Health Care Medicing- Complexity and Error in Medicine

Health Care Medicing- Complexity and Error in Medicine

Harvard Business School

9-699-024

Rev. April 28, 2000

Complexity and Error in Medicine
The practice of medicine centers on caring for people’s health and, where possible, curing
them of disease. It is a basic tenet that the caregivers should not themselves cause harm. This is
enshrined in the adage learned by all first-year students, “Primum non nocere,” first do no harm.
Nonetheless, it is increasingly apparent that error is endemic in medicine and that at least some of
these errors result in preventable harm to patients. It has been said that if the airline industry
tolerated the error rate that the health care industry does, there would be three “jumbo” jet crashes
every two days.1 Widespread recognition within the health care industry of the extent of medical
error has been slow in coming, as have the development of technological and, especially,
management responses to manage error risk. This note will examine the issues underlying the error
rate in medicine and the range of possible action that managers can take to decrease the likelihood of
a harmful error.

The Extent of the Problem
When a patient suffers some kind of injury that is the result of his or her care (either the care
that was given or the care that was omitted), it is termed an “adverse event.” Thus, adverse events
are a subset of the potential outcomes of health care. They can be classified according to the type of
care with which they were associated (e.g., adverse drug events, surgical complications) or by
whether they are potential, preventable or nonpreventable adverse events. Potential adverse events
are incidents with the potential to cause injury (regardless of whether the patient actually suffered an
injury). En example of this is the delivery of a dose of penicillin to a patient known to be penicillin
allergic, even if the patient did not have a reaction to the drug. Preventable adverse events are those
where there has been an error (e.g., when a patient is given the wrong drug) or that were preventable
by any means currently available. Nonpreventable adverse events are those that are not related to an
error. An example of a nonpreventable adverse event is when a patient, not previously known to be
allergic to penicillin, has a reaction to penicillin. In this case the allergic reaction could not have been
predicted and therefore the prescription and delivery of penicillin was not an error. Nonetheless, the
patient did suffer harm and therefore had an “adverse event.”2

1 Leape L. Error in medicine. JAMA 1994; 272 (23): 1851-1857
2 Bates DW, Cullen DJ, Laird N, et al. Incidence of adverse drug events and potential adverse drug events:

Implications for prevention. JAMA, 1995; 274: 29-34.

This note was prepared by Professor Richard Bohmer for the purpose of aiding class discussion. It may be used in
association with “Dana-Farber Cancer Institute,” HBS Case #699-025.
Copyright © 1998 by the President and Fellows of Harvard College. To order copies or request permission to
reproduce materials, call 1-800-545-7685, write Harvard Business School Publishing, Boston, MA 02163, or go to
http://www.hbsp.harvard.edu. No part of this publication may be reproduced, stored in a retrieval system,
used in a spreadsheet, or transmitted in any form or by any means—electronic, mechanical, photocopying,
recording, or otherwise—without the permission of Harvard Business School.
1

699-024

Complexity and Error in Medicine

Most adverse events involving an error are potentially preventable. Not all errors result in
adverse events, because many errors do not cause harm to the patient; these are called potential
adverse events. Although there is no universally accepted taxonomy of error, the one illustrated in
Figure A is commonly used.
Medical error is notoriously difficult to quantify. As noted, not all adverse events are the
result of a medical error. When an error does occur it may go unrecognized or unreported. Patients
who are seriously ill are already at high risk for an adverse outcome. Adverse outcomes that are the
result of an error may not be recognized as such because an adverse outcome was likely. Errors that
do not result in immediate harm to patients may go unnoticed or unreported. Clinical staff
(physicians, nurses therapists) may be reluctant to report the errors they observe for fear that they
may be reprimanded by the institution or be sued for malpractice.

Figure A: The relationship between errors, potential adverse events, adverse events, and preventable
adverse events. The ratio of all errors to serious errors [potential adverse events+adverse events] is
10:1. There are 2 potential adverse events to every preventable adverse event and 2 preventable
adverse events to every nonpreventable adverse event.
A number of recent studies have attempted to quantify the extent of errors and adverse
events and reported substantial variation in these outcomes. This variation is due, in part, to the
different definitions they use for an adverse event and an error. It also reflects the difficulties in
collecting data on error. The most widely quoted study in recent years is the Harvard Medical
Practice Study.3 In this population-based study the charts of over 30,000 patients hospitalized in
New York State in 1984 were screened by nurses, and suspicious cases then were reviewed by two
physicians. The definition of an adverse event used was very stringent. An adverse event was
defined as an injury that prolonged hospital stay by one or more days or resulted in measurable
disability (including death). Adverse events that did not result in increased length of stay or

3 Brennan TA, Leape LL, Laird N, et al. Incidence of adverse events and negligence in hospitalized patients:

results of the Harvard Medical Practice Study I. N Engl J Med. 1991;324: 370-376.

2

Complexity and Error in Medicine

699-024

significant disability, such as short-lived allergic drug reactions, were not included. It is likely,
therefore, that this study identifies the lower bound of the frequency of adverse events.
Overall, 3.7% of patients suffered an adverse event. Of these, 69% were judged by the
physician reviewers to be due in part to errors in patient management. Over 13% of the injuries were
fatal, a rate that if extrapolated to the Unites States as a whole suggested that approximately 180,000
deaths a year were, at least partly, the result of injuries received during the course of care. Table A
shows the sources of the adverse events.4
Other studies have found higher adverse event rates, e.g., 16.6% (51% of these were
considered preventable) in an Australian population-based study5 and 17.7% in a single Chicago
hospital.6 A further study by the Harvard researchers went into greater detail for a subset of adverse
events, adverse drug events (ADEs).7 In this study, based at two large Boston academic teaching
hospitals, an adverse drug event was classified as either an ADE (an injury resulting from medical
intervention related to a drug) or a potential ADE (incidents with the potential for injury related to a
drug). Adverse drug events were detected using a combination of chart review, staff report, and
daily site visits to the study wards. The rates measured were 6.5 ADEs and 5.5 potential ADEs per
100 patient admissions (excluding obstetric admissions). Twenty-eight percent of the ADEs were
judged to be preventable.
Error Rate

Frequency (%)

Overall

3.7

Non operative

52.3
Drug related

19.4

Diagnostic mishap

8.1

Therapeutic mishap

7.5

Procedure related

7.0

Other non operative

10.3
47.7

Operative
Wound infection

13.6

Technical complication

12.9

Late complication

10.6

Other operative

10.6

Table A: Sources of adverse events
Hence, while it not clear what the “true” adverse event, adverse drug event, and error rates
actually are, it is clear that the harder one looks the more one finds. Rates of the order of 10% are
very plausible.

4 Leape LL, Brennan TA, Laird N, et al. The nature of adverse events in hospitalized patients: results of the

Harvard Medical Practice Study II. N Engl J Med. 1991;324: 370-376.

5 Wilson RM, Runciman WB, Gibberd RW, et al. The quality in Australian health care study. Med J Aust. 1995;

163: 458-471.

6 Andrews LB, Stocking C, Krizek L, et al. An alternative strategy for studying adverse events in medical care.

Lancet. 1997; 349: 309-313.

7 Bates DW, Cullen DJ, Laird N, et al. Incidence of adverse drug events and potential adverse drug events:

Implications for prevention. JAMA, 1995; 274: 29-34.

3

699-024

Complexity and Error in Medicine

Errors are more likely to give rise to an adverse event in some clinical areas than others. For
example, some classes of drugs are simply more dangerous and are more likely to be associated with
an adverse event than others (the most common drugs associated with an adverse event in the Boston
hospital study are shown in Table B).

Drug class

Frequency (%)

Frequency (%)

USA

Australia

Antibiotic

16.2

12.9

Antitumour

15.5

9.4

Anticoagulant

11.2

10.7

Cardiovascular

8.5

11.6

Antiseizure

8.1

-

Antihypertensive

5.0

8.2

Other

35.5

-

Table B: Sources of adverse events

Where Do Errors Originate?
If it is a difficult task to define and quantify the extent of error in medicine, it is even more
difficult to unravel the underlying causes of medical error. Discussions of error tend to focus on one
of three levels of analysis: human factors (the functioning of individuals), the functioning of complex
systems, and the contribution of organizational factors to the potential for making and identifying
errors. When thinking about error risk, it is also useful to differentiate those factors associated with
the probability of committing an error and those associated with the likelihood that an error will be
detected and redressed.
(1) Human Factors
Human factors, in particular the ways individuals process information, contribute to both
error commission and error detection. In his classic work, “Human Error,”8 James Reason identifies a
number of ways in which individuals make errors, each of which has a differing underlying cognitive
precursor (Table C).
Cognitive “mode”

Focus of attention

Underlying “cause”

Slip

Error type

Automatic
“unintentional”

On something other than
the task in hand

Inattention

Mistake

Problem solving
“intentional”

Directed at problem
related issues

Inappropriate rules
applied or appropriate
rules misapplied

Problem solving
“intentional”

Directed at problem
related issues

Inadequate knowledge or
failure of heuristic

Rule-based
Mistake
Knowledge-based

Table C: Types of error

8 Reason J, 1990. Human Error. Cambridge University Press, Cambridge England.

4

Complexity and Error in Medicine

699-024

Broadly speaking, humans function cognitively on two levels—automatic or problem
solving. Thinking in the “automatic mode” is rapid, unconscious, and effortless. The thinking
required to drive a car is often automatic. In more complex situations, or situations that we have not
encountered previously, we move into “problem-solving mode.” Here, thinking is conscious, slower,
and more arduous. Rule-based problem solving is for dealing with familiar problems by applying
stored rules, based on prior experience (i.e., if [situation X exists] then [do Y]) that have previously
been successful to new situations. Knowledge-based problem solving is reserved for out of the
ordinary and unfamiliar situations where there are no rules to apply and new solutions must be
created in “real time” using conscious analysis and stored knowledge.
The importance of these types of cognitive processes is that failure in each of them gives rise
to different kinds of error, each with implications for different error-prevention strategies. Slips are
errors of automatic processing and occur during routine actions. The most common sub-type is
called “double capture” and occurs when you intend one action but habit intrudes and you do
another. An example of this is setting out to drive to the beach and finding yourself driving to work
instead. Other sub-types include “perceptual confusion” (the microwave beeps and you start to
answer your pager) and “omissions associated with interruptions” (you go into the living room to get
a book, the phone rings and you answer it. When you hang up you leave the living room without the
book). The likelihood of these types of error is increased by, among other things, fatigue,
interruptions, and anxiety. All of these are common in the working lives of nursing and junior
medical staffs of modern hospitals. Slips are fairly common but are usually detected and intercepted.
While they provide many opportunities for error they actually contribute to relatively few accidents,
though when they do occur these can be serious and even fatal. There are many examples of this
kind of error in modern hospitals, for example:
At 4.00 AM a junior doctor is told by a patient that she is allergic to penicillin.
Moments later he writes an order for the drug.
A nurse inadvertently connects a gastrostomy feeding solution to a central
venous line. The patient suffers irreversible brain damage.9
An oncology fellow writes a chemotherapy dose for the total 4-day course of
treatment on the form for the daily drug administration. The patient receives 4
times the daily dose each day for 4 days and subsequently dies.
Mistakes are failures of problem solving. These are less common and less easily detected.
Detection usually requires external intervention. They are also less easily intercepted because they
are purposeful and it is therefore difficult for an individuals to recognize their own mistakes without
external reference. Rule-based mistakes occur when the situation or problem is misclassified and
then either the wrong rule is applied to this situation or the right rule is misapplied. According to
Tversky and Kahneman,10 when faced with an uncertain situation we do not dispassionately review
it, rationally evaluate the data and possible diagnoses and courses of action, and finally come to a
reasoned judgment. Rather, we apply one or more rapid rules of thumb.
The most well known of these rules are the representativeness heuristic (the assumption that
like cause like) and the availability heuristic (judging things as more frequent because they spring to
mind more readily). In using the representativeness heuristic we compare the current situation to
past experience and training. In medicine the “classical case” described in textbooks is a frequent

9 Gastrostomy feeding solution is placed directly into the highly acidic stomach and is used to provide

nourishment for patients who cannot eat. An alternative route for delivering nourishment is directly into a vein,
but this solution is of a very different composition to that placed into the stomach.
10 Tversky A, Kahneman D. Judgment under uncertainty: heuristics and biases. Science 1974; 185: 1124-1131.
5

699-024

Complexity and Error in Medicine

reference point. This allows the clinician to aggregate a large number of variables into a clinical
decision. However, it presumes that the past is an adequate predictor of the future. This
presumption may narrow a clinician’s thinking and blind him or her to the possibility of new
diseases. In fact, few cases are truly classic. A clinician faced with a new or uncertain situation who
falls back on the representativeness heuristic to guide decision making may be inadvertently limiting
his or her range of options.
The availability heuristic describes the way in which familiar and common events guide our
decision making. A given set of clinical information should suggest a more common rather than a
rarer diagnosis. However, frequently we are prone to recall, instead, rare but vivid events and use
these as the basis for interpreting a new and uncertain situation. For example, a cardiac surgeon may
use the drug lidocaine as prophylaxis for ventricular arrhythmias on all post-coronary artery bypass
graft patients because he recalls a case in 1973 when he did not and the patient died of a ventricular
arrhythmia. Another example would be a physician estimating the probability of bacteria in the
bloodstream of an ill patient. Physicians tend to overestimate this probability (and order tests
accordingly) if they have recently cared for a patient with a bloodstream infection. 11 Knowledgebased mistakes occur because the human mind has limited capacity as a processing unit and can only
deal with a limited number of variables and relationships at any one time. Imperfect and incomplete
knowledge and the tendency to attend to psychologically salient rather than logically important
information (the availability heuristic) compound this problem.
In summary, at one level individual humans are prone to make errors because of the ways in
which they process information and make decisions. Under normal operating conditions these
cognitive mechanisms are entirely appropriate and are in fact the reason that we make so few errors.
In certain circumstances however, such as physical and mental stress (e.g., fatigue, anxiety),
situations of ambiguity, and lack of information and knowledge, these same cognitive mechanisms
contribute to an increased risk of error.
Gaba12 has applied some of this body of work to anesthesiologists working in the operating
room. He notes that in anesthesiology at least 18% of cases will involve an unanticipated problem
that will require an intervention by an anesthesiologist, and 3% to 5% will involve a serious
unplanned event requiring a major intervention. At any time an anesthesiologist may face an
unexpected and unfamiliar situation, that requires rapid decision making. In the context of routine
work the anesthesiologist must observe a data stream, recognize that there is a deviation from what is
expected and, furthermore, that this deviation may lead to a bad outcome, propose and implement a
corrective action, and evaluate the effects of that action. There are two possible ways to plan
corrective action. “Pre-compiled” responses are those that are based on prior experience (e.g., “the
blood pressure is dropping, therefore I do the following …”). This rule-based decision making is
rapid. Alternatively, where the situation is one the anesthesiologist has not faced before, the solution
may require reasoning from fundamental medical knowledge and will be slower. Failures can occur
at any stage in this process; not observing the data stream, not recognizing deviations from what is
expected, not recognizing the deviation to be a potential problem, and selecting the wrong
intervention. Anesthesia is one area that lends itself to the use of training simulators. These give
practitioners the opportunity to practice managing unexpected events. “Prior experience” can be
gained and pre-compiled responses tested in the simulator. Interestingly, Gaba notes that when
anesthesiologists are observed responding to simulated crises they predominantly use pre-compiled
responses.

11 Poses RM, Anthony M. Availability, wishful thinking, and physicians' diagnostic judgments for patients with
suspected bacteremia. Medical Decision Making 1991; 11: 159-168
12 Gaba DM. Human error in dynamic medical domains. In Bogner MS (ed.), Human error in medicine.
Lawrence Erlbaum Associates. New Jersey, 1994.

6

Complexity and Error in Medicine

699-024

(2) System Complexity
Individuals do not function in isolation. They interact with the systems in which they work.
The design and complexity of these systems are in and of themselves a contributor to the potential for
error. Systems can fail and systems can induce humans to fail. Importantly, systems can be designed
to be more resistant to human failure and less likely to allow humans to fail. In one sense the
inadvertent connection of a gastrostomy feeding solution to a central venous line, noted above, is not
a human slip at all; it is a failure of system design in that the system for delivering feeding solution
was constructed in such a way as to allow this slip to happen. Systems analysis might have predicted
that this error might occur and thoughtful design might have prevented it.
How do systems contribute to error potential? Reviews of errors frequently identify operator
error, mechanical failures, lack or failure of backup systems, and inadequate procedures as root
causes. In his book “Normal Accidents,” Charles Perrow13 argues that the sheer complexity of a
system and the potential for unpredictable interactions between its many components makes
accidents unavoidable. At the core of the accidents Perrow examines (including the Three Mile
Island nuclear disaster) is the concurrent failure of more than one sub-system or component. These
sub-system failures interact in a way that could not have been predicted; Perrow calls this
“interactive complexity.” Creating new routines or safety measures after the fact may prevent that
particular failure and interaction from happening again but not prevent some other combination. In
fact, new routines, safety systems, and technologies will only further increase the complexity of the
system as a whole and may actually contribute to accident potential.
Sub-systems and components of a complex system can be more or less tightly coupled.
When there is a lot of time between steps in a process, or one step does not immediately trigger the
next, the system is loosely coupled. Much of the patient care process is loosely coupled. For example
missing a dose of a bowel motion-softening agent will not inevitably result in constipation, and
missing a single dose of antibiotic may not slow the resolution of bronchitis. Mechanical systems on
the other hand are more often tightly coupled. Processes move quickly and cannot be isolated so that
when one process fails, those “down stream” cannot be turned off. This is the problem in nuclear
reactors. Furthermore, the interactions among components of a complex system may literally be
invisible to the operators and incomprehensible for some crucial period of time.
Of course there are other system characteristics, beyond interactive complexity itself, that
contribute to the potential for an adverse event. Poor system design, inadequately specified or
contradictory procedures, lack of role definition and accountability and faulty and failing equipment
all contribute to increased error and adverse event potential. The adverse drug events study
identified a number of system failures, rather than failures of individual human functioning, that
underlay the preventable and potential ADEs.14 The most common of these are shown in Table D.
Errors that occurred earlier in the drug ordering process were more likely to be intercepted.
Many of the systems failures listed in Table D relate to processes that are inadequately
designed, unnecessarily complex (with many hand-offs that are opportunities for
miscommunication), or simply not “processes” at all, but a collection of tasks for many different
caregivers that have not been designed into an integrated process.

13 Perrow C. Normal accidents: Living with high risk technologies. Basic Books. 1984
14 Leape LL, Bates DW, Cullen DJ et al. Systems analysis of adverse drug events. JAMA. 1995; 274: 35-43.

7

699-024

Complexity and Error in Medicine

System

Attributed
errors (%)

(MD
Drug
knowledge
dissemination
knowledge about drug choice, dose, frequency,
etc.)

29

Dose and identity checking (correct drug in
correct dose delivered to correct patient)

12

Patient information availability (at the time of
drug ordering and delivery)

11

Drug order transcription

9

Allergy defense (preventing ordering and
delivery of drugs to which the patient is known
to be allergic)

7

Medication order tracking (from the time of
writing to the time of drug delivery)

5

Interservice communication (between nurses,
pharmacist, and physicians)

5

Device use (e.g., drug infusion pumps)

4

Standardization of doses and frequencies
(among physicians and patient care units)

4

Other

14

Table D: Systems underlying adverse drug events
When hospitals undertake process improvement, they frequently discover that there is not a
specified process to improve. Care has been delivered in an ad hoc fashion that varies from patient to
patient. That catastrophes have not previously occurred is the result of the care and attention of
highly committed professionals, not well-designed processes and systems of care that decrease the
base probability of error.
(3) Organizational Factors
Organization-level factors also contribute to the potential for making errors and the
likelihood that an error will be detected and intercepted. Perhaps the most important of these is
organizational culture. As Lucian Leape points out, medical error has not been a topic of importance
until recently. When asked why their organizations have not done more to address the risk of error,
CEOs reply that they did not know it was a problem.15 Physicians and nurses are reluctant to discuss
their errors and near misses, partly because they risk sanction or prosecution and partly because the
culture of medicine does not promote open discussion about any form of failure. In one study of 254
internal medicine house officers, 144 reported making a mistake.16 The reported mistakes resulted in
serious adverse patient outcomes in 90% of cases (including death in 31%). Only 54% of these house
officers reported discussing the mistake with their senior attending physician, though 88% discussed
the mistake with some other doctor. The mistakes were discussed during attending physician rounds
in 57% of cases and during morning report of morbidity and mortality conference in 31% of cases.
The house officers felt that the hospital atmosphere prevented them from talking about mistakes in
27% of cases. Reported error rates are in fact higher in more supportive team environments, ones

15 Lucian Leape. Personal communication.
16 Wu AW, Folkman S, McFee SJ, Lo B. Do house officers learn from their mistakes? JAMA. 1991; 265: 2089-2094.

8

Complexity and Error in Medicine

699-024

characterized by openness in discussing error and by high degrees of goal setting and coaching by
nurse managers, in short where team members feel safe to report error and failure.17
External factors clearly play a role in creating such a culture of silence. Individual staff
members may be sued and hospital records may be subpoenaed. This stands in sharp contrast to the
airline industry in which near misses are discussed openly and individual crew members have
immunity if they report errors within 48 hours.
Organizational culture contributes not only to the likelihood that an error will be reported
but also to the likelihood that one will be made in the first place. In high-risk environments when
individuals “get away” with taking risks, for example by not following established procedures, this
expands their notion of what is an acceptable risk.18 Retrospective review interprets this risk taking
in a positive light if the outcome was good. So risk becomes “normalized.” Over time increased
levels of risk taking become acceptable. This tendency may be greater in areas that are perceived to
be high risk in the first place, such as cancer chemotherapy and ICUs, and may be exacerbated in
loosely coupled systems. Diane Vaughn, in an analysis of the Challenger accident, points out that
this tendency is exacerbated in circumstances where there are unclear or conflicting signals of
potential danger. In the NASA of the 1980s, problems were not only common but to be expected so
any individual problem was not necessarily a signal of potential danger. Mixed signals (when a sign
of potential danger is followed by evidence that all is well), weak signals (that do not clearly predict
danger) and routine signals (ones that occur frequently) all encouraged engineers to underestimate
the risks. In hospital environments where alarms sound frequently (routine signals – common in
ICUs and operating rooms) or policies are not followed routinely (weak signals, such as not
documenting and checking drug dosages), risk-taking behavior may become the norm.
Organizational structure is also an contributing factor. Errors are most frequent during the
handoffs between house officers that occur at shift changes. One study in a Boston teaching hospital
found that patients with potentially preventable adverse events (54 of 124 errors) were more likely
than controls to be cared for by a physician other than one from their usual care team.19 This was not
so for those adverse events judged to be unpreventable. The authors noted that an interesting
implication of this study is that the patient’s usual but fatigued intern may render more appropriate
care than a well-rested intern who has less detailed knowledge of that particular patient.
Standardization of the data exchanged by physicians when “signing out” using a computerized
system eliminated the excess risk associated with “cross coverage.”20
The way teams are configured and managed is important, especially when a successful
outcome requires coordination and communication among different disciplines whose roles and
tasks are interdependent. Some of the best-studied teams are airline cockpit crews. Like the health
care industry, the airline industry has become more complex over time. Early passenger aircraft were
flown manually by highly skilled pilots who were completely responsible for the flight. Modern
passenger jets are complex and flown by a team. Analysis of recent airline accidents indicates that
70% are caused by team actions. Examples include when one member of the team identifies a
problem and either does not communicate it to the other members of the crew or is not heard by
them. The leadership role of the captain and the tone she or he sets is particularly important.

17 Edmondson AC. Learning from mistakes is easier said than done: group and organizational influences on the
detection and correction of human error. J App Beh Sci. 1996; 32: 5-28.
18 Vaughn D. The trickle down effect: Policy decisions, risky work, and the Challenger tragedy. California
Management Review. 1997; 39 (2): 80-102.
19 Petersen LA. Brennan TA. O'Neil AC. Cook EF. Lee TH. Does housestaff discontinuity of care increase the risk
for preventable adverse events? Annals of Internal Medicine. 1994; 121(11): 866-872.
20 Petersen LA, Orav EJ, Teich JM et al. Using a computerized sign-out program to improve continuity of
inpatient care and prevent adverse events. Joint Commission Journal on Quality Improvement. 1998; 24: 77-87.

9

699-024

Complexity and Error in Medicine

Successful captains are those who make it clear that they are both in charge and also respectful of
their subordinates’ expertise and willing to challenged by them.21
Evidence from other industries suggests that part-time and “contingent” workers (i.e., those
who work occasionally or on-demand) are more prone to accidents resulting in harm to themselves
or others.22 Long-term employees are more familiar with the work setting, local procedures,
personnel, and equipment than contingent workers, and this familiarity contributes to lower accident
rates. Organizations, including hospitals, invest in the ongoing training of long-term staff. In the
mining and petrochemical industries contingent workers tend to be younger and less experienced
and more likely to have an accident in their first year of employment.
Finally, in complex organizations comprising many processes, disciplines, teams, and sites,
good coordination is an essential skill. Evidence of the importance of coordination and its relation to
patient outcome is only just emerging in health care. Several studies have suggested that
coordination within and among individuals and teams is an independent predictor of patient
outcome.23 This coordination may be achieved by a number of methods, including a specific person
acting as a coordinator, or peer-to-peer interaction between or across disciplines.24

Managing the Error Potential in Medicine
The above discussion of the various sources of error has implications for the strategies that
may be employed to decrease the likelihood of an error occurring and to increase the likelihood that
when one does occur it will be detected and intercepted. Managing error potential has two
components: assessing the risk of error, and intervening to reduce that risk.
(1) Risk Assessment
Given that reported accidents and adverse events are relatively rare, the assessment of error
potential tends to be as much “hypothetical” as based on data from previous adverse events. As
Rasmussen points out, “for hazardous large scale installations, design cannot be based on experience
gained from accidents …”25
One set of techniques used to assess error risk involves analyzing individual processes and
attempting to predict the ways in which they will fail and the possible consequences of such failures.
The general structure of these assessments, established in the nuclear power industry, is (1) identify
the potential adverse outcome, (2) identify the possible initiating events that might lead to this
outcome, (3) establish all the possible sequences of events that might follow from the initiating event,
(4) quantify each sequence, and (5) thereby calculate the overall risk. The quantification of risk

21 Helmreich RL, Fouchee HC. Why crew resource management? Empirical and theoretical bases of human
factors training in aviation. In Weiner EL, Kanki BG, Helmreich RL (eds), Cockpit Resource Management.
Academic Press, San Diego, 1993.
22 Rousseau DM, Libusser C. Contingent workers in high risk environments. California Management Review.
1997; 39 (2): 103-123.
23 E.g., Argote L.
Input uncertainty and organizational coordination in hospital emergency units.
Administrative science quarterly. 1982; 27: 420-434.
24 Young GJ, Charns MP, Daley J et al. Best practices for managing surgical services: The role of coordination.
Health Care Management Review, 1997; 22: 72-81.
25 Rasmussen J. Interdisciplinary workshops to develop a multi-disciplinary research programme based on a
holistic system approach to safety and management of risk in large-scale technological operations. Paper
commissioned by the World Bank, Washington, D.C. 1988. Quoted in Reason J, 1990. Human Error. Cambridge
University Press, Cambridge England.

10

Complexity and Error in Medicine

699-024

depends on the likelihood of an initiating event and the likelihood that backup systems will fail.26
Calculating the likelihood of an initiating event or the failure of a backup system relies on data that
may not exist, in which case the probabilities have to be estimated, usually using expert judgment.
Such an analysis can be performed either by working backwards from an adverse outcome
(fault analysis) or working forward from a specific event (event analysis). For example, the analysis
might begin by asking “How can an overdose occur?” and identifying all the possible root causes of
an overdose. One such root cause will obviously be writing an order for a dose that is too high. The
event analysis then asks “What is the effect of writing an inappropriately high dose?” and identifies
all the ways this error can propagate, the backup systems that must function to prevent the error
propagating, and the ways in which these backup systems can fail. The next step is to apply
probabilities to these failures and values to each failure (i.e., is this failure minor or very serious?).
This kind of analysis can be done at the process level (“how can the process fail?”) or the
human level. A human factors analysis regards human actions in the same way as a process or
system event. The actions of each staff member can be broken down into a set of tasks and each task
treated as an event that can “fail.” The probabilities for a human factors analysis can be derived from
measurements of actual error rates in simulations. Yet another version of this analysis evaluates the
effects of cognitive errors (i.e., mistakes rather than slips) that occur after the actual event. The
failures of cognition are (1) failure to perceive that an event has occurred, (2) failure to diagnose the
nature of the event and take appropriate remedial action, and (3) failure to correctly implement those
actions.27
These techniques all share the same basic limitations. The probabilities are, at best, estimates,
usually derived through consensus of expert judgments or statistical analysis of previous events. In
complex tightly coupled systems it is usually impossible to predict all the modes of systems failure.
Finally, health care facilities are complex institutions with many potentially harmful processes, and it
may not be practical to subject each to this kind of analysis.
There are two alternatives to this painstaking process evaluation. One is to evaluate only
high-risk processes. In reality there are a limited number of processes that are particularly high risk
for patients, such as chemotherapy, blood transfusion, opiate pain relief, anesthesia and surgery, and
resuscitation. The other is to externally benchmark essential processes. In fact, this is one purpose of
the Joint Commission on the Accreditation of Healthcare Organizations (JCAHO). The current
JCAHO accreditation process assesses hospitals along five patient-care functions (patient rights and
organization ethics, assessment of patients, care of patients, education, and continuum of care), six
organization-level functions (improving organizations’ performance, leadership, management of the
environment of care, management of human resources, management of information, and
surveillance, prevention and control of infection), and four “structures with functions” (governance,
management, medical staff, and nursing). In preparing for the triennial accreditation review,
hospitals review their own internal processes to see if they meet the JCAHO’s published standards.
The JCAHO’s process benchmarking is useful but far from foolproof. The ability of such an external
benchmarking of core processes to accurately predict adverse events was called into question recently
when Newton-Wellsley Hospital in Boston had two maternal deaths a year after receiving the Joint
Commission’s highest commendation.28

26 This methodology is discussed in some detail in Reason J, 1990. Chapter 8.
27 Op. cit. Reason J, 1990, p. 224
28 Tye L. Clashing evaluations of hospital spur worry. Boston Globe, 20 August, 1997

11

699-024

Complexity and Error in Medicine

(2) Intervening to Reduce Risk
Like the causes of error themselves, interventions may be made at the level of (a) the
individual human, (b) the system level, or (c) the organization. Broadly speaking, if error is thought
of as a function of system complexity—or the interactions between humans and complex systems—
then the strategies to reduce the potential for medical error are to either reduce system complexity or
to create processes and systems to manage that complexity.
a. Human factors and the human-system interface
As noted, the potential for slips is increased when staff are fatigued and the working
environment is busy and stressful. This has been the rationale for placing limits on the length of
shifts and the minimum number of hours between duties. For example, airline crews are obliged to
take a minimum of eight hours between duties. Limiting junior doctors’ hours (both the number of
consecutive hours and the total number of hours) became increasingly important after a famous case
in New York State in which an 18-year-old patient died within hours of admission to a New York
hospital. Although the individual doctors involved were ultimately exonerated, the grand jury
investigation found that the long hours worked by often unsupervised interns contributed to the
patient’s death.
Interestingly, fatigue by itself is not a sine qua non of error. In the Boston teaching hospital
study,29 the patient’s usual but fatigued junior doctors made fewer errors. Fouchee et al., in studying
airline crews, noted that fatigued crews that had been flying together for several days were more
prone to errors but also intercepted those errors so that the net effect was fewer errors overall.30 The
crew’s superior team functioning compensated for the fatigue of the individual team members.
Recent surprise inspections of 12 New York hospitals by the Department of Public Health found that
all institutions violated the 80-hour work week and 24-hour consecutive shift limits. However the
inspectors did not find evidence that the overworked residents “made any mistakes that jeopardized
patients.”31
As already noted, mistakes are errors of problem solving often caused by inadequate
information or inappropriate interpretation of information. Memory aids and formal procedures for
reviewing them prevent mistakes due to inadequate information and information overload. For
example, the pre-flight checklist has been an essential component of airline safety. Guidelines and
critical paths can serve the same function in health care.
Human errors do not occur in isolation but when humans confront systems and processes
whose design either invites error or, at the least, does not prevent it. Forcing functions are systems
designed in such a way to disallow slips. An example of a forcing function in health care is a
computer order entry system. Physicians write the orders for a patient’s drugs on the hospital
computer. The computer checks that the dose is within acceptable range, and checks that the drug
choice and dosage are compatible with the patient’s special characteristics (such as drug allergies and
concurrent medical conditions that might require dose modification). If the physician has chosen an
incorrect dose or a drug that is not appropriate for the patient, the computer will not allow the order,
but will ask the physician to order another dose or drug. The physician can override the computer
recommendation but must record the reason for doing so. The systems also function as a checklist

29 Petersen LA, Brennan TA, O'Neil AC, Cook EF, Lee TH. Does housestaff discontinuity of care increase the
risk for preventable adverse events?. Annals of Internal Medicine. 1994; 121(11): 866-872.
30 Fouchee HC, Lauber JK, Baetge MM, Acomb DB (1986). Crew factors in flight operations III: The operational
significance of exposure to short-haul air transport operations. (Technical Memorandum No. 88342). Moffett
Field, CA. NASA Ames Research Center.
31 Kennedy, R. Study finds that hospitals overwork young doctors. New York Times. May 19, 1998.

12

Complexity and Error in Medicine

699-024

and a teaching aid. A simple forcing function to prevent the inadvertent delivery of gastrostomy
feeding solution into a central line would be to redesign the connector on the bag of gastrostomy
solution so that it does not fit the central line.
Finally, training has an important role in decreasing error potential. Military organizations
and airlines routinely undertake sophisticated training and simulation exercises designed to ensure
that when an individual or a team face an unusual and unpredicted set of circumstances it will not be
for the first time.
b. System Complexity
Clearly an important way to manage complexity is to reduce it wherever possible. Simple
processes that generate the required outcome are to be preferred over complex ones. For example,
drug-ordering processes may involve a number of steps, disciplines, and individuals, and therefore
have the potential for miscommunication and transcription error. They can be redesigned to be
simpler and involve fewer handoffs between individuals. Of course where there is no explicit
process, staff members will create their own so that complexity is born of a myriad of infinitely
varying processes. One way to decrease complexity is to create a process!
Standardization is another strategy for coping with complex processes. Standardized
processes are easier to learn (“practice makes perfect”) and teach. Each staff member can know what
their specific tasks are and how these tasks fit into the overall process. Standardization is also a
means of resolving conflicts between processes. If the strategy for treating pneumonia, whether it is
primary pneumonia or pneumonia associated with asthma or chronic obstructive airways disease, is
the same then nursing and medical staff will not have to reconcile potentially conflicting treatment
strategies as they move through the hospital. An associated strategy is to create homogeneous
patient populations by placing similar patients together in the same ward (patient “aggregation”).
This way staff see the same issues in patient care every day.
Where process simplification is not possible, there are a number of available strategies for
managing complexity. Redundancy of staff and systems allows for both checking during routine
operations and backup in the case of failure. This does not mean simply having more people (though
more people allows staff to “buddy” and check each other’s work—a strategy used by air traffic
controllers) or technological backup systems (e.g., power generators) but designing and dividing
tasks so that more than one person or discipline can perform them. If staff members understand not
only their own but others’ tasks, then many people can monitor the complex care process. In the
Navy this cross-training is combined with broad “ownership” of and accountability for the process.
If there is a problem then “you own the problem, until either you can fix it or you can find someone
who can.”32
Redundancy of staff and tasks and an associated sense of ownership and accountability for
outcomes requires that all staff concerned share the same goals. If staff have conflicting goals, then
they cannot substitute for one another nor check up on each other’s work. Task redundancy also
requires that individual staff are skilled in negotiation and conflict resolution, and there is a
mechanism for resolving those conflicts that staff cannot deal with one-on-one. For example, case
managers often have a named senior physician to whom they can turn if they are unable to resolve
issues of patient management and clinical resource utilization with the physician concerned.
Complex processes and systems, especially when they comprise staff of different disciplines
and training, give rise to the potential for miscommunication. This is compounded by each discipline

32 Roberts KH. Managing high reliability organizations. California Management Review, 1990; 32(4) Summer:
101-113

13

699-024

Complexity and Error in Medicine

having its own particular perspective and specialized terminology. Medicine is full of ambiguous
language. There are many similar but subtly different terms, drug names, and abbreviations. For
example, MTX is an abbreviation for methotrexate—a common anti-cancer agent. It can also be taken
to mean mitoxantrone—another anti-cancer agent. ADR is a common abbreviation for adriamycin,
an anti-cancer agent, that could be taken to indicate aredia, a drug for treating the high calcium levels
that are often found in cancer patients. Both adriamycin and aredia have similar dosage ranges (60 to
90 milligrams). Furthermore, physicians are notoriously variable in the way they interpret
expressions of probability. For example, in one study a group of physicians ascribed probabilities
ranging from 0.25 to 0.95 to the phrase “consistent with.” This phrase is commonly used in
laboratory and radiology reports.33
Standardized communication and specialized language reduces ambiguity. For example, in a
cardiac operating room the cardiac surgeon and perfusionist establish a pattern of repeating orders
back to each other (e.g., “flow to 50,” “flow is at 50”). The existence of specialized language that is
either discipline or organization-specific is one of the reasons that contingent workers may have an
increased error rate. To combat this, hospitals try to increase the familiarity of part-time staff with
their care system. They do this by creating their own pool of “float” nurses or contracting with an
agency that deliberately rosters the same nurses onto the same wards.
c. Organization Level
Creating an organizational culture that supports open discussion of errors and near-misses is
perhaps the single most effective intervention. This is clearly part of the culture of airlines. Crews
have immunity if they report errors and near-misses. Error potential and its precursors are an
explicit part of crew training. Contrary to our expectations of a hierarchical military organization,
some decisions on an aircraft carrier are deliberately decentralized. Even the lowest-level
crewmember has the right to halt flight operations if that member perceives a hazard. Incorrectly
calling a halt is not punished.34 As such a culture is created, it is likely that the error rate will
increase, not because more errors are being made but because more are being reported.
Creating such a culture implies requiring leadership at the senior level. Boards and CEOs
must want to know about errors and near-misses, and those who tell them must be made to feel safe
to do so. Of course lower-level members of the organization will be watching carefully to see what
senior leaders do with these data. Are error reports treated as “opportunities for improvement” or
reasons to fire an individual staff member? Do senior leaders watch for errors as they occur and
react, or do they prospectively identify the highest risk areas and intervene prior to an event? Do
they personally get involved in error prevention or delegate it to someone more junior to “fix”? The
importance of the clarity and consistency of senior leadership is supported by the finding that in the
Union Carbide plant accident in Bhopal, India, poor staff motivation and top management
discontinuity were important antecedents.35
Such a culture is reinforced by the routine collection and analysis of error-related data. This
means collecting data about errors, near-misses, and error proxies. These are events that themselves
do not represent errors but are markers for situations in which error is more likely. For example, in
an attempt to avert a larger incident, the navy monitors the “crunch rate” on aircraft carriers—the
frequency with which planes bump into each other as they are moved around the deck.36 Ideally,

33 Bryant GD, Norman GR. Expressions of probability: Words and Numbers. N Engl J Med. 1980; 302: 411.
34 Roberts KH. Some characteristics of one type of high reliability organization. Organization Science, 1990; 1:

160-167.
35 Shrivastava P. Bhopal. New York, NY. Basic Books, 1986. Quoted in Roberts, 1990, ibid.
36 Roberts, 1990. Op. Cit.
14

Complexity and Error in Medicine

699-024

error-related data should be fed into the institution’s continuous improvement program so that
whenever an error risk is identified it is acted upon.

Conclusion
In the past there has been no systematic management intervention to decrease medical error.
Traditionally, when error in health care has been addressed at all, the focus has been retrospective
and on individual cases. When cases come to light, interventions have usually been at the level of the
individual physician or nurse associated with the error. Individual adverse events are discussed in
monthly “Morbidity and Mortality” meetings held by individual hospital clinical departments.
Serious cases are reported to the hospital’s Patient Care Assessment Committee which decides
whether the case must be reported to the state level Board of Registration in Medicine. A parallel
system exists for nurses. Individual physicians and nurses who make an error are liable to be
sanctioned by the hospital and the state (to say nothing of being sued in civil court). They may be
demoted, fired, or even lose their license to practice.
Error prevention in medicine thus far has concentrated on training professionals thoroughly
and instilling in them a culture of perfection. Responsibility for error has been presumed to rest at
the level of the individual clinician, not the system in which he or she works. Such a punitive and
litigation-oriented system has done nothing to foster a culture of openness about error. So there have
been few attempts to address error risk in health care prospectively by designing “error proof”
processes and systems of care delivery and creating organizations that support ongoing error
identification and prevention efforts.
Recent highly publicized hospital errors may change this as CEOs and Boards of Governors
realize that an approach based primarily upon post hoc reaction to errors as they occur is insufficient
for managing an increasingly complex process of care. Furthermore, the management of error in
health care has, to date, occurred largely out of the public eye. This too is changing so that individual
clinicians and provider organizations may come under public pressure to decrease the risk of a
preventable medical error. This will require that hospitals not rely simply on weeding out those
individuals who have been associated in the past with an adverse events,37 but take a system view of
error potential and both reduce system complexity and create processes and routines explicitly for
managing that complexity.
A successful strategy for decreasing error potential will need to include interventions at all
three levels—human factors and the human-system interface, system complexity, and organization.
It will need to combine actions to both decrease and manage complexity with the creation of an
organizational environment that is at one time open about error and intolerant of system failure.

37 It is of note that in the Bates et. al. Adverse drug-event study, the median number of errors per physician was
one. This supports the notion that errors are not propagated by a limited number of individuals whose
performance is consistently egregious.

15