BACKGROUND AND OBJECTIVES: Hospitalized infants, children, and adolescents are typically exposed to numerous distinct medications during inpatient admissions, increasing their risk of potential drug−drug interactions (PDDIs). We assessed the prevalence and characteristics of PDDI exposure of pediatric patients treated in children’s hospitals.
METHODS: This retrospective cohort study included patients <21 years old hospitalized in children’s hospitals throughout the United States. PDDIs were identified by using the MicroMedex DRUG-REAX system. We calculated the patients exposed to PDDIs, stratified according to the seriousness of the interaction; daily and cumulative counts of PDDI exposures; and characterization of the cited potential adverse effects.
RESULTS: Of 498 956 hospitalizations in 2011, 49% were associated with ≥1 PDDI, with a “contraindicated” PDDI occurring in 5% of all hospitalizations, a “major” PDDI present in 41%, a “moderate” PDDI in 28%, and a “minor” PDDI in 11%. Opioids were involved in 25% of all PDDIs, followed by antiinfective agents (17%), neurologic agents (15%), gastrointestinal agents (13%), and cardiovascular agents (13%). One-half of all PDDI exposures were due to specific drug pairs occurring in ≤3% of patients per hospital day. The most common potential adverse drug events included additive respiratory depression (in 21% of PDDIs), bleeding risk (5%), QT interval prolongation (4%), reduced iron absorption/availability (4%), central nervous system depression (4%), hyperkalemia (3%), and altered diuretic effectiveness (3%).
CONCLUSIONS: Exposure to PDDIs is common among hospitalized children. Empirical data are needed to determine the probability and magnitude of the actual harm for each specific PDDI, particularly for less common drug pairs.
What’s Known on This Subject:
Hospitalized pediatric patients are often exposed to many medications during an inpatient admission. Drug–drug interactions may increase the risk of developing medication-related adverse drug events, leading to serious clinical morbidity and mortality.
What This Study Adds:
Exposure to “major” potential drug−drug interactions occurs in 41% of pediatric hospitalizations in children’s hospitals. One-half of all these exposures were due to less common specific drug pairs (≤3% of patients exposed per hospital day) and thus may be less clinically familiar.
Hospitalized pediatric patients are often exposed to an extensive array of distinct medications, with counts exceeding 25 unique medications for children with longer stays or rare conditions.1 Polypharmacy, defined as the concurrent use of multiple medications, has been implicated as a significant risk factor in the pediatric population for developing medication-related adverse drug events (ADEs),2 likely as a consequence of exposure to drug–drug interactions. In addition, hospitalized pediatric patients are vulnerable to medication-related issues because of off-label prescribing of drugs,3 lack of therapeutic profiles for rarer drugs, and weight-based dosing schemes.4
The Institute of Medicine has identified medication safety in children, including the recognition and prevention of ADEs, as an important priority for the delivery of effective health care.5 With computerized physician order entry systems, standardized prescribing and checking of potential drug–drug interactions (PDDIs) at the time of order have become more prevalent.4 The epidemiology of pediatric PDDIs, however, is largely unknown: we lack estimates of the prevalence of pediatric PDDIs, reliable knowledge regarding the risk posed by specific PDDIs (especially important given the countervailing risks posed by the condition under treatment), or an adequate understanding about pathways by which PDDIs may lead to harmful ADEs in pediatric patients (eg, as a direct physiologic effect on an organ system or through modulation of another medication’s effect).6–8 Each of these critical pieces of information is integral to developing and studying strategies to mitigate clinically important ADEs associated with specific PDDIs.
In the present study, our goal was to advance this research and quality improvement agenda by assessing the prevalence and characteristics of PDDI exposure of pediatric patients treated in children’s hospitals. To do so, we performed a retrospective cohort study of hospitalized pediatric patients by using a large administrative database, including comprehensive clinical and pharmacy information. We then used a well-established compendium of PDDIs to perform a high-volume analysis of all PDDI exposures for each patient in the cohort on each day of their hospitalization.
Human Subjects Protection
The institutional review board at The Children’s Hospital of Philadelphia determined that this study of de-identified data did not constitute human subjects research.
The study used the Pediatric Health Information System (PHIS) database (Children’s Hospital Association [CHA], Kansas City, KS), composed of 43 freestanding children’s hospitals at the time of the study.9 The PHIS database includes patient demographic characteristics, diagnosis, and procedures as well as detailed pharmacy information, including International Classification of Diseases, Ninth Revision, Clinical Modification, codes and Clinical Transaction Classification (CTC) codes for each procedure, generic drug entity dispensed, and clinical services for each day of hospital stay of each patient. Data quality and reliability are assured through a joint effort between CHA and the participating hospitals. Every participating hospital performs coding audits monthly, and CHA performs data quality checks to assure that classified errors occur in <2% of a hospital’s data.1,10,11 Before distribution to end-users, CHA deletes all patient identifiers. One hospital without detailed pharmacy information was excluded from the present study.
Identification and Classification of Study Population
We included all hospitalized patients <21 years of age admitted to participating hospitals between January 1, 2011, and December 31, 2011. Complex chronic conditions (CCC) were defined as medical conditions that can be reasonably expected to last at least 12 months and that involve either several different organ systems or 1 organ system severely enough to require specialty pediatric care and probably some period of hospitalization in a tertiary care center. To identify whether an individual was diagnosed with a CCC, we used our previously published classification scheme based on International Classification of Diseases, Ninth Revision, Clinical Modification, codes.12
Drugs and therapeutic agents were recorded by using CTC codes and drug descriptions. We converted CTC codes into National Drug Codes (NDCs) by using a CTC–NDC crosswalk table provided by CHA.13 We compared a subset of CTC codes for the most frequently used drugs versus their corresponding NDC codes and found that they correctly specified the same generic drugs. A standardized dictionary of 1227 generic drug entities was implemented. These generic drugs were grouped into 27 major categories, principally by using the American Hospital Formulary System Pharmacologic Therapeutic Classification hierarchy of drug and therapeutic agent classes and prevalent subclasses.1
Potential Drug–Drug Interactions
PDDIs were identified by using the DRUG-REAX system (Thomson Micromedex, Truven Health Analytics Inc., Greenwood Village, CO). The software has been previously validated and described in detail.14–16 The software classifies the drug interactions into 5 levels of seriousness: contraindicated, major, moderate, minor, and unknown. The seriousness category represents the seriousness of the PDDI if it occurs and does not assess the likelihood that the interaction may occur. “Contraindicated” drug pairs should not be used concurrently. “Major” interactions may be life-threatening or require medical intervention to minimize or prevent ADEs. “Moderate” interactions may result in exacerbation of the patient’s condition or require an alteration in therapy. “Minor” interactions have limited clinical effect that may include an increase in the frequency or severity of the adverse effects but generally would not require a major alteration in therapy. The scientific documentation status of the PDDI is classified as excellent, good, fair, poor, or unlikely; we did not consider PDDIs in either the poor or unlikely categories. “Excellent” indicates controlled studies have clearly established the existence of the drug interaction; “good” means the documentation strongly suggests that a drug interaction exists, but well-controlled studies are lacking; and “fair” indicates that although the available documentation is scarce, pharmacologic consideration may lead clinicians to suspect the existence of a drug interaction.
Potential ADEs of PDDIs
For each PDDI combination, we determined the organ systems affected by the PDDI, as well as the specific physiologic end-effects or alterations of other pharmacologic agents.7 We made these classifications by directly parsing the text in the drug–drug interaction compendium provided by Truven Health Analytics for each PDDI observed. For example, text containing “respiratory depression” was mapped to the respiratory system or “acute renal failure” was mapped to the renal system. The general classification scheme is provided in Supplemental Appendix 1; the authors will provide the complete scheme and software upon request.
The demographic and clinical characteristics of the patients in the study cohort are described by calculation of percentages. To describe the prevalence of PDDIs, the percentages of specific PDDIs were calculated according to patient and each hospital day; we then calculated the percentage/proportion of patients with PDDI exposure (subsequently referred to as “exposure percentage/proportion”) with 95% confidence intervals by interaction level and by patient demographic and clinical characteristics. Patient-level percentiles were calculated of the following: (1) the number of exposures of distinct PDDI on each day of hospitalization; (2) the cumulative number of exposures of distinct PDDI on each successive day of hospitalization; and (3) the PDDI exposure percentage with 95% confidence intervals in each day up to 30 days’ in-hospital stay. Trend analyses were performed by using the Cochran-Armitage test. The PDDI patterns are also described according to age groups. For each of the PDDIs, we next calculated counts and percentages of the implicated medication pairs, the severity of the interaction, and the level of evidence for the interaction. Finally, we calculated the percentage of all PDDIs associated with each of the different categories of potential ADEs, as described in the previous section. Each PDDI could be associated with ≥1 potential ADE.
All data management and statistical analyses were conducted by using SAS version 9.3 (SAS Institute, Inc, Cary, NC) and Stata version 13.1 (StataCorp, College Station, TX).
In 2011, a total of 498 956 pediatric hospitalizations occurred in the 43 member children’s hospitals (Table 1). The majority of patients were aged <5 years (51%), male (54%), and without a CCC (53%). Hospitalizations were most frequently due to respiratory diagnoses (19%), <4 days in length (60%), funded by Medicaid (49%), and resulted in home discharge (93%).
PDDIs and Exposure Proportions
Among these hospitalizations, we identified a total of 4 497 448 PDDI exposures, comprising 5292 distinct PDDIs. The level of evidence supporting PDDIs was excellent for 7% of all identified PDDIs, good for 55%, and fair for 38%. Overall, 49% of all hospitalized pediatric patients were exposed to a PDDI (Table 1). A contraindicated PDDI exposure occurred in 5% of all hospitalizations, a major PDDI exposure in 41%, moderate in 28%, and minor in 11%. The likelihood of PDDI exposure increased for patients who were older, had longer hospital stays, or had CCC. The presence of a CCC increased the likelihood of any PDDI exposure, with a 2.5-fold increase in children with >3 CCC compared with those without (85% vs 34%).
Most Frequent Specific PDDIs by Severity Level
Certain classes of drugs were commonly implicated in PDDIs, including opioids, antiinfective agents, neurologic agents, gastrointestinal agents, and cardiovascular agents (Table 2). Among contraindicated pairs, exposure to ibuprofen and ketorolac (with the potential for enhanced adverse gastrointestinal effects) occurred in 1.8% of hospitalizations, followed by fluconazole and ondansetron (0.9%) and calcium chloride and ceftriaxone (0.7%). Among pairs with major interactions, fentanyl and morphine (with the potential for additive respiratory depression) occurred in 13.2% of hospitalizations, followed by fentanyl and midazolam (11.2%), midazoloam and morphine (9.2%), and bupivacaine and propofol (5.8%). The pairs with moderate reactions included dexamethasone and rocuronium (3.3%), heparin and vitamin A (2.1%), and propofol and succinylcholine (2.0%). Opioids were involved in nearly 25% of all PDDIs, followed by antiinfective agents (17%), neurologic agents (15%), gastrointestinal agents (13%), and cardiovascular agents (13%). Of all 4 497 448 PDDIs, morphine was involved in 10%, followed by fentanyl (8%), midazolam (8%), furosemide (6%), and aspirin (6%). Complete lists of all distinct PDDIs, along with lists of most commonly involved specific generic drug pairs and individual drugs, are provided in Supplemental Appendix 2 and 3.
PDDI Exposure Proportions Over the Course of Hospitalization
Although the likelihood of exposure to PDDIs increased over the course of hospitalizations, the patterns varied according to patient age (Fig 1). For infants, 21.8% of patients were exposed to a PDDI on hospital day 1; by hospital day 30, 32.0% had been exposed (trend test, P < .05). Infants at the 90th percentile of PDDI exposure had 2 unique PDDIs on hospital day 1, and 3 by hospital day 30; cumulatively, this finding corresponded to 2 PDDIs on hospital day 1, increasing to 16 distinct PDDIs by day 30. For patients aged >1 year, 34.7% were exposed to a PDDI on hospital day 1, increasing to 66.3% by day 30 (P < .05). Patients aged >1 year at the 90th percentile of exposure to PDDIs had 3 unique PDDIs on hospital day 1, and 8 by hospital day 30; cumulatively, this finding corresponded to 3 PDDIs on hospital day 1, increasing to 30 distinct PDDI exposures by day 30.
Distribution of Distinct PDDI Exposures Among Patients
Several more common distinct PDDIs, such as those involving combinations of opiates and/or hypnotic agents), had exposure proportions exceeding 15% of patients. However, a large proportion of less common PDDI drug pairs were associated with contraindicated and major interactions. Examining the cumulative distribution of all PDDI exposures across the range of exposure proportions for distinct PDDIs (from rarer to more common), 50% of all major PDDI exposures were due to specific drug pairs occurring in ≤2.8% of patients per hospital day, and 10% were due to specific drug pairs occurring in ≤0.2% of patients per hospital day. The patterns differed for infants compared with older patients, with a larger proportion of infant PDDI exposures being due to more common PDDIs (Fig 2). For example, in patients aged <1 year, 50% of major PDDI exposures were due to specific drug pairs occurring in ≤7.5% of patients per hospital day. In comparison, for patients ≥1 year of age, 50% of major PDDI exposures were due to specific drug pairs occurring in ≤1.6% of patients per hospital day.
Potential ADEs Associated With PDDIs
In the source compendium of PDDIs, each distinct PDDI was cited as having ≥1 specific harmful end-effect. Among all the PDDIs identified for this study’s cohort of hospitalized patients, the most common physiologic end-effects or specific alterations of pharmacologic agents due to PDDIs (Table 3) were respiratory depression (21% of PDDIs), followed by bleeding risk (5%), QT interval prolongation (4%), iron absorption/availability (4%), central nervous system depression (4%), hyperkalemia (3%), and diuretic effectiveness (3%).
We classified all specific harmful end-effects regarding the organ system or systems likely to be harmed (Fig 3), which varied based on the exposure percentages for PDDIs. Significantly more contraindicated exposures occurred as the exposure percentage for a distinct PDDI decreased (exposure >5%: 0 contraindicated exposures; exposure 1%–5%: 27 903 contraindicated exposures; and exposure <1%: 34 981 contraindicated exposures.
Having identified all PDDIs for a large cohort of pediatric patients treated at 42 children’s hospitals in the United States, we found that nearly one-half of patients were exposed to a PDDI during their admission. Contraindicated PDDIs occurred in 5% of hospitalizations and major PDDIs in 41% of hospitalizations. The likelihood of PDDI exposure increased for patients who were older, had longer hospital stays, or had CCC (all likely due to exposure to more drugs). Certain classes of drugs were commonly implicated in PDDIs, including opioids, anti-infective agents, neurologic agents, gastrointestinal agents, and cardiovascular agents, and were associated with a small subset of all potential PDDI consequences, such as additive respiratory depression or gastrointestinal disturbances, which may (due to the frequency with which they are encountered) be anticipated or recognized and appropriately managed. By contrast, a large proportion of less common PDDI drug pairs were associated with contraindicated and major interactions distributed across many diverse consequences, including physiologic end-effects, alterations of other drugs, and different body systems affected.
Our findings need to be interpreted keeping in mind that our data identify only potential drug–drug interactions: our study used an administrative database that is insufficient for detecting actual harm.17 The high prevalence of observed PDDI exposures does not correspond with reported rates of ADEs, which either occur or are recognized far less frequently.2,8,18,19 At the same time, ADEs can result in significant morbidity and mortality.20–22 Our findings therefore point to a central problem or tradeoff of warning systems, encapsulated by the story of the boy who cried wolf: the current compendiums of PDDIs suffer from identifying so many PDDIs that the signal of true potential patient harm may get lost in the noise of too many alarms. This situation is analogous to the growing concern that an overabundance of bedside patient monitors and their separate alarms does not necessarily improve patient safety.23
Improving the system of PDDI identification and management, accordingly, requires responses in 3 complementary directions. First, we need to identify currently listed PDDIs that do not warrant a red flag. The prevalence of PDDI exposures resulting in clinically significant ADEs is likely substantially lower than the percentage of patients exposed to a PDDI that we observed.19 Furthermore, the level of evidence for assignment of severity to a PDDI was good to excellent for only 62% of the PDDIs, indicating that the remaining PDDIs have a theoretical basis for inferring the possibility of an ADE, but that these inferences have not been empirically substantiated in clinical practice. Classifying drug interactions into various levels of seriousness (severity) is challenging. Thus, interpreting the results according to level of seriousness may be misleading. Even medication combinations labeled as contraindicated are not necessarily an absolute contraindication. Labels applied by various compendia to the same interactions have been shown to vary substantially and may contribute to clinician alert fatigue.24–27 Clinicians may be oversaturated by computer-based warnings of PDDIs, which can result in a general disregard of these warnings in clinical practice. Even when prompted by computer-based alerts, clinicians disregard certain PDDIs and do not complete the recommended monitoring for subsequent ADEs.28 In addition, if clinicians avoid the use of specific drugs due to unwarranted concerns for PDDIs, the effectiveness of therapy may be compromised. For these reasons, future research should continue to discern specific PDDIs that can be downgraded or altogether omitted from alerting systems regarding clinical decision support. Doing so will enable us to combat PDDI alert fatigue and enable clinicians to focus on a more limited set of PDDIs, which is important because their retention of PDDI knowledge declines without practical reinforcement of specific PDDIs.29,30 Effort will also be required to address the medico-legal concerns or consequences from delisting a PDDI.31
Second, given that severe ADEs do occur secondary to PDDIs, we need to prioritize PDDIs based on the probability of occurrence and the magnitude of harm to patients, and subsequently develop remediation plans for specific high-priority PDDIs; therefore, the likelihood of hospitalized patients experiencing significant ADEs is reduced. The most common contraindicated and major PDDIs were due to more commonly used medications, such as opiates and antiinfective agents. For these types of common PDDIs, clinicians may understand clinical situations in which coadministration is acceptable, monitor for potential ADEs, and adjust medications regimens when ADEs do develop. For example, a combination of an opioid and a benzodiazepine is often used to provide analgesia/sedation for patients who are critically ill or are undergoing invasive procedures; the potential ADE of additive respiratory depression is well known, with a conscious assessment of the potential benefits and harms of administering this combination of drugs. We therefore hypothesize that the majority of unanticipated harm, in particular that from more severe interactions, may arise from pairs of less common medications, prescribed by clinicians with limited experience or insight into the potential adverse effects and the necessary monitoring for a wide range of adverse effects.
Finally, given the rarity of certain specific drug–drug combinations, we need to develop and use customized strategies to identify potential harm in small samples of patients. Specifically, we should consider combining the information from large data sets (eg, PHIS) with more detailed information obtained either from other clinical data sources (eg, electronic medical records) or from research data sources (eg, pharmacologic studies or active adverse event surveillance studies) by using 2-stage sampling or other methods.32
Hospitalized patients are commonly exposed to PDDIs, but the subsequent probability of occurrence and magnitude of patient harm requires further empirical substantiation. Less common PDDIs represent the majority of all PDDI exposures, and these less common PDDIs may provide a high-yield target to improve patient safety.
The authors thank Truven Health Analytics and Brian Cohan, RPh, for their assistance in performing the en masse identification of PDDIs by using the DRUG-REAX system.
- Accepted October 27, 2014.
- Address correspondence to Chris Feudtner, MD, PhD, MPH, CHOP North, Room 1523, The Children's Hospital of Philadelphia, 34th and Civic Center Boulevard, Philadelphia, PA 10194. E-mail:
Dr Feinstein participated in the concept and design of the study, performed the statistical analyses and initial interpretation of data, drafted the initial manuscript, and critically revised the manuscript; Dr Dai conceptualized and designed the study, performed the statistical analyses and initial interpretation of data, and critically revised the manuscript; Dr Zhong and Dr Freedman participated in the concept and design, interpretation of the data, and critical revision of the manuscript; Dr Feudtner conceptualized and designed the study, performed the statistical analyses and interpretation of data, and critically revised the manuscript; and all authors approved the final manuscript as submitted.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: Supported by the Agency for Healthcare Quality and Research, Comparative Effectiveness and Safety of Hospital-Based Pediatric Palliative Care (1R01HS018425).
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
- Aspden P,
- Institute of Medicine
- ↵Children’s Hospital Association. PHIS+: augmenting the Pediatric Health Information System with clinical data. Available at: www.childrenshospitals.org/phisplus/index.html. Accessed September 9, 2014
- ↵Rizkalla NA, Feudtner C, Dai D, Zuppa AF. Patterns of medication exposures in hospitalized pediatric patients with acute renal failure requiring intermittent or continuous hemodialysis. Pediatr Crit Care Med. 2013;14(9):e394–e403
- Strom BL,
- Kimmel SE,
- Hennessy S
- ↵US Food and Drug Administration. National Drug Code directory. Available at: www.fda.gov/drugs/informationondrugs/ucm142438.htm. Accessed September 9, 2014
- ↵Barrons R. Evaluation of personal digital assistant software for drug interactions. Am J Health Syst Pharm. 2004;61(4):380–385
- ↵Leone R, Magro L, Moretti U, et al. Identifying adverse drug reactions associated with drug-drug interactions: data mining of a spontaneous reporting database in Italy. Drug Saf. 2010;33(8):667–675
- Fulda TR, Valuck RJ, Zanden JV, Parker S, Byrns PJ, The USPDURAP. Disagreement among drug compendia on inclusion and ratings of drug-drug interactions. Curr Ther Res. 2000;61(8):540–548
- Wong CM,
- Ko Y,
- Chan A
- White JE
- Copyright © 2015 by the American Academy of Pediatrics