View Full Version : A Digital Matter of Life and Death

Herman J. Woltring
01-12-1989, 09:02 PM
Dear BIOMCH-L readers,

While the item below is somewhat outside the target domain of our list (as
Biomechanics and Kinesiology are still very much a research endeavour, with
limited confidence on the part of many physicians), we all hope that our work
will be of benefit in public health matters. Therefore, I think that the sad
history of the Therac-25 Radiation Therapy Machine warrants dissemination over
our list as much as over the other lists mentioned below.

Regards -- Herman J. Woltring (List owner/editor, BIOMCH-L@HEARN)

[EARN/BITNET/NETNORTH users can subscribe to the Risks Digest at the listserver
LISTSERV@FINHUTC via the SUBSCRIBE RISKS "full_personal_name" command. Each
week, about two issues are distributed with contributions from individuals


health-l@irlearn !International Discussion on Health Research
!healthco@rpicicge !Communication in Health/Medical Context
medinf-l@dearn !Medical Informatics Discussion List
oncogene@irlearn !BIOSCI Oncogenes Bulletin Board
radsig@uwavm !Radiology Special Interest Group
support@rpicicge !Moderators for Healthco

RISKS-LIST: RISKS-FORUM Digest Wednesday 11 January 1989 Volume 8 : Issue 5

ACM Committee on Computers and Public Policy, Peter G. Neumann, moderator


Date: Tue, 10 Jan 89 12:04 EST
Subject: Medical software

A Digital Matter of Life and Death
by Ivars Peterson
Science News, 12 March 1988

The radiation-therapy machine, a Therac 25 linear accelerator, was designed to
send a penetrating X-ray or electron beam deep into a cancer patient's body to
destroy embedded tumors without injuring skin tissue. But in three separate
instances in 1985 and 1986, the machine failed. Instead of delivering a safe
level of radiation, the Therac 25 administered a dose that was more than 100
times larger then the typical treatment dose. Two patients died and a third
was severely burned.
The malfunction was caused by an error in the computer program controlling
the machine. It was a subtle error that no one had picked up during the
extensive testing the machine had undergone. The error surfaced only when a
technician happened to use a specific, unusual combination of keystrokes to
instruct the machine.
The Therac incidents and other cases of medical device failures caused by
computer errors have focused attention on the increasingly important role
played by computers in medical applications. Computers or machines with
built-in microprocessors perform functions that range from keeping track of
patients to diagnosing ailments and providing treatments.
"The impact of computers on medical care and the medical community is the
most significant factor that we have to face," says Frank E. Samuel Jr.,
president of the Health Industry Manufacturers Association (HIMA), based in
Washington, D.C. "Health care will change more dramatically in the next 10
years because of software-driven products than for any other single cause."
Samuel made his remarks as a recent HIMA-sponsored conference on the regulation
of medical software.
At the same time, reports of medical devices with computer-related
problems are appearing more and more frequently. In 1985, the Food and Drug
Administration (FDA) reported that recalls of medical devices because of
computer faults had roughly doubled over the previous five years. Since then,
the number of such complaints has risen further.
The problems range across a wide spectrum of computer-based medical
devices. A system designed for monitoring several patients at once was
recalled because it kept mixing up the patients. A programmable heart
pacemaker suddenly "froze" while it was being adjusted by a doctor. A device
for dispensing insulin delivered the drug at an inappropriate rate. An expert
system gave the wrong diagnosis, resulting in a patient receiving a drug
overdose. An ultrasound scanner sometimes underestimated fetal weight.
"No one can deny that allowing computers to perform some of the functions
normally carried out by trained and licensed medical professionals raises
questions concerning the personal health and safety of citizens," Michael
Gemignani of the University of Maine in Orono comments in ABACUS (Vol. 5, No.
1). "But even if we agree something more needs to be done to protect society
in the face of these technological innovations, we are still left with the
question: What should be done and by whom?"
The FDA, in its mandated role as guardian of public health and safety, is
now preparing to regulate the software component of medical devices. The
agency's effort has already raised questions about what kinds of products,
software and information systems should be regulated.
Last fall, the FDA published a draft policy for the regulation of computer
products marked for medical use. In that policy, the concept of "competent
human intervention" sets the dividing line between what is and is not
regulated. In other words, the computer product in question is subject to
regulation if a qualified doctor or nurse cannot effectively intervene to
override the machine's actions. Devices such as software-driven cancer therapy
machines, programmable heart pacemakers and automatic drug dispensers clearly
fall into that category.
On the other hand, the FDA states that it would not regulate computer
products that simply store, retrieve and disseminate information analogous to
that traditionally provided by textbooks and journals. In addition, the
agency's regulations would not apply to computer products used only for
communications, general accounting or teaching.
For example, a physician may use a computer program known as an expert
system to help make a diagnosis. Because the expert system does not directly
drive another medical device that, say, could dispense a drug when needed, and
because the doctor can make an independent judgment, such an expert system
would be exempt from FDA rules governing medical devices.
However, the greatest advantage of software - its flexibility - is also,
from a regulatory point of view, one of its biggest problems. Computer
programs are easy to change and can be used in many different ways. If
corrections are made or new features added, how much scrutiny should the
modified version of a previously approved computer product undergo? That
question is still unresolved.
Furthermore, it's sometimes hard to make a clear distinction between
programs that perform a "library" function and those that can be classified as
being part of a medical device. A case in point is the patient medical record,
traditionally a file folder containing various sheets of paper listing
treatments, medical observations and other pieces of information vital for the
patient's proper care.
Many hospitals are now moving toward medical records that are stored on a
computer. The difficulty arises when such information systems are connected
directly to machines that, for example, record patient blood pressure and heart
rate. If a nurse takes down the data and then enters the figures into a
computer, the information system software would not be subject to FDA rules.
But if the machine sends the data directly to the computer, then the
information system is considered by the FDA to be an "accessory" to a medical
device and subject the same level of regulation as the machine itself.
Information system vendors disagree with the FDA's position. They argue
that the FDA does not presently have rules governing the quality and content of
paper medical records. There's no reason for the FDA to start regulating such
records, they say, just because the records happen to be in a computer's memory
rather than on paper. In fact, using a computer-based system would
dramatically reduce the incidence of errors in patient records, the vendors
claim. The benefits of improved record keeping would clearly outweigh the need
for burdensome regulation.
The FDA's James S. Benson concedes that "regulation is not the automatic
solution to problems in hospitals and elsewhere." Nevertheless, the agency
must comply with a 1976 law that contains a broad definition of what
constitutes a medical device. Interpreted in its broadest sense, the
definition encompasses practically everything used in a hospital, from X-ray
machines to pencils.
FDA officials say they recognize the difficulties involved in regulating
medical software. "The agency fully appreciates the revolution occurring in
medicine with the introduction of computers and microprocessors," says Frank E.
Young, FDA commissioner. "We're taking a reasoned, structured approach with a
minimum of oversight. We have tried to give general guidelines. The policy
has been deliberately made flexible."
The flexibility allows the FDA to consider applications for approval on a
case-by-case basis. That limits the "chilling fear of undue regulation," says
Young. Furthermore, as technologies change and experience with computers in
medical applications grows, decisions on how much regulation is needed may also
To many manufacturers and users of medical products, the FDA's idea of
flexibility leaves too much uncertainty and opens up the possibility of
increased regulation in the future. "The FDA casts too wide a net," says
Edward M. Basile of King & Spalding, a law firm in Washington, D.C. "Their
basic assumption is that everything should be regulated."
"There's no disagreement about the extremes," says Harold M. Schoolman of
the National Library of Medicine in Bethesda, Md. "The question is how and
where to draw the line between the extremes." The important issue, he says, is
maintaining a balance between appropriate safeguards and incentives for
Even in situations where it's clear that certain software ought to be
reviewed, the FDA faces the additional difficulty of how to go about verifying
that a particular computer program does what it's supposed to do -- nothing
more, nothing less. As experience with software for other applications has
shown, the task of checking software quality can be overwhelming (SN: 9/13/86,
p. 171).
A few years ago, when most medical devices did not contain computers, it
was relatively easy to foresee all possible inputs and to check the
consequences of each one, says James Howard of General Electric's Medical
Systems Group in Milwaukee, Wis. With computers, the number of possible paths
is greatly increased. "It's more important than ever to build safe products
that perform as required," he says. But because a detailed analysis takes so
long, it often can't be done. "This is a major concern to both manufacturers
and the FDA," says Howard.
The FDA defines software as a "set of instructions that enables a
computing machine to control, monitor or otherwise interact with a medical
device." The proposed regulations require a software developer to show that the
algorithm, or mathematical recipe, used in the computer program is appropriate
and has been implemented correctly in the software. The FDA also requires
assurance that any software failure would not injure the patient.
How that assurance can be provided is still unclear. Techniques for
evaluating software safety are relatively new. Who does the checking, how much
evidence is enough and whether the FDA can perform an independent check are
also unresolved issues. Furthermore, software developers are wary of
submitting complete listings of the instructions in their computer programs
because competitors may get a look at this "source code" by making a request to
the FDA under the Freedom of Information Act.
The trouble with the FDA approach, says Howard, is that it doesn't
consider under what conditions software is used. Instead, the FDA ought to
focus on the idea that not all computer errors are equally serious. Using a
kind of hazard analysis to focus on situations that could lead to
life-threating computer failures would be one way to eliminate the most serious
potential faults and to shorten testing times.
Software developers also need to improve the methods they use for
constructing computer programs. We need to "industrialize" software
development so that programs are written in a consistent way, says James
Dobbins of Verilog USA, Inc., in Alexandria, Va. Too often, programmers
include a description of what each part of a program does only as an
afterthought. They rarely go back to clean up or polish a program to make it
more understandable.
Software development can be standardized and automated, says Dobbins.
"The tools are there to industrialize the whole process. You just have to go
find them."
Programmers, on the other hand, complain that they're in a no-win
situation. Software is continually modified as it evolves, often to meet
demands for new features to make the product more competitive. In the rush to
market, when delays can put a company at a competitive disadvantage, software
testing often loses out. Delays in completing a software package are balanced
against the possibility of failing to root out potentially embarrassing errors.
This is the kind of situation that can lead to lawsuits, says Vincent
Brannigan, an attorney in Adelphi, Md. Software is clearly a product, he says.
If it's defective and injures a consumer, then the manufacturer is liable.
Among the faults Brannigan lists is the tendency of software and computer
companies to promise more than they can fulfill and to cut costs by doing less
testing. This is the only field, he says, where the customer is expected to
pay for finishing a product through the purchase of periodic updates and
corrections to the software.
"Disclaimers don't mean anything," Brannigan says. "The product should
have been right in the first place." That means paying much more attention to
how software is written and tested. "The software must look as shiny and clean
as the rest of the machine," he says.
So far, software developers have generally escaped damaging lawsuits and
settlements, but that may change. To many medical-device producers, the threat
of litigation may be even more effective than proposed FDA regulations for
assuring the quality of products.
Even finding out what went wrong is a time-consuming process. The FDA and
other groups are still investigating aspects of why the Therac 25, manufactured
by Atomic Energy of Canada Ltd. in Kanata, Ontario, failed. What's evident is
that the problem could probably have been avoided if an appropriate safety
analysis had been done.
The Therac 25 delivers two forms of radiation: either a high-energy
electron beam or, when a metal target intercepts the electron beam, a
lower-energy X-ray beam. It turns out that when a nimble, experienced
technician punches in a particular sequence of commands faster than the
programmers had anticipated, the metal target fails to swing into place.
A safety analysis would have identified the missing target as a
potentially dangerous situation. The machine could have been programmed so
that it couldn't operate if the target, as confirmed by a sensor, were not in
Perhaps such a complex, computer-driven machine wasn't even necessary. By
sacrificing a little convenience and flexibility, a machine with a simple
on-off switch and a timer could probably have done the same job - with a much
smaller chance of failure.

[This is a familiar topic to RISKS readers, but this particular article
is extremely well written and seems worth including, even if old.
(RISKS has reported one additional death involving the Therac.) PGN]


End of RISKS-FORUM Digest 8.5