Supportive care medications coinciding with chemotherapy among children with hematologic malignancy

Abstract

Pharmacokinetic (PK) conflicts can arise between supportive care medications (SCM) and chemotherapy in children with hematologic malignancy (HM). In this retrospective study, medical records for children (28 days–18 years) diagnosed with HM and receiving an SCM antimicrobial were collected from a hospital network between 1 May 2000 and 31 December 2014. PK drug-gene associations were obtained from a curated pharmacogenomics database. Among 730 patients (median age of 7.5 (IQR 3.7–13.9) years), primarily diagnosed with lymphoid leukemia (52%), lymphoma (28%), or acute myeloid leukemia (16%), chemotherapy was administered in 2846 hospitalizations. SCM accounted for 90.5% (n = 448) of distinct drugs with 93% (n = 679) of children, receiving ≥5 different SCM/hospitalization. Same-day SCM/chemotherapeutic PK gene overlap occurred in 48.3% of hospitalizations and was associated with age (p = 0.026), number of SCM, HM subtype, surgery, and hematopoietic stem cell transplant (p < 0.0001). A high and variable SCM burden among children with HM receiving chemotherapy poses a risk for unanticipated PK conflicts.

Introduction

Hematologic malignancies (HM) are the most commonly diagnosed cancers affecting children [1]. Features of HM, such as compromised immune function, and the associated treatment regimens put patients at extreme risk of infection [2]. Antimicrobial therapy is often initiated among patients with HM for suspected or confirmed infection as well as chemotherapy-induced neutropenia [3]. Antimicrobials and other supportive care medications (SCM) are essential to improving the quality of life and survival among those undergoing treatment for cancer [4]. SCM and chemotherapy are often administered within the same episode of clinical care, but little is known about how this may affect drug disposition.

Polypharmacy, typically considered a problem affecting the elderly, is common for children with cancer [5]. The inherently complex drug regimens coupled wide hospital-to-hospital variation in prescribing practices elevates risks around drug-drug interactions (DDIs) and adverse drug reactions (ADRs) [6–8]. Detecting ADRs and DDIs in children with cancer is challenging [9,10]. They can manifest distinct and substantial differences in drug disposition compared to their cancer-free peers or adults and present with severe illness that can mask an ADR [11–13]. While therapeutic drug monitoring is performed for a few select drugs, concentration data outside of a target therapeutic range may trigger a dosing change, but is unlikely to prompt an investigation into potential pharmacokinetic (PK) interactions [14].

Nonetheless, for children with HM, DDIs between narrow therapeutic index chemotherapeutics and SCM could be sources of unwanted, and possibly unnecessary, PK variation of high clinical significance. SCM and chemotherapeutics which converge on the same drug metabolism and transport pathways can lead to PK drug-drug-gene interactions (DDGIs) [15]. DDGIs can alter drug disposition leading to changes in efficacy and/or toxicity and stem from a variety of direct and indirect mechanisms [16]. As an example, many anticonvulsants can increase clearance of chemotherapeutics by engaging xenobiotic-activated transcription factors that upregulate drug metabolizing enzymes and transporters which can be harmful [17]. Moreover, genetic variation affecting the function of drug metabolizing enzymes and transporters (as well as drug targets) can profoundly impact drug PK [18]. Preemptive pharmacogenetic screening has been suggested as a means to aid efforts to improve drug safety and efficacy within dynamic and complex drug regimens [16]. Yet little is known about the prescribing of SCM which intersect (as substrates, inhibitors, or induces of elements in the xenobiotic detoxification system) with antineoplastic regimens for children with HM.

This study describes patterns of contemporary SCM use, availability of corresponding population-specific PK data, and the potential for SCM-chemotherapeutic PK interactions among children with HM undergoing active chemotherapy. The aim is to highlight areas of clinical priority related to drug therapy for children with HM [19,20].

Methods

Study design and setting

A multicenter retrospective study was conducted utilizing the enterprise data warehouse of Intermountain Healthcare (IH). IH is comprised of 23 hospitals, serving ~1.5 million people within the Intermountain West region (Utah, Idaho, Wyoming, Nevada, and Montana. University of Utah and IH IRBs approved this study.

Participants and clinical variables

Patients, aged 28 days to <19 years, diagnosed with HM who received any antimicrobial drug (e.g. antibiotic, antiviral, antifungal, or antiparasitic agent) during hospitalization between May 2000 and December 2014 were included [21]. HM subtype was defined per the International Classification of Childhood Cancer (ICCC), International Classification of Diseases for Oncology, 3rd Ed., and WHO Classification of Tumors of Hematopoietic and Lymphoid Tissues (ICD-O-3/WHO 2008). Age stages were defined using National Institute of Child Health and Human Development (NICHD) pediatric terminology: Infant [28 days to <12 months], toddler [12 months to <2 years], child [2 to 12 years], adolescent [12 to <19 years] [22]. All Patient Refined-Diagnosis Related Group inpatient classification system (APR-DRG) was used to define the severity of illness (SOI) and risk of mortality (ROM) categories [23]. Surgical procedures were categorized as ‘minor’ if related to recovery from surgery (e.g. dressing changes, local incision care), sutures, intravenous lines, biopsy, blood transfusions, or diagnostic tests and procedures [24].

Generic drug names were linked to pharmacy records using a drug dictionary containing World Health Organization (WHO) International Nonproprietary Names (INN) [25]. SCM were defined as drugs administered via enteral or parenteral routes not used in the direct treatment of malignancy (i.e. antineoplastic/chemotherapy). Topical and ophthalmic preparations were excluded in addition to intravenous fluids (e.g. normal saline, electrolyte solutions (WHO ATC: B05XA)), heparin, parenteral nutrition, and antineoplastic agents. ‘SCM polypharmacy’ was defined as ≥5 concomitant SCM [26].

Genes associated with SCM pharmacokinetics

Gene-chemical/drug combinations associated with a PK phenotype (e.g. biotransformation and transport) were compiled from files (i.e. Clin_ann_metadata, Var_drug_ann, Relationships, and Clinical variants) accessed on the PharmGKB website (1/5/2019) and contained 1033 PK gene-chemical/drug associations [27,28]. Analyses were limited to parent drug only. Venn diagram was created using the eulerAPE program [29].

Systematic review: Search strategy, selection criteria, and data extraction

A systematic search of PubMed (accessed 6/11/2019) for all PK studies of SCM administered to ≥25% of patients was conducted using the base query: pharmacokinetic [All Fields] OR metabolism [MeSH Terms] OR metabolic networks and pathways [MeSH Terms]) AND neoplasms [MeSH Terms]. The search was filtered by age groups ≤19 years of age (i.e. Newborn, Infant, Preschool Child, Child, and Adolescent) and names of each SCM were separately added (Supplementary Table S5) [30]. A priori definitions for study relevance were: 1) Included subjects <19 years of age; 2) PK parameters or measures for the SCM of interest reported; 3) Exclusion of publications as reviews, duplicates, or not in English; and 4) Study population was wholly or in part comprised of patients diagnosed with cancer. For publications of potential relevance, full manuscripts were retrieved. The references of relevant publications were additionally evaluated.

Statistical methods

Analyses performed based on episodes of care (admit to discharge) are designated as ‘hospitalizations,’ those at the patient level were inclusive of all individual hospitalizations in the study period. As appropriate, chi-square or Fisher’s exact test was used to compare categorical variables while continuous variables with normal distributions were expressed as the mean (± standard deviation [SD]) and compared using Student’s t-test (with or without a prior transformation based on a Box-Cox test) or the Wilcoxon log-rank test. For multivariate logistic regression analyses, a threshold p-value <0.2 was used for covariate selection in univariate analyses with α at <0.05. Statistical analyses were performed using R version 3.4.4 (R Foundation for Statistical Computing, Vienna, Austria), or Prism version 6 (GraphPad Software, La Jolla, CA, USA).

Results

Study parameters and clinical variables

A total of 2399 unique patients (15,779 hospitalizations) diagnosed with malignancy were identified within the study period (Figure 1). Of these, 813 patients (4624 hospitalizations) were diagnosed with HM, had their care primarily managed within the IH network, and received an antimicrobial. Antineoplastic therapy was prescribed in 61.5% (n = 2846) of the hospitalizations for 730 patients, which comprised the final analysis cohort. Demographics of the study population are presented in Table 1.

Figure 1.

Patient selection flowchart for final analysis cohort.

Table 1.

Baseline clinical characteristics, by unique patient.

	Overall (n = 730)
	n	(%)
Sex
Female	326	(44.7)
Male	404	(55.3)
Race
White	655	(89.7)
Other	33	(4.5)
Not reported	42	(5.8)
Ethnicity
Hispanic/Latino	70	(9.6)
Non-Hispanic/Non-Latino	417	(57.1)
Unknown	243	(33.3)
Hematological malignancy subtype
LL	378	(51.8)
Lymphoma	206	(28.2)
AML	117	(16.0)
cMPD	11	(1.5)
MDS/MPD	10	(1.4)
OL	8	(1.1)
Hematopoietic stem cell transplant
LL	45	(11.9)
Lymphoma	20	(9.7)
AML	46	(39.3)
MDS/MPD	9	(90.0)
OL	4	(50.0)

AML: acute myeloid leukemia; MDS: myelodysplastic syndrome; cMPD: chronic myeloproliferative diseases; LL: lymphoid leukemias; MDS/MPD: myelodysplastic syndrome and other myeloproliferative diseases; OL: unspecified and other specified leukemias.

By hospitalization, lymphoid leukemia (LL) was the most common type of HM reported (51%), followed by lymphoma (24%), acute myeloid leukemia (AML; 21%), myelodysplastic syndrome and other myeloproliferative diseases (MDS/MPD; 2%), chronic myeloproliferative disease (cMPD; 1%), and unspecified and other specified leukemias (OL; 1%) (Table 2). AML was the most common HM subtype among ‘Infant’ (41%) and ‘Toddler’ (45%), whereas LL was most prevalent for ‘Child’ (60%) and ‘Adolescent’ (44%) groups. Overall, the median (IQR) age at admission was 7.5 (3.7–13.9) years and the time from initial cancer diagnosis to admission date was 133 (30–572) days. The rate of ICU admission was 11%, with 4% requiring intubation, and 51% having a surgical procedure. Of the surgical procedures, 4.3% of the interventions were classified as ‘major.’ Hematopoietic stem cell transplant (HSCT) was performed in 17.7% (n = 129) of patients.

Table 2.

Baseline clinical characteristics, by hospitalization and hematological malignancy subtype.

	Overall (n = 2846)		AML (n = 595)		cMPD (n = 35)		LL (n = 1447)		Lymphoma (n = 689)		MDS/MPD (n = 49)		OL (n = 31)		p Value
Age at admission, years															0.0001
Mean (SD)	8.6	(5.6)	8.0	(6.1)	7.4	(5.6)	8.3	(5.1)	10.1	(5.7)	5.9	(5.0)	3.0	(5.2)
Median (IQR)	7.5	(3.7–13.9)	7.4	(1.9–13.9)	5.3	(1.7–12.4)	7.2	(4.1–12.4)	9.8	(5.0–15.7)	3.6	(2.4–63)	1.4	(0.6–2.6)
Age group at admission, n (%)															<0.0001
Infant	111	(3.9)	46	(7.7)	0	(0.0)	34	(2.4)	17	(2.5)	0	(0.0)	14	(45.2)
Toddler	238	(8.4)	108	(18.2)	11	(31.4)	76	(5.3)	30	(4.4)	9	(18.4)	4	(12.9)
Child	1514	(53.2)	230	(38.7)	13	(37.1)	905	(62.5)	327	(48.5)	29	(49.2)	10	(32.3)
Adolescent	983	(34.5)	211	(35.5)	11	(31.4)	432	(29.9)	315	(45.7)	11	(22.5)	3	(9.7)
Length of stay, days															0.0001
Mean (SD)	11.0	(14.7)	21.9	(16.7)	11.1	(14.2)	7.9	(12.8)	6.6	(8.0)	21.0	(22.5)	27.4	(30.3)
Median (IQR)	4.9	(2.8–13.4)	23.1	(6.9–29.7)	4.4	(2.6–11.4)	4.0	(23–6.7)	4.2	(2.7–6.5)	9.5	(5.0–33.4)	26.2	(5.4–31.9)
ICU admission, n (%)															<0.0001
Yes	304	(10.7)	81	(13.6)	2	(5.7)	140	(9.7)	59	(8.6)	14	(28.6)	8	(25.8)
No	2542	(89.3)	514	(86.4)	33	(94.3)	1307	(90.3)	630	(91.4)	35	(71.4)	23	(74.2)
Intubation, n (%)															<0.0001
Yes	109	(3.8)	35	(5.9)	0	(0.0)	50	(3.5)	12	(1.7)	7	(14.3)	5	(16.1)
No	2737	(96.2)	560	(94.1)	35	(100.0)	1397	(96.5)	677	(98.3)	42	(85.7)	26	(83.9)
Surgery, n (%)															0.0005
Major	122	(4.3)	28	(4.7)	1	(2.9)	64	(4.4)	20	(2.9)	7	(14.3)	2	(6.5)
Minor	1331	(46.8)	300	(50.4)	18	(51.4)	656	(45.3)	310	(45.0)	26	(53.1)	21	(67.7)
No surgery	1393	(48.9)	267	(44.9)	16	(45.7)	727	(50.2)	359	(52.1)	16	(32.7)	8	(25.8)
HSCT, n (%)															<0.0001
	144	(5.1)	50	(8.4)	5	(14.3)	49	(3.4)	21	(3.0)	13	(26.5)	6	(19.4)
Distinct SCM prescribed															0.0001
Mean (SD)	16.5	(11.8)	22.6	(13.1)	15.2	(11.9)	14.8	(11.1)	13.9	(9.0)	24.9	(16.1)	24.3	(15.5)
Median (IQR)	13	(8–21)	20	(13–29)	11	(8–18)	12	(8–18)	12	(8–17)	21	(13–34)	21	(14–33)
Distinct antineoplastic agents prescribed															0.0001
Mean (SD)	2.9	(1.6)	2.8	(1.3)	2.8	(1.9)	2.7	(1.6)	3.4	(1.8)	2.3	(1.5)	3.7	(1.6)
Median (IQR)	3	(2–4)	3	(2–3)	3	(1–4)	2	(1–4)	3	(2–5)	2	(1–3)	4	(2–5)

AML: acute myeloid leukemia; MDS: myelodysplastic syndrome; cMPD: chronic myeloproliferative diseases; LL: lymphoid leukemias; MDS/MPD: myelodysplastic syndrome and other myeloproliferative diseases; OL: unspecified and other specified leukemias; HSCT: hematopoietic stem cell transplant; SD: standard deviation; ICU: intensive care unit; IQR: interquartile range.

Patterns of SCM utilization

In total, 495 distinct ‘systemic’ drugs were prescribed, of which 90.5% (n = 448) were SCM. The median (IQR) number of distinct agents used per hospitalization was 13 (8–21) for SCM and 3 (2–4) for antineoplastic agents (Table 2, Distinct antineoplastic agents prescribed; Supplementary Table S1 describes specific antineoplastic agent use by HM subtype). Table 3 lists SCM used in the treatment of ≥25% of patients with any HM subtype, and includes, acetaminophen (71%), diphenhydramine (65%), ondansetron (54%), and propofol (41%) among the most commonly prescribed non-antimicrobial SCM in the overall cohort. The most frequently used antimicrobial agents (Supplementary Table S2) were sulfamethoxazole and trimethoprim (67%), ceftazidime (42%), vancomycin (27%), gentamicin (24%), and acyclovir (23%) and their use varied by HM subtype. SCM burden was significantly associated (p < 0.0001) with higher age at admission, longer hospital stay, ICU admission, increased SOI, intubation, surgical procedures (major and minor), and increasing number of antineoplastic agents by multivariate regression analysis for the overall cohort (Figure 2(A) and Supplementary Table S3). However, increased ROM was associated with decreased SCM use (p = 0.005).

Table 3.

SCM percent exposure per hospitalization based on a threshold of ≥25% for any HM subtype.

Drug	Overall	AML	cMPD	LL	Lymphoma	MDS/MPD	OL
Acetaminophen*	71.3	82.5	65.7	71	61	79.6	93.6
Acyclovir	23.3	34.3	40	21.6	12.6	65.3	38.7
Allopurinol	20.7	17.7	25.7	23.6	17.4	12.2	29
Alteplase	16.5	35.3	8.6	10.3	11.6	30.6	41.9
Amlodipine*	14.1	18.5	14.3	13.6	9.7	32.7	19.4
Amoxicillin	4.5	5.7	5.7	3.4	1.6	51	22.6
Amphotericin B	12.2	25.4	31.4	9.7	2.9	30.6	29
Caspofungin	11.6	20.8	11.4	9.1	5.8	42.9	29
Ceftazidime	42.2	51.9	22.9	48.5	23.1	16.3	48.4
Cyclosporine*	5.2	12.6	20	2.7	1.3	30.6	0
Diphenhydramine	64.8	83.2	51.4	56.9	64.4	77.6	80.7
Dolasetron	38.7	44.7	40	42.5	26.9	18.4	35.5
Fentanyl*	24.7	27.6	22.9	23.2	23.8	38.8	35.5
Filgrastim	18	22.4	5.7	18.2	12.8	24.5	45.2
Fluconazole	20.8	23.7	25.7	22.1	15	18.4	35.5
Furosemide	17.5	26.7	11.4	15.4	11.6	38.8	45.2
Gentamicin	24.2	41.2	22.9	24	9.4	26.5	38.7
Hydrocortisone	18	22.5	17.1	14.5	19.2	42.9	25.8
Lansoprazole*	36.7	28.9	20	42.4	31.5	55.1	22.6
Lorazepam	39.3	60.2	25.7	29.1	41.8	46.9	64.5
Mannitol	17.7	31.3	8.6	10.7	21.6	12.2	16.1
Meropenem	12.9	25.4	31.4	9.7	4.5	49	29
Metoclopramide	10.7	15	17.1	10.2	5.5	36.7	19.4
Metronidazole	13.2	23.5	8.6	0.5	8.7	8.2	25.8
Midazolam*	17.7	23	11.4	17.4	13.2	24.5	29
Morphine*	34.9	44.2	25.7	33.9	27.1	57.1	51.6
Nystatin	7.5	12.4	17.1	5.2	4.8	34.7	25.8
Ondansetron*	53.8	60	34.3	44.9	67.1	61.2	67.7
Oxycodone	35.2	45.6	25.7	27.7	41.8	36.7	51.6
Piperacillin and enzyme inhibitor	12.7	23	11.4	10.9	7.8	4.1	25.8
Promethazine	33.4	46.6	31.4	27	33.7	53.1	41.9
Propofol*	40.9	44.2	40	40.5	38.2	44.9	54.8
Ranitidine	21.8	28.7	17.1	18.5	23.1	14.3	29
Rocuronium bromide*	11.1	13.6	11.4	11	7.1	24.5	32.3
Sevoflurane	25.5	29.4	25.7	27.6	17.3	30.6	25.8
Sulfamethoxazole* and trimethoprim	67.1	82.9	65.7	61.2	65	71.4	80.7
Tacrolimus*	2.7	2.7	0	2.8	0.9	28.6	0
Ursodeoxycholic acid	5.9	10.8	8.6	4.5	2.8	28.6	9.7
Vancomycin	27.4	51.3	14.3	22	16.6	42.9	54.8
Voriconazole*	16.1	49.1	5.7	7.8	5.5	4.1	35.5

Notes: All numbers are reported as percentages.

AML: Acute myeloid leukemia; MDS: myelodysplastic syndrome; cMPD: chronic myeloproliferative diseases; LL: lymphoid leukemias; MDS/MPD: myelodysplastic syndrome and other myeloproliferative diseases; OL: unspecified and other specified leukemias.

^*Denotes SCM associated with ‘same-day SCM–Chemotherapeutic DDGIs.’ Gray shaded SCM are anti-infectives.

Figure 2.

Coinciding SCM and chemotherapy. (A) Multivariable regression analysis of clinical covariates associated with increasing SCM burden. The dashed vertical line is the line of no difference. Boxes represent coefficients and horizontal error lines represent 95% confidence interval. (B) Cumulative distinct drug exposures by hospital day, designated as SCM (green circles) or chemotherapeutics (blue circles). Error bars are mean and standard deviation. (C) Patient level drug exposure and coinciding PK-related genes. For each patient in the cohort, the total number of distinct drugs to which they have been exposed, that is, a drug is counted once, across all records of clinical care in which chemotherapy was prescribed (x-axis). The y-axis displays the corresponding number of distinct PK-related genes where each gene is counted only once. The total number of PK-related genes based on total distinct SCM are indicated by the black circles. The number of PK genes that intersect for both SCM and chemotherapy are presented as circles. Horizontal and vertical red dashed lines represent the median number of total distinct PK-related genes or SCM per patient, respectively. Abbreviations/acronyms: ICU (Intensive care unit), SCM (supportive care medications), PK (pharmacokinetic).

Drug-gene and drug-drug-gene interactions

For patients receiving chemotherapy, SCM comprise a significant proportion of unique chemical entities to which a patient is exposed (Figure 2(B)). Of all medications prescribed, 26.5% (131 of 495 drugs) were associated with 159 PK-related genes. Of these medications, 113 were SCM and 18 chemotherapeutics. At the patient level, a median (IQR) of 26 (16–48) distinct SCM were administered (across all hospitalizations in which chemotherapy and any antimicrobial were prescribed) and this is represented by the vertical dashed red line in Figure 2(C). These patient level distinct SCM exposures were associated with a median (IQR) of 41 (27–60) distinct PK-related genes (horizontal dashed red line, Figure 2(C)) and reflect drug-gene associations which are also referred to as drug-gene interactions (DGI; Figure 3(A)). Of these, a median (IQR) of 15 (13–26) genes (Figure 2(C), orange circles) were also associated with the disposition of chemotherapy, creating the potential for SCM-chemotherapy drug-drug-gene interactions (DDGI).

Figure 3.

Potential PK SCM-chemotherapy DDGIs among children diagnosed with HM. (A) Co-prescribed drugs, including SCM and chemotherapy, which share pathways of biotransformation and/or transport can converge on single genes facilitating these processes. Many of these genes are highly polymorphic, leading to functional alterations in their activity. This can lead to drug-gene interactions. Likewise, two (or more) drugs (e.g. Drug A and Drug B) interacting at the same drug metabolizing enzyme or transporter can profoundly influence drug PK affecting drug safety and efficacy and are considered as potential DDGIs. If drug A is an SCM and drug B a chemotherapeutic and both are substrates, inhibitors, or inducers of an enzyme or transporter, this would be considered as a potential SCM-chemotherapy DDGI. (B) The percentage of SCM and chemotherapy drug pairs prescribed on the same hospital day which coincide with a PK-related gene (i.e. DDGI). PK-related genes mediating phase I and phase II biotransformation and drug transport are grouped into one of three ovals - Phase I as a light red oval, Phase II as a hashed oval, and transporters as the open oval. In the text boxes adjacent to the ovals, are listed the overall percentage of SCM-chemotherapy drug pairs associated with each PK function type (Phase I, II, and transport) and the contribution of individual genes to that specific PK function type. Overlap between ovals represents the percentage of SCM-chemotherapy drug pairs subject to combinations of Phase I, II and/or transport processes. SCM-chemotherapy DDGIs are based on the parent compounds only, metabolites were not included in the analysis. ‘Drug metabolizing enzyme or transporter’ represented by ribbon diagram of CYP3A4: https://www.rcsb.org/structure/6bdh; ‘Xenosensor’ represented by ribbon diagram of human retinoic x receptor/liver x receptor heterodimer on DNA (http://www.rcsb.org/structure/4NQA); Diagrams modified using PyMOL Molecular Graphics System, Version 2.0 Schrödinger.

In a single hospital day, patient regimens often contained >1 drug coinciding with a specific PK-related gene (DDGI; Figure 3(A)). Potential same day SCM-antineoplastic and SCM-SCM DDGIs were present in 48.3% (n = 1375) and 83.7% (n = 2383) of hospitalizations, respectively. The median (IQR) for SCM-antineoplastic DDGI was 3 (1–6) and SCM-SCM DDGI was 9 (3–26). In univariate analysis, hospitalizations with HSCT, intubation, ICU admission, surgical procedure (major or minor), HM subtype, older (Adolescent) or younger (Infant) ages, and increased SOI and ROM scores were significantly associated (p < 0.05) with SCM-chemotherapy DDGI. SCM-chemotherapy DDGI, as a dichotomous variable in a final logistic regression model, was significantly associated (odds ratio (OR), 95% confidence interval (CI), p-value) with NICHD age group (OR 1.14, CI 1.02–1.27, p = 0.026), degree of polypharmacy (OR 1.37, CI 1.26–1.48, p < 0.0001), HM subtype (OR 1.49, CI 1.38–1.62, p < 0.0001), surgical procedure (OR 2.76, CI 2.28–3.36, p < 0.0001), and HSCT (OR 10.60, CI 5.23–25.40, p < 0.0001). The number of SCM-chemotherapy DDGI decreased in hospitalizations after HSCT as compared to encounters prior to receiving HSCT (p < 0.0001) or as compared to those who did not receive HSCT (p < 0.0001).

The total number of potential SCM-antineoplastic DDGIs (n = 7788), encompassed 60 drugs (50 SCM and 10 chemotherapeutics; Table 3, SCM indicated with an asterisk, and Supplementary Figure S1) and 22 genes classified as drug transporters, phase 1, or phase 2 drug metabolizing enzymes (Figure 3(B)). The exception to this classification (and excluded from Figure 3(B)) was the xenosensor nuclear receptor (NR1I2; pregnane x receptor), which, upon activation, influences the expression of a diverse array of genes, including the upregulation (Figure 3(A), Induction) of xenobiotic biotransformation enzymes and transporters [31].

Changes in prescribing over time could affect estimates of DGI and DDGI. Approximately 12% (n = 56) of SCM had >10% year-to-year variation in utilization within the study period. Among these, there was significant multi-year fluctuations in trends of use with 8 agents exceeding and 5 declining by ≥25% change in use (Supplementary Table S4). For SCM polypharmacy, year-to-year rates, based on the number of distinct agents per hospitalization, remained consistent (Supplementary Figure S2). Overall, 93% of hospitalizations had ≥5 SCM prescribed, 67% with ≥10, 42% with ≥15, and 26% with ≥20.

Availability of literature with SCM population-specific PK parameters

Assessing the consequences of potential DGI and DDGI will require population-specific PK information. A systematic review of the literature (Supplementary Table S5; query syntax) for those SCM prescribed to ≥25% of patients as presented in Table 3, population-specific PK information was available for 8 of 26 non-antimicrobial SCM (Supplementary Table S6) and 11 of 14 of the antimicrobials. Several studies failed to meet inclusion criteria due to the inability to disaggregate PK data between adults and children or from patients without a cancer diagnosis. The majority of the non-antimicrobial studies involved cyclosporine (n = 13), followed by morphine (n = 9), filgrastim (n = 6), loraze-pam (n = 4), fentanyl (n = 2), acetaminophen (n = 1), dolasetron (n = 1), and tacrolimus (n = 1). Of these, 5 studies had >50 patients enrolled. The median number of patients per study was 15 (range 1–91).

Discussion

SCM are an essential, and sometimes, life-saving part of treatment for patients with cancer [32,33]. Prescribing practices for SCM vary markedly across children’s hospitals in the United States, including the oncology wards [6]. This study assessed SCM utilization for pediatric patients with HM within a healthcare system representative of the Intermountain West region of the United States and the corresponding population-specific PK information to identify areas needing study. We found that for children with HM, ‘SCM polypharmacy’ (defined as ≥5 different SCM) was common and nearly half had potential SCM-chemotherapy DDGIs.

Dai and colleagues, utilizing the Pediatric Health Information System (PHIS) database, reported that pediatric patients diagnosed with neoplasms, especially those admitted to the ICU, had the highest risk for DDIs [7]. Within our study cohort, 11% were admitted to the ICU, but many patients experienced other risk factors for potential DDIs such as frequent polypharmacy and hospitalizations >7 days. For individual patients, increased SCM burden was associated with surgical procedures, length of hospitalization, and severity of illness. As such, SCM polypharmacy may be viewed as a ‘modifiable factor,’ as opposed to altering an antineoplastic regimen, to reduce the potential for DDIs or ADRs [34]. In contrast, we did find that increased ROM was associated with a decrease in SCM burden in pediatric patients with HM. This appears to be in accord with initiatives to decrease drug burden (‘thoughtful discontinuation’) in the care for those with terminal malignancies [35]. As a limitation of this retrospective study and sample size, we cannot determine prescriber intent or evaluate if discontinuation intended to improve quality of life and/or limit costs. This warrants further investigation.

Our focus was on patterns of SCM use that might change the disposition of chemotherapy through DDGIs, and thereby identify specific drugs and genes of high clinical relevance for targeted PK and genetic studies. While it may be possible, with sufficient information, to perform individualized dose adjustments to counterbalance known DDIs in clinical care, it may be more effective to reduce the overall concomitant use of SCM and chemotherapeutics. A place to begin would be with SCM used most commonly, such as those agents we identified as being administered to ≥25% of patients with HM.

Differences in methodology, setting, and time make comparisons of adult studies evaluating potential chemotherapeutic-comedication PK DDIs with our own analysis difficult [36,37]. In particular, there is a lack of adult data in the context of HM and the patterns and types of drugs prescribed (e.g. warfarin), and associated with risk of DDI, are frequently related to the treatment of comorbid illness (e.g. cardiovascular, diabetes, and depression) more common among elderly patients [38,39]. Additionally, the PK of a given drug can differ substantially between children with cancer and that of their cancer-free peers or adults. In part, this PK variability stems from factors, often dynamic in nature, such as cancer biology, chemotherapeutic regimen, age (maturation), and nutritional status [40–43]. Pro-inflammatory conditions, caused by malignancy and/or infection, influences the expression patterns of drug metabolism enzymes and transporters [44]. These characteristics ensure most drugs administered to children with cancer will exhibit distinct, although variable, PK.

Genetic variation in the processes underpinning the kinetic principles of drug ADME, can also substantially contribute to PK variability. However, the degree to which genetic polymorphisms contribute to variability in drug PK among pediatric patients with cancer remains unknown [8,19]. We used the PharmGKB resource, a curated pharmacogenomics database intended to connect individual genetic variation with differences in drug disposition and effect, to approximate the magnitude of PK-related gene-drug associations in our cohort. Although, our estimate is restricted to the parent compound as we did not evaluate genes associated with metabolites of SCM or antineoplastics. Our analyses did not extend to parsing or interpreting specific ‘clinical annotations’ of chemical-gene pairs based on individual variants or gene haplotypes, as these can also include in vitro studies, instances of non-replicated, or contradictory associations. These associations are further confounded by studies with combination drug therapy (polypharmacy) leaving it unclear as to which drug(s) are associated with a particular genetic variant [45]. Even in circumstances for which an explicit connection between gene and drug is made, attributing the consequences of the variant (e.g. nonsynonymous SNP), and defining a null result (i.e. no relationship) complicates interpret-ability [16,46].

Large databases with information pertaining to potential DDIs, gene-drug, and disease-drug interactions and ADR risks are increasing [47,48]. Yet, a lack of population-specific PK data hinders their clinical utility to provide meaningful dosing guidance for pediatric patients diagnosed with HM. It is unclear, and often controversial, about how to interpret the information from these resources and apply it to pediatric patients with cancer [16]. For instance, among patients diagnosed with cancer with PHIS records, 70% of the potential DDIs reported could be classified as contraindicated or severe [5]. The need to re-calibrate the risk-to-benefit for drug combinations (to diminish DDI ‘alert fatigue’) in a population-specific manner has been recognized [49].

The selection criteria for what constituted SCM in this study, was oriented to agents achieving systemic exposure, but some agents taken orally, such as nystatin, largely remain in the intestinal lumen and are considered to have minimal risk for clinically significant DDIs [50]. Yet, SCM-antineoplastic interactions within the gastrointestinal tract can significantly influence drug exposures. Several case reports of a potential intestinal interaction between co-administered metronidazole (a frequent SCM in this study) with busulfan leads to increased busulfan plasma concentrations and associated toxicity [51]. While metronidazole was prescribed in 10 of 39 hospitalizations where busulfan was used, there was only a single instance of co-administration. As a limitation of this retrospective study and sample size, we cannot determine prescriber intent or whether potential DDIs influenced therapeutic decisions.

Our systematic literature review agrees with earlier studies demonstrating that clinical trials provided limited PK data, even for the most frequently used SCM [52]. For the studies that do exist, differences in study design, clinical environment, PK estimation methods, drug assay methodology, as well as differences in drug formulation and route of administration complicate comparisons of PK and outcome [53]. However, obtaining pediatric PK data for SCM has many challenges such as limited patient recruitment among rare cancer subtypes, exclusion of patients due to organ dysfunction, and challenges of studies in infants and adolescents [52,54]. As an example, a study assessing the PK of morphine in children with cancer took 3.5 years to enroll 40 patients across seven centers [55]. There is even less data available for those children with cancer presenting with renal or hepatic dysfunction on which to base rational dosing decisions [56]. As our study showed, patterns of SCM use varied by HM subtype, this highlights the heterogeneity and complexity inherent in the clinical care of children with blood cancer.

This study has several important limitations. This analysis was restricted to a subset of medications administered to inpatients via oral, enteral, and/or parenteral routes and may have missed important outpatient prescriptions as well as self- or parent provided OTC medications and/or supplements. Our study was restricted to a single geographic region of the Intermountain West and limited to pediatric patients receiving antimicrobial therapy, which additionally narrows the generalizability of our findings [57]. Regional formulary decisions or drug supply shortages could have impacted our results on SCM utilization. Although retrospective in design, our study may better reflect standard-of-care patterns of SCM utilization not captured in clinical trial settings, which tend to underestimate supportive care resource utilization [58].

In conclusion, this study demonstrates that pharmacologically diverse SCM exposure, concurrent with active antineoplastic therapy, was frequent among pediatric patients with HM. For a subset of drugs, notably SCM and chemotherapy prescribed on the same day, there was often overlap with specific elements of xenobiotic detoxification pathways creating the potential for altered drug exposures. A lack of population-specific PK data for SCM hinders the ability to investigate the consequences of polypharmacy in children with HM. The Best Pharmaceuticals for Children Act (BPCA) of 2002 set out to reduce medication errors affecting hospitalized children. BPCA and the Pediatric Research Equity Act stress the need to conduct pediatric-specific drug studies. In order to translate these goals into reality, contemporary and population-specific drug utilization information will be necessary [7,59].