Re: Control charting for insulin monitoring.

[kromkowski@aol.com]

In response to Robert Shaw (quoted below):

I presume all of this you know given your background, but I
state it to create a context:

1.  As a matter of context, it is important to remember:
the diabetic is already "out of control" with respect to 
insulin; hence, they are a diabetic (even a temporary one) 
rather than a non-diabetic.

2. As Deming pointed out the use statistically information 
must also be in the context of the underyling 
scientifically understood causal mechanisms at work.  
For example, in a chemical plant the use of statistical 
information without an understanding of the laws of
physics and chemistry involving time, pressure, temperature, acid/base chemistry, etc., etc. will be 
futile.  Here, there are known causal mechanisms primarily 
involving diet, activity and the sleep cycle (because you 
don't normally eat while you are sleeping), weight,
size.

3. The life and death of people is potentially at issue so 
once again we see Shewart's and Deming's claim that applied 
science must be more exacting than what might be called 
normal science.  

4. The purpose of the control chart is to help us
distinguish between special causes of variation and 
systemic causes of variation each of which require 
different approaches.  

5.  There is a difference between specifications or 
tolerances and what is statistically out of control.  You 
speak of data falling outside the 80 to 110 glucose range 
as being out of control.  But this is an inaccurate 
description.   Control limits are derived from the system 
(the person in this case) producing the data.  The 80-110 
range is not a control limit, it is a specification
or tolerance.  For example,  a sytem may produce widgets 
that need to have an opening between 80 and 110 mm.  If the 
widget is beyond these specifications, it is defective and 
something must be done to the widget or the machine that
needs the widget will not work.  The control limits for the 
widget producing system are not necessarily 80-110 
(although it would be nice if was such or 90-100).
The control limits for the widget producing system can only 
come from data produced by the widget producing system.

6.  In many ways, the adminstering and adjustment of 
insulin doses is akin to re-aiming the funnel (I presume 
you are familiar with the funnel experiment).  Theory
and practice without the special context of medical 
knowledge would suggest that one should not re-aim the 
funnel because this will make variation worse and that
the only thing that should be worked on are system causes 
(the height of the funnel, the rollyness of the table, 
etc.  BUT here the underlying science/medicine
hits fan, because if an adjustment (insulin dose) is not 
made then people die.  In that regard, we/you really need 
to understand what a 111 or 79 means. 

So what to do?

First, make some actual data available to this forum.  I 
seriously doubt that you need to transform your data.  I 
suspect that an Ind X/Mr will do the trick as a starting 
place.

Second, there needs to be a better understanding of how the 
80-110 range was derived and the medicin of what a 111 or 
79 once in a while really means.  If you measured non-
diabetic people what would the control chart look like.  
Measure yourself or have the nurses measure themselves at 
the same time that they measure the patients to get 
some data. 

Third, we need to understand the possibilites and the 
medicine of the data. E.g.,We can have data showing:

Ideal Patient A.   Centered at 100 and with control limits 
within 80 and 110.

Patient B.   Centered above 100 but with control limits 
still within 80 and 110.
Is this medically acceptable based on the known science. Is 
a patient between 100 and 110 "better", "worse" or 
medically no different than the patient between 80
110?

Patient C.   Centered below 100 but with control limits 
still within 80 and 110.
The 80-100 patient.  Is this medically acceptable?

Patient D.   Centered above or belaw 100 and with control 
limits at or wider than  30 apart (110-80=30).  I.e.; part 
in the 80-110 range and part out of the range. Put
another way one control limit within tolerance and another 
outside. 

This raises the possibility of a static like % within the acceptable range.  

For example, is a patient who is data predictably falls within 81 and 111 really a problem or not, based upon the 
medicine/science.  What is the likelihood that changing 
insulin protocal will push a statiscally stable patient
into a statistically unstable situation.

Patient E.  Centered at or near 100 but with both control 
limits beyond specifications.

Patient F.  Both control limits outside on the low side or 
both outside on the high side.
I.e.,  0% falling within the tolerances or specifications.  
Here you might in theory (but I doubt in reality) have a 
pretty tiny variation always between 110.1 and 115 or 
between 77 and 79.9.  What does the medicine/science say 
about this condition?  It suggests to me at least that the 
body may well be appropriately self regulating but that 
there is a normal variant, who really knows. Not me.

As to chart, a few present themselves but all are dependant 
on the underlying science. 

Ind X/ Mr -   

X-bar/R   Does the 80 to 100 represent a daily range 
accounting for food and time of day, if so the subgrouping 
might be as simple as the day.  I have coefficients for N 
up to 10, they go higher.  How many readings are taken a 
day?

And then a p-chart. Where you'd be measuring percentage of 
data within the range. Here the problem is the N is 
probably going to be small on a daily basis unless the 
technology is now such that with little discomfort to 
patient and cost for testing you are getting readings every 
few minutes.

Again, it would be great learning experience if some raw 
data were available, especially data regarding what a 
normal person's reading look like.

John David Kromkowski
Attorney at Law
6600 York Road - Suite 108
Baltimore, Maryland 21212-2028
Kromkowski@aol.com Telephone: (410) 377-6248 Facsimile: (410) 372-0624

Message posting through the Clemson CQI Web Server.