---------------------------------------------------------
ROY M. POSES MD
BROWN UNIVERSITY CENTER FOR PRIMARY CARE AND PREVENTION
MEMORIAL HOSPITAL OF RI
111 BREWSTER ST.
PAWTUCKET, RI 02860
USA
401 729-2383
FAX: 401 729-2494
[log in to unmask]
----------------------------Original message----------------------------
2x2 tables are a numeric representation of single covariation. For example
the relationship of the pulse to the outcome of an asthma attack can be
analysed by 2x2 tables. Naturalist epistemologists would argue that such
representations are abnormal for humans. They argue humans see the world
in terms of multiple co-variation rather than single covariation. In other
words a GP will see a child is "sick and needing to be admitted", but will
not see a particular sign in that child. I did an audit of asthma care in
our area and found that the majority of GPs referred patients to hospital
who were in the appropriate range of severity for asthma (no admitted
patients were "well"), but when we looked at the GP notes we found that
there were many pertinent details not recorded - for example the pulse was
rarely recorded. Now when we asked GPs about the pulse, everyone claimed
the pulse is an important sign to record. Perhaps these results reflects a
universal inability by gps to record single items of data. Indeed we found
individual data items were recorded more often when we introduced an asthma
data recording form - a change from 48% to 68% of patients with acute
asthma had their pulse recorded. However many GPs commented that "we dont
go by the pulse" or "I judge what to do by how sick the child looks". So I
guess the influence of observation at a level of multiple covariation is
still present.
----------------------------------------------------------------------
Studies in judgment and decision psychology suggest that people have poor
insight about the factors they actually use to make judgments. Clare Harries'
studies of physicians in the UK showed similar results: the variables doctors
say are important to making a particular judgment are not the same as the
variables they seem to use when making judgments for individual cases. (I can
find the reference if anyone wishes.) So I guess I'm not surprised
Another question is whether these physicians' judgments were accurate, and
whether they lead to good decision making. If no admitted patients were well,
were any patients who were not admitted sick?
-----------------------------------------------------------------------
It takes training to focus on a single co-variation. 2x 2 tables are
"abnormal" in the sense that we do not naturally see single co-variation.
It is for this reason that it is up to those of us who want to improve on
the use of 2x 2 tables or likelihood ratios, to make them easy to use and
make them part of the "habits needed to be a doctor".
Unfortunately statistical manipulation of multiple co-variation is
extremely complex. Furthermore, such modelling is likely to remain as
academic interest for a while yet - at least as long as it remains slower
than a GP noticing (almost instantly) that a child is "sick".
------------------------------------------------------------------------
I'm not sure it's a good idea to focus on single co-variation. It makes more
sense to try to use all relevant variables when making a judgment (or a
decision.) Multivariate statistical models are complex. There are at least
some that have proven to be good predictors of specific clinical outcomes.
Although it takes a lot of effort to construct such a model, using such a
model is only slightly slower than making an intuitive judgment. If such a
model is more accurate than intuitive judgment, and it is important to judge
the outcome in question to make a decision, maybe physicians should learn to
use such models.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|