[Reply posted to OP and to edstat. -- DFB.]
Difficult to know how to respond. Apparently this is not an experiment,
in which quantities of particular foods ingested, and the times at which
they are ingested, were systematically controlled by the experimenter.
So we have what we might call "casual" data, recorded as in a narrative.
Among the things that are unclear: Do you have data like this for a
number of diabetic persons, or only for one? If for several, do you
seek a general model that would apply (one might hope!) to the general
class of diabetics, or do you seek individual models for each person,
which might differ substantially from person to person?
To carry out any kind of sensible analysis, some kind of model (or set
of models) is needed. You write, for instance, that "The periodicity
of an insulin response to food is around an hour to three hours.", which
I interpret to mean that the possible lag time between ingesting insulin
and a response to it (in the form of glucose level?) is between 1 and 3
hours. (You may have meant something else, but presumably the idea of
"lag time" would apply in any case.) Do you wish to model this lag time
as constant, or as variable depending on other conditions (and what
conditions?)?
The column headed "item" appears to identify types of food -- that is,
it contains a list of variable names. Better to assign one column to
each variable; it's really difficult to disentangle values under the
structural arrangement shown. I'd propose something like the following,
with a variable labelled "other" because surely all the food a person
eats cannot be classified into {sugars / proteins /complex carbonates}
and nothing else; and possibly there may be interesting effects of some
as yet undefined food substances on glucose level, which you would want
to know about, at least partially, if they exist.
date time ------------ substance ingested --------------- glucose
sugars protein cmplx-crb INS_A INS_B other level
gm gm gm units units
3/13 06:15 223
3/13 06:32 9 245 432
3/13 06:46 248
3/13 07:54 265
3/13 09:47 232
3/13 11:45 187
3/13 13:19 25
3/13 13:23 196
3/13 13:32 385
3/13 13:32 425
3/13 14:46 247
3/13 15:45 20
3/13 16:30 187
... and so on. Note the 24-hour clock. (You may want to revise the
way in which time is recorded, or the form in which it is used in your
analyses, but at least you need to be able to distinguish AM from PM.)
On Tue, 30 Mar 2004, Jim Kroger wrote:
> Hello, I'm trying to develop a way to determine the effect of
> different kinds of food and insulin on glucose (blood sugar) for
> diatetics. Different foods elevate it to different degrees, and
> different kinds of insulin lower it to different degrees. What kind of
> analytic method would allow one to determine
> 1) what the effect of each kind of food or insulin alone is on blood
> glucose
What do you mean, "alone"? Do you have instances of only one (type of)
food being ingested at any particular time? If not, you can hardly
assess the effect of any kind of food "alone", but only as the effect of
assorted combinations.
> and 2) what the interaction is between two elements (for example,
> perhaps a gram of sugar has less effect when consumed with a gram or
> protein than when consumed alone)?
Without clearer experimental control, I doubt whether you can assess
this at all. E.g., at 6:32 on 3/13 you have 9 grams of sugar being
consumed along with 245 grams of protein and 432 grams of complex
carbonates: nothing at all like "a gram of sugar ... consumed with a
gram of protein". For the data you report, you surely need to model
some kind of dose/response, AND some kind of dose/response interactions
for the various combinations you want to consider.
> Also, I'm curious if there is a need to sample glucose with any
> particular time resolution to use the analysis, such as at least every
> hour.
This is not a statistical question, and cannot be answered by
statistical considerations. What does your understanding of diabetic
biochemistry tell you about that? You can of course try various levels
of temporal resolution, and see whether interesting patterns emerge in
data taken, say, hourly, that are not visible in data taken, say, daily.
But whether the hourly data will miss noteworthy effects that can only
be seen in measurements made at (say) 10-minute intervals: who can say,
without measuring at 10-minute intervals?
> The periodicity of an insulin response to food
> is around an hour to three hours.
> Thanks. Here's an example of a (tab delimited) data file, though the
> structure is somewhat arbitrary at the moment.
>
> date time item grams units glucose
> 3/13 6:15 223
> 3/13 6:32 sugars 9
> 3/13 6:32 protein 245
> 3/13 6:32 cmplx carb 432
> 3/13 6:46 248
> 3/13 7:54 265
> 3/13 9:47 232
> 3/13 11:45 187
> 3/13 1:19 INSULIN A 25
> 3/13 1:23 196
> 3/13 1:32 protein 385
> 3/13 1:32 cmplx carb 425
> 3/13 2:46 247
> 3/13 3:45 INSULIN B 20
> 3/13 4:30 187
>
> and so on for many days
------------------------------------------------------------
Donald F. Burrill [EMAIL PROTECTED]
56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
