You are on page 1of 10

Identification o f Serious D r u g - D r u g Interactions: Results of t h e Partnership t o P r e v e n t D r u g - D r u g Interactions

Daniel C. Malone, Jacob Abarca, Philip D. Hansten, Amy J. Grizzle, Edward P. Armstrong. Robin C. Van Bergen, Babette S. Duncan-Edgar. Steven L. Solomon. and Richard B. Lipton
Received April 21, 2003, and in revised form September 9, 2003. Accepted for publication November 15,2003. Daniel C. Malone, PhD, and Edward P. Armstrong, PharmD, are Associate Professors; Jacob Abarca, PharmD, is Assistant Research Scientist; and Amy J. Grizzle, PharmD, is Assistant Director, Center for Health Outcomes and PharmacoEconomic Research, College of Pharmacy, University of Arizona, Tucson. Philip D. Hansten, PharmD, is Professor, School of Pharmacy, University of Washington, Seattle. Robin C. Van Bergen was formerly Manager and Babette S. Duncan-Edgar, PharmD, is Vice President, Center for Health lmprovement. AdvancePCS Center for Healthier Aging, Hunt Valley, Md. Steven L. Solomon, MD, is Acting Director, Division of Health Care Quality Promotion, Centers for Disease Control and Prevention, Atlanta, Ga. Richard 6. Lipton, MD, is Professor, Departments of Neurology and Epidemiology and Population Health, Albert Einstein College of Medicine, New York, N.Y.


Objective: To develop a list of clinically important drugclrug interactions (DDls) likely to be encountered in community and ambulatory pharmacy settings and detected by a computerized pharmacy system. Design: Cross-sectional, one-time evaluation. Setting: United States in fall 2001. PartiCipant~: expert panel comprising two physiciat~s, clinical pharAn two macists, and an expert on DDIs. Interventions: Systematic review of drug interaction compendia and published literature, ratings (on a 1 to 10 scale) of various clinical aspects of DDIs (e.g., clinical importance, quality and quantity of evidence, causal relationship, risk of morbidity and mortality), and a modified Delphi consensus-building process. Main Outcome Measure: Panelists' opinions about clinical iniportance of DDIs. Results: The expert panel considered 56 DDIs. Of these, 28 had a mean clinical importance score of 8.0 or more. The ratings for clinical importance ranged from 3.2 to 9.6, with a mean + SD of 7.5 + 1.5 across the combinations examined. The mean score for the quality of literature suggesting the interaction exists ranged from 1.0 to 9.6, with a mean + SD of 5.8 + 2.5. In terms of substantiation of the interactions evaluated, the mean + SD rating was 6.3 + 2.2, with a range from 1.4 to 9.2. Through the modified Delphi process, the panel determined that 25 interactions were clinically important. Conclusion: Using an expert panel and a standard evaluation tool, 25 clinically important drug interactions that are likely to occur in the community and ambulatory pharmacy settings were identified.Pharmacists should take steps to prevent patients from receiving these interacting medications, and computer software vendors should focus interaction alerts on these and similarly important DDIs. Key words: Drug4rug interactions, drug utilization review, medication errors, patient safety. J Am Phann Assoc. 2004;44: 142-1 51.

Correspondence: Daniel C Malone, PhD, College of . Pharmacy, University of Arizona, PO Box 210207, Tucson, AZ 85721-0207. Fax: 520-626-7355. E-mail: See related articleson pages 128 and 136. Disclosure: Dr. kmsten is an editor of and declares financial interests in the textbook, Hansten and Horn's Drug lnteractjonAnalysjsandManagement,Dr. Lipton has financial interests in AdvancePCS, and as noted above, Ms. Van Bergen is a former employee and Dr. Duncan-Edgar is a current employee of that firm. The other authors declare no conflicts of interest or financial interests in any product or sewice mentioned in this article, including grants, employment, gifts, stock holdings, or honoraria. Funding: Supported by the Centers for Disease Control and Prevention, Contract 2001-N-08014. Acknowledgment: The authors would like to thank John Palmer, MD, for participatingon the expert panel used in this study. Presented previously at the Academy of Managed Care Pharmacy Annual Meeting, Minneapolis, Minn., April 10. 2003. and the American Academv of Familv ~ G s i c i a n s National Ambulatory primary car; Research & Education Conference on Patient Safetv. ,. Chicago, Ill., September 18,2003.

142 Journal of the American F'hannwkts Association

MamMApdI 2004 Val. 44, No. 2

Identifying Serious Drug-Drug Interactions


e 1999 Institute of Medicine's (IOM) report on medical errors, To Err Is Human, thrust into the spotlight numerous inadequacies of the American health care system.' This report suggested that between 44,000 to 98,000 persons die annually from medical errors within hospitals. The IOM specifically identified medication e m r s as a major subcategory of all medical emrs. Further, the economic burden of drug misadventuring is substantial. Using an expert panel and decision analysis model, Johnson and Bootman estimated in 1995 that the direct costs of drug-related morbidity and mortality were $76.6 billion.' An updated analysis of this model estimated the direct costs of drug misadventuringin 2000 had risen to $177.4 b i l l i ~ nWith the total .~ cost of prescription drug expenditures at $145 billion," more than $1 is consumed for direct medical costs related to drug misadventures for every dollar spent on the medications themselves. Medication errors, a subset of medical errors, are estimated to account for more than 7,000 deaths annually in the United States? The contribution of drug-drug interactions (DDIs) to the total of these e m r s in the United States is unknown. One case report involving an interaction of fluoxetine and selegiline required a 15day hospitalization, emergency room visits, ambulance services, magnetic resonance imaging, electrocardiogram, laboratory tests, and consultations. The resultant total medical expenditures for treatment of thi5 single case of an interaction-induced illness was $17,213.6 The aging of the population and increasing complexity of medication regimens used to treat ambulatory patients and a fragmented health care system with multiple prescribers managing each patient make the occurrence of serious DDIs even more likely. Community pharmacists rely on their knowledge of interactions

Synopsis: An expert panel-two physicians, two clinical pharmacists, and an expert on drug interactionssystematically reviewed the available evidence and compiled a list of 25 clinically important drug-drug interactions (DDIs) in the community and ambulatory pharmacy ~ettings. These interactions should be emphasiled by increased pharmacists' vigilance and changes to computer software program to protect patients from harmful DDIs. Analysis: As demonstrated in a companion article and discussed in an editorial in this issue of JAPhA, DDI compendia vary considerably in their ratings c?f the clinical severity of interactions. Tlirough a proces F sirch us this one, clinically relevant DDIs can be identijied. Similar efforts directed at improving the uniformity and agreement among major DDI compendia are needed for interactions that are possible in all patient-cure settings.

and dispensing software to identify potential DDIs. In addition, community pharmacists have been at the forefront in applying technology designed to reduce medication-related errors, including DDls. Most pharmacies use in-store prospective drug utilization review (PDUR) software that provides real-time, concurrent review of medication regimens to identify potential drug therapy problems and prevent adverse events, including D D k 7 In addition, many pharmacy benefit managers (PBMs) and claims processors operate online concurrent PDUR software that provides review of medication regimens. One advantage of PBM-initiated PDUR is that all of the patients' medication claims can be reviewed, not just those filled by a particular pharmacy or chain? Though concurrent PDUR systems identify medication-related problems before prescriptions are dispensed, evidence indicates that these systems need improvement? The evidence base for the DDls targeted by PDUR is inconsistent, and recommendations on managing DDIs is often not helpful. Pharmacists often have difficulty in distinguishing between important messages about patient safety and the "noise" created by false alerts in many of these systems. Many PDUR warnings, for example, do not require intervention or are not clinically relevant. Given the tremendous volume of messages provided to pharmacists and pharmacy personnel, the majority of patient safety messages are overridden. One study found that 88% of PDUR alerts in community pharmacies were ignored or overridden by pharmacists? Therapeutic duplication and DDIs accounted for 20% and 15% of these alerts, respectively. To improve electronic systems and patient safety in the area of DDIs, we created the partnership to prevent DDIs (PP-DDIs) with funding from the Centers for Disease Control and Prevention (CDC). The partnership consists of a large PBM (AdvancePCS), three academic institutions (University of Arizona, Albert Einstein College of Medicine, University of Washington), and CDC. The PP-DDIs' overall objectives were to identify clinically important DDIs based upon a systematic review, identify and assess current strategies for preventing DDIs. and to design, implement, and assess better strategies for preventing DDZs.

0 bjectives
The purpose of this study was to develop a list of the clinically important DDIs that are likely to be encountered in community and ambulatory pharmacy settings and detected by a computerized pharmacy system.

A three-stage process was undertaken to identify DDIs of the highest clinical importance. Clinically important interactions were defined a those interactions most likely to cause hann if not 5 detected. This process began with the selection of candidate interJournal o the American Phamdsts Awnciation f

VoL 44, No. 2





Identifying Serious Drug-Drug Interactions a literature search was conducted to identify recently published articles for each DDI. Index Medicus (1995 to January 2002; searched electronically through Ovid MEDLINE) and International Pharmaceutical Abstracts ( 1995 to January 2002; searched electronically through Ovid) were queried for primary literature using the names of the medications listed for each drug interaction as search tenns. The resulting literature searches were reviewed, and the relevant literature was identified and assembled. For each drug interaction, the assembled literature was examined by one of the authors, a clinical pharmacist (JA). A one-page evidence report was prepared for each DDI, including a summary of the interaction, the evidence supporting the interaction, and a brief synopsis of the supporting information for each interaction. In addition, several key articles supporting the DDI were identified. Another author (either DCM, AJG, or in a few instances EPA) verified the contentand accuracy containedin each evidence report. The evidence report and key articles for each candidate drug interaction were distributed to an expert panel for review. This panel was comprised of an expert in drug interactions (PHD), two practicing physicians with expertise in internal medicine (not an author) and neurology (RBL), and two clinical pharmacists (EPA and BDE).

actions from a review of drug interaction compendia followed by a systematic literature review and the development of an evidence report for each candidate DDI. These evidence reports were then evaluated and rated by an expert panel using a modified Delphi process.'0 For each DDI combination, the object and precipitant drugs were defined for each when possible. For a particular DDI, the object drug was defined as the medication whose therapeutic effect is modified leading to the misadventure associated with the interaction. The precipitant drug is the medication that is responsible for affecting the pharmacologic actions or phmacokinetics of the object drug (e.g., competition for active site. inhibition of metabolism).

Selection of Candidate DDIs The first stage in our process was to select candidate DDIs from existing drug interaction compendia. Details of this stage, summarized here, are described in full in a companion paper being published in thisissue of JAPhA.I1 Interactions rated as most severe or "major" were selected from each of the four commonly used DDI Drug compendia (Evaluarion of Drug I n t e r a c r i o n ~ , ~ ~ Interaction Fa~ts,~"rug Interactions: Analysis and M a n ~ ~ e r n e n tand ,'~ MICROMEDEX (DRUG-REAX) s stem' 5, and then cross-referenced to each other. These compendiaprovide inconsistentclassification to determine which interactions are clinically important1 and were chosen because their focus is on DDIs, as compared to other tertiary references that include a drug monograph that may include a section pertaining to interactions. Since each compendium uses a slightly different rating system to classify DDIs. criteria specific to each was developed. Interactions were selected for further review if they were listed in three of the four compendia. For interactions at the class level, specific drugs were aggregated to create a drug-class group (e.g., nonsteroidal anti-inflammatory drugs [NSAIDs], sulfbnarnides, monoamine oxidase inhibitors [MAOIs], selective serotonin reuptake inhibitors [SSRIs]). The focus of this research was to prevent DDIs in community and ambulatory pharmacy settings. Thus, several interactions were excIuded as previously mentioned1]:(1) the drug was not available for human use in the United States: (2) the drug was not routinely dispensed in the conlmunity pharmacy setting (e.g., dopamine, halothane); or (3) the interaction involved a product not likely to be captured by a computerized database, such as nonprescription medications, alcoholic drinks, or foods. Two additional criteria were applied in this analysis to exclude DDIs in which (1) the medications were commonly used together to achieve a therapeutic benefit or (2) the interaction occurred upon discontinuation of one of the medications.


Gathering Evidence About DDls In the second stage, evidence about those DDIs identified in the first stage was collected. The references cited for each DDI in each compendium were obtained and assembled for review. In addition,
Journal of the American Pharmadrts Awocialion

Evaluation of Candidate DDls The third stage of this investigationwas designed to narrow the list of DDIs to those that were clinically important, well-supported interactions that should become primary targets for future surveillance and intervention. The expert panel was provided a summary of each interaction, copies of the key articles, and a bibliography for each interaction. Panel members could request fulltext copies of additional references as desired. Expert panel members were asked to evaluate each DDI using a I ditem instrument developed by members of the research team (DCM, AJG, EPA), based in part upon a previously published instrument (copies of the instrument available from the corresponding author).16 Four constructs related to DDls were initially identified. These constructs included: severity, evidence, modifying factors, and probability of occurrence. Each of these constructs was discussed via conferelice calls involving all of the investigators and expert panel members, and a final version of the evaluation form was agreed upon. I'he final four sections of the questionnaire included the following: 1 . Three items about the evidence supporting the interactionquantity of data, quality of the data, biological plausibility. 2. Four items about the severity of the interaction-clinical importance, likelihood of mortality, likelihood of irreversible morbidity, likelihood of a prescriber taking action to prevent harm once an interaction is identified. 3. Five items on the probability of the interaction-likelihood of the interaction producing adverse consequences in randomly selected patient, effect of timing of administration, effect of age, effect of gender, effect of concurrent disease states.
mvwjaphaorg MarchIApd 2004 Vol. 44, No. 2

Identifying Serious Drug-Drug lnteractions


4. Four items about the probability of co-administration of the interacting drugs-likelihood of combination being used togedier f-fkaihyh1dicombi ------ nationp-7 difkgntpractitioners, clinician m w at G of clinician work required to evaluate and manage the interaction. &&Ions wereGchored using a semantic differential response scale when appropriate. The response choices for each question ranged from I to 10. Intraclass correlation coefficients (ICC) for each question were calculated using the method of Shrout and ~ 1 e i s s .An overall measure of the 16-item instrument was calcul~ lated for each DDI by summing the responses to each question. An overall ICC was calculated for the summary measure.

Table 1. Major Interactions as Listed by Drug Interaction Compendiaa

No. Maior Interactions


- -

Modifications Made During Delphi Process This expert panel participated in a modified Delphi survey approach to rank order the interactions. In Round 1 of the process, each of the experts independently assessed the DDls using the evaluation tool described above. The mean response was calculated for each item by DDI. For Round 2, the experts were given their own scores and the group mean scores for each question and were asked to re-evaluate their scores in light of the group mean scores. Panel members discussed the ratings via conference call during this stage of the research. By a consensus decision, DDIs could be eliminated from further consideration or have additional limitations (e.g., nonselective beta-blockers only, low-dose methotrexate only) placed on how each DDI would be evaluated. Also during this round, DDIs not identified by at least three of the interaction compendia could be added by any member of the expert panel if the expert believed that such an interaction met the criteria of this investigation. Summaries of the added DDIs were distributed to the expert panel for evaluation using the same 16-item instrument. An additional conference call was held to discuss the ratings for the new DDIs. Any of the new DDIs could be eliminated from further consideration by a consensus decision. For the third and final round, the interactions were rank-ordered using five different scales based on single or multiple questions from the evaluation tool. The primary rank order was based on clinical importance. However, the expert panel had access to the other four rank orders, which included the severity of the interaction, the likelihood that the combination would be prescribed for therapeutic benefit, the likelihood of clinician awareness of the interaction, and the published evidence supporting the interaction. The panel members were asked to determine which DDIs represented serious and clinically important interactions about which clinicians and pham~acists should always be infom~ed.

Table 1 lists the number of severe or major interactions identified in each compendium. The listed interactions were then aggreVoL 44, No. 2

gated at the medication class level when appropriate. Sixty-two DDIs were found in at least 3 out of the 4 compendia. As detailed below, 18 DDIs were excludedfrom further consideration for one of the five reasons for exclusion presented in the Methods. Eight DDIs were excluded because at least one of the medications was no longer availablein the United States (the nonsedating antihistamines terfenadine and astemizole interacting with macrolide antibiotics, azole antifungal agents, the SSRIs fluoxetine or fluvoxarnine, or rnibefradil; anticoagulants-phenylbutazones; cisapride-nefazodone; cisapride-protease inhibitors; cisapride*ole antifungal agents). Four DDIs were excluded because at least one of the medications was not likely to be dispensed in a community pharmacy (nondepolarizing muscle relaxants-aminoglycosi&s; metformin-iodinated contrast agents; theophyllines-halothane;and hydantoinsdopamine). An additional four DDIs involved products that were not usually captured in a pharmacy claims database (disulfiram+thanol; acitretin+thanol; chlorpropamide-ethanol; tyramine-MAOIs). One DDI was excluded because the interaction occurs upon discontinuation of the medication (clonidine-beta-blockers) and another one because the combination was likely to be used for therapeutic benefit (phenothiazinesSSR1). Thus, a total of 44 DDIs identified through the compendiareview were evaluated by the expert panel during Round 1 (Table 2). Because of the wide disparity among the four compendia in classifying the interactions, expert panel members were asked to suggest additional interactions that they deemed clinically important. These additional 15 interactions were proposed by individual members of the expert panel: Carbamazepine-macrolides Carbamazepine-propoxyphene Pimozide-macrolides Pimozide-az~leantifungal agents Theophyllinequinolones Theophyllinefluvoxamine Phenytoin-fluoxetine or fluvoxamine Rifarnpin-oral contraceptives Rifampin-diltiazem
Journal o the h r i c a n P h a r m a h h ~ t i o n 145 f


Identifying Serious Drug-Drug Interactions

Table 2. Mean Expert Panel Ratings on a 1 to 10 Scale of Clinical Aspects of Major Drug-D~g Interactions

Object Drug

Precipitant Drug llopurinol acrolide antibiotics 9.2 8.6 9.0 .6 6.4 3.0 4.4 9.8 9.4 8 8.0 6.0 8.4 2.0 48 4.6 1.0 1.8 1.0 1.8 1.2 1.8 2.6 1.2 1.2


6.4 2.2 4.4

6.0 5.2 8.4


6 6

4.0 34

4.4 3.0

8.8 8.8

6.0 6.2

7.8 3.4

4.2 1.0

3.4 1.0

7.0 2.0


7.4 5.4


2.8 5.6




1.8 1.




















7.6 Androgens 7.4


1.8 5.2 6.8 3.2 3.6 9.4 5.0 3.2 1.4

1.6 1.4 2.4 2.0 6.2 3.6 4.0


146 Journal d Ule Amrican Pharnmbb Association


M a d A p d 2004 Val. 44, No. 2

Identifying Serious Drug-Drug lnteractions


Table 2. Mean ~ x p e s Panel Ratings on a 1 to 10 Scale of Clinical Aspects of Major Chug-Drug Interactions (continued)

Object Drug

Precipitant Drug Sympathomimetics Clarithrornycin 7.2 6.8 4.2 3.6 4.6 5.2 4. .6 8.4 7.4 4.8 4.4 5.0 3.4 1.0 2.4





Theophyllines Allopurinol Sibutramint, Serotonin Sibutramine (5-hydroxytryptamine-1) agonists tript tan^)^




7.0 2.0 1.0

7.2 2.6

1.8 2.2

1.4 2.6 1.8 2.0

8.0 5.4

5.8 2.4 2.6 2.0 4.0



1.0 3.8


6.6 5.8

6.4 2.4









1.4 6.0 3.0 1.0

3.2 9.8 8.4 1.6

1.8 8.4 5.8 1.7


1. 4.




lntraclass correlation a m o n g expert raters

0.39 0.71 0.65 0.60

Abbreviations used: MAO, m o n o a m i n e oxidase; SSRls, selective serotonin reuptake inhibitors; NSAIDs. nonsteroidal anti-inflammatory drugs; H M G Co-A, 3-hydroxy-3-methylglutaryl coenzyme A.

- - -I

.32 0.26 G 3 0.16 0.16

0.29 0.17

0.20 0.53 0.24 0.61 0.07

Rifarnpin4fedipine Rifampin-simvastatin or lovastatin Rifampin-triazolam Rifampin-verapamil Digoxin~larithromycin Warfarin-NSAIDs After discussing these additional interactions, the panel decided to eliminate all the interactions involving rifampin, except its interaction with oral contraceptives and its previously included interaction with cyclosporine. The rationale for this was that interactions involving rifampin lead to a decreased therapeutic effect of the object drug. With the exception of oral contraceptives and cyclosporine, the result of a decreased therapeutic effect would likely be less severe than the occurrence of a supratherapeutic or toxic effect observed in other interactions. For the remaining addiV L 44, No. 2 o

tional interactions, a literature search was undertaken and a summary of the interaction was written as was done for the first set of interactions. The expert panel then rated these interactions (Table 2). Including the interactions derived from the compendia plus the additionalinteractionsrecom~nended the expert panel, a total of by 54 DDls were evaluated. During Round 2, an additional 12 DDIs were eliminated because of various mitigating factors: aspirin-heparin; levodopa-MAOIs, potassium-potassium-sparing diuretics; methotrexate-salicylates; gemfibrozil-HMG-CoA reductase inhibitors; anticoagulantssalicylates; and six interactions involving sibutrarnine. Aspirin, the most commonly used salicylate, is typically not captured by community pharmacy databases and, thus, would not be amenable to a computerized PDUR intervention. The interaction involving levodopa and MAOls is tnitigated
Journal d the American Pharmackb hwoclation



Identifying Serious Drug-Drug Interactions than 8.0 were also added to this list of interactions, including theophylline-quinolones, theophylline-fluvoxamine, and digoxinclarithromycin. However, the expert panel believed that the severity of these interactions was not adequately represented by the clinical importance score alone. The panel also recommended that three interactions be removed from this list: epinephrine-betablockers, TCA-MAOJs, and methotrexate-NSAIDs. The first interaction was removed because the benefit of epinephrine administration (most likely for anaphylaxis) would outweigh the risk associated with concurrent administration. The second interaction was eliminated because published studies have documented that this combination can be safely administered together, making computerized alerts for this interaction a nuisance. Similarly, the combination of methotrexate.NSATDs is frequently prescribed together for a variety of disorders (e.g., rheumatologic disorders). The ZCCs for agreement among the expert panel ranged from 0.07 to 0.71, depending upon the question (ICC values closer to 1 represent greater levels of agreement). Questions pertaining to quantity of the evidence, quality of the evidence, causal relationship, and use in combination for therapeutic use had ICCs greater than 0.5. The ICC for clinical importance was 0.39. When the score for each question for each interaction was summed, the ICC was 0.63, suggesting a moderate to good level of agreement between the expert panel members. After all deliberations, the expert panel believed 25 interactions to be "clinically important" (Table 3). Most interactions were initially grouped at the drug class level when two or more products from a therapeutic class were considered, but when only a single entity from a therapeutic class was involved, only that medication name is listed.

when levodopa is administered with carbidopa, as is commonly diuretics done. The combinations of potassium-potassium-sparing and gemfibrozil-HMG-CoA reductase inhibitors were eliminated since they are sometimes used for therapeutic benefit. In addition, the expert panel members had rated interactions involving sibutramine low across the four constructs, mainly because these interactions are theoretical and there was a lack of evidence relative to the other interactions evaluated. The expert panel also agreed to reevaluate specific interactions using additional criteria. For instance, interactions involving methotrexate were reevaluated based on the assumption that a patient would be on low-dose methotrexate. The experts also decided to reevaluate the interaction between epinephrine and beta-blockers by considering only the nonselective beta-blockers. For the interaction between tricyclic antidepressants (TCAs) and MAOJs, the panel limited its analysis to clomipramine and imipramine since these entities have more serotonin activity than do other TCAs. The interaction involving azole antifungal agents and benmdiazepineswas reevaluated considering only two benzodiazepines, triazolam and alprazolam. The results of the DDl rankings for interactionsfrom Round 3 are listed in Table 2. Of the 54 DDls evaluated, 28 had a mean rating of 8.0 or more on our 10-pointclinical importance scale. The range for clinical importance ranged from 3.2 to 9.6, with a mean f SD of 7.5 + 1.5 across the combinations examined. The mean score for the quality of data that suggested a potentially harmful interaction ranged from l .O to 9.6, with a mean + SD of 5.8 + 2.5. In terms of substantiationof the interactionsevaluated, the expert panel's mean + SD rating was 6.3 2.2, with range from 1.4 to 9.2. The likelihood of mortality if the patient received the combination was rated fairly low overall by the expert panel members (mean SD = 3.1 1 .I). The risk of mortality was rated highest for pimozide-macrolide antibiotics, with an average rating of 6.4 by the expert panel. When asked to rate the likelihood of serious. irreversible morbidity, the combination of cyclosporinerifamycins had the highestoverall swre (6.0) fiom the expert panel. When the expert panel was asked how likely they would be to take action to prevent harm in the event they became aware of a patient taking the combination, the mean + SD score across all interactions was 7 . 9 2 1.6. Seventeen interactions had mean scores of 9.0 or more. The probability that different practitioners would prescribe the object and precipitant medications averaged (+ SD) 5.9 2 1 .4, with the combination of oral contraceptives-rifampin rated as being most likely (8.5) to be prescribed by different practitioners. The combinations of dextromethorphan-MAOIs, pimozide-azole antifungal agents, and bromocriptine-sympathomimetics were all rated as having a high likelihood to be prescribed by different practitioners. To identify a finite number of clinically important interactions, the expert panel agreed to first rank the interactions based on a group mean for clinical importance. The initial list of important DDIs included those interactions with a mean clinical importance score of 8.0 or higher. Three interactions with mean scores lower

We used a systematic evaluationof the literature and an expertpanel process to identify clinically important DDIs. Our panel evaluated 54 DDIs and assigned a mean score for clinical importance of 7.5 on a 1 to 10 scale. After excluding products that were not on the U.S. market, products not likely to be used in an outpatient setting. and combinations that can be used for therapeutic benefit, a total of 25 clinically important DDls were identified. Hundreds of drug interactions are listed in the various drug interaction compendia; however, these references vary considerably, and this is the first effort to systematically rank order interactions in temis of their clinical importance. The identification of new DDIs continues as we learn more about drug metabolism and as cellular and biochemical analytical techniques evolve.18 The number of new molecular entities approved by the Food and Drug Administration with DDI data increased fiom 30% for those prnducts approved from 1987 to 199 I, to 53% for those approved from 1992 to 1 997.19 Pharmaceutical manufacturers appear to be increasingly diligent in identifying and reporting ~ ~ 1 However, DDl compendia appear to rely on individual authors or

f 148 Joumal o the American Phannarists Aaociation

MarchIApIi12004 Val. 44, No. 2

Identifying Serious Drug-Drug Interactions



Table 3. DN-N~

Interactions Selected by Panel as Having Greatest Clinical Importance

Azole antifungal agents Propoxyphene Rifamycins (rifampin, rifabutin, rifapentine)

, ketoconazole)

Digoxin tamine, ergotamin Estrogen-progestin products (oral contraceptives) Ganciclovir M A 0 inhibitors (isocarboxazid, phenelzine, selegiline. tranylcypromine) ergide)


Rifampin Zidovudine Anorexiants (amphetamine, benzphetamine, dexfenfluram~ne, dextroamphetamine, diethylpropion, fenfluramine, mazindol, methamphetamine, phendimetrazine, phentermine,

paroxetine, sertraline. ventaf Theophyllines

Thiopurines (azathioprine, mercaptopurine)



Nonsteroidal anti-inflammatory drugs (celecoxib, diclofenac, etodolac, flurbiprofen, fenoprofen, ibuprofen, indomethacin, ketoprofen, ketorolac, meclofenamate, mefenamic acid, nabumetone, naproxen, oxaprozin, piroxicam, rofecoxib, sulindac, tolmetin)


Fibric acid derivatives (clofibrate, fenofibrate, gemfibrozil)

Abbreviations used: MAO, m o n o a m i n e oxidase; SSRls, selective serotonin reuptake inhibitors.

a small number of editors to evaluate interactions, and the result is considerable inconsistency among the compendia on what represents a serious DDI.11,21 In addition, many PDUR alerts are issued for medications that while potentially interacting, are commonly prescribed together
Vo1.44, No. 2 MarchIApril 2OW

for therapeutic benefit. Pharmacists have not surprisingly been conditioned to ignore the vast majority of drug interaction alerts? Even pharmacy computer systems vary considerably as to which interactions trigger warnings! Few published articles examine pharmacists' knowledge of
Joumsd nf tbe Amerienn Phannnckb Assodation




Identifying Serious Drug-Drug Interactions

DDIs, but one study suggested that pharmacists' recall of serious prey to the same signal-to-noise problems that currently plague DDIs is approximately 50% to 70%. In this analysis, pharmacists' community pharmacies during the dispensing process. The high number of alerts that do not require pharmacist action ability to detect serious DDIs was a function of the number of medications in the patient profile.22 diminishes the effectiveness of clinically important alerts.29 Alert A study of prescribers' knowledge of interactions found that messages sent to pharmacies are often viewed as burdensome and clinicians correctly classified drug-drug combinations as interact- are many times overridden or PBMs and pharmacy ing or not about 50% of the timeF3 More importantly, 28% were computer software vendors provide comprehensive alert mesnot sure whether sildenafil and isosorbide interacted, 27% were sages, in part, to prevent litigation? However, issuing too many uncertain whether there was an interaction between cisapride and alerts has negative consequences. The goal of our study was to erythromycin, and 43% were not sure whether concomitant use of identify those DDIs that should always generate an alert to a physiphenelzine and sertraline presented problems. All of these interac- cian or other prescriber and the pharmacist. As a starting point to tions are potentially life-&~reatenin~?~ improve computer systems that support patient care, alert mesThe consequences of being exposed to an interaction are not sages should occur only when DDIs are clinically relevant and trivial. Juurlink et al.24recently reported that the risk of hospital- have a high probability of causing harm. The literature has demonstrated pharmacists and prescribers ization substantially increases for those patients exposed to a DDI. Computerized physician order entry (CPOE) has been touted as cannot rely on computerizedsystems alone to identify interactions. the best method to prevent many types of medication errors, [n addition, other studies have found that health professionals' Automated prescribing systems knowledge of serious interactions is lacking. Therefore, multiincluding illegible h a n d ~ r i t i n g ? ~ have been suggested as a possible solution to decreasing medica- faceted interventions are important, including continuing education errors.26 Advances in computer technology, however, have tion. retrospective drug utilization review, and letters sent to preexpanded the role of these systems, providing opportunities to scribers with patient-specific information. Other strategies include reduce a variety of potential medication errors. Physicians make changing the claims-processing system to provide unique alerts to prescribing decisions based upon individual patient characteristics, pharmacies for those interactions considered clinically important. One diseases, and numerous social and economic f a c t ~ r s . ' ~ , ~ ~ such approach may be the institution of hard edits for some Technological systems, though sophisticated in their own right. critically important DDls (that is, preventing a claim for a medicalack much of the patient data and the medical reasoning that is part tion involved in a serious interaction from being processed). However, these measures should not be implemented without of the prescribing process?9 Nonetheless, previous research has found beneficial effects of thorough testing and evaluation to prevent other unknown or unanCPOE systems in institutional settings. Serious medication errors ticipated consey uences that may cause patient harm. were reduced in an inpatient setting by more than 50% (from 10.7 to 4.86 events per 1,000 patient days) following implementationof direct CPOE." A computerized anti-infective management pro- Limitations gram resulted in significant reductions in errors caused by allergies (76%) and excessive drug dosages (78.5%)." Findings such as The reader should keep severdl limitations in mind when interthese have led the National Patient Safety Partnership to recorn- preting our findings, most notably the use of an expert panel to rate the DDIs. Expert panels make subjective judgments bawd on the mend implementing direct CPOE strategiesz6 Electronic prescribing systems may be able to link to patient- best evidence available, but these conclusions may be imperfect. s level information, such as health plan eligibility, formulary list- The ~ s u l t are likely to differ depending upon the membership of ings, patient history, medication records, allergies, and concomi- the panel. By using a modified Delphi approach, we were able to tant condition^?^ Some systems allow physicians to enter patient minimize the potential for coalescence around one or two members' diagnosis and receive instantly a list of possible alternative treat- opinions and to ensure that all members were equally recognized. ments with appropriate dosing i n f ~ r m a t i o n ? ~ the introduc- The ICC values indicated only moderate to good agreement among With tion of the personal data assistants (PDAs), such data are available members of the expert panel, with substantially lower levels of at the point of care, providing the clinician with all the pertinent agreement on certain questions, reflecting divergence of opinions. The interactions rated by the expert panel consisted of both information needed to make an appropriate prescribing decision. PDAs permit the generation of printed, faxed, or electronically those derived from a systematic review of the compendia as well submitted prescriptions. Many systems are designed to perform as Fro~nindividual members. We noted in Table 2 whether the PDUR functions, issuing alerts regarding potential DDIs, incorrect interaction was a suggested interaction from the expert panel or dosing, drug allergies, therapeutic duplication, drug-disease con- from the drug compendia. Readers should remember that these traindications, and excessive utilization. These warnings may be interactions represent a selected set of all possible DDIs. In addiprovided to the clinician at the time of order entry, or may be tion to those we identified for the community and ambulatory issued after transmittal of a prescription to the electronic prescrib- pharmacy setting, other interactions are clinically important and ing vendor's server. However, such CPOE technologies may fall should be prevented.
150 Journal dthe American Phameists AaocinticIn
MaretJApd 2004 Vol. 44, No. 2

Identifying Serious Drug-Drug Interactions


Also, expert panel members were presented with the published evidence supporting each interaction, but these studies were not explicitly rated for study quality. An alternate approach might have been to rate these studies and place more weight on those studies that provided a higher level of evidence. This approach was not taken because the vast majority of DDI literature consists of case reports. Thus, most studies would have been assigned the same weight. To overcome this limitation, a summary of the literature was assembled for each interaction to be evaluated, and individual articles were made available to experts. We also did not include every drug-reference book or information source that contains information about DDIs. We explicitly chose those compendia that were believed to be most commonly used by health care professionals to determine whether an interaction exists. Other compendiathat are commonly available in pharmacies have information on DDIs, including the USP DI Drug Information ,for the Health Care Provider and the American Hospital Formulary Service Drug Information, but these tertiary texts do not consistently contain detailed information on DDIs along with references supporting the interaction. Finally, the 25 DDls that were deemed to be clinically important have not been systematically evaluated in terms of health outcomes and expenditures. A recent study found that persons exposed to a warfarin-NSAID interaction were 5.8 times more likely to experience a bleeding eventP5 Additional similar studies are needed to assess the frequency with which our 25 interactions occur and to quantify the harm, if any, induced by these interactions.

7. Chrischilles EA, Fulda TR, Byrns PJ, et al. The role of pharmacy computer systems in preventing medication errors. J Am Pharm Assoc. 200 1;42:439-48. 8. Hazlet TK, Lee TA, Hansten PD, Horn JR. Performance o f community pharmacy drug interaction software. J A m Pharm Assoc. 200 1;41:200-4. 9. Chui MA, Rupp MT. Evaluation of online prospective DUR programs i n community pharmacy practice. J Manag Care Pharm. 2000;6:2732. 10. Harman AJ. Collecting and analyzing expert group judgment data. Santa Monica, Calif.: RAND Corporation; 1975. 11. Abarca J, Malone DC, Armstrong EP, et al. Concordance of major drug interaction classifications among drug interaction compendia. J A m Pharm Assoc. 2004;44:13641. 12. Evaluations of drug interactions. St. Louis, Mo.: First DataBank, The Hearst Corporation; 2001. 13. Drug interaction facts. St. Louis, Mo.: Facts and Comparisons; 2001. 14. Hansten PD, Horn JR. Drug interactions: analysis and management. St. Louis, Mo.: Facts and Comparisons; 2001. 15. DRUG-REAX system. Greenwood Village, Colo.: MICROMEDEX; 2001. 16. Roberts JS, Watrous ML. Schulz RM, et al. Quantifying the clinical significance of drug-drug interactions: scaling pharmacists' perceptions o f a common interaction classification scheme. Ann Pharmacother. 1996: 30:92-4. 17. Shrout PE, Fleiss JL. lntraclass correlations: uses in assessing rater reliability. Psychol Bull. 197986:420-8. 18. Rodrigues AD, Lin JH. Screening of drug candidatesfortheir drug-drug interaction potential. Curr Opin Chem Biol. 2001;5:396-401. 19. Marroum PJ. Uppoor RS, Parmelee T, et al. In VNO drug-drug interaction studies-a survey of all new molecular entitiesapproved from 1987 t o 1997. Clin Pharmacol Ther. 2000;68:280-5. 20. Huang SM, Lesko LJ, Williams RL. Assessment of the quality and quantity of drug-drug interaction studies in recent NDA submissions: study design and data analysis issues. J Clin Pharmacol. 1999;39:100&14. 21. Fulda TR, Valuck RJ, Vander Zanden J, et al. Disagreementamong drug compendia on inclusion and ratings of drug-drug interactions. Curr Ther Res. 2000;61:540-8. 22. Weideman RA, Bernstein IH, McKinney WP. Pharmacist recognition of potential drug interactions. A m J Health Syst Pharm. 1999;56:15244. 23. Glassman PA, Simon B, Belperio P, Lanto A. Improving recognition of drug interactions benefits and barriers to using automated drug alerts. Med Care. 2002;40:1161-71. 24. Juurlink DN, Mamdani M. Kopp A, et al. Drug-drug interactions among elderly patients hospitalized for drug toxicity. JAMA. 2003;289:1652-8. 25. Schiff GD. Computerized prescriber order entry: models and hurdles. A m J Health Syst Pharm. 2002; 59:145&60. 26. lnstitute for Safe Medication Practices. A call to eliminate handwritten prescription within 3 years [report]. Huntingdon Valley, Pa.: Institute for Safe Medication Practices; 2000:l-12. 27. Manasse HR. Medication use i n an imperfect world: drug misadventuring as an issue o f public policy. Part 2. A m J Hosp Pharm. 1989;46:114152. 28. Manasse HR. Medication use i n an imperfect world: drug misadventuring as an issue of public policy. Part 1. A m J Hosp Pharm. 1989;46:929-44. 29. Soumerai SB, Lipton HL. Computer-based drug-utilization review-risk benefit, or boondoggle? N Engl J Med. 1995;332:16415. 30. Bates DW, Leape LL, Cullen DJ, et al. Effect of computerized physician order entry and a team intervention on prevention of serious medication errors. JAMA. 1998;280:1311-6. 31. Evans RS, Pestotnik S L Classen DC, et al. A computer-assisted management program for antibiotics and other antiinfective agents. N Engl J Med. 1998;338:232-8. 32. Armstrong EP, Chrischilles EA. Electronic prescribing and monitoring are needed to improve drug use. Arch Intern Med. 2000;160:2713-4. 33. Venot A. Electronic prescribing for the elderly: will it improve medication usage? Drugs Aging. 1999;15:77-80.

Using an expert panel and a standard evaluation tool, we identified 25 drug interactions deemed to be clinically important and likely to occur in the community and ambulatory pharmacy settings. These interactions represent a subset of all interactions that have been noted by drug interaction compendia as being of high severity or major importance. We encourage pharmacists to take steps to prevent patients from receiving these interacting medications and computer software vendors to focus interaction alerts on these and similarly imponant DDIs.

1. National Academy of Sciences Institute of Medicine. To err is human. Washington, DC: National Academy Press; 1999. 2. Johnson JA, Bootman JL. Drug-related morbidity and mortality. Acostof-illness model. Arch Intern Med. 1995;155:194%56. 3. Ernst FR, Grizzle AJ. Drug-related morbidity and mortality: updating the cost-of-illness model. J A m Pharm Assoc. 2001;41:1924. 4. Anonymous. IMS health reports 14.9 percent growth in US prescription sales to $145 billion in 2000 [news release]. Fairfield, Conn.: IMS Health; 2001. 5. Phillips DP, Christenfeld N, Glynn LM. Increase in US medication-error deaths between 1983 and 1993. Lancet. 1998;351:643-4. 6. Shad MU, Marsh C Preskorn SH. The economic consequences of a , drug-drug interaction. J Clin Psychopharmacol. 2001;21:11~20.
Vn1.44, No. 2 March/Apri12004

34. Lyles A, Zuckerrnan IH, DeSipio SM, Fulda T. When warnings are not
enough: primary prevention through drug use review. Health Aff. 1998;17:175-83. 35. Knijff-Dutmer EA, Schut GA, van de Laar MA. Concomitant coumarin-NSAID therapy and risk for bleeding. Ann Pharmacother. 2003;37:126.

Journal d the American Pharmacistsh d t i o n