Prevalence of Clinically Significant Drug-Drug Interactions Across US Children’s Hospitals

We conducted a multicenter retrospective cohort study of individuals admitted (inpatient and observation status) to children’s hospitals that contribute data to the Pediatric Health Information System (PHIS) database. The PHIS administrative database includes billing and use data from 52 tertiary care children’s hospitals across the United States that are affiliated with the Children’s Hospital Association (Lenexa, KS). Data quality and reliability are ensured through a joint effort between the Children’s Hospital Association and participating hospitals. Hospitals submit discharge data including demographics, diagnoses, procedures using the International Classification of Diseases, 10th Revision, Clinical Modification, as well as detailed daily pharmacy and claims information.

We included all hospitalized patients <26 years of age who were discharged from a PHIS hospital between January 1, 2016 and December 31, 2018. An age cutoff of 26 years was chosen because young adults admitted children’s hospitals represent those with chronic comorbidities stemming from birth or childhood. These individuals are likely a high-risk group for DDIs, and the risk of ADEs are likely more reflective of those experienced by children and adolescents in comparison with healthy young adults or older adults with comorbidities. We excluded normal newborns and those receiving neonatal intensive care and peri- or postpartum obstetrical care.

Medication pairings with clinically significant DDIs were identified by using a published expert consensus definition that includes 53 medication pairings with significant safety implications in pediatric populations.⁴  These DDI pairings were chosen on the basis of their clinical relevance, including the prevalence of prescribing, severity of interaction, and likelihood of resulting adverse event, in individuals treated at US children’s hospitals. Individuals were considered exposed to a DDI pair if there was ≥1 hospital day in which both medications were administered on the same hospital day. Medications captured include those from any hospital setting while inpatient, including operating rooms. Medications with topical, ophthalmic, and otic routes of delivery were excluded because these are unlikely to achieve the serum concentration levels necessary to be involved in a DDI. If the drug pairing involved a drug class or pharmacologic mechanism effect, exposure was determined for all drugs in the class or mechanism associated with the DDI and denoted with a “+” sign. Drugs in the class not associated with the DDI were excluded from the analysis. For example, for the “citalopram+, linezolid+” pair, selective serotonin reuptake inhibitors associated with the DDI include citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, sertraline, vilazodone, and vortioxetine but not dapoxetine. Therefore, all the selective serotonin reuptake inhibitors except dapoxetine were included in the analyses.

DDIs and associated risk grades were further classified by using the Lexicomp-Interactions System (Lexicomp, Inc, Hudson, OH). The Lexicomp software classifies the drug interactions into levels of DDI risk. The risk grades also provide a recommended action on the basis of the relevance of the interaction, level of evidence, and iatrogenic risk. Actions associated with risk grades are as follows: “A” (no interaction), “B” (no action is necessary), “C” (monitor therapy), “D” (modify regimen), and “X” (avoid combination). The DDI effects associated with each pairing were categorized in the following 5 groups on the basis of the Lexicomp description of the DDI: increased serum concentration, decreased serum drug concentration, increased risk of ADEs, altered drug metabolism, and multiple effects (constituting ≥2 effects).

Demographic characteristics were summarized by using frequencies and percentages for categorical variables and median and interquartile range (IQR) for continuous variables. Bivariate comparisons were made between patients with and without DDI exposures.

To define variation in DDI prescribing across hospitals, we first calculated the unadjusted proportions of drug pairings in aggregate and by individual drug pairings for each hospital. Proportions were calculated by dividing the number of hospitalizations with ≥1 DDI exposure by the total number of hospitalizations during the study period. The DDI exposure rate was calculated by dividing the number of hospital days with DDI exposures by the total number of hospital days and expressed per 1000 hospital days. Proportions and rates were stratified by risk grade and summarized by All Patients Refined Diagnosis Related Groups (3M Company, St Paul, Minnesota) service line assignment.

We calculated the adjusted hospital-level proportions of DDI exposure using a generalized linear mixed-effects model, with hospital included as a random effect. Covariates in the model included age, sex, race and ethnicity, payer, number of complex chronic conditions¹⁰  (CCCs), length of stay (LOS), service lines, geographic region, ICU status, and case mix index for each hospitalization. Factors independently associated with clinically significant DDI were evaluated by using multivariable logistic regression incorporating the covariates listed above. All analyses were performed by using SAS version 9.4 (SAS Institute, Inc, Cary, NC), and P < .05 was considered statistically significant. The Institutional Review Board at Vanderbilt University Medical Center determined that this study of deidentified data did not constitute human subjects research (Institutional Review Board 160244).

A total of 2 331 305 hospitalizations were included in the study. During the study period, 2.0% (47 414) of hospitalizations had an exposure to a clinically significant DDI drug pair. Of the total study population, 42 735 (1.83%), 3661 (0.16%), and 1018 (0.04%) were exposed to 1, 2, and ≥3 clinically significant DDI pairs, respectively. Children who were exposed to DDI pairs were older, compared with those not exposed to a DDI (median age of 10 vs 6 years; IQR 2–15; P < .001) and had a longer median LOS (5 vs 2 days; IQR 2–18; P < .001). Of those exposed to a DDI, 82.1% (38 911) had a CCC, 34.9% (16 556) had an ICU stay, and 51.4% (24 350) were non-Hispanic white (Table 1).

In multivariable analysis, increasing age, number of CCC, and longer LOS were associated with increasing odds of clinically significant DDI pair exposure (Table 2). Surgical service line, compared to medical service line, was associated with a 42% increase in the odds of DDI exposure. Factors associated with a decreased odds of clinically significant DDI pair exposure included female sex (adjusted odds ratio [aOR]: 0.91; 95% confidence interval [CI] 0.90–0.93), non-Hispanic Black race (aOR: 0.81; 95% CI 0.79–0.83), and private payer insurance (aOR: 0.89; 95% CI 0.87–0.91).

Across the 2-year study period encompassing 328 585 total hospital days, there was a DDI exposure rate of 34.9 per 1000 hospital days. The most common DDI pairings were nicardipine and tacrolimus (14 893; 27.8%), fentanyl and linezolid (10 596; 19.7%), lamotrigine and valproic acid (2918; 5.5%), and haloperidol and quetiapine (2706; 5.1%). Individual drug pairing exposure rates and associated DDIs are listed in Table 3.

The most common risk grade for DDIs was “C: monitor therapy,” with a rate of 20.84 per 1000 hospital days. For pairings with a risk grade of “D: consider treatment modification,” there was an exposure rate of 11.03 per 1000 hospital days. For drug pairings with the highest risk grade of “X: avoid combination,” there were 11 734 exposures with a rate of 3.04 rate per 1000 hospital days. Conversely, the rate of clinically significant DDI exposure per 1000 DDI patient hospital days increased with increasing risk grade. The DDI exposure rates per 1000 DDI patient hospital days was 32.3, 117.1, and 221.27 for C, D, and X risk grades, respectively, suggesting that clinically significant DDI pairings with higher risk grades were prescribed for a longer duration than lower risk DDI pairings. The DDI effects associated with each pairing were categorized into 5 groups, with increased serum concentration being the most common (32.1%) followed by an increased risk of ADE (30.2%), decreased serum drug concentration (26.4%), altered drug metabolism (9.43%), and multiple effects (1.89%) (Table 3).

Among patients with a clinically significant DDI exposure, 55.1% (26 121) had care with a diagnosis in the medical service line. The medical services lines with the highest percentage of DDI exposures were solid organ transplant, bone marrow transplant, and nephrology and urology. The most common DDI pairing across service lines was nicardipine and tacrolimus (Table 4). The surgical services with the highest proportion of DDI exposures were craniofacial and plastic, nephrology and urology, and other trauma. Fentanyl and linezolid was the most common DDI pairing for these surgical service lines.

The proportion of exposure to DDIs varied widely across US children’s hospitals, ranging from 0.65% to 7.33%; adjusted proportions ranged from 1.05% to 4.92% (Fig 1). When compared by individual clinically significant DDIs, there was significant variation across hospitals (Fig 2). For example, compared to other hospitals, hospital 1 prescribed DDIs more frequently (as indicated by the red squares across DDIs on the heatmap), whereas hospital 52 was a frequent prescriber of DDI pair number 22 (atorvastatin and erythromycin) but not other DDIs. There was also significant variation among the prescribing of individual drug pairings. For example, DDI pair number 44 (nicardipine and tacrolimus) had a large variation, with every hospital with at least some DDI prescribing (ranging from 0.01% to 3.6%). In contrast, DDI pair number 24 (carbamazepine and cyclosporine) has little variation. Out of 52 hospitals, only 1 hospital had any exposure to the combination.

The 5 most frequent drug combinations had a wide distribution in prescribing rates across children’s hospitals: specifically, fentanyl and linezolid (median rate per 1000 patient-days: 0.36; IQR 0.25–0.59), lamotrigine and valproic acid (0.11; IQR 0.08–0.17), nicardipine and tacrolimus (0.54; IQR 0.28–0.76), haloperidol and quetiapine (0.06; IQR 0.03–0.14), and tacrolimus and voriconazole (0.05; IQR 0.01–0.11; Fig 3).

In this large multicenter study in which we evaluate the exposure of individuals hospitalized at US children’s hospitals to medications with clinically significant DDIs, there are 3 main findings. First, the proportion individuals exposed to clinically significant DDIs is substantial, encompassing 2% of all those hospitalized at US children’s hospitals. We found that older age, CCC, LOS, and surgical service line diagnoses were associated with a greater odds of DDI exposure. Second, one-quarter of DDI exposures were considered contraindicated (risk grade: “X”) and were prescribed across a broad range of medical and surgical service lines. These drugs were prescribed for longer durations than lower risk grade DDIs. Third, there was wide variation in prescribing patterns of DDIs across children’s hospitals, suggesting exposure to some of these DDIs may be unwarranted and preventable.

In previous studies, researchers reported that 49% of all hospitalized children,³  41.7% of hospitalized children with epilepsy,¹¹  and 75% of PICU patients⁶  were exposed to drugs with potential DDIs. However, in these studies, researchers used adult-based drug classification software to identify potential DDIs, and these could vary in their clinical relevance. In this study, we evaluated 53 medication pairings with known DDIs of clinical importance in those admitted to children’s hospitals. We found that clinically significant DDIs had a lower prevalence compared to all potential DDIs.^3,6,11  Nonetheless, these DDIs were prescribed in 2% of hospital encounters, exposing >47 000 children to potential ADEs during our 2-year study period.

Medications with DDIs are often categorized by the severity of a potential DDI, rather than risk grade, by using drug classification DDI software.^3,6  Severity classification (typically categorized as minor, moderate, major, and contraindicated) are similar but distinct from risk grade. Severity classifications indicate the potential severity of the interaction if it occurs but not the likelihood that it would occur. There is also significant variation in the interrater reliability of severity classifications between drug software programs as well as compared to clinicians’ assessment of DDI severity.^4,7  Feinstein and colleagues assessed all potential DDI exposures in children’s hospitals and found the following severity classifications: 5% contraindicated, 41% major, 28% moderate, and 11% minor. The risk grade provides a recommended action based on the relevance of the interaction, level of evidence, and iatrogenic risk. Risk grade, therefore, is likely a better indicator of the potential DDI risk in children than severity stratifications. Of the DDIs included in this study, the recommended actions were contraindicated (risk grade X: 25%), consider therapy modification (risk grade D: 27%), and monitor therapy (risk grade C: 56.2%).

Finally, after adjusting for hospital and clinical factors, we found significant variation in the rates of DDI prescribing across US children’s hospitals, with a greater than fourfold difference between hospitals. The reasons for this differential prescribing pattern are unclear. Some hospitals may have measures in place to identify and prevent unnecessary DDI exposures, such as embedded ICU pharmacists or electronic medical record prescription clinical decision support tools. It is possible that there is unmeasured confounding that was not accounted for in the statistical model, such specialization of cases (eg, dedicated transplant centers) not accounted for in case mix index classification. Almost certainly, a portion of DDI prescribing, such as immune suppression in transplant patients, is unavoidable. However, it is likely that a number of DDI exposures are preventable, indicating an opportunity for improvement in care. For example, in DDIs in which there are multiple medication options (eg, quetiapine and haloperidol or citalopram and linezolid, among others) choosing a safer alternative medication combination may be feasible.

Our results align with others that reveal exposure to potential DDIs is common in US children’s hospitals.^6,11–15  Although DDI safety alerts are available in several systems, not all DDIs identified may be clinically relevant. There are some drug pairings that, because of the frequency and severity of their DDI, should not be used together (such as those with a risk grade of X).⁸  There is a larger set of other potential DDIs that might be unavoidable or have clinical reasons for concomitant use (those with risk grades of C and D) but are likely to cause harm due their frequency of use. In this study, we found that drugs with risk grades of C and D were prescribed more often and for longer durations than those with a risk grade of X. Finally, there are other drug combinations not included in this study that are rated as having DDIs in safety databases but are likely not clinically relevant to those cared for in children’s hospitals. Most institutions use a limited set of existing medication safety databases to inform their safety alerts, and, to our knowledge, none of these databases are pediatric specific. When these vendor-based alerts are perceived to be of limited value, physicians may discount all alerts, including appropriate ones.⁹  In US children’s hospitals, >95% of DDI alerts are overridden.^15,16 

Therefore, it is important to identify pediatric-specific clinically significant drug combinations to prioritize in drug alert systems. Factors that may lead to increased rates of preventable DDI prescribing include DDI-alert fatigue, insufficient communication across care providers, decreased accessibility of prescribing information across electronic medical records, and limited dissemination of information relaying the appropriate risk of medication combinations. By using the fentanyl and linezolid combination as an example, there are multiple providers who could potentially be involved in prescribing 1 drug but not the other, including members of the surgery, infectious disease, ICU, and anesthesiology teams, among others. Efforts to improve the efficacy of DDI alerts and better communication of the medication management plan across prescribing providers and locations of care (emergency department, inpatient floor, ICU, operating room, etc) may aid in preventing unnecessary DDI exposures.

Given that severe ADEs can occur as a result of DDIs, prioritizing DDIs on the basis of frequency of exposure, probability of occurrence, and magnitude of harm is essential. Identifying children at high risk for DDI exposure provides a potential target for care improvement. We found that older age, CCC, LOS, and surgical service line diagnoses were associated with greater odds of DDI exposure. Rather than focusing efforts on all hospitalized children, targeting clinical decision support interventions and DDI alerts to patients and providers involved in frequent DDI exposures is of high value. Implementing DDI alerts incorporating patient-specific information, (including complexity, laboratory values, illness severity, etc) and customizing alerts may better relay DDI risk to practitioners.¹⁷  In evidence, it is also suggested that embedding a pharmacist within the physician care team caring for high-risk patients improves the identification of DDIs before patient exposure.¹⁸ 

Limitations of our findings include the retrospective nature of the study as well as incomplete clinical patient-level data in the PHIS data set. The PHIS data set only includes children’s hospitals, which may limit the generalizability of our results to community hospitals. We also excluded newborns, NICU patients, and those admitted peri- or postpartum obstetrical care, and our findings may not be generalizable to these populations. However, children’s hospitals generally care for those at higher risk for DDIs, suggesting this hospital population is of particular importance. Furthermore, we only evaluated 53 clinically significant DDI medication pairings. Although we selected these interactions given their known clinical relevance, there are likely other clinically significant DDI combinations not included in our analysis, and, thus, we may have underestimated the frequency of clinically significant DDI exposure in children. In addition, in our study, we did not measure the rates and severity of ADEs resulting from DDI exposures, and pediatric-specific data are lacking. Additional studies specifically designed to better understand the association between DDIs and ADEs in the pediatric population would aid practitioners in understanding the risk to benefit ratios for these drug pairings. Despite these limitations, our findings represent an important step in understanding clinically significant DDIs in hospitalized children, and, with those findings, we provide several potential targets in which to prioritize efforts to improve safe medication prescribing in hospitalized children.

Approximately 2% of those treated at US children’s hospitals are prescribed clinically significant DDI combinations, with a greater odds of exposure in older individuals, those with increased number of CCCs, and those with surgical service line diagnoses. We found a wide variation in the rates of DDI prescribing across children’s hospitals, suggesting an opportunity to improve safe medication use in children. Further study is needed to determine the rate of adverse events due to DDI exposures and factors amenable for interventions done to promote safer medication use.

ADE

adverse drug event

aOR

adjusted odds ratio

CCC

complex chronic condition

CI

confidence interval

DDI

drug-drug interaction

IQR

interquartile range

LOS

length of stay

PHIS

Pediatric Health Information System