Full text of DOI:10.1111/j.1553-2712.2000.tb00483.x

ACADEMIC EMERGENCY MEDICINE • November 2000, Volume 7, Number 11
1321
Physician versus Computer Knowledge of Potential
Drug Interactions in the Emergency Department
M
ARK I. LANGDORF, MD, MHPE, JOHN C. FOX, MD,
AKESH S. MARWAH, BS, BRIAN J. MONTAGUE, BS,
ICHAEL M. HART, MD, P
R
M
HD
Abstract. Introduction: Proliferation of Food and ber. Results: Seventeen percent of the patients had
Drug Administration–approved drugs makes it im- potential drug–drug interactions, and 25% of these
possible for emergency medicine (EM) faculty to stay were judged to be clinically significant. Up to 52% of
current on potential interactions between drugs, and the patients had potential drug–ethanol interactions,
with diseases, laboratory tests, and ethanol. A com- while 38% of the patients could have potential drug–
puter database may augment physician knowledge. lab interactions. Sensitivity, specificity, and positive
Objectives: To compare the performance of EM fac- and negative predictive values of the EM faculty for
ulty and an ‘‘expert’’ emergency physician (EP) with potential drug–drug interactions compared with the
that of a criterion standard computer database in computer were poor, at 14%, 58%, 6%, and 23%, re-
identifying potential drug interactions, and to report spectively. The corresponding values for the ‘‘expert’’
the incidence of drug–ethanol and drug–laboratory EP were 25%, 86%, 26%, and 85%. The ‘‘expert’’ EP
test interactions. Methods: This was a retrospective was statistically better than the EM faculty, but still
review of 276 emergency department charts for drug, less sensitive and predictive than the computer. Con-
ethanol, lab, and medical history. Evaluation by both clusions: A computer can aid the physician in avoid-
EM faculty and an ‘‘expert’’ EP of patient history was ing potential drug interactions. Prospective valida-
done to identify potential interactions, and compari- tion of these findings should be done. Key words:
son with the Micromedex Drug-Reax database for po- emergency department; computer database; drug in-
tential interactions (graded minor, moderate, or ma- teraction; medical education; drugs; physician knowl-
jor) was made. Clinical significance of potential edge. ACADEMIC EMERGENCY MEDICINE 2000;
interactions was judged by a second EM faculty mem- 7:1321–1329
HERE have been 19,313 drugs approved by rious interactions. Furthermore, drugs interact not
T
the Food and Drug Adminisration (FDA) over
the years, and an average of 340 new drugs are
added to the formulary each year (personal com-
munication, Scott, GR, FDA, Rockville, MD, 1997).
A major drug interaction software program lists
more than 100,000 potential interactions between
these drugs (personal communication, Micromedex
Corporation, Denver, CO, 1997). It is clearly im-
possible, then, for the practicing emergency phy-
sician (EP) to stay current with all potentially se-
only with each other, but also with patients’ dis-
eases, ethanol, and certain laboratory tests.
For 20 years, various computerized systems
have been used, mostly by pharmacists and mostly
for inpatients, to try to identify and avoid potential
interactions. In one study1 28% of adverse drug
events were judged preventable, and some of these
were drug–drug or drug–disease interactions.
(Others related to dose and frequency of adminis-
tration.) Organized medicine has called attention
to the problem in position papers,2,3 but there has
been no agreement to use a standard computer
system that identifies potential interactions before
the prescription is written. Proprietary pharmacy
systems exist that perform this function, but re-
quire that all of a patient’s prescriptions be filled
at the same pharmacy or chain. Many software
programs are commercially available and have
been reviewed in the literature,4–19 but these have
not become widely used in the emergency depart-
ment (ED).
From the University of California, Irvine Medical Center,
Orange, CA (MIL, JCF, RSM, BJM, MMH).
Received May 8, 2000; revision received June 23, 2000; ac-
cepted July 3, 2000. Presented at the California Chapter,
American College of Emergency Physicians Annual Scientific
Assembly, Rancho Mirage, CA, May 1997; and the National
American College of Emergency Physicians Annual Scientific
Assembly, October 1998, San Diego, CA, October 1998.
Supported in part by the University of California, Irvine, Un-
dergraduate Research Opportunities Program Grant designa-
tion ‘‘Marwah.’’
Address for correspondence and reprints: Mark I. Langdorf,
MD, MHPE, Department of Emergency Medicine, Route 128,
UCI Medical Center, 101 City Drive, Orange, CA, 92868, Fax:
714-456-5390; e-mail: milangdo@uci.edu
There have been four analyses of the utility of
computer-generated potential drug interactions in
ED patients, three dealing with only drug–drug

1322
DRUG INTERACTIONS
Langdorf et al. • DRUG INTERACTIONS IN THE ED
interactions,20–22 and one including drug–disease of the study, had neither medical training in nor
interactions.23 These have identified high-risk knowledge of drug interactions. All charts were re-
groups, such as elders21–23 and those taking mul- viewed prior to accessing the computer database,
tiple medications.22,23 Underlying these studies is to eliminate bias in data collection. They recorded
the assumption that the computer does a better job all medications, prior to the ED visit, in the ED,
than the physician, but to the best of our knowl- and upon discharge. We did not verify medication
edge, this has never been systematically assessed. history. The patient’s alcohol history was a re-
Furthermore, while potential drug–drug interac- quired element on the ED chart under social his-
tions have been studied, we believe the incidence tory, so this was frequently recorded.
of potential interactions between drugs and etha-
We entered these data in the 1997 Micromedex
nol or laboratory tests have not been quantified. Drug-Reax database, and recorded potential drug–
We know of only one report of potential drug–dis- drug, drug–disease, drug–lab, and drug–ethanol
ease interactions.23
interactions. We define a drug–disease interaction
when a drug exacerbates a pre-existing medical
The objectives of this study were twofold:
1. to compare the performance of emergency med- condition. This is distinct from an adverse effect,
icine (EM) faculty and an ‘‘expert’’ EP physician where a drug brings about a new medical condi-
with that of a criterion standard computer data- tion. Further, a drug–laboratory interaction occurs
base
2. to report the incidence of potential drug–etha- ment, thus giving a false reading.
nol and drug–laboratory test interactions. The computer program identifies potential drug
when a drug interferes with a laboratory measure-
interactions with myriad diseases without regard
to whether the patient actually has the disease.
The computer grades potential interactions as mi-
nor, moderate, or major. The software program de-
fines ‘‘minor’’ interactions as those that would have
limited clinical effects, and manifestations may in-
volve an increase in frequency or severity of side
effects, but would generally not necessitate a major
alteration of therapy. ‘‘Moderate’’ interactions may
result in an exacerbation of patient condition and/
or necessitate an alteration of therapy. ‘‘Major’’ in-
teractions are those that are life-threatening and/
or necessitate medical intervention to minimize or
prevent serious adverse effects.
M
ETHODS
Study Design. We conducted a retrospective
chart review of 276 consecutive patients present-
ing to a university hospital ED. The study qualified
for exempt registration by the UC Irvine Institu-
tional Review Board.
Study Setting and Population. The ED is a
Level I trauma center with an annual census of
38,000. The hospital is a 463-bed, former county
institution serving a significant proportion of in-
digent patients. Medicaid and county insurance
patients comprise 35% of ED patients, while 40%
are self-pay and 21% have commercial insurance.
We then gave the data sheets to eight board-
certified EPs (the EM faculty), approximately 35
patients each, and asked them to record both the
Study Protocol. Research assistants were presence and the severity of the potential interac-
trained to abstract data from ED charts, including tions. We also asked them whether they thought
where to find drug history, alcohol use, laboratory potential interactions were ‘‘clinically significant,’’
tests, and past and present disease states. These defined as necessitating a change in management.
were recorded in predefined areas of the ED med- All 276 data sheets were also analyzed by an
ical record. Urgent care charts lacked these ‘‘expert’’ physician, an EM faculty member with a
prompts to record data, and so were excluded. Be- PhD degree in pharmacology, 20 years of practice
cause of the standardized form of the ED chart, experience, and a particular interest in emergency
assessment of interobserver variability was not medications.
thought to be necessary. An EM resident and fac-
We assessed whether the potential interactions
ulty assisted as needed. In the ED, charts were identified by the computer, but missed initially by
filed as they were completed, by date, but in no the physician, were clinically important. A second
other particular order. The ED records from three physician reviewed the data sheets and recorded
consecutive days were analyzed according to ex- which potential interactions were important, i.e.,
plicit criteria on standardized data sheets. While those that would change patient management, and
exact patient census was not available on the days which were trivial. There was no attempt to as-
chosen, approximately 280 patients would have sess whether any actual drug-related problems oc-
been seen in the main ED over three days, exclud- curred.
ing the urgent care census. Therefore, we believe
we achieved a consecutive series of ED patients.
Measurements and Key Outcome Measures.
The research assistants, while aware of the aim We calculated the sensitivity, specificity, and posi-

ACADEMIC EMERGENCY MEDICINE • November 2000, Volume 7, Number 11
1323
TABLE 1. Most Common Potential Drug–Drug Interactions, and Their Level of Severity, Identified by Computer
Incidence
Drug
Drug
Verapamil
Effect
Severity
(n = 168)
Aspirin
Haloperidol
Potassium
Increased risk of bleeding
Anticholinergic effect
Hyperkalemia
Moderate
Moderate
Major
2
2
1
Cogentin
Hydrochlorothiazide
ϩ triamterene
Heparin
Atenolol
Glyburide
Ciprofloxacin
Ranitidine
Fentanyl
Phenobarbital
Cimetidine
Phenytoin
Nitroglycerin
Ibuprofen
Gemfibrozil
Warfarin
Decreased anticoagulant effect
Decreased atenolol effect
Hypoglycemia
Increased risk of bleeding
Increased risk of bleeding
Respiratory depression
Decreased carbamazepine effect
Lidocaine toxicity
Moderate
Moderate
Moderate
Moderate
Moderate
Moderate
Minor
Minor
Minor
Minor
Minor
1
1
1
1
1
1
1
1
1
1
1
Warfarin
Midazolam
Carbamazepine
Lidocaine
Ranitidine
Cimetidine
Fluoxetine
Phenytoin toxicity
Diazepam toxicity
Trazodone toxicity
Diazepam
Trazodone
tive and negative predictive values for the general while 168 (61%) were exposed to at least two
EM faculty vs the criterion standard computer and
for the ‘‘expert’’ EP vs the computer. We assessed
whether the ‘‘expert’’ physician achieved closer
agreement with the computer than the general EM
faculty. Because the computer reports any possible
potential drug interaction with myriad diseases,
and the faculty did not, we do not report compar-
isons between the faculty identification of potential
drug–disease interactions and that of the com-
puter.
drugs, giving rise to a possible drug–drug inter-
action. Pre-ED visit drugs, ED drugs, and dis-
charge drugs averaged 2.4, 1.8, and 1.2 per patient,
respectively. We found 32 potential drug–drug in-
teractions for 29 of the 168 patients who took or
received at least two drugs (17% of patients). Of
these, eight interactions were classified as major,
17 moderate, and seven minor. Table 1 lists 15 of
the most common and potential drug–drug inter-
actions identified by computer.
To assess the ability of the computer to aid phy-
sician recognition of potential drug interactions,
we report the proportion of computer-generated,
but physician-missed, interactions that were
judged clinically significant.
There were 131 potential drug–disease inter-
actions for 109 of the 227 patients exposed to at
least one drug (48%). These 131 interactions were
classified by the computer as major in 28, moder-
ate in 99, and minor in four. Table 2 lists the most
common potential drug–disease interactions iden-
tified by computer. However, the patient actually
had the disease, either pre-existing or identified
during the ED visit, in only five of these 227 cases
(2.2%).
For 87 of the 227 patients exposed to at least
one drug (38%), the computer identified 111 poten-
tial drug–lab interactions. Table 3 describes the
most common potential drug–lab interactions, re-
gardless of whether the patient had the laboratory
test ordered. However, the patient had the sus-
pected interacting laboratory test ordered in the
ED in only five of 227 cases (2.2%). These are listed
in Table 4.
Data Analysis. We used chi-square (True Epi-
stat, Version 5.0, Epistat Services, Richardson, TX)
to determine whether the proportions of correct an-
swers regarding drug interactions were statisti-
cally different between the general EM faculty and
the ‘‘expert’’ physician, with the computer serving
as the final criterion standard. Statistical signifi-
cance was set at p < 0.05 by convention. We used
Kendall’s tau-b (SAS Institute, Cary, NC) as a mea-
sure of concordance between the responses of the
general EM faculty and the ‘‘expert’’ physician,
ranging from Ϫ1 to 1, where Ϫ1 indicates complete
discordance, and 1, complete concordance.
For 119 of the 227 patients exposed to at least
one drug (52%), the computer identified 132 poten-
tial drug–ethanol interactions. The most common
interactions are listed in Table 5. Fifty patients
(18%) were recorded on the ED chart as ethanol
users/abusers. Of these 50 patients, 29 (58%) used
or were given drugs that had potential interactions
with alcohol. Because of the retrospective nature
R
ESULTS
Sixty-five percent of the patients said they took
medications prior to the ED visit. Thirty-nine per-
cent were medicated in the ED, while 61% received
a discharge prescription for a new drug. Overall,
227 patients (82%) were exposed to at least one
medication prior to, during, or after the ED visit, of this study, we could not determine whether pa-

1324
DRUG INTERACTIONS
Langdorf et al. • DRUG INTERACTIONS IN THE ED
TABLE 2. Most Common Potential Drug–Disease Interactions Identified by Computer
Incidence
Drug
Disease
Effect
Severity
(n)
Nonsteroidal anti-inflammatory drugs
Trimethoprim–sulfamethoxazole
Peptic ulcer disease (PUD)
Systemic Lupus erythe-
matosus (SLE)
Worsens PUD
Worsens SLE
Moderate
Moderate
50
16
Prednisone/dexamethasone
Prednisone/dexamethasone
Prednisone/dexamethasone
Prednisone/dexamethasone
Nonsteroidal anti-inflammatory drugs
Albuterol/metaproterenol
Furosemide
Diabetes mellitus
Hypertension (HTN)
Congestive heart failure
Peptic ulcer disease
Gout
Hypertension
Diabetes mellitus
Gout
Hyperglycemia
Worsens HTN
Fluid retention
Worsens PUD
Hyperuricemia
Worsens HTN
Hyperglycemia
Hyperuricemia
Worsens glaucoma
Major
Major
11
11
11
11
10
10
5
Moderate
Moderate
Moderate
Moderate
Moderate
Moderate
Major
Furosemide
Dexamethasone
5
3
Glaucoma
TABLE 3. Most Common Potential Drug–Lab Interactions
Incidence
Drug
Lab Test
Effect
Severity
(n)
Trimethoprim–sulfamethoxazole
Cephalexin
Methylprednisolone
Aspirin
Theophylline level
Theophylline level
Digoxin level
Serum glucose
Acetaminophen level
False increase
Moderate
Moderate
Moderate
Moderate
Moderate
15
11
11
10
10
False increase
False increase
False increase
False increase
Aspirin
TABLE 4. Patients with Potential Drug–Lab Interactions Where Lab Was Actually Ordered (One Occurrence Each)
Drug
Lab Test
Effect
Severity
Iron
Aspirin
Cimetidine
Trimethoprim–sulfamethoxazole
*Phenergan
Fecal occult blood
Serum glucose
Fecal occult blood
Serum theophylline
Urine salicylate
False positive
False elevation
False positive
False elevation
False positive
Moderate
Moderate
Moderate
Moderate
Minor
*Patient overdosed on ibuprofen. Urine toxicology screen ordered.
tients used alcohol concurrently with the interact- significance where the computer identified an in-
ing drug, or whether they abstained from alcohol teraction (‘‘positive interaction’’) and where it did
when taking medications. Table 6 describes the not (‘‘no interaction’’).
most common potential drug–ethanol interactions
in documented ethanol users.
Regarding Tables 7 and 8, the total number of
patients at risk for potential drug–drug interac-
Tables 7 and 8 show the comparison between tions was 167, while for potential drug–lab and
the physicians and the computer. We list sensitiv- drug–ethanol interactions, it was 227. The tables
ity, specificity, and predictive values, with the com- list sample sizes appropriate for the calculations
puter serving as the criterion standard for identi- in question. For example, there were 29 patients
fying interactions.
with computer-identified potential drug–drug in-
In addition, the tables compare the general EM teractions, so the EM faculty and ‘‘expert’’ physi-
faculty with one ‘‘expert‘‘ physician, and answer cian each had 29 opportunities to agree or disagree
the question, ‘‘Did the ‘expert’ physician come with the criterion standard computer. Thus, the
closer to the computer criterion standard than the sample size (number of decisions) for chi-square is
general EM faculty?’’ To compare sensitivities, double the patient sample size, or 58. Conversely,
specificities, and predictive values between the there were 138 (167 Ϫ 29) patients with no inter-
‘‘expert’’ physician and the EM faculty, both rela- action, an equally important determination. Both
tive to the computer criterion standard, we ana- the EM faculty and the ‘‘expert’’ physician had 138
lyzed physician decisions regarding whether there chances to agree or disagree with the computer,
was a potential interaction for each patient. We hence, the doubling of the sample size to 276.
used chi-square analysis to compare the propor-
For potential drug–drug interactions, sensitiv-
tions of correct responses, and report statistical ity for the EM faculty was no different from that

ACADEMIC EMERGENCY MEDICINE • November 2000, Volume 7, Number 11
1325
T
ABLE 5. Most Common Potential Drug–Ethanol
for the ‘‘expert’’ physician (p = 0.50). However,
specificity was significantly better for the ‘‘expert’’
physician compared with the EM faculty (p =
0.000001). Accuracy of the EM faculty was 47%,
while the ‘‘expert’’ physician was accurate 77% of
the time. The advantage of the ‘‘expert’’ physician
rested entirely in identifying the absence of poten-
tial drug interactions when the computer said none
existed. The EM faculty and ‘‘expert’’ physician
were equally poor at identifying true-positive po-
tential interactions.
Interactions by Computer
Incidence
Drug
Effect
Severity
(n)
Acetaminophen
Codeine
Hydrocodone
Meperidine
Aspirin
Hepatotoxicity
Sedation
Sedation
Sedation
Gastrointestinal
bleeding
Moderate
Moderate
Moderate
Moderate
Moderate
87
18
18
10
10
For potential drug–ethanol interactions, the
EM faculty and the ‘‘expert’’ physician were similar
in their concordances with the computer in iden-
tifying patients who truly had potential interac-
tions (p = 0.99), but the ‘‘expert’’ physician was sta-
tistically better at correctly identifying the absence
100% because both groups rarely, if ever, identified
an interaction.
In Tables 9 and 10, we report only those poten-
tial interactions that the physicians judged to be
of potential drug–ethanol interactions. In the ag- clinically significant.
gregate, the ‘‘expert’’ physician was statistically
The Kendall’s tau-b statistic for potential drug–
better (p = 0.0l) than the EM faculty. For potential drug interaction, comparing the EM faculty with
drug–lab interactions, the EM faculty and the the ‘‘expert’’ physician, was Ϫ0.235 for all patients.
‘‘expert’’ physician were equally poor in sensitivity, This indicates a minor to moderate discordance of
identifying only two and none, respectively, of the opinion, with the ‘‘expert’’ physician being correct
87 potential drug–lab interactions found by the more often. The advantage of the ‘‘expert’’ physi-
computer. Specificities were equivalent and almost cian over the EM faculty lies almost entirely in
TABLE 6. Most Common Potential Drug–Ethanol Interactions in Documented Ethanol Users
Incidence
Frequency
(% of 50)
Drug
Effect
Severity
(n)
Acetaminophen
Hydrocodone
Codeine
Nitroglycerin
Morphine
Hepatotoxicity
Sedation
Sedation
Hypotension
Sedation
Gastrointestinal bleeding
Moderate
Moderate
Moderate
Moderate
Moderate
Moderate
17
5
4
3
3
34%
10%
8%
6%
6%
Aspirin
3
6%
T
ABLE 7. Comparison of Emergency Medicine (EM) Faculty and ‘‘Expert’’ Emergency Physician (EP) with Computer
Database*
Sensitivity (%)
Specificity (%)
Positive Interactions
p
No interactions
p
Interaction
Drug–drug
Drug–ethanol 238
Drug–lab 174
n
EM Faculty ‘‘Expert’’ EP
n
EM Faculty ‘‘Expert’’ EP
58
14
70
2
25
71
0
0.50
0.99
0.48
276
216
280
58
34
98
86
59
99
0.000001
0.00039
0.61
*n represents the number of decisions by faculty and the ‘‘expert’’ EP.
T
ABLE 8. Comparison of Emergency Medicine (EM) Faculty and ‘‘Expert’’ Emergency Physician (EP) with Computer
Database*
Positive Predictive Value (%)
Negative Predictive Value (%)
Interaction
n
Faculty
‘‘Expert’’ EP
n
Faculty
‘‘Expert’’ EP
Drug–drug
Drug–ethanol
Drug–lab
89
282
6
6
54
40
26
66
0
245
172
448
23
51
62
85
65
62
*n represents the number of decisions by faculty and the ‘‘expert’’ EP.

1326
DRUG INTERACTIONS
Langdorf et al. • DRUG INTERACTIONS IN THE ED
Percentage of Computer-identified,
T
ABLE 9. Proportion of Potential Interactions Identified by Computer, but Missed by an Emergency Medicine Faculty
Member, that Were Then Judged to Be Clinically Significant.
Number of Computer-identified,
Physician-missed Interactions
Number of Interactions that
Are Clinically Significant (n)
Physician-missed Interactions
that Are Clinically Significant
Interaction
Drug–drug
Drug–ethanol
Drug–lab
24
78
119
221
5
20
2
21%
26%
2%
All interactions
27
12%
T
ABLE 10. Proportion of Potential Interactions Identified by an Emergency Medicine Faculty Member, but Missed by the
Computer, that Were Judged to Be Clinically Significant
Percentage of Physician-identified,
Computer-missed Interactions
that Are Clinically Significant
Number of Physician-identified,
Computer-missed Interactions
Number of Interactions that
Are Clinically Significant (n)
Interaction
Drug–drug
Drug–ethanol
Drug–lab
80
94
3
36
17
0
45%
18%
0%
All interactions
177
53
30%
predicting the lack of potential interaction when glyburide causing hypoglycemia; 4) warfarin and
ciprofloxacin causing an increased risk of bleeding;
and 5) warfarin and ranitidine causing an in-
creased risk of bleeding.
Table 10 describes the proportion of potential
interactions identified by physicians, but missing
from the computer database, which a second fac-
ulty member confirmed to be clinically significant.
It is the converse of Table 9.
Table 11 summarizes the proportion of patients
at risk for potential drug interactions, and cate-
gorizes the severity of these (note that these are
patients, not numbers of interactions as reported
above in the text).
The faculty identified potential drug–drug in-
teractions that the computer missed. These
omissions fell into several categories. Regarding
oversedation, the computer frequently did not re-
port potential interactions between two different
opioids (n = 22), opioids and antiemetics (n = 11),
two anticholinergic/antiemetic agents (n = 7), and
benzodiazepines and either opioids (n = 3) or bar-
biturates (n = 1). These potential interactions were
validated in a common medical text.24
one does not exist. The Kendall’s tau-b for patients
with potential interactions was Ϫ0.043, indicating
little difference of opinion between the ‘‘expert’’
physician and the EM faculty. However, the Ken-
dall’s tau-b for patients without potential drug in-
teractions was Ϫ0.286, indicating that the ‘‘expert’’
physician differed in opinion from the EM faculty
to a moderate degree. From the specificity analy-
sis, it is clear that the ‘‘expert’’ physician was su-
perior to the EM faculty, identifying the absence of
a potential interaction when none existed (specific-
ity 86% vs 58%). The two groups did not substan-
tially agree with each other in patients with po-
tential interactions, and both groups fell equally
short of the computer (sensitivities of only 14% and
25%, respectively).
Table 9 outlines the proportion of potential in-
teractions missed initially by one physician, but
caught by the computer, which a second faculty
member then judged to be clinically significant. Po-
tential drug–lab interactions were rarely judged
clinically significant because the laboratory test
was ordered only twice in patients with a possible
interaction. However, potential drug–lab interac-
tions were identified by the computer 119 times,
40 of which (33.6%) were judged by a second fac-
ulty as clinically significant. Excluding potential
drug–lab interactions then, the rate of clinically-
significant potential drug–drug and drug–ethanol,
computer-identified but physician-missed, interac-
tions was 25% (25/102). Examples of these are: 1)
aspirin and verapamil causing an increased risk of
bleeding; 2) nitroglycerin and heparin causing a
The faculty did not find significant omissions
with potential drug–disease interactions. How-
ever, they identified that the computer missed sig-
nificant potential problems with ethanol. Its poten-
tial interactions with myriad other sedatives,
including
benzodiazepines,
antidepressants,
opioids, anticholinergics/antiemetics/antipsychot-
ics (n = 27), were not contained in the database.
Furthermore, ethanol’s effect on the development
or exacerbation of gastritis or peptic ulcers was
similarly overlooked by the computer (n = 27). Po-
decreased anticoagulant effect; 3) gemfibrozil and tential interactions that cause or exacerbate gas-

ACADEMIC EMERGENCY MEDICINE • November 2000, Volume 7, Number 11
1327
TABLE 11. Proportion of Patients at Risk for Potential Drug Interactions, and Corresponding Severity
Potential
No (%) of Patients
Total No. (%) with
Interaction
(n = 276)
No. (%) Minor
No (%) Moderate
No. (%) Major
Potential Interactions
Drug–drug
Drug–disease
Drug–ethanol
Drug–lab
168 (60.9%)
227 (82.2%)
227 (82.2%)
227 (82.2%)
6/168 (3.6%)
3/227 (1.3%)
n/a
16/168 (9.5%)
88/227 (38.8%)
7/168 (4.2%)
18/227 (7.9%)
n/a
29/168 (17.3%)
109/227 (48.0%)
119/227 (52.4%)*
87/227 (38.3%)†
n/a
n/a
n/a
n/a
*Micromedex reports potential drug–ethanol interactions for all patients, regardless of presence and quantity of ethanol use.
Severity, therefore, is not relevant. Only 50 patients were recorded as alcohol users, and 29/50 (58%) of these had potential
drug–alcohol interactions.
†Micromedex reports potential drug–lab interactions for all patients, regardless of whether the suspect laboratory test was
ordered. In reality, only 5/87 (5.7%) patients had the suspect lab ordered.
trointestinal bleeding were omitted as well (n = 2). hypothesis, because the study aim became evident
when one physician assessed another’s discrep-
ancy with the computer. This could lead to bias by
D
ISCUSSION
either judging ‘‘missed’’ interactions as clinically
insignificant to protect one’s colleagues, or affirm-
ing the interaction because of the computer’s au-
thority.
Drugs can interact with one another through myr-
iad mechanisms, including impaired or enhanced
absorption, induction or inhibition of enzymes,
changes in first-pass metabolism, protein binding
displacement, effects on renal excretion, and direct
and indirect receptor effects.25 It is unrealistic to
expect the EP to be familiar with all of these mech-
anisms, and the multitude of drugs that interact.26
The actual harm done to patients by drug inter-
actions is difficult to quantify, but probably small.
Only three studies, all of hospitalized patients,
have identified potential drug–drug interactions,
and then performed chart review to assess clinical
effect.27–29 Potential interactions in these studies
occurred in 7.7%, 4.7%, and 11.1% of patients, re-
spectively, while the rates of actual interactions
were 0.7%, 0.1%, and 1% (aggregate n = 8,626). To
generalize, available evidence shows that 2.1% of
potential interactions are clinically important by
retrospective review by implicit clinical criteria.
The identification and severity of potential drug
interactions from a database represent a judgment
by the consultant to the software maker. In the
Micromedex database, documentation for some
listed interactions is ‘‘poor’’ and based on case re-
ports or small retrospective case–control studies.
As such, the computer may overstate the severity
Three other studies have assessed the fre-
quency of drug–drug interactions as a subset of all
adverse drug reactions.33–35 They found 2.7%, 6.5%,
and 2.6%, respectively (aggregate n = 620), for a
weighted average of 3.5%. Clearly, drug interac-
tions are a small subset of all adverse drug events.
We know of no studies in the ED that have doc-
umented clinical harm due to drug interaction. A
surrogate for this is whether an experienced phy-
sician would alter treatment based on knowledge
gained from a computer. We found 32 potential
drug interactions for 29 of the 168 patients who
had at least two drugs (17%). The EM faculty iden-
tified only one-fourth of these potential interac-
tions, and five of the 24 missed interactions (21%)
were judged clinically significant. A similar study
found potential drug interactions for 30% of ED
patients (61/201), and reported changes in man-
agement for 25% (15/61) as well.22
In the study design, we assumed the computer
program would be an appropriate criterion stan-
dard against which to measure the physicians.
This assumption deserves scrutiny. We were sur-
prised that the EM faculty identified 174 potential
interactions that the computer missed (80 drug–
of interaction and gives no estimate of frequency. drug, 94 drug–ethanol). A second physician vali-
While some of the drug combinations listed in the dated the clinical importance of these interactions
tables are clinically used together, the interactions in 30% of cases (45% of drug–drug, 18% of drug–
do exist on both a physiologic and clinical basis. ethanol). Conversely, the computer generated 102
For example, the verapamil–aspirin interaction (24 drug–drug and 78 drug–ethanol) potential in-
causing increased bleeding is based on two case teractions of which the physician was unaware,
reports and a small study. Eighteen patients had and 25% of these were clinically significant. There-
impaired platelet aggregation with verapamil,30 fore, it was more likely that the physician identi-
and three had significant hemorrhages and pro- fied a clinically important potential interaction
longed bleeding times when verapamil was added missing from the database, than the computer
to a stable aspirin regimen.31,32
The physicians could not be blinded to the study computerized information is a useful adjunct to
identified one that the physician missed. Clearly,

1328
DRUG INTERACTIONS
Langdorf et al. • DRUG INTERACTIONS IN THE ED
physician knowledge and judgment, not a substi- overstated. Advice against driving while taking se-
tute. Though the computer is flawed as a criterion dating medicines is routine. Alcohol use should be
standard, absent a better one, it must serve this similarly discouraged.
function at present.
Reporting potential drug–lab interactions, as in
In an evaluation of ED drug interactions in pa- Table 3, does not imply that the physician should
tients 65 aged years and older, Hancock et al.20 forgo ordering the test. Nevertheless, the result
found that 2.7% of the patients visiting an ED had must be interpreted with the knowledge that the
drug interactions between their existing medica- value may be falsely increased.
tions and those newly prescribed. Our study found
We recognize that this study tested the knowl-
a much higher rate (17%), but we assessed all po- edge of only one physician ‘‘expert,’’ and his perfor-
tential interactions: among pre-existing drugs, mance is not generalizable. Furthermore, our
those given in the ED, and discharge drugs.
study design judged potential interactions merely
It appears that computer programs used to qualitatively. We recognize that medication dose,
alert physicians to potential interactions have alcohol levels, and disease severity would have sig-
varying capabilities. This is well documented in nificant impact on the magnitude of potential in-
a series of software evaluations by Poirier and teractions. The judgment of potential drug inter-
Giudici.4,5,7–18 The most recent comparison found actions by physicians was subjective. Following
programs that identified all of ten important po- computer identification of a potential interaction,
tential interactions, or as few as five of ten.5
a physician judged clinical significance. This was
In a similar study using a program called Drug confirmed or refuted by a second physician, but
Master Plus, Goldberg et al.23 found an almost disagreements between the two physicians were
threefold higher incidence of potential drug–drug not adjudicated by a third. In addition, some drug
interactions (47% vs 17%) in a county and com- combinations, while potentially harmful, are used
munity hospital population. They also found a clinically because of a positive benefit–risk ratio.
higher number of potential interactions per patient Our study identified these combinations as poten-
(2.5 vs. 1.1) than the current study. Conversely, the tially problematic, despite their clinical use.
same study found less than half of the potential
Future research should focus on a prospective
drug–disease interactions (21% vs 48%). While it program, where real-time potential drug interac-
is possible that differing physician expertise is re- tions are identified and corrected during the ED
sponsible, the discrepancy is more likely due to in- visit. This will require a user-friendly database
consistent content of computer databases.
with prescription writing capability, and automatic
It is useful to examine the power of the study warnings for interactions, improper dose, or pre-
to discriminate between the EM faculty and the viously noted allergy. A web-based version is cur-
‘‘expert’’ physician. We found no difference between rently under commercial development, but not yet
groups in the ability to correctly identify potential ready for widespread use.
drug–drug interactions for patients who had them
(n = 29). We also found the EM faculty to have an
C
ONCLUSIONS
accuracy of approximately 50%. If a clinically im-
portant improvement of the ‘‘expert’’ physician over
the EM faculty is 30%, ␣ = 0.05 and ␤ = 0.80, a
sample size of 45 interactions would be required to
demonstrate a statistically significant difference in
performance. Therefore, in this subset of patients,
the study has a high likelihood of type II error.
With the immense formulary of FDA-approved
drugs, it is beyond the scope of the EP to accu-
rately identify all potentially harmful drug inter-
actions. In this test of physician knowledge of po-
tential drug interactions, all physicians fared
poorly compared with the criterion standard com-
puter database. Conversely, the computer database
has significant omissions, and may not be an ap-
propriate criterion standard. An ‘‘expert’’ EP was
LIMITATIONS AND FUTURE QUESTIONS
Our drug–ethanol interaction analysis deserves more specific but just as poorly sensitive as the
comment. Since documentation of alcohol use on general EM faculty for potential drug–drug and
the ED chart was inconsistent, and a high per- drug–ethanol interactions, indicating that further
centage of general ED patients use alcohol, we in- education is not likely to raise the level of physi-
structed the EM faculty reviewing cases to assume cian knowledge to that of the computer. Poten-
all patients were drinkers. We did not confirm tially, up to 52% of patients have drug–ethanol in-
whether patients drank or abstained from drink- teractions, while 38% of patients could have drug–
ing while taking medications that interact with al- lab interactions. A computer can aid the physician
cohol. Therefore, our finding of 52% of patients in avoiding potential interactions. Prospective val-
with a potential drug–ethanol interaction is likely idation should follow.

ACADEMIC EMERGENCY MEDICINE • November 2000, Volume 7, Number 11
1329
crocomputer software: Part 2—Drug Interactions Advisor.
Hosp Pharm. 1989; 24:125–8, 144.
18. Poirier TI, Giudici RA. Drug interaction microcomputer
software: PDR’s drug interactions and side effects diskettes.
Hosp Pharm. 1990; 25:839–42, 851.
Statistical analysis was performed by Robert Newcombe, PhD,
Department of Statistical Consulting, University of California,
Irvine (compensated). The authors give special thanks to the
EPs whose knowledge they tested.
19. Goldberg RM, Mabee J, Mammone M, et al. A comparison
of drug interaction software programs: applicability to the
emergency department. Ann Emerg Med. 1994; 24:619–25.
20. Hancock D, Kennington JM, Beckner RR, Quick G. Emer-
gency department medication and drug interaction evaluation.
Hosp Pharm. 1992; 27:129–32.
21. Beers MH, Storrie M, Lee G. Potential adverse drug in-
teractions in the emergency room: An issue in the quality of
care. Ann Intern Med. 1990; 112:61–4.
References
1. 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.
2. American Society of Hospital Pharmacy. 1992 draft guide-
lines on preventable medication errors. Am J Hosp Pharm.
1992; 49:640–8.
3. AMA Board of Trustees Report: Medication (Drug) Errors
in Hospitals, April 1994, Chicago, IL, Resolution 512, I-93.
4. Poirier TI, Giudici RA. Evaluation of drug interaction mi-
22. Herr RD, Caravati EM, Tyler LS, Iorg E, Linscott MS. Pro-
spective evaluation of adverse drug interactions in the emer-
gency department. Ann Emerg Med. 1992; 21: 1331–6.
crocomputer software: comparative study. Hosp Pharm. 1991; 23. Goldberg RM, Mabee J, Chan L, Wong S. Drug–drug and
26:30–3, 37.
drug–disease interactions in the ED: analysis of a high-risk
population. Am J Emerg Med. 1996; 14:447–50.
24. Gilman AG, Rall TW, Nies AS, Taylor P. Goodman and Gil-
man’s The Pharmacological Basis of Therapeutics, 8th ed. New
York: Pergamon Press, 1990, pp 362, 427, 428, 499.
25. McInnes GT, Brodie MJ. Drug interactions that matter: a
critical reappraisal. Drugs. 1988; 36:83–110.
26. Lesar TS, Briceland L, Stein DS. Factors related to errors
in medication prescribing. JAMA. 1997; 277:312–7.
27. Ford DR, Rivers NP, Wood GC. A computerized detection
system for potentially significant adverse drug–drug interac-
tions. J Am Pharm Assoc. 1997; 17:354–7.
28. Puckett WH Jr, Visconti JA. An epidemiological study of
the clinical significance of drug–drug interactions in a private
community hospital. Am J Hosp Pharm. 1971; 28:247–53.
29. Kwan TC, Wahba WW, Wildeman RA. Drug interactions:
a retrospective study of its epidemiology, clinical significance
and influence upon hospitalization. Can J Hosp Pharm. 1979;
32:12–6.
30. L-Lacoste L, Lam JY, Hung J, Waters D. Oral verapamil
inhibits platelet thrombus formation in humans. Circulation.
1994; 89:630–4.
31. Verzino E, Kaplan B, Ashley JV, Burdette M. Verapamil–
aspirin interaction. Ann Pharmacother. 1994; 28:536–7.
32. Ring ME, Martin GV, Fenster PE. Clinically significant
anti-platelet effects of calcium channel blockers. J Clin Phar-
macol. 1986; 26:719–720.
33. Leape LL, Bates DW, Cullen DJ, et al. Systems analysis
of adverse drug events. ADE Prevention Study Group. JAMA.
1995; 274:35–43.
34. Leape LL, Brennan TA, Laird N, et al. The nature of ad-
verse events in hospitalized patients. Results of the Harvard
Medical Practice Study II. N Engl J Med. 1991; 324:377–84.
35. Bates DW, O’Neil AC, Boyle D, et al. Potential identifia-
bility and preventability of adverse events using information
5. Poirier TI, Giudici RA. Evaluation of drug interaction mi-
crocomputer software: an updated comparison. Hosp Pharm.
1995; 30:888–90, 893–4.
6. Jankel CA, Martin BC. Evaluation of six computerized drug
interaction screening programs. Am J Hosp Pharm. 1992; 49:
1430–5.
7. Poirier TI, Giudici RA. Drug interaction microcomputer
software evaluation: Drug Therapy Screening System (DTSS).
Hosp Pharm. 1990; 25:738–41, 4.
8. Poirier TI, Giudici RA. Drug interaction microcomputer
software evaluation: the medical letter drug interactions pro-
gram. Hosp Pharm. 1990; 25:458–61.
9. Poirier TI, Giudici RA. Drug interaction microcomputer
software evaluation: S-O-A-P. Hosp Pharm. 1990; 25:342–7.
10. Poirier TI, Giudici RA. Drug interaction microcomputer
software evaluation: Rx/Triage. Hosp Pharm. 1990; 25:136–40.
11. Poirier TI, Giudici RA. Evaluation of drug interaction mi-
crocomputer software: Dambro’s Drug Interactions. Hosp
Pharm. 1990; 25:30–3, 72.
12. Poirier TI, Giudici RA. Drug interaction microcomputer
software evaluation: Drug Master 89. Hosp Pharm. 1989; 24:
1003–6.
13. Poirier TI, Giudici RA. Drug interaction microcomputer
software evaluation: Therapeutic Software’s Adverse Interac-
tions. Hosp Pharm. 1989; 24:936–40.
14. Poirier TI, Giudici RA. Evaluation of drug interaction mi-
crocomputer software: drug interactions III. Hosp Pharm.
1989; 24:522–6.
15. Poirier TI, Giudici RA. Evaluation of drug interaction mi-
crocomputer software: Medicom Micro Plus. Hosp Pharm.
1989; 24:273–7.
16. Poirier TI, Giudici RA. Evaluation of drug interaction mi-
crocomputer software: Part 1—Hansten Drug Interaction
Knowledge Base Program. Hosp Pharm. 1988; 23:1062–5,
1070, 1079.
17. Poirier TI, Giudici RA. Evaluation of drug interaction mi- systems. J Am Med Inform Assoc. 1994; 1:404–11.