patient's data - including the

bandwidth deviation.

Karsten

 no, I don't think about things that way, and agree with Jim.

I understand what Karsten is getting at, and encourage ongoing possible "eurekas" on things that may be clinically informative. Unfortunately, the bandwidth idea IMO needs to be retired, I am sorry to have to say :-(

The concept may be that a result that is only slightly outside of a test range that is normally tight, it is more confidently pathological than a result that deviates, with higher amplitude, from a range that is wider. However, I have seen many results of the anticoagulation test INR which has a reference range of 0.9-1.0 or 0.9-1.1 come back at 1.2 or even 1.3 at which I sometimes raise an eyebrow as to the latter but never did it end up connecting to any pathology and that is a result deviating by 3x (300%) of the bandwidth.

Also the concept being pursued is something like a standard deviation but the problem is one cannot judge dispersion from a knowledge of only the limits for, in the case of one test, most humans may be nearer the centre of the range and yet, and in another instance, more widely dispersed within the limits depending on how many internal (genetic) and external factors are operating. Also we do not readily know whether, in the case of any one test, its upper and lower limits were set based on 95% (2 standard deviations *assuming* Gaussian distributions) or whether they were set by the laboratory based on thresholds determined from pathologic study for example most tumor markers... there is no normal or reference range, only an upper limit with lower limit zero. So in summary I would say: keep thinking, more (and sometimes brilliant) ideas will come !

PS after enough people had been using GNUmed, we could always contemplate a research study that could be based on database queries of whether test abnormality as determined relative to bandwidth was a predictor of reliability or severity of pathology. And I would then gladly concede to the prescience. But maybe only then ;-)