View Full Version : Data pooling: the other side of the medallion

Herman J. Woltring
01-06-1992, 09:39 AM
Dear List readers,

Following my recent posting on Jim Gage's highly interesting book on Gait
Analysis in Cerebral Palsy and his enthousiasm (shared by myself) on improving
reliable diagnostics by pooling data from multiple data collection centres, I
think that the following posting from the Risks Digest (Volume 13, issue 01)
as received today deserves to be read, too.

Herman J. Woltring, Eindhoven/NL


Date: Sun, 5 Jan 92 13:21 EST
From: "Warren M. McLaughlin"
Subject: Life-and-Death Computer

The Washington Post, 5 Jan 1992, page C6 (the editorial page):

As technologies become more powerful, the distinction between a helping tool
and a decision-making tool keeps gaining importance. Nowhere is this clearer
than in the case of the new diagnosis-aiding computers, which offer doctors the
benefit of a gigantic data base -- far larger than their own experience could
be -- compiled from the results of many thousands of cases nationwide. By
conglomerating and analyzing the results of these cases, the computer can read
out a series of alternative treatments, a probability rating on the success of
a given procedure or -- most controversially -- the statistical risk of a
patient's dying upon arrival in an intensive care unit in a given condition.
Physicians with access to such a machine now bear a responsibility at least as
weighty as that of diagnosis itself: that of balancing the computer's seemingly
precise numbers and instant certainties with the knowledge that its results are
dependent upon human judgement.

According to the staff in a Michigan hospital using a program of this
type called APACHE, the computer's predictions of a patient's statistical
probability of dying -- calculated to two decimal points -- are used strictly
as tools, rather as any doctor might estimate, say, a 10 percent chance of
survival from a given operation. A better description of risk, in that sce-
nario, need not govern the doctor's (or the family's) decision as to whether
the risk should be taken, only inform it better than individual experience
ever can. But the incomplete results of a different study performed in France
suggested that doctors with access to that kind of risk data were more likely
than others to terminate care. The fear among practitioners is that hospital
administrators or health bureaucrats, all increasingly beleaguered and pushed
by public pressure toward cost-cutting, might see computer-confirmed statis-
tics on death risk as a road to easier triage.

Given the capability for vastly enhanced diagnosis by means of computers, the
medical profession will be stuck with the same responsibility -- also vastly
enhanced -- as before: first, to recognize that a computer can serve the cause
of accurate diagnosis only on the basis of properly entered information by the
physician using his or her senses; second, to keep in mind a fact much of the
general public has trouble with, which is that a statistic about the probabi-
lity of an event bears no causal relationship to that event. A person with a
95 percent chance of dying under a procedure is not the same thing as a person
whom that procedure cannot help, or a person from whom care can be withheld
with no compunctions. Obscure that distinction, and you take a step toward
making the computer the master -- a bad one.
- Mike
PO Box 54, Bridgewater, VA 22812-0054

(Xposted without permission -- HJW)