Institutional and Regional Variation in Opioid Prescribing for Hospitalized Infants in the US

Key Points Question How does the prescribing of short-acting opioids and methadone in high-risk infants vary across institutions and regions of the US? Findings In this cohort study of 132 658 high-risk infants, there was significant hospital-level variation in opioid and methadone exposure and cumulative days received. The study estimated that 16% of the variability in any opioid exposure and 20% of the variability in methadone treatment was attributable to the individual hospital. Meaning These findings suggest that institution-level variation in opioid and methadone prescribing in hospitalized infants persists across US children’s hospitals, underscoring the need for standardized prescribing in this vulnerable population.


Introduction
2][3] Infants exposed to painful procedures experience acute physiologic responses and increased morbidity, and opioids reduce these poor outcomes. 4,51][12][13][14] Ultimately, regional and hospital-level differences in hospitalized infants' opioid exposure may have significant effects on both short-and long-term clinical outcomes.
Regional opioid prescribing differences in the US are well described in adults. 157][18][19][20] Infants infrequently receive outpatient opioids, with most opioid exposure occurring during hospitalization. 212][23][24][25][26] Understanding regional and institutional variation can inform best-practice initiatives and encourage practitioners to carefully consider when opioids are the appropriate medications for the patient.
This study examined regional and institutional variation in overall opioid exposure and methadone treatment in high-risk infants cared for at children's hospitals in the US.We hypothesized that significant variation exists across US regions and among institutions in opioid exposure and methadone treatment among high-risk hospitalized infants.

Study Design
A cohort of high-risk infants younger than 1 year at the time of admission from January 1, 2016, to December 31, 2022, was identified using the Pediatric Health Information Systems (PHIS) database.
The PHIS database is maintained by the Children's Hospital Association and includes clinical and resource utilization data for both inpatient and outpatient encounters for more than 47 children's hospitals across the US.Hospitals registered in the PHIS database include freestanding children's hospitals and large pediatric hospitals within an academic health system. 27All data are deidentified, and data integrity is checked by the Children's Hospital Association data quality program, which issues quarterly reports to participating hospitals that detail any quality concerns.The Children's Hospital Los Angeles Institutional Review Board approved the study and waived the need for informed consent given the use of deidentified data.This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
2][3] We defined high-risk infants using International Statistical Classification of Diseases and Related Health Problems, extracorporeal membrane oxygenation (ECMO), and other abdominal surgery.These categories were defined by ICD-10 codes for each patient encounter and thus are not mutually exclusive.Each patient may fall into multiple categories.
Patients with the following ICD-10 codes were excluded: P96.1 (neonatal withdrawal symptoms from maternal use of drugs of addiction), P96.2 (withdrawal symptoms from therapeutic use of drugs in newborn), and P90.49(newborn affected by maternal use of other drugs of addiction) (n = 7055) (eFigure in Supplement 1).Patients receiving methadone alone and no other opioid exposure were excluded because receipt of methadone alone likely captured infants with in utero opioid exposure undergoing treatment for withdrawal prevention (n = 110).Additionally, patients with malignant tumors as defined by Feudtner et al 28 were excluded because pain management in these patients is complex and can be chronic and/or palliative in nature (n = 1895).Infants who did not undergo a cardiac or abdominal surgical intervention and had a CHD or abdominal surgical diagnosis code alone were excluded (cardiac: n = 139 647, abdominal: n = 35 495) (eFigure in Supplement 1).
The cohort was divided into US Census and PHIS regions based on hospital location.The regional distribution of the 47 institutions was as follows: Northeast, 6; South, 17; Midwest, 14; and West, 10.Individual hospitals were deidentified so that variation among institutions could be examined.Demographic and clinical factors were characterized and included the following: gestational age at admission (weeks), biologic sex, birth weight (kilograms), race, ethnicity, insurance type, mechanical ventilation, intensive care unit (ICU) stay, and number of complex chronic conditions.0][31] Race categories were defined according to the PHIS and included American Indian, Asian, Black, Pacific Islander, White, other (a defined race category in the PHIS database), and unknown. 32Similarly, ethnicity categories (Hispanic, non-Hispanic, and unknown) were determined via the PHIS. 32Hospitals submit race and ethnicity data to the PHIS for each visit according to hospital-specific practices, which may include parent or guardian report at the time of hospital registration.Additionally, the literature suggests that sex-based differences in behavior and gene expression are associated with neonatal opioid exposure. 33,34Insurance or payer type is also associated with opioid prescribing both postoperatively and in hospitalized pediatric patients. 29,35,36

Outcome Measure
The primary outcome of interest was any opioid exposure (including short-acting opioids and methadone) during hospitalization.Opioids were identified based on pharmacy billing codes for opioid analgesic.Additional outcomes of interest included cumulative days of opioid exposure, methadone treatment after short-acting opioid exposure during hospitalization, and cumulative methadone days.

Statistical Analysis
Continuous variables were described using median (IQR), and Wilcoxon-Mann-Whitney tests were used to compare findings among different regions.Categorical variables were described using numbers (percentages), and χ 2 or Fisher exact tests were used to compare findings among groups.
The Kruskal-Wallis test was used to examine the difference in distributions between the cumulative days of exposure to opioids and methadone.
A 2-level hierarchical generalized linear model (HGLM) with hospitals as the random intercepts was used to evaluate variation in opioid prescribing; patients (at level 1) were nested within institutions (at level 2). 37Institution-level variation in any opioid prescribing was estimated by computing the intraclass correlation coefficient, which indicates how much of the total variation in the probability of use during a patient encounter is attributed to the hospital.Due to the nonnormal distribution of data, an HGLM model was executed to examine the effects of demographic and clinical factors on natural log-transformed cumulative days of opioid exposure during hospitalization.Results were then back-transformed for ease of interpretation.Because most of the cohort was composed of infants with CHD requiring intervention, these infants were used as the reference group.Covariates (sex, race, ethnicity, insurance type, prematurity, high-risk category, mechanical ventilatory support, ICU stay, and complex chronic conditions) were chosen a priori, 14,[29][30][31][32][33][34][35][36] on the basis of clinical expertise, and from significant bivariate associations.Model fit was checked via the Akaike information criterion and the Bayes information criterion, where lower values indicate improved fit. 38All analyses were conducted with an α = .05.Data were analyzed using SAS software, version 9.4 (SAS Institute Inc).

Results
Following exclusions, a cohort of 132 658 high-risk infants was evaluated (0.4% American Indian, procedural intervention.There were significant regional differences in demographic and clinical factors, including high-risk diagnosis category, proportion of patients undergoing surgical intervention, and number of high-risk procedures.The Northeast and West regions had a higher percentage of patients who were surgically treated during the index hospitalization.
During hospitalization, 76.5% of high-risk infants were prescribed opioids.When examining specific opioid type during hospitalization, 66.5% of high-risk infants were exposed to fentanyl, 60.6% to morphine, 5.8% to hydromorphone, and 7.9% to methadone (Table 2).On bivariate comparison, opioid and methadone exposure significantly varied by US region ( Outcome was back-transformed.
b "Other" is a defined race category in the Pediatric Health Information Systems database.
support when compared with VLBW infants (69.3% vs 37.6%).Additionally, cumulative days of opioid exposure increased for each 1-unit increase in the number of complex chronic conditions (1.36; 95% CI, 1.36-1.37)(Table 3).Additionally, all but 1 (VLBW) of the high-risk infant diagnosis categories were associated with increased cumulative opioid days.Cumulative opioid days in patients with surgical NEC were 104% (95% CI, 99%-112%) higher compared with patients without surgical NEC.
Significant hospital-level variation in opioid and methadone exposure was demonstrated within each region (Figure 1).Similar wide variation in cumulative days of opioid and methadone exposure was seen within each US region (Figure 2).Results from our HGLM unconditional model were used to calculate the intraclass correlation coefficient.The probability of opioid exposure   at a typical US children's hospital was 78%, but the probability of opioid exposure varied considerably across institutions.The probability of methadone exposure after short-acting opioid exposure in high-risk infants at a typical US children's hospital was 6%, but the probability of methadone exposure similarly varied considerably across hospitals.Approximately 16% of the variability in opioid prescribing and 20% of the variability in methadone treatment were attributable to the individual hospital.Therefore, 84% of the variability in opioid prescribing and 80% of the variability in methadone treatment were accounted for by individual patient-level characteristics and possible unobserved or unknown characteristics.

Discussion
In this retrospective cohort study of 132 658 high-risk infants admitted to 47 children's hospitals in the US, we found that most received opioids during hospitalization with wide variation across US regions and between hospitals.Furthermore, 16% of the variability in any opioid prescribing and 20% of the variability in methadone treatment was attributable to the individual hospital.This is the first study, to our knowledge, to describe regional variation and quantify institutional variation in inpatient opioid exposure and duration of prescribing in high-risk infants.Our findings underscore the magnitude of variation at children's hospitals across the US, which likely has significant implications for short-and long-term clinical outcomes and resource use.
Adult data indicate that state and regional patterns in prescribing have remained relatively consistent, with areas of high prescribing persisting despite national efforts to decrease opioid exposure. 15Similar findings have been reported in the pediatric literature examining outpatient prescription opioids. 180][41][42] Our study found that once hospital-and patient-level characteristics were controlled for, regional variation was no longer significantly associated with opioid prescribing, with the exception of the Northeast region.These findings imply that local and institutional quality improvement efforts are needed to minimize variation in infant opioid prescribing.
Notably, opioids can decrease morbidity that can result from the physiologic response to pain in infants. 4,5However, prolonged exposure in neonatal surgical populations is associated with increased health care use, including prolonged hospitalization, total parenteral nutrition use, and mechanical ventilation. 98]11 Viewed collectively, our results underscore the need for individual hospitals to track cumulative opioid exposures and methadone use for hospitalized infants to improve clinical and possibly neurodevelopmental outcomes.
9,39 Importantly, despite the significant institutional variation in opioid prescribing, 84% of the variability in opioid exposure and 80% of the variability in methadone treatment were accounted for by individual patients and other unknown characteristics.4][45][46] Investigations into genetic predisposition to poor opioid response have been conducted in infants receiving mechanical ventilation.A study by Elens et al 47 concluded that specific genetic alleles predispose premature infants to diminished opioidinduced pain relief.Further investigation of genetic factors that influence opioid response in high-risk infants could explain the variability in opioid prescribing attributed to unknown factors.
In the hierarchical analysis, patients requiring mechanical ventilation demonstrated higher cumulative opioid days of exposure.Use of opioids as sedation in infants receiving mechanical ventilation has been increasing over time despite illness severity not increasing. 4835,36 Additionally, emerging preclinical data suggest that there are sex-based differences in behavior and gene expression associated with neonatal opioid exposures and metabolism. 33,34Importantly, our findings highlight that specific subpopulations within high-risk infants are expected to have more cumulative days of receiving opioids, including those with surgical NEC, ECMO, CHD, and ELBW and those undergoing abdominal surgery.Infants with surgical NEC are at risk for increased opioid exposure and methadone use, 9 as well as inadequate pain control despite opioid administration. 49Similarly, initiation of ECMO is associated with substantial sedative exposure and increased frequency of iatrogenic withdrawal syndrome. 50,51[54] Interestingly, although prematurity and VLBW were both associated with decreased cumulative opioid days, ELBW was associated with increased opioid days.Previous studies examining the effects of opioids on premature infants support judicious use of opioids and may have influenced current practices. 55,56However, ELBW infants often require prolonged hospitalization and mechanical ventilation; thus, their clinical need for opioids could be greater.In our cohort, ELBW infants had a median (IQR) length of stay of 76 (17-119) days, and VLBW infants had a median (IQR) length of stay of 36 (19-62) days.Additionally, a higher percentage of ELBW infants received ventilation when compared with VLBW infants (69.3% vs 37.6%), which likely explains the differential effect on cumulative opioid days.Notably, a higher cumulative opioid dose is associated with worse cognitive outcomes at 20 months in ELBW infants. 11Our findings highlight that certain infant populations are at increased risk for prolonged opioid exposure and thus would likely benefit the most from standardized opioid protocols.
[59][60][61][62][63] However, these efforts largely exclude infants, despite opioid exposure having unique neurodevelopmental risks in this age group.Future studies are needed to examine the long-term effects of prolonged opioid exposures in high-risk infants and further understand the impact of variations in institutional prescribing.Ultimately, opioid stewardship efforts encouraging judicious opioid prescribing should also extend to infants.

Limitations
This study has some limitations.Of note, use of an administrative database (PHIS) leads to a nonrandom sample of hospitals, which, although large in number, may have different opioid prescribing practices than hospitals not included in the PHIS.The PHIS database contains data on patients cared for at children's hospitals, which limits generalizability because this may not be representative of high-risk infants admitted to other types of hospitals.Additionally, the PHIS does not include medication dosage or frequency.Thus, it is possible that the observed hospital-level variation may reflect differences in institutional prescribing trends.For example, an institution that administers opioids daily could be administering fewer total morphine equivalents than a center that uses larger doses but for fewer days.Furthermore, there is little context for the variation observed, and conjecture is needed to deduce prescribing choices.Similarly, clinician variability and clinicianlevel characteristics and practices are not examined within this study and may have contributed to the observed institutional variation.The use of multimodal pain management strategies, including nonopioid pain medications, regional anesthetics, and nonpharmacologic interventions, was also not examined and may have influenced variation.

Figure 1 .
Figure 1.Institutional Variation in Any Opioid and Methadone Exposure by US Region

Table 2
). Examining any opioid prescribing within regions showed that of high-risk infants, 74.2% in the Northeast, 78.6% in the South, 71.6% in the Midwest, and 81.2% in the West were exposed to opioids.Similarly, of high-risk infants, 4.6% in the Northeast, 10.3% in the South, 6.6% in the Midwest, and 7.1% in the West were exposed to methadone.Notably, when infants were stratified by preterm vs term birth, we found that premature infants had less exposure to opioids and methadone than term infants (eTable 2 in Supplement 1).Across the total cohort, the median (IQR) duration of any opioid exposure was 5(2-12)cumulative days, and the median (IQR) duration of methadone exposure was 19(7-46)cumulative days.Despite median days of opioid exposure between US regions being equal (5 days), cumulative days of opioid use significantly varied by US region due to differences in data distribution and IQR among regions (eTable 3 in Supplement 1).Median (IQR) cumulative days of methadone exposure was23 (11-54)days in the Northeast, 18(7-43)days in the South, 21(7-53)days in the Midwest, and 16(6-43)days in the West (eTable 3 in Supplement 1).When groups were stratified by prematurity, White race (0.94; 95% CI, 0.91-0.98),andprivatevs public insurance (0.96; 95% CI, 0.94-0.97)(Table3).Although the Northeast region was associated with decreased days of opioid exposure when compared with the West (0.82; 95% CI, 0.70-0.97),no other US region was significantly associated with days of opioid exposure.Notably, cumulative days of opioid exposure in premature infants was 13% lower (95% CI, 16% to 11%) lower compared with infants who were full term.Other

Table 1 .
Demographic and Clinical Characteristics of the Total Cohort and by US Region a Abbreviations: CHD, congenital heart disease; ECMO, extracorporeal membrane oxygenation; ELBW, extremely low birth weight; HIE, hypoxemic ischemic encephalopathy; ICU, intensive care unit; NEC, necrotizing enterocolitis; NICU, neonatal intensive care unit; VLBW, very low birth weight.aDataare presented as number (percentage) of patients unless otherwise indicated.b"Other" is a defined race category in the Pediatric Health Information Systems

Table 2 .
Opioid and Methadone Exposures in Total Cohort and US Region

Table 3 .
Association of Demographic and Clinical Characteristics With Cumulative Days of Opioid Exposure a Reference diagnosis of congenital heart disease.

eTable 3 .
Cumulative Days of Opioid and Methadone in Study Cohort and Stratified by U.S. Region eTable 4. Cumulative Days of Opioid and Methadone Exposure in Premature vs Term Infants and Stratified by U.S.