A Machine Learning–Based Clustering Analysis to Explore Bisphenol A and Phthalate Exposure from Medical Devices in Infants with Congenital Heart Defects

Abstract

Background:

Plastic-containing medical devices are commonly used in critical care units and other patient care settings. Patients are often exposed to xenobiotic agents that are leached out from plastic-containing medical devices, including bisphenol A (BPA) and phthalates. Given the potential health implications, there is an urgent need for a comprehensive understanding of this exposure.

Objectives:

This multi-institutional study aimed to determine the time-dependent concentrations and analyze the exposure patterns of bisphenol A (BPA) and phthalate metabolites in urine obtained from infants with congenital heart defects (CHD) undergoing cardiac surgery during the perioperative period.

Methods:

We collected daily urine samples from infants with CHD undergoing cardiac surgery during the perioperative period (from birth to 21 d) and measured BPA, di(2-ethylhexyl)phthalate (DEHP) metabolites (MEHP, MEHHP, MEOHP, MECPP), and non-DEHP phthalate metabolites (MBP, MBzP, MMP, MEP, MCPP) using ultra-high-performance liquid chromatography–tandem mass spectrometry. Machine learning–based cluster analysis was utilized to analyze these time-dependent data.

Results:

Utilizing a machine learning–based clustering approach, six distinct clustering groups were identified among infants exhibiting similar time-series toxicokinetic exposure patterns. These distinct clustering groups correlated with the utilization of extracorporeal membrane oxygenation (ECMO) and cardiopulmonary bypass (CPB), as well as the intensity of medical care. Notably, clustering groups associated with ECMO use demonstrated elevated levels of urinary BPA and DEHP metabolites compared to those without ECMO use, a trend not observed with non-DEHP metabolites. Moreover, peak concentrations in toxicokinetic profiles were associated with intensity of medical care.

Discussion:

Our findings suggest that dynamic changes of urinary BPA and DEHP metabolites corresponded to the type and number of medical devices used in infants. Further studies are needed to investigate the potential toxicological risks of infants with CHD undergoing cardiac surgery exposed to these chemicals in medical devices. https://doi.org/10.1289/EHP15034

Introduction

Multiple endocrine-disrupting chemicals such as bisphenol A (BPA) and phthalates can not only mimic the biological actions of hormones, but can also interfere with and disrupt components of the endocrine system, negatively impacting fetal growth and development as well as the onset of puberty, fertility, and general neurodevelopment.^1–3 Studies in animal models and epidemiological research have shown that exposure to BPA or phthalates can alter neurotransmitter homeostasis in different regions of the brain, contributing to neurodevelopmental effects in infants and children.⁴ In recent years, exposure to BPA and phthalates through the use of medical devices has received increasing attention,⁵ especially in newborns deemed highly vulnerable to environmental toxins and stressors. Previous studies reported that high levels of BPA, di(2-ethylhexyl)phthalate (DEHP),^6–9 and other non-DEHP metabolites [e.g., mono-n-butyl phthalate (MnBP), mono-isobutyl phthalate (MiBP), and mono-benzyl phthalate (MBzP)]^10–12 are present in the urine and blood of infants and children undergoing pediatric intensive care following exposure to life-saving medical devices.

Patients in neonatal and pediatric intensive care units are exposed to multiple plastic-containing medical devices such as intravenous (IV), intraarterial, and bladder catheters; IV fluid bags; endotracheal and nasogastric tubes; cardiopulmonary bypass (CPB) circuits; and extracorporeal membrane oxygenation (ECMO) circuit tubing. In addition to these plastic-based devices, stored blood products and transfusions are recognized as another major source of phthalate exposure, particularly in critically ill neonates requiring multiple transfusions during surgery.^13–15 These devices often incorporate polymer materials, such as polycarbonate plastics and polyvinyl chloride (PVC), which contain significant amounts of BPA and DEHP. BPA is primarily used as a building block in the production of polycarbonate plastics, whereas DEHP is not covalently bound to PVC and can leach out into bodily fluids.^7,13,16 Additionally, dermal absorption is another potential route of exposure, particularly from devices that come into contact with the skin, such as electrocardiogram (ECG) leads.¹⁷ Urinary levels of DEHP metabolites in infants spending time in intensive care settings often exceed concentrations reported for children ($<6$ years) in the US National Health and Nutrition Examination Survey (NHANES) (NHANES 2001–2002).^13,16 A recent study¹⁸ further demonstrated that phthalate metabolites were detectable in all pediatric cardiac surgery patients and that CPB circuits, particularly those primed with red blood cells (RBCs), significantly contributed to post-surgical phthalate exposure. Their findings underscore the importance of assessing CPB-related exposure to neonates undergoing cardiac surgery. We have previously reported⁷ that high BPA and DEHP metabolite levels are found in the urine of neonates undergoing cardiac operations, suggesting that infants’ exposure to BPA and/or DEHP metabolites is associated with the use of medical devices.

Given the increasing concern of exposure to BPA and phthalates from medical devices in infants receiving intensive care, this study aimed to identify and quantify the unique sources of BPA and phthalate exposure from medical devices in neonates receiving cardiac surgery. The specific objectives of this study were to determine a) whether specific medical treatment procedures, such as ECMO and CPB, significantly contribute to BPA and phthalates exposure; b) if the temporal patterns of BPA and phthalates levels correlate with the intensity and duration of medical device use; and c) if neonates undergoing cardiac surgery have significantly higher urinary levels of BPA and phthalates compared to the general neonatal population from the literature. To achieve these objectives, this study collected daily urine samples from birth to week 3 of life from 18 infants with congenital heart defects (CHD) undergoing cardiac surgery at two institutions. For each patient, the use of different medical devices and the characteristics of each device along with whether each device contains BPA, DEHP, or other phthalates (e.g., non-DEHP phthalates metabolites) were recorded. Machine learning clustering analysis techniques were used to cluster the time-course urine concentrations of BPA, DEHP metabolites, and non-DEHP phthalate metabolites, investigating the relationship between medical device use and BPA or phthalate exposure. Given the distinct chemical properties and potential differential health effects of DEHP and non-DEHP phthalates, it is crucial to differentiate between these phthalate metabolites when analyzing exposure patterns. DEHP is a commonly used phthalate in medical devices, while non-DEHP phthalates, although also present, have different sources and usage patterns. This distinction motivated our analytic approach to separately cluster DEHP and non-DEHP metabolites to more accurately identify and characterize the sources and dynamics of exposure. Additionally, critical features were used to cluster patients’ groups with similar toxicokinetic (TK) profiles, which describe the temporal changes in chemical concentrations within the body, to quantify variation in exposure to BPA, DEHP, and non-DEHP metabolites following the use of different medical devices.

Materials and Methods

Study Population

A prospective, observational cohort study was conducted to investigate the patterns and magnitude of BPA and phthalates exposure in neonates undergoing cardiac surgery. Infants were recruited from two institutions: the Children’s Hospital of Philadelphia (CHOP) and the Medical University of South Carolina (MUSC). A total of 22 infants were initially enrolled in the study (17 at CHOP and 5 at MUSC). All enrolled infants were $<21$-day-old neonates. However, four infants were excluded—two infants withdrew, and two had insufficient urine samples for analysis. Therefore, the final study cohort consisted of 18 infants (Figure S1). Inclusion criteria were specific and limited to infants with hypoplastic left heart syndrome (HLHS) or transposition of the great arteries (TGA) and an expected operation with CPB at 44 wk postconception or younger. No additional exclusion criteria were applied beyond withdrawal or sample insufficiency. The Children’s Hospital of Philadelphia institutional review board and the Medical University of South Carolina institutional review board approved the study. Written informed consent was obtained from the parent or guardian. Because pediatric data were analyzed at the University of Florida, the study was also approved by the University of Florida institutional review board.

Urine samples and electronic device clinical record forms (CRFs) were collected daily from birth up to 60 d of life or hospital discharge, whichever occurred first. For this study, we focused on the first 20 d of life, during which 327 urine samples (271 from CHOP and 56 from MUSC) were analyzed for BPA and phthalate metabolites (Figure S1). Patients were enrolled between 1 May 2020 and 31 December 2021. A total of 18 infants with congenital heart defects (CHD) undergoing cardiac surgery were included in the study. Operations were performed by teams of cardiac surgeons and anesthesiologists at each institution. Patient-related and perioperative variables were collected from medical records, including sex (male/female), race (Caucasian, African American, or other/not reported), ethnicity (Hispanic/Latino, non-Hispanic/non-Latino, or other/not reported), gestational age at birth (weeks), age at surgery (days), cardiac diagnosis [hypoplastic left heart syndrome (HLHS) or transposition of the great arteries (TGA)], initial surgical procedure (Norwood, hybrid, or arterial switch operation), and use of circulatory support [cardiopulmonary bypass (CPB) and/or extracorporeal membrane oxygenation (ECMO)]. Details on study follow-up, sample collection, and exclusions are provided in Figure S1. The characteristics of the study cohort are presented in Table 1.

Table 1.

Characteristics of infants ($n=18$) undergoing surgery for hypoplastic left heart syndrome or transposition of the great arteries at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC), 2020–2021.

	Total ($\mathit{n}=18$)	CHOP ($\mathit{n}=15$)	MUSC ($\mathit{n}=3$)
Sex
Female	3	2	1
Male	15	13	2
Race
Caucasian	16	13	3
African American	2	2	0
Other races/not reported	0	0	0
Ethnicity
Hispanic/Latino	3	3	0
Non-Hispanic/non-Latino	14	11	3
Other ethnicity/not reported	1	1	0
Age
Mean GA (weeks)	38.3	38.3	38.5
Mean age at surgery (days)	4	4	3
Diagnosis
HLHS	12	9	3
TGA	6	6	0
Initial procedure
Norwood procedure	9	8	1
Hybrid procedure	3	1	2
Arterial switch operation	6	6	0
Circulatory support
CPB	15	14	1
ECMO	4	4	0

Note: CPB, cardiopulmonary bypass; ECMO, extracorporeal membrane oxygenation; GA, gestational age; HLHS, hypoplastic left heart syndrome; TGA, transposition of the great arteries.

Urine Collection and Analysis

A daily urine sample was collected from each infant using an indwelling Foley catheter placed for clinical indications or from a cotton ball placed in a diaper. The cotton balls were placed into a $50\text{-mL}$ polypropylene syringe (Becton-Dickinson), and the urine was transferred into a polypropylene transport tube (Globe Scientific). The time of urine collection was noted in the study log. Note that exposure to CPB occurred during the cardiac operation and lasted a few hours. The first postoperative sample was usually obtained the day after surgery (12–24 h after exposure). The timing also depended on the urinary output of the child and was not standardized for this study. The urine was continuously drained into a collection bag (urine bag) attached to the catheter for samples collected using an indwelling Foley catheter. The nurse caring for the child emptied the collection bag at intervals (usually every 1–2 h). Study samples were obtained from this collection bag whenever a urinary catheter was used. The collected specimens were sealed inside a zip bag and placed in a refrigerator until picked up by study personnel and delivered to the Center for Human Phenomics Sciences Laboratory (CHOP) or the Pediatric Research Laboratory (MUSC) for processing and storage. Specific gravity (SG) was measured at room temperature using a handheld refractometer. Samples were allowed to come to room temperature before SG measurement. Urine samples were placed in $4\text{-mL}$ glass vials (Supelco), frozen at $–20°\mathrm{C}$, and shipped overnight on dry ice to SGS AXYS Analytical Services LTD for analysis. Sterile water control samples were collected in a similar fashion. Sterile water supplied by SGS AXYS Analytical Services LTD was placed in each individual collection device used for urine sampling (e.g., cotton balls and Foley catheter collection bags). The sterile water was allowed to dwell in the devices for $\sim 1$ to 4 h before being transferred to $4\mathrm{mL}$ glass vials (Supelco) and shipped overnight in coolers with ice packs. This process was implemented to assess potential contamination from the collection devices.

Urine Sample Preparation

Urine samples were thawed and gently mixed using a vortex mixer. A $1\text{-mL}$ portion of the urine sample was placed into a glass centrifuge tube and pH adjusted with $250\mathrm{\mu }\mathrm{L}$ of sodium acetate buffer ($1\mathrm{M}$, pH 6.5). The sample was spiked with isotopically labeled standards (see Table S1 and Table S2) and vortex mixed. A $50\text{-}\mathrm{\mu }\mathrm{L}$ volume of 4-methylumbelliferyl glucuronide solution ($1\text{\hspace{0.17em}ng}/\mathrm{\mu }\mathrm{L}$, 100% aq.) was added to each sample and quality control (QC) sample to monitor enzymatic deconjugation. The sample was then spiked with $20\mathrm{\mu }\mathrm{L}$ of $\mathrm{\beta }$-glucuronidase enzyme solution to each sample and incubated for 3 h at 37°C. The urine samples were then diluted to $5\mathrm{mL}$ using high-performance liquid chromatography (HPLC) water and pH adjusted to 2 using drops of concentrated HCl.

Urine Sample Extraction

Urine sample extraction was conducted using Phenomenex Strata-X, $200\mathrm{mg}/12\mathrm{mL}$ solid phase extraction (SPE) cartridges. The SPE cartridges were conditioned with $10\mathrm{mL}$ of acetonitrile:acetone 1:1, $5\mathrm{mL}$ of acetonitrile, $3\mathrm{mL}$ of HPLC water, and $3\mathrm{mL}$ of HPLC water at $\text{pH\hspace{0.17em}}2$. The deconjugated sample extract was loaded onto the SPE cartridges, and the cartridge was washed with $3\mathrm{mL}$ of 5% acetonitrile in HPLC water. The cartridges were dried under vacuum for 5 min, and the analytes were eluted with $2\times 5\mathrm{mL}$ of acetonitrile:acetone 1:1. The extracts were then reduced to dryness under a gentle stream of nitrogen and a water bath temperature of 40°C. The extracts were reconstituted with $450\mathrm{\mu }\mathrm{L}$ of 0.05% acetic acid in acetonitrile. The extracts were then spiked with $25\mathrm{\mu }\mathrm{L}$ of ¹³${\mathrm{C}}_4$-mono-(3-carboxypropyl) phthalate (¹³C₄-MnOP) recovery standard for the mono phthalates and $25\mathrm{\mu }\mathrm{L}$ of ²${\mathrm{H}}_6$-bisphenol A. This extract was split into two portions, $100\mathrm{\mu }\mathrm{L}$ for mono phthalates analysis and $400\mathrm{\mu }\mathrm{L}$ for bisphenol A analysis. The first portion was diluted with $100\mathrm{\mu }\mathrm{L}$ of HPLC water and analyzed by ultra-high-performance liquid chromatography–tandem mass spectrometry (UHPLC-MS/MS). The second portion was reduced to dryness and reconstituted with $100\mathrm{\mu }\mathrm{L}$ of methanol and analyzed by LC-MS/MS.

Analysis of Phthalate Ester Metabolites

Analysis for phthalate ester metabolites (PEMs) was performed by UHPLC-MS/MS using a Waters TQ-S micro tandem mass spectrometer with Waters ACQUITY UPLC I-class sample and binary solvent managers. The mass spectrometer was operated in negative ion electrospray multiple reaction monitoring (MRM) mode. A list of precursor/product ions used for quantification of PEMs is provided in Table S2. Analyte separation was achieved using a Waters HSS T3, $1.8\text{-}\mathrm{\mu }\mathrm{m}$, $50\text{-}\times 2.1\text{-mm}$ column. The mobile phase consisted of 0.1% acetic acid in UPLC water (solvent A) and 0.1% acetic acid in the acetonitrile organic phase (solvent B). The starting mobile phase composition was 5% B, which was linearly increased to 10%, 20%, 30%, 31.7%, 70%, and 100% at 0.5, 4, 5, 7, 12, and 12.2 min, respectively. The gradient was held at 100% B for 0.6 min and returned to initial conditions. The column was allowed to equilibrate for 1 min before the next injection. Quantification was achieved by isotope dilution using the average response factor. Due to rapid degradation of standard solutions for mono-(3-carboxypropyl) phthalate (MCPP) and mono-(2-ethyl-5-carboxypentyl) phthalate (MECPP), these analytes were quantified by bracketing calibration using a freshly prepared standard that was run at the beginning and end of the analysis run.

Analysis of BPA

Analysis for BPA was performed by high-pressure liquid chromatography–tandem mass spectrometry (HPLC-MS/MS) using a Waters 2690 or 2795 liquid chromatograph coupled to Waters Micromass Quattro Ultima Triple Quadrupole mass spectrometer. The mass spectrometer was operated in negative ion electrospray multiple reaction monitoring (MRM) mode. A list of precursor/product ions used for quantification of BPA is provided in Table S1. Analyte separation was achieved using a Waters Xterra C18MS, $3.5\text{-}\mathrm{\mu }\mathrm{m}$, $10.0\text{-}\times 2.1\text{-mm}$ column. The mobile phase consisted of HPLC water at $\text{pH\hspace{0.17em}}10$ (solvent A) and a 1:1 acetonitrile:methanol organic phase (solvent B). The starting mobile phase composition was 20% B, which was linearly increased to 50% and 100% at 2 and 7 min, respectively. The gradient was held at 100% B for 3 min and returned to initial conditions. The column was allowed to equilibrate for 3 min prior to the next injection. Quantification was achieved by isotope dilution using the average response factor. Further details of the analytical method are provided in the supplemental material (Tables S1 and S2). The limits of detection (LOD) of analyzed BPA and phthalate metabolites are provided in the supplemental material (Table S3). Specifically, the LOD values for each compound were as follows: BPA ($0.25\text{\hspace{0.17em}ng}/\mathrm{mL}$), MECPP ($0.5\text{\hspace{0.17em}ng}/\mathrm{mL}$), mono-(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP) ($0.75\text{\hspace{0.17em}ng}/\mathrm{mL}$), mono(2-ethyl-5-oxohexyl) phthalate (MEOHP) ($1\text{\hspace{0.17em}ng}/\mathrm{mL}$), mono(2-ethylhexyl) phthalate (MEHP) ($0.5\text{\hspace{0.17em}ng}/\mathrm{mL}$), monomethyl phthalate (MMP) ($1\text{\hspace{0.17em}ng}/\mathrm{mL}$), monoethylphthalate (MEP) ($1\text{\hspace{0.17em}ng}/\mathrm{mL}$), monobutyl phthalate (MBP) ($1.5\text{\hspace{0.17em}ng}/\mathrm{mL}$), monobenzyl phthalate (MBzP) ($0.5\text{\hspace{0.17em}ng}/\mathrm{mL}$), mono-cyclohexyl phthalate (MCHP) ($0.5\text{\hspace{0.17em}ng}/\mathrm{mL}$), MCPP ($0.5\text{\hspace{0.17em}ng}/\mathrm{mL}$), and monoisononyl phthalate (MINP) ($0.5\text{\hspace{0.17em}ng}/\mathrm{mL}$). Detailed information about the concentrations measured in these sterile water controls is provided in Tables S4–S9. Study data were collected and managed using REDCap electronic data capture system hosted at the Children’s Hospital of Philadelphia.^19,20 Because of the very small sample sizes available, there was not enough sample to perform duplicate analyses. The performance of the method was monitored by the sample surrogate recoveries, lab blanks, and the on-going precision and recovery (OPR) samples within each batch of study samples. The batch-to-batch OPR sample results provided a measure of reproducibility for each analyte. The spiked OPR reference material is prepared by fortifying a synthetic urine sample with known amounts of the target analytes. These OPR samples were processed alongside the study samples and used to demonstrate the ongoing analyte reproducibility.

Contamination Control Assessment

To assess potential contamination from urine collection materials, sterile water was placed in cotton balls, syringes, cylinders, pipets, Cryovials, and urine bags and allowed to dwell for $\sim 1$ to 4 h before being transferred to $4\text{-mL}$ glass vials (Supelco). These control samples were analyzed to determine whether any BPA or phthalate metabolites leached from the collection materials. The results of this contamination assessment are detailed in the supplemental material (Tables S4–S9).

Assessment of Exposure to BPA, DEHP, and non-DEHP Metabolites

Laboratory analysis for phthalate ester metabolites (PEMs) and bisphenol A (BPA) was conducted at SGS AXYS using UHPLC-MS/MS according to the laboratory’s protocol, SGS AXYS method MLA-059 “Analytical procedure for the analysis of bisphenol A and phthalate ester metabolites in urine by UHPLC-MS/MS,” which is based on a method developed by the US Centers for Disease Control and Prevention.²¹ The method used for this study was validated using an EPA tier I level validation.²² The analytical method is accredited by the Canadian Association for Laboratory Accreditation Inc. (CALA). The method was also evaluated for performance through regular involvement in performance testing (PT). In a recent PT organized by the German External Quality Assessment Scheme, ILC, GEQUAS-67, 2021, the results generated by this method were within the acceptance criteria of the study for all of the analytes tested. In this study, the measurements of BPA refer to total BPA, which includes both BPA itself and its metabolites.

Before formal analysis, if the measured concentration for each analyte was between 0 and its LOD and if the machine read value was available, we used the machine measured concentration but indicated it with a flag that the value was below the LOD. This approach allows us to retain quantitative information from low-level detections, even if they are below the formal LOD. However, if a measured concentration was reported as 0.0 (indicating nondetectable), then the value was imputed as LOD/sqrt(2).²³ This imputation method provides a reasonable estimate for nondetected values, minimizing bias and maintaining statistical integrity. In addition, urinary biomarker concentrations were adjusted for dilution using specific gravity (SG). We computed SG-corrected biomarker concentrations using the following formula: $Pc=P\left[\left(x-1.000\right)/\text{SG}-1.000\right]$, where Pc is the SG-corrected biomarker concentration (mg/L), P is the measured biomarker concentration (mg/L), x is the SG value for a given participant (equivalent to the term “subject” or “patient” in this study), and SG is the median SG value for the specific study population.⁷

Quantity of BPA- and Phthalate-Containing Medical Device Use

Daily assessment of medical device exposure was performed by documenting the numbers and types of medical devices used in the critical care units (e.g., endotracheal tube, ECMO circuit components, IV tubing, stopcocks, feeding tubes, etc.). DEHP and BPA disclosure information was obtained in two ways. First, we inspected each device’s packaging for disclosure of BPA or phthalates. If this information was not disclosed on the packaging, we contacted the manufacturer of each device used at both CHOP and MUSC by phone, email, or through the company inquiry portal to determine whether or not the devices contained BPA, DEHP, and/or any other phthalate. There are no standardized methods for quantifying the intensity of medical care. We classified intensity of medical care based on clinical judgement and the complexity of care. Typically, a child with only a peripheral IV and/or an EKG lead is not receiving critical care services and is stable. Conversely, a child requiring ECMO support is critically ill and unstable, requiring constant care and intervention. In addition, we performed a count of the devices used each day. For the device count, we did not separate into BPA, DEHP, and non-DEHP devices, as there was overlap and we were not confident in the available information concerning the composition of each device. The device intensity score was calculated using a nominal scale from 0 to 5, based on the number and type of devices used daily. The levels were defined as follows:

• •0: No device
• •1: Electrocardiogram (EKG) leads and/or peripheral IV present
• •2: Noninvasive ventilation and/or feeding tube present, plus any device in level 1
• •3: Central line and/or chest tube present, plus any device in lower levels
• •4: Endotracheal tube and/or blood transfusion, plus devices in lower levels
• •5: CPB and/or ECMO plus devices from levels below

The same system was used to calculate the device intensity score for both BPA- and DEHP-containing devices. Devices were categorized based on their potential to leach BPA or DEHP, and the intensity scores were assigned accordingly (non-DEHP chemicals were not specifically in this calculation). For instance, devices such as CPB and ECMO, which involve significant blood contact and longer duration of use, were assigned to the highest intensity level due to their higher potential for exposure. In cases where the urine sample was collected early in the day before the use of CPB or ECMO, the measured analyte levels would not reflect exposure to these devices. Thus, those days were designated as non-CPB or non-ECMO days. These criteria ensured a comprehensive and systematic assessment of medical device utilization, facilitating a more accurate analysis of the relationship between device use and exposure levels of BPA and DEHP. The intensity of care was calculated daily, allowing for variations in the level of care. As such, a patient could be classified as low intensity 1 day and high intensity the next, depending on the types of devices being used.

Medical Device Usage

In this study, the devices used for each infant were determined based on the discretion of the clinical team caring for each infant at the two institutions, not by the study protocol. The specific devices and timing of use were chosen based on the clinical condition and the unique practices of each institution. The infants in our study were all neonates undergoing care for congenital heart disease and might be critically ill. They were typically admitted shortly after birth and generally underwent cardiac surgery within the first few days of life. This period was associated with a significant increase in the number of devices used. As the infants recovered, the devices were removed when they were no longer needed for clinical care. Conversely, if a child’s condition became worse, there would likely be an increased number of devices used. In general, the number of devices decreases as the infants improve and are prepared for discharge from the hospital. The use of specific devices, timing of use, and timing of removal were clinical decisions made by the team caring for the child and were not determined by the study protocol. This dynamic use of devices could explain the reduction in exposure over time and help reconcile the clusters identified in the study. Detailed information on the baseline devices and usage timeline is provided in the section below.

Baseline Devices and Usage Timeline

The clinical team determined the devices used for each infant based on the infant’s medical condition and institutional practices, not by the study protocol. As a result, there was no standardized set of baseline devices. However, common devices used included the following:

1. Intravenous (IV) lines: Used for medication and fluid administration, typically inserted at birth, and removed when the infant is stable and no longer requires intravenous therapy.

2. Endotracheal tubes: Used for mechanical ventilation, inserted at birth for respiratory distress or for surgery, and removed once the infant can breathe independently.

3. Catheters (e.g., Foley): Used for urine collection and measurement, typically inserted intraoperatively, and removed as soon as the infant can urinate independently.

4. Feeding tubes: Used for nutrition delivery, typically inserted after surgery for infants unable to feed orally, and removed once the infant can feed on its own.

5. CPB and ECMO circuits: Used during cardiac surgery and critical care for circulatory and respiratory support, inserted during surgery, and removed postoperatively once the infant is stable. The ECMO circuits utilize different components at each institution. For this analysis, all ECMO cases were from CHOP and CHOP ECMO circuit tubing is heparin coated. An intraoperative cell saver was used in 14/15 CPB cases. There were three non-CPB cases and a cell-saver was not used. Therefore, the use of CPB and cell saver are highly correlated. The overall results reflect the use of a cell-saver. The magnitude of exposure from the cell-saver cannot be determined for our data.

The timeline for device use varied dynamically with the infant’s clinical status. Specifically, devices were gradually removed as the infant recovered, reducing exposure levels, whereas additional devices were introduced if the infant’s condition worsened, potentially increasing exposure. To accurately assess the relationship between medical device use and urinary levels of BPA and phthalates, we compared the urinary biomarker levels with the corresponding day’s medical device utilization. Specifically, the urinary concentrations measured on a given day were directly compared to the number and type of medical devices used on the same day. This approach ensured that the timing of exposure assessment aligned precisely with the use of medical devices, providing a more accurate understanding of the exposure dynamics in neonates undergoing cardiac surgery.

Statistical Analysis

The geometric mean and standard deviation (SD), median, and interquartile ranges were used to describe the distributions of urinary concentrations of BPA, DEHP, and non-DEHP phthalate metabolites. To explore the bivariate associations between medical device utilization and exposure levels of BPA, DEHP, and non-DEHP phthalate metabolites, we calculated the geometric mean of the urinary concentrations across all time points for each individual. We also conducted a bivariate association analysis by categorizing daily medical device usage into three groups based on the distribution of device intensity scores across all measured days: low ($\le 25$th percentile), medium (26th–50th percentile), and high ($>50$th percentile). While based on quartile thresholds, the upper two quartiles were combined into a single “high” category for analytical simplicity due to sample size considerations. This classification facilitated comparison across varying levels of device usage. A Kruskal–Wallis nonparametric test was used to estimate the bivariate association of urinary BPA and phthalate metabolites’ concentrations with BPA- and DEHP-containing product use groups, ECMO and CPB treatments, as well as differences between institutions.

Additionally, a linear mixed-effects (LME) regression model using the lmer function implemented in R was used to estimate the associations between medical device utilization and urinary concentrations of BPA, DEHP, and non-DEHP phthalate metabolites. In the LME analysis, the dependent variables were the urinary concentration of BPA, DEHP, and non-DEHP phthalate metabolites, while the independent variables included the utilization of medical devices (e.g., Foley catheter, device intensity score) and specific treatments (e.g., ECMO and CPB). The model included both fixed and random effects. Additionally, three separate linear mixed-effects (LME) models were used to examine the associations between medical device utilization and urinary concentrations of BPA, DEHP, and non-DEHP phthalate metabolites. In each model, the dependent variable was the urinary concentration of BPA, DEHP, or non-DEHP phthalate metabolites, while the independent variables included the utilization of medical devices (e.g., Foley catheter, device intensity score) and specific treatments (e.g., ECMO and CPB). The model included both fixed and random effects. Fixed effects in the model included the device intensity score, which quantifies medical device utilization on a nominal scale from 0 to 5 (as detailed in “Quantity of BPA- and Phthalate-Containing Medical Device Use”), and the type of medical treatment (e.g., ECMO, CPB). Random effects accounted for the variability between individuals, allowing us to control within-infant correlation over time. The LME model can be simplified as follows:

$$C_{ij}={\mathrm{\beta }}_0+{\mathrm{\beta }}_1{\text{DIS}}_{ij}+{\mathrm{\beta }}_2{\text{Treatment}}_{ij}+u_{0i}+{ϵ}_{ij},$$

where $C_{ij}$ is the urinary BPA/DEHP/non-DEHP metabolite concentration for an individual i at time j, $\mathrm{\beta }$ is the fixed effects coefficient, DIS is the device intensity score, Treatment is the type of medical treatment, $u_{0i}$ is the random intercept for an individual i, and ${ϵ}_{ij}$ is the residual error for an individual i at time j. We accounted for the correlation of repeated urinary measurement within participants using random intercept for each infant. To account for the potential lag effects of medical device utilization and treatment, lagged versions of the predictor variables were included in the model. All statistical analyses were performed in R software (version 4.2.3; R Development Core Team), with statistical significance set at $p<0.05$. $p$-Values are reported as exact values unless $<0.001$, in which case they are reported as “$<0.001$.”

Time-Series Hierarchical Clustering

Time-series clustering, an unsupervised machine learning technique, can organize time-series data into groups based on their similarity. This approach was chosen because it allows us to identify distinct patterns in the TK profiles of urinary BPA and DEHP metabolites, which are essential for understanding the exposure dynamics in neonates undergoing cardiac surgery. By grouping similar exposure profiles, we can better characterize the variability in exposure levels and their potential sources. In this study, similarity in the time-series data was characterized primarily by the trajectory and overall shape of the TK profiles, including features such as peaks, valleys, and trends over time. The agglomerative hierarchical clustering method was used to cluster the time-series TK profiles for urinary BPA concentration, molar sum of DEHP metabolites ($\sum \text{DEHP}$), and non-DEHP phthalate concentrations [i.e., monobutyl phthalate (MBP), monobenzyl phthalate (MBzP), monomethyl phthalate (MMP), and monoethylphthalate (MEP)]. The $\sum \text{DEHP}$ was calculated by dividing each DEHP metabolite concentration [i.e., mono(2-ethylhexyl) phthalate (MEHP), mono-(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), mono(2-ethyl-5-oxohexyl) phthalate (MEOHP), and mono-(2-ethyl-5-carboxypentyl) phthalate (MECPP)] by its respective molecular weight and then summing the molar concentrations of metabolites. To avoid any significant differences in the urinary non-DEHP phthalate metabolite concentrations between institutions, the clustering analysis for urinary non-DEHP phthalate metabolites only considered the TK profiles from CHOP. This decision was based on an initial assessment, which indicated higher variability in non-DEHP phthalate metabolite concentrations between institutions compared to DEHP metabolites. A detailed comparison of institutional differences is presented in “Results.” Additionally, to avoid any potential release of MEHP from urine bags, we also measured and compared the clustering results based on the TK profiles with and without MEHP (Figures S2 and S3). A complete linkage method²⁴ was used to determine the intercluster distances through the farthest distance data pairs. Agglomerative hierarchical is a clustering algorithm capable of merging the objects from any cluster to a single cluster from the bottom to the top of the dendrogram tree.²⁵ The dendrogram graphically presents the informative clustering tree that includes the merging process as well as the intermediate clusters.

Implementation with Cluster Analysis and Parameters

The time-series clustering analysis was implemented using the “dtwclust” package in R,²⁶ which employs dynamic time warping (DTW) to align time-series data. DTW is particularly useful for clustering patterns that exhibit similar trends but are not exactly aligned in real-time. This method stretches and compresses segments of the time series to achieve optimal alignment, allowing for more accurate clustering based on the overall shape and trajectory of the profiles rather than their exact temporal alignment. For this analysis, we used a window size of 10% of the series length to control the extent of warping allowed, and the symmetric one-step pattern was chosen to ensure symmetric treatment of the time series, allowing for both insertions and deletions. The time-series data were z-normalized to ensure that the clustering focused on the shape of the series rather than absolute values, and the DTW distance metric was used to compute the pairwise distances between the time-series profiles.

Inclusion of Medical Device Data and Device Counting

The clustering analysis was based on the TK profiles of urinary BPA, DEHP, and non-DEHP phthalate metabolites. While medical device use and treatments such as ECMO and CPB were not directly included in the clustering analysis, they were considered in the subsequent analysis of the clusters. Due to the uncertainty regarding the exact composition of many medical devices, we included all devices in our calculations, regardless of their presumed contribution to BPA or DEHP exposure. This comprehensive approach allowed us to compare the number of devices used with the TK profiles of each cluster. Additionally, the devices were counted daily. For instance, if a patient had an endotracheal tube (ETT) removed and then replaced on the same day, it would have been counted as a single device for that day. This approach was used to avoid overestimating the number of devices used due to multiple replacements or changes within a single day. Our focus was on capturing the presence and usage of medical devices on a daily timeline to better understand the exposure patterns throughout the patient’s treatment period.

Cross-Correlation Analysis

A cross-correlation analysis was conducted to investigate the temporal relationship between urinary concentrations of BPA and DEHP metabolites and the number of medical devices used. The analysis was performed using the TSA package in R, examining correlations at various lags ranging from $-5$ to $+5$ days. The cross-correlation function (CCF) was calculated using a correlation coefficient, such as Pearson’s correlation coefficient or Spearman’s rank correlation coefficient. This correlation coefficients at each lag provides insights into the potential delays in exposure and subsequent metabolite detection in urine samples. A value of 1 indicates a perfect positive linear relationship, $-1$ indicates a perfect negative linear relationship, and 0 indicates no linear relationship.

Results

Distribution of BPA and Phthalate Metabolites

Among the urine samples collected from these infants, only $<10\%$ of detectable metabolite concentrations of DEHP [MEHP, mono(2-ethyl-5-hydroxyhexyl) phthalate (MEEHP), MEOHP, MECPP], dibutyl phthalate (DBP) (MBP), and benzyl butyl phthalate (BzBP) (MBzP) were less than LOD, compared with 17%, 19%, 20%, and 66% of concentrations of MEP, BPA, MCPP [mono-(3-carboxypropyl) phthalate], and MMP less than the LOD (Table 2). Control sample analysis indicated potential contamination of BPA and MEHP from urine bags, while no detectable levels were found in other collection materials, including cotton balls, syringes, cylinders, pipets, and Cryovials (Tables S4–S9). The 75th percentiles of concentrations of BPA, MEHP, MEHHP, MEOHP, MECP, MBP, MBzP, MMP, MCPP, and MEP were 1.6, 9.5, 146, 119, 818, 48, 12, 4.6, 14, and $5.4\text{\hspace{0.17em}ng}/\mathrm{mL}$, respectively. Overall, the geometric mean concentrations of urinary DEHP metabolites MEHP, MEHHP, and MEOHP were 4.0, 59, and $501\text{\hspace{0.17em}ng}/\mathrm{mL}$, respectively. Median concentrations of urinary DBP and BzBP metabolites were $8.1\text{\hspace{0.17em}ng}/\mathrm{mL}$ for MBP and $3.9\text{\hspace{0.17em}ng}/\mathrm{mL}$ for MBzP. Median BPA concentration was also reported. Median values for DEHP metabolites (MEHP, MEHHP, MEOHP, and MECPP) among infants undergoing cardiac surgery in the NICU were 3.7, 55, 49, and $302\text{\hspace{0.17em}ng}/\mathrm{mL}$, respectively (Table 2).

Table 2.

Distribution of bisphenol A (BPA) and phthalate metabolite concentrations in urine samples ($n=356$) from infants ($n=18$) undergoing surgery for hypoplastic left heart syndrome or transposition of the great arteries at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC), 2020–2021.

	No.	BPA	DEHP				DBP	BzBP	DMP	DBP	DEP
			MEHP	MEHHP	MEOHP	MECPP	MBPa	MBzP	MMP	MCPP	MEP
Distribution of BPA/phthalate measurements	356
No. < LOD ($n$)		19	7	0	3	0	4	9	66	20	17
Geometric mean		1.1	4.0	59	501	292	17	5.5	2.4	3.8	2.9
Geometric SD		4.1	3.9	4.5	4.7	4.9	7.8	5.6	2.3	6.0	2.8
25th percentile		0.4	1.5	23	19	127	3.5	1.8	1.3	0.8	1.4
Median		0.7	3.7	55	49	302	8.1	3.9	2.3	3.4	2.5
75th percentile		1.6	9.5	146	119	818	48	12	4.6	14	5.4
Median urinary phthalate concentrations from other studies
United States
Children, 6–11 years of age (CDC, 2015)27,b	389	—	2.5	2.4	16	—	27	15	—	6	49
Infants, 1–8 months of age28	19	—	7.1	11.8	11.9	49	—	27	6	2.1	21
Infants, 28 weeks of age15	132	—	3.2	1.2	0.8	9.2	—	1.9	—	—	5.0
Canadian
Children, 6–11 years of age29,c	1,038	—	1.3	3.3	19	—	33	21	—	2.7	26
Infants, 3 months of age30	55	0.27	—	0.8	0.8	—	4.9	2.5	—	0.9	5.9
China
Infants, 128 days of age31	748	—	0.2	—	0.5	—	18	—	—	—	3.3
NHANES (1999–2000), persons $\ge 6$ years of age	2,782	1.2	4.1	20	14	—	26	16	—	—	—

Note: —, not available; BPA, bisphenol A; BzBP, benzyl butyl phthalate; CDC, US Centers for Disease Control and Prevention; DBP, dibutyl phthalate; DEP, diethyl phthalate; DEHP, di(2-ethylhexyl) phthalate; DMP, dimethyl phthalate; LOD, limit of detection; MBP, mono-n-butyl phthalate; MBzP, mono-benzyl phthalate; MCPP, mono(3-carboxypropyl) phthalate; MECPP, mono(2-ethyl-5-carboxypentyl) phthalate; MEHP, mono(2-ethylhexyl) phthalate; MEHHP, mono(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono(2-ethyl-5-oxohexyl) phthalate; MEP, monoethyl phthalate; MMP, monomethyl phthalate; NHANES, US National Health and Nutrition Examination Survey; SD, standard deviation.

^a MBP is also a metabolite of BzBP.

^b From NHANES, 2013–2014. The concentration is presented as geometric mean.

^c From The Canadian Health Measures Survey 2007–2009 with the concentration presented as geometric mean. The unit of all concentrations is ng/mL.

Bivariate Associations between Urinary Concentrations of Phthalates or BPA and Medical Device Exposure Groups

With the increasing use of medical devices, the distributions for higher-intensity groups of BPA- and DEHP-containing devices tend to exhibit longer tails, indicating a higher likelihood of elevated concentrations of urinary levels of BPA, MEHHP, MEOHP, and MECPP (Table 3). Specifically, while the median concentration of urinary BPA was the lowest for the highest-intensity group, the highest 75th percentile urinary BPA concentration was observed in the high-intensiveness group ($5.96\text{\hspace{0.17em}ng}/\mathrm{mL}$) compared with infants in the low- and medium-intensiveness groups (1.07 and $1.33\text{\hspace{0.17em}ng}/\mathrm{mL}$). Median urinary MEHHP concentrations (25th, 75th percentiles) among infants in the low-, medium-, and high-intensiveness of DEHP-containing medical device use groups were 54.4 (20.9, 121), 35.6 (13, 52.1), and 82.3 (32.4, 244) ng/mL, respectively. Similarly, median urinary MEOHP concentrations among infants in these groups were 41.8 (19.9, 90.6), 30.7 (11.6, 53.4), and 71 (30.3, 193) ng/mL, respectively, and median urinary MECPP concentrations among infants in these groups were 300 (140, 581), 172 (89.2, 376), and 394 (155, 1,260) ng/mL, respectively. In contrast, the pattern of associations between the intensiveness of medical device use and non-DEHP phthalates (e.g., MBP, MBzP, MMP, and MEP) did not follow a consistent trend (Table 4). For instance, the medians for the high-intensity group for all non-DEHP metabolites were generally higher than those for the medium-intensity group, and MMP showed the highest median concentration in the high-intensity BPA exposure group. This indicates that the relationship between the intensity of medical device use and non-DEHP phthalate exposure is more variable and does not exhibit a straightforward inverse pattern. In contrast, a reverse pattern of associations was observed between the intensiveness of medical device use and urinary concentration of MBP, MBzP, and MEP (Table 4). A summary table showing the number of uses of BPA- or DEHP-containing medical devices between institutions is provided in Table S10. We observed higher concentrations of MBP [median = 44.3 (25th, 75th: 10.5, 474) ng/mL], MBzP [median = 4.94 (25th, 75th: 2.59, 16.6) ng/mL], and MEP [median: 3.8 (25th, 75th: 2.1, 6.6) ng/mL] in the lower-intensiveness group than the median- and high-intensiveness DEHP groups. However, the pattern of association between intensiveness of medical device use and MMP did not follow a consistent trend (Table 4). The median concentration of MMP was the highest in the high-intensity BPA exposure group, indicating variability in the relationship between device use and exposure levels.

Table 3.

Bivariate associations between urinary DEHP phthalate metabolites and BPA/DEHP exposure group ($n=356$ samples) in 18 infants undergoing surgery at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC), 2020–2021.

	Urinary BPA				Urinary MEHP				Urinary MEHHP				Urinary MEOHP				Urinary MECPP
	25th	Median	75th	$p$-Valuea	25th	Median	75th	$p$-Value	25th	Median	75th	$p$-Value	25th	Median	75th	$p$-Value	25th	Median	75th	$p$-Value
Intensiveness of BPA-containing medical devices use (nb)
Low (77)	0.54	0.73	1.07		3.17	6.04	14.8		19.9	57.3	144		19.8	46.2	109		169	357	750
Medium (100)	0.49	0.81	1.33		1.31	2.58	4.24		19.3	40.9	78.5		14.9	37.9	70.2		90.3	185	370
High (179)	0.40	0.66	5.96		1.09	3.39	9.6		29.4	66.5	233		25.4	61.6	176		127	337	1,207
				0.01				0.02				0.06				0.06				0.03
Intensiveness of DEHP-containing medical devices use ($n$)
Low (77)	0.55	0.75	1.18		2.14	3.63	8.87		20.9	54.4	121		19.9	41.8	90.6		140	300	581
Medium (100)	0.39	0.62	0.95		1.39	3.39	5.46		13.0	35.6	52.1		11.6	30.7	53.4		89.2	172	376
High (179)	0.44	0.76	6.61		1.25	4.02	13.3		32.4	82.3	244		30.3	71.0	193		155	394	1,260
				0.01				0.16				0.01				$<0.01$				$<0.01$
ECMO used
Yes (79)	11.4	14.8	29.5		6.49	20.0	27.5		335	533	795		327	456	653		1,715	2,695	4,219
No (277)	0.42	0.66	1.18		1.41	3.39	7.86		21.3	49.1	101		18.6	41.6	90.6		113	252	571
				$<0.001$				$<0.001$				$<0.001$				$<0.001$				$<0.001$
CPB used
Yes (293)	6.15	8.29	13.24		9.47	17.61	48.31		502	956	1,244		357	920	1,143		432	1,864	2,446
No (63)	0.43	0.68	1.35		1.46	3.49	8.55		22	53	125		20	46	103		123	291	746
				$<0.001$				$<0.001$				$<0.001$				$<0.001$				$<0.001$
Institutions (nc)
CHOP (15)	0.44	0.73	1.81		1.37	3.49	9.03		23.4	52.5	142		19.2	48.5	122		107	250	836
MUSC (3)	0.51	0.72	1.08		2.45	4.69	12.11		31.1	69.9	159		29.4	53.2	113		257	402	767
				0.69				0.06				0.26				0.39				0.01

Note: Values are presented as percentiles (25th, 50th, and 75th). BPA, bisphenol A; CPB, cardiopulmonary bypass; DEHP, di(2-ethylhexyl) phthalate; ECMO, extracorporeal membrane oxygenation; MECPP, mono(2-ethyl-5-carboxypentyl) phthalate; MEHP, mono(2-ethylhexyl) phthalate; MEHHP, mono(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono(2-ethyl-5-oxohexyl) phthalate.

^a $p$-Values are derived from the Kruskal–Wallis nonparametric test for differences in BPA/DEHP metabolites distribution across exposure groups. $p$-Values $<0.001$ are reported as “$<0.001$.” The unit of all concentrations is ng/mL.

^b Exposure intensity groups were derived from the distribution of medical device use. Specifically, participants were grouped as follows: low ($\le 25$th percentile), medium (26th–50th percentile), and high ($>50$th percentile), with the upper two quartiles (51st–100th percentiles) combined into a single “high” category due to sample size considerations. Cutoffs for intensity score were low at $\le 1$, $\text{medium}=2$, and high at $\ge 3$ based on the distribution of daily device use scores (0–5 scale).

^c Under “Institutions,” the reported numbers refer to the number of participants (infants) rather than the number of collected samples.

Table 4.

Bivariate associations between urinary non-DEHP phthalate metabolites and BPA/DEHP exposure group in urine samples ($n=18$) from 18 infants undergoing surgery at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC), 2020–2021.

	Urinary MBP				Urinary MBzP				Urinary MMP				Urinary MEP
	25th	Median	75th	$p$-Valuea	25th	Median	75th	$p$-Value	25th	Median	75th	$p$-Value	25th	Median	75th	$p$-Value
Intensiveness of BPA-containing medical devices use (nb)
Low (77)	10.5	194	855		3.19	8.9	33.9		1.4	2.2	4.1		2.2	4.0	6.7
Medium (100)	4.71	9.29	19.3		2.31	4.61	26.8		1.2	1.8	2.8		1.4	2.2	4.4
High (179)	2.65	4.50	13.1		1.47	2.63	5.71		1.2	2.8	5.0		1.2	2.0	5.3
				$<0.01$				$<0.001$				0.62				0.2
Intensiveness of DEHP-containing medical devices use ($n$)
Low (77)	10.5	44.3	474		2.59	4.94	16.6		1.8	2.5	4.4		2.1	3.8	6.6
Medium (100)	2.82	5.17	9.00		1.51	2.66	12.91		1.2	1.6	3.7		1.2	2.0	2.6
High (179)	2.98	5.78	33.6		1.62	3.36	7.9		1.3	2.1	4.9		1.3	2.6	6.0
				$<0.001$				0.002				0.402				0.006
EMCO used
Yes (79)	32.03	47.88	105.6		2.68	5.38	11.56		1.92	3.26	5.15		5.78	9.13	24.48
No (277)	3.29	6.92	27.20		1.64	3.82	12.11		1.27	2.09	4.10		1.36	2.24	4.54
				$<0.001$				0.132				0.149				$<0.001$
CPB used
Yes (293)	12.97	29.01	61.84		15.09	25.02	43.40		0.60	0.92	1.20		1.62	3.31	5.26
No (63)	3.41	7.51	46.11		1.70	3.74	10.57		1.47	2.61	4.88		1.42	2.46	5.35
				$<0.001$				$<0.001$				$<0.001$				0.81
Institutions (nc)
CHOP (15)	3.12	6.03	15.9		1.54	3.04	6.80		1.27	2.74	5.17		1.29	2.19	4.59
MUSC (3)	325	698	1,689		8.87	25.5	155		1.40	2.03	2.56		2.53	5.42	9.95
				$<0.001$				$<0.001$				0.09				$<0.001$

Note: 20th, 50th, and 75th are percentiles. BPA, bisphenol A; CPB, cardiopulmonary bypass; DEHP, di(2-ethylhexyl) phthalate; ECMO, extracorporeal membrane oxygenation; MBP, mono-n-butyl phthalate; MBzP, monobenzyl phthalate; MEP, monoethyl phthalate; MMP, monomethyl phthalate.

^a From the Kruskal–Wallis nonparametric test for differences in phthalate distribution. The unit of all concentrations is ng/mL.

^b The units $n$ shown in parentheses represent the number of urine samples per group; the groups were divided based on quartiles of the intensity of medical device use (i.e., low: $\le 25\%$; medium: 25–50%; high: $>50\%$).

^c The counts of measurements associated with each institution represent the number of participants rather than the total number of collected samples.

Similar to the pattern of associations between the intensiveness of medical device use and urinary phthalates, urinary DEHP metabolite concentrations [e.g., MEHP, MEHHP, Mono(2-ethyl-5-hydroxyhexyl) octyl phthalate (MEHOP), and MECPP] were significantly higher among infants undergoing ECMO. Specifically, the median values with ECMO vs. without ECMO were as follows: MEHP: $20\text{\hspace{0.17em}ng}/\mathrm{mL}$ vs. $3.39\text{\hspace{0.17em}ng}/\mathrm{mL}$; MEHHP: $533\text{\hspace{0.17em}ng}/\mathrm{mL}$ vs. $49.1\text{\hspace{0.17em}ng}/\mathrm{mL}$; and MEHOP: $456\text{\hspace{0.17em}ng}/\mathrm{mL}$ vs. $41.6\text{\hspace{0.17em}ng}/\mathrm{mL}$. In contrast, a reverse pattern of associations was observed between the intensiveness of medical device use and urinary concentrations of some non-DEHP phthalates (e.g., MBP and MEP), which were significantly higher in the lower-intensiveness group. The concentrations of MBzP and MMP were not significantly different between groups with and without ECMO, and urinary MEP concentrations were not significantly different between groups with and without CPB (Table 4). Overall, six of the eight measured metabolites (MEHP, MEHHP, MEHOP, MECPP, MBP, and MEP) were higher in the ECMO group, indicating a substantial exposure associated with ECMO treatment.

The bivariate associations were used to estimate the differences in urinary BPA, DEHP, and other non-DEHP concentrations between institutions (Tables 3 and 4). Results presented here indicate that the concentrations of BPA and most of the DEHP metabolites such as MEHP, MEHHP, and MEOHP in infants’ urine were similar between the two institutions, but urinary MECPP values were significantly higher among infants at MUSC ($p=0.01$) (Table 4). Urinary non-DEHP phthalate metabolite, including MBP, MBzP, and MEP (Table 5) ($p<0.001$), concentrations were also significantly higher among infants at MUSC.

Table 5.

Mixed-effect associations between medical treatment and devices with urinary BPA and $\sum \text{DEHP}$ concentrations in urine samples ($n=18$) from 18 infants undergoing surgery at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC), 2020–2021.

	Urinary BPA concentrations (ng/mL)			Urinary $\sum \text{DEHP}$ concentrations (mM)
	Estimatesa	95% CI	$p$-Valueb	Estimates	95% CI	$p$-Value
Fixed effectsc
Intercept	4.7	$-7.99$–17.5	0.464	0.36	$-4.02$–4.74	0.871
CPB	2.66	$-7.69$–13.01	0.614	$-0.35$	$-4.48$–3.78	0.867
ECMO	51.5	41.93–60.97	$<0.001$	13.4	9.88–16.9	$<0.001$
Foley Catheter	6.50	1.85–11.16	$<0.001$	6.15	4.26–8.04	$<0.001$
Device intensity score	$-1.81$	$-5.42$–1.80	0.324	0.138	$-1.18$–1.46	0.837
Random effectsd
Variance	101			3,686
Residual	253			45,177
Marginal $R^2$e	0.393			0.367
Conditional $R^2$e	0.566			0.414

Note: The unit for the sum of DEHP metabolites’ concentrations is mM because we had to convert the unit from ng/mL to mM in order to calculate the sum of DEHP metabolites’ concentrations. On the other hand, the unit for BPA remains to be ng/mL. BPA, bisphenol A; CI, confidence interval; CPB cardiopulmonary bypass; ECMO, extracorporeal membrane oxygenation; $\sum \text{DEHP}$, sum of di(2-ethylhexyl) phthalate metabolites, including MEHP, MEHHP, MEOHP, and MECPP; MECPP, mono(2-ethyl-5-carboxypentyl) phthalate; MEHP, mono(2-ethylhexyl) phthalate; MEHHP, mono(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono(2-ethyl-5-oxohexyl) phthalate.

^a Estimates represent the regression coefficients from the mixed-effects model. Bold text indicates statistically significant regression coefficients ($p<0.05$).

^b $p$-Values are derived from the mixed-effects model and represent the statistical significance of each fixed-effect predictor. $p$-Values $<0.001$ are reported as “$<0.001$.”

^c Fixed effects represent the estimated associations between medical treatment, medical devices, and urinary BPA/$\sum \text{DEHP}$ concentrations.

^d Random effects account for interindividual variability.

^e Marginal $R^2$ represents the variance explained by fixed effects, while conditional $R^2$ represents the variance explained by both fixed and random effects.

Mixed Effects Associations between Urinary Concentrations of Phthalates or BPA and Medical Device Exposure Groups

Table 5 shows the results of mixed effects associations between medical treatments or devices and urinary BPA and $\sum \text{DEHP}$ values (i.e., the molar sum of MEHP, MEHHP, MEOHP, and MECCP). Results indicate that urinary BPA concentrations (ng/mL) increased by 51.5 [95% confidence interval (CI): 41.9, 60.9] and 6.5 (95% CI: 1.85, 11.2) ng/mL, respectively, with each one-unit increase in ECMO treatment and the use of Foley catheter. $\sum \text{DEHP}$ concentrations were correspondingly increased by 13.4 (95% CI: 9.88, 16.9) and 6.15 (95% CI: 4.27, 8.04) mM, respectively, with each one-unit increase in ECMO treatment and the use of Foley catheter. The random effects were significant in both BPA and $\sum \text{DEHP}$, as the high variance among participants was observed (Table 5). There were no statistically significant associations between non-DEHP phthalate metabolites and ECMO treatment except for MMP [5.69 (95% CI: 2.70, 8.68)] and MEP [22.28 (95% CI: 17.65, 26.91)]; however, there was a statistically significant within-participants effect (Tables S11 and S12). Note that the raw data for Tables S11–S14 are provided in Excel Tables S5–S8, respectively.

Furthermore, the LME results incorporating lag effects (1-day delay) for urinary BPA and $\sum \text{DEHP}$ values indicated significant associations for ECMO and Foley catheters (Table S13). For DEHP, each one-unit increase in ECMO treatment was associated with a significant increase in $\sum \text{DEHP}$ concentrations by $14.11\text{\hspace{0.17em}mM}$ (95% CI: 10.74, 17.49; $p<0.01$), and the use of Foley catheters was associated with an increase in $\sum \text{DEHP}$ concentrations by $5.28\text{\hspace{0.17em}mM}$ (95% CI: 3.41, 7.15; $p<0.01$). For BPA, each one-unit increase in ECMO treatment was associated with a significant increase in urinary BPA concentrations by $55.49\text{\hspace{0.17em}ng}/\mathrm{mL}$ (95% CI: 46.47, 64.51; $p<0.01$), and the use of Foley catheters was associated with an increase in urinary BPA concentrations by $4.69\text{\hspace{0.17em}ng}/\mathrm{mL}$ (95% CI: 0.24, 9.14; $p=0.0389$). Associations for CPB and intensity of device intensity score were not significant for either DEHP or BPA. The high variance among participants highlighted the significant random effects for both BPA and $\sum \text{DEHP}$ (Table S13).

Clustering of BPA Concentration Data

The hierarchical clustering analysis was used to group the number of medical devices used (Figure 1A), the treatment of ECMO or CPB (Figure 1B), and time-series urinary BPA TK profiles (Figure 1C) based on the exposure patterns of urinary BPA concentrations among 18 neonates with CHD. The six clusters with different dynamic patterns were identified, and the dendrogram (Figure 1C) perfectly reflected the differences between clustering groups (clusters 1–6), indicating two major clusters (light gray and orange color clusters) and four outliers (participants 1, 2, 4, and 9) (right panel). The BPA levels in the groups with treatment with ECMO (participants 1, 2, 4, and 9) were higher than the levels in the group without the treatment of ECMO (clusters 1 and 2). The time-series plots for selected individuals within each cluster are shown in Figure 2. These plots illustrate representative TK profiles of BPA for each cluster. It is important to note that these plots represent selected individuals within the clusters and provide a snapshot of typical patterns observed. Based on these figures, while the representative TK profiles of BPA in clusters 1 (Figure 2A) and 2 (Figure 2B) appear similar, there is considerable variability within each cluster, as evidenced by the broader analysis in Figure 1. This variability suggests that the clustering could be influenced by different timing patterns with respect to device use, as well as the absolute magnitude of the BPA concentrations, which tend to be lower in cluster 1 compared to cluster 2. In the cluster groups with ECMO treatments (Figures 2C–F), the maximum peak concentrations corresponded to the use of ECMO, CPB, and other medical devices. In summary, the cluster groups with ECMO treatments (clusters 3, 4, 5, and 6) showed that the higher levels of BPA and the maximum peak concentrations corresponded to the trend of ECMO use, yet these results were unseen in the cluster groups without ECMO treatment.

Figure 1.

Hierarchical clustering analysis for time-series: (A) the number of medical devices used, (B) CPB or ECMO procedures, and (C) toxicokinetic (TK) profiles of BPA following day of life (DOL) in infants ($n=18$) undergoing surgery for hypoplastic left heart syndrome or transposition of the great arteries at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC). The study period was 2020–2021. Each row corresponds to the same infant across all panels (A–C), ensuring consistency in data representation. In the plot (B), the solid black lines represent the cardiopulmonary bypass (CPB) procedure and the red dashed stripes represent the extracorporeal membrane oxygenation (ECMO) procedure. The use of continuous multiple red dashed stripes for ECMO reflects its continuous usage during the treatment period. The y-axis of the plot in panel B does not have a quantitative scale because it represents a yes/no indicator. The plot in panel C consists of TK profiles of BPA and dendrogram, and the colors of gray, yellow, purple, red, light-yellow, and blue represent cluster 1 to cluster 6, respectively. The numbers along the x-axis in the plot in panel C represent the dynamic time warping–based clustering distance used in the hierarchical clustering dendrogram, which reflects the similarity between different infants’ TK profiles of BPA. Shorter distances indicate more similar exposure patterns, while longer distances reflect greater differences in TK profiles. Complete linkage was used to determine intercluster distances. Further elaborated data can be viewed in Excel Table S1. Note: BPA, bisphenol A; DTW, dynamic time warping.

Figure 2.

Clustering-based representative TK profiles of urinary bisphenol A (BPA) for (A) cluster 1, (B) cluster 2, (C) cluster 3, (D) cluster 4, (E) cluster 5, and (F) cluster 6 following day of life (DOL) in infants undergoing surgery for hypoplastic left heart syndrome or transposition of the great arteries at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC). The study period was 2020–2021. Data for participant_12, participant_11, participant_2, participant_9, participant_4, and participant_1 were used to be the representative TK profiles for clusters 1 to 6, respectively. The solid gray and red dashed vertical lines represent the day of the use of cardiopulmonary bypass (CPB) and extracorporeal membrane oxygenation (ECMO), respectively. The gray point curve indicates the number of medical devices used per day. The representativeness of participants was determined based on the similarity of the overall shape and trajectory of their TK profiles with other participants within each cluster. Further elaborated data can be viewed in Excel Table S2. Note: TK, toxicokinetic.

Clustering of $\sum DEHP$ and Non-DEHP Phthalates Metabolites Concentration Data

The time-series TK profiles of $\sum \text{DEHP}$ (i.e., the molar sum of MEHP, MEHHP, MEOHP, and MECCP) could be organized into six clustering groups, including four major clusters (clusters 1, 2, 3, and 4) and two outliers (clusters 5 and 6) (Figure 3). Higher levels of DEHP metabolites were observed in cluster 5 (participant 9) and cluster 6 (participant 1). Figure 4 shows the clustering-based representative time-series TK profiles of $\sum \text{DEHP}$. For clusters 1 (Figure 4A), 3 (Figure 4C), and 4 (Figure 4D), these patients did not receive ECMO, and thus the levels of DEHP metabolites were significantly lower than the clusters with ECMO [cluster 2 (Figure 4B), cluster 5 (Figure 4E), and cluster 6 (Figure 4F)]. The maximal peak concentrations were in agreement with the use of ECMO in the clusters with ECMO treatment (clusters 2, 5, and 6). The dynamic changes of DEHP metabolite concentrations for all clusters (clusters 1–6) corresponded to the numbers of medical devices used. Overall, the high levels of DEHP metabolites were observed in outlier groups in which the participants were frequently treated with ECMO, and the maximal peak concentrations in the TK profiles of $\sum \text{DEHP}$ were observed in the periods when there was a frequent use of ECMO treatments.

Figure 3.

Hierarchical clustering analysis for time-series: (A) the number of medical devices used, (B) CPB or ECMO procedures, and (C) toxicokinetic (TK) profiles of $\sum \text{DEHP}$ following day of life (DOL) in infants ($n=18$) undergoing surgery for hypoplastic left heart syndrome or transposition of the great arteries at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC). The study period was 2020–2021. Each row corresponds to the same infant across all panels (A–C), ensuring consistency in data representation. In the plot in panel B, the black solid lines represent the cardiopulmonary bypass (CPB) procedure and the red dashed stripes represent the extracorporeal membrane oxygenation (ECMO) procedure. The y-axis of the plot in panel B does not have a quantitative scale because it represents a yes/no indicator. The plot in panel C consists of TK profiles of $\sum \text{DEHP}$ and a dendrogram, with colors gray, yellow, purple, red, light-yellow, and blue representing cluster 1 to cluster 6, respectively. The numbers along the x-axis in the plot in panel C represent the DTW-based clustering distance used in the hierarchical clustering dendrogram, which reflects the similarity between different infants’ TK profiles of $\sum \text{DEHP}$. Shorter distances indicate more similar exposure patterns, while longer distances reflect greater differences in TK profiles. Complete linkage was used to determine intercluster distances. $\sum \text{DEHP}$ is the molar sum of DEHP metabolites: mono(2-ethylhexyl) phthalate (MEHP), mono(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), and mono(2-ethyl-5-oxohexyl) phthalate (MEOHP). The use of continuous multiple red stripes for ECMO accurately reflects its continuous usage during the treatment period. Further elaborated data can be viewed in Excel Table S3. Note: BPA, bisphenol A; CPB, cardiopulmonary bypass; DEHP, di(2-ethylhexyl) phthalate; DTW, dynamic time warping; ECMO, extracorporeal membrane oxygenation; MEHP, mono(2-ethylhexyl) phthalate; MEHHP, mono(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono(2-ethyl-5-oxohexyl) phthalate.

Figure 4.

Clustering-based representative TK profiles of urinary di(2-ethylhexyl) phthalate (DEHP) metabolites for (A) cluster 1, (B) cluster 2, (C) cluster 3, (D) cluster 4, (E) cluster 5, and (F) cluster 6 following day of life (DOL) in infants undergoing surgery for hypoplastic left heart syndrome or transposition of the great arteries at Children’s Hospital of Philadelphia (CHOP) and Medical University of South Carolina (MUSC). The study period was 2020–2021. The Participant_6, Participant_4, Participant_2, Participant_10, Participant_9, and Participant_1 were used to be the representative TK profiles for clusters 1 to 6, respectively. The gray and red bars represent the day of the use of cardiopulmonary bypass (CPB) and extracorporeal membrane oxygenation (ECMO), respectively. The gray point curve indicates the number of medical device use. $\sum \text{DEHP}$ is the molar sum of DEHP metabolites: mono(2-ethylhexyl) phthalate (MEHP), mono(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), and mono(2-ethyl-5-oxohexyl) phthalate (MEOHP). The representativeness of participants was determined based on the similarity of the overall shape and trajectory of their TK profiles with other participants within each cluster. Further elaborated data can be viewed in Excel Table S4. Note: CPB, cardiopulmonary bypass; DEHP, di(2-ethylhexyl) phthalate; DOL, day of life; ECMO, extracorporeal membrane oxygenation; MEHP, mono(2-ethylhexyl) phthalate; MEHHP, mono(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono(2-ethyl-5-oxohexyl) phthalate; TK, toxicokinetic.

Because metabolites of MEHP might be released from the urine bag (Table S4), to avoid bias of the analysis results, we also measured and compared the clustering results based on the TK profiles with and without MEHP (Figures S2 and S3). The results indicate that MEHP contaminations did not influence the cluster groups, and the results after adjusting MEHP concentrations were similar to those without adjusting MEHP concentrations (Figure S3). For the clustering analysis of non-DEHP phthalates, such as MBP (Figure S4), MBzP (Figure S5), MMP (Figure S6), and MEP (Figure S7), we only considered data from CHOP due to significant differences observed between institutions in some non-DEHP phthalates (e.g., MBP, MBzP, and MEP) based on Table 3. To avoid potential confounders, samples from MUSC ($n=3$) were excluded from this specific analysis. The clustering groups for non-DEHP phthalates were different from those for BPA and $\sum \text{DEHP}$. Only three clusters were identified among these metabolites, and they did not have a significant trend to correlate with CPB or ECMO treatment (Table 3). Note that the raw data used to generate Figures S2–S7 are provided in Excel Tables S9–S14.

Cross-Correlation Analysis

The cross-correlation analysis revealed the temporal relationship between urinary concentrations of BPA and DEHP metabolites and the number of medical devices used (Table S14). For BPA, the highest positive correlation ($\text{mean\hspace{0.17em}correlation}=0.38$) was observed at lag 0, indicating that BPA levels in urine were most strongly correlated with the number of medical devices used on the same day. Positive correlations were also observed for lags $-1$ and $+1$, suggesting a temporal association with device use within a day. For $\sum \text{DEHP}$, the highest positive correlation ($\text{mean\hspace{0.17em}correlation}=0.55$) was observed at lag $-1$, indicating that $\sum \text{DEHP}$ metabolite levels in urine were most strongly correlated with the number of medical devices used 1 day prior. Positive correlations were also observed at lag 0 and $+1$, highlighting a slightly delayed response compared to BPA.

Discussion

In our study of 18 neonates undergoing cardiac operations from a two-institution prospective and observational cohort study, we measured time-intensive TK profiles of urinary BPA, DEHP, and other non-DEHP metabolites in infants from birth to 3 wk of life, which included before, during, and after cardiac operations. The substantive exposures to BPA, DEHP, and other phthalates were observed in our study participants by comparing with previous studies.^28–31 Clustering of TK profiles stratified participants and showed different patterns and concentrations of urinary biomarkers due to different use of medical devices and medical operation processes (i.e., ECMO and CPB) as well as different patterns in the two institutions. We identified that significantly higher levels of urinary BPA and DEHP metabolites were observed in the groups with ECMO compared with the groups without ECMO therapy, while the peak concentrations of the time-course TK profiles of BPA and DEHP metabolites among those groups were correlated with the number of the medical devices used (this observation was based on a visual inspection of Figures 1 and 3). Comparably, these results were unseen in the TK profiles of urinary non-DEHP metabolites. Compared to our earlier study⁷ that did not quantify urinary BPA or phthalate concentrations over the entire exposure duration or characterize the specific type of medical devices used for each participant, the result from the present study substantially improves our understanding of neonatal exposure to BPA, phthalates, and their metabolites in infants with CHD undergoing heart surgery.

The observed urinary concentrations of phthalate metabolites in our study were considerably higher than those observed in recent studies of other cohorts from different countries, including the US, Canada, and China, who were not exposed to medical devices. For example, the geometric median concentrations of DEHP metabolites MEHP, MEHHP, and MEOHP in our study were 4.0, 59, and $501\text{\hspace{0.17em}ng}/\mathrm{mL}$, respectively, substantially exceeding those reported in the US NHANES 2013–2014 (2.5, 2.4, and $16\text{\hspace{0.17em}ng}/\mathrm{mL}$) and in the Canadian Health Measures Survey (CHMS 2007–2009; 1.3, 3.3, and $19\text{\hspace{0.17em}ng}/\mathrm{mL}$). Similarly, the median concentrations of urinary MEHHP and MEOHP in our participants (55 and $49\text{\hspace{0.17em}ng}/\mathrm{mL}$) were higher than those in infants 1–8 months of age from a US cohort study (11.8 and $11.9\text{\hspace{0.17em}ng}/\mathrm{mL}$),²⁸ in 3-month-old Canadian infants (0.8 and $0.8\text{\hspace{0.17em}ng}/\mathrm{mL}$),³⁰ and in Chinese infants 128 wk of age ($0.5\text{\hspace{0.17em}ng}/\mathrm{mL}$ for MEOHP).³¹ Median urinary levels of MBP ($8.1\text{\hspace{0.17em}ng}/\mathrm{mL}$) and MBzP ($3.9\text{\hspace{0.17em}ng}/\mathrm{mL}$) in our cohort were also elevated compared to Canadian infants (MBP: $4.9\text{\hspace{0.17em}ng}/\mathrm{mL}$; MBzP: $2.5\text{\hspace{0.17em}ng}/\mathrm{mL}$).³⁰ Additionally, DEHP metabolite concentrations were higher in our cardiac surgery cohort (MEHP: $3.7\text{\hspace{0.17em}ng}/\mathrm{mL}$; MEHHP: $55\text{\hspace{0.17em}ng}/\mathrm{mL}$; MEOHP: $49\text{\hspace{0.17em}ng}/\mathrm{mL}$; MECPP: $302\text{\hspace{0.17em}ng}/\mathrm{mL}$) than those reported in NICU infants who did not undergo surgery in a separate study (MEHP: $3.2\text{\hspace{0.17em}ng}/\mathrm{mL}$; MEHHP: $1.2\text{\hspace{0.17em}ng}/\mathrm{mL}$; MEOHP: $0.8\text{\hspace{0.17em}ng}/\mathrm{mL}$; MECPP: $9.2\text{\hspace{0.17em}ng}/\mathrm{mL}$).¹⁵ Our results suggest that infants undergoing high-intensity medical treatments might be exposed to BPA and multiple phthalate compounds, resulting in higher levels of urinary BPA and phthalate metabolites than the infants in the general population. Previous studies for infants in the NICU also showed similar results of high urinary levels of BPA and phthalate metabolites.^8,13,16,32 Our findings align with those of Guerrelli et al.,¹⁸ who investigated phthalate exposure in pediatric cardiac surgery patients using blood-based biomonitoring. Their study demonstrated that CPB circuits, particularly those primed with RBCs, significantly increased postsurgical phthalate levels, with higher exposure associated with postoperative complications. While their findings highlight the acute elevations in circulating phthalate levels immediately after surgery, our study extends these observations by capturing time-series TK profiles in urine, allowing us to evaluate exposure trends over multiple days. Additionally, Guerrelli et al.¹⁸ found that RBC washing effectively reduced phthalate levels in CPB prime solutions, a factor not explicitly assessed in our study but worth considering in future investigations. Our findings further emphasize the importance of CPB and ECMO as major sources of phthalate exposure and highlight the need for mitigation strategies to reduce exposure in neonates undergoing cardiac surgery. However, it is important to note that in this study, infants’ BPA and DEHP metabolites (i.e., MEHP) median levels ($0.7\text{\hspace{0.17em}ng}/\mathrm{mL}$ for BPA and $3.7\text{\hspace{0.17em}ng}/\mathrm{mL}$ for MEHP) were still below the European Human Biomonitoring Initiative guidance values in children (BPA: $137\text{\hspace{0.17em}ng}/\mathrm{mL}$; MEHP: $340\text{\hspace{0.17em}ng}/\mathrm{mL}$),^33,34 the peak concentration of BPA for participant 1 ($231\text{\hspace{0.17em}ng}/\mathrm{mL}$) was the only value that exceeded the human guidance value in children. Nevertheless, the peak urinary concentrations of other metabolites were relatively high (e.g., $\mathrm{5,701}\text{\hspace{0.17em}ng}/\mathrm{mL}$ for MEHHP, $\mathrm{5,954}\text{\hspace{0.17em}ng}/\mathrm{mL}$ for MEOHP, and $\mathrm{22,554}\text{\hspace{0.17em}ng}/\mathrm{mL}$ for MECPP in participant 1). Human biomonitoring guidance values for these metabolites are not available. The human health implications of the urinary concentrations of these metabolites remain to be investigated.

DEHP is commonly found in polyvinyl chloride (PVC) applications such as vinyl flooring, adhesives, sealants, car-care products, and some personal care products.³⁵ Some of the medical device products, such as cotton balls, Foley bags, pipets, syringes, and cylinders, that are used to collect urine samples may also contain DEHP. In this study, we also specifically examined the concentrations of DEHP metabolites in several control samples related to urine collection, including cotton balls, syringes, cylinders, pipets, Cryovials, and urine bags. Our results indicated that there is possible contaminated exposure of BPA and MEHP from urine bags (Table S4), but it is unseen in other medical devices such as cotton balls (Table S5), syringes (Table S6), cylinders (Table S7), pipets (Table S8), and Simport Cryovials (Table S9). However, while we account for potential contamination from urine collection bags, we acknowledge that Foley catheters themselves could also be a source of contamination. Since the flexible material of Foley catheters may contain DEHP, additional contamination controls should be considered in future studies to evaluate potential chemical leaching from these devices. Including dedicated control samples from Foley catheters would provide a more comprehensive assessment of potential contamination sources in urine biomonitoring studies. Moreover, the infants’ median MBP ($8.1\text{\hspace{0.17em}ng}/\mathrm{mL}$) and MBzP ($3.9\text{\hspace{0.17em}ng}/\mathrm{mL}$) concentrations in our study were higher than those observed among 3-month-old infants (MBP: $4.9\text{\hspace{0.17em}ng}/\mathrm{mL}$ and MBzP: $2.5\text{\hspace{0.17em}ng}/\mathrm{mL}$) in a Canadian population³⁰ (Table 2). However, the unexpected inverse association between the intensiveness of product use and urinary concentrations of MBP was observed (Table 5). For other non-DEHP phthalate exposure, limited non-DEHP phthalate was detected in medical devices (Table S10), indicating that the medical devices might not be the major exposure source of DBP and other non-DEHP phthalate or non-DEHP phthalates are in the devices but are not disclosed.

There are two significant findings from the cluster analysis of urinary BPA and DEHP metabolites in our study. First, although there were different groups in the cluster analysis of urinary BPA and $\sum \text{DEHP}$ concentrations, the participants with the highest urinary BPA and $\sum \text{DEHP}$ concentrations were attributed to be outliers in both analyses. For example, “Participant_1” (Figures 1 and 3) had the highest urinary BPA and $\sum \text{DEHP}$ concentrations among all participants, which was categorized as a single-participant cluster and was different from other participants. In this participant, the high levels of BPA and $\sum \text{DEHP}$ in urine were correlated with prolonged exposure to ECMO. Our results were consistent with previous findings that patients or individuals who had the highest exposure to DEHP were those undergoing ECMO.^36,37 ECMO is a special extracorporeal medical therapy that provides prolonged cardiac and respiratory support. However, its operation requires heavy use of plasticized PVC products such as blood bags, tubes, and long-lasting blood contact. The patient’s blood is circulated through the circuit for several days and occasionally weeks, resulting in a continuous exposure to the components and any plasticizers they contain. Because DEHP is the favored additive in PVC components, PVC materials have a high leaching potential of DEHP into the blood.^38–40 Due to toxicological implications, DEHP exposure from the use of medical devices should be further monitored, in particular for patients who are exposed to PVC-containing products such as the ECMO circuit. Second, the participants with lower urinary DEHP and BPA levels were categorized as the same cluster (cluster 1 in Figures 1 and 3). The common features of this cluster were that participants did not undergo ECMO therapy but received CPB. There are many similarities between the ECMO and CPB circuits; however, the duration of CPB support is much less, hours rather than days. Consistent with this difference in duration of exposure, our results suggest that, compared with ECMO therapy, exposure to CPB is a significant but short source of DEHP or BPA. Overall, the dynamic use and removal of these medical devices during medical treatment are essential factors in understanding the reduction in exposure over time and reconciling the identified clusters. The clusters were identified based on the temporal patterns of exposure, which were influenced by the intensity and duration of device use. Additionally, it is important to discuss the observed dissimilarity in the patterns within cluster 1 (Figure 1). This can occur due to several reasons. Time-series data often have inherent variability and noise. While DTW attempts to align patterns that may not be perfectly synchronized, significant differences in shape or trend can still result in clusters containing dissimilar patterns. Another reason might be due to the sensitivity of DTW algorithm to the local alignment of data points. Although DTW can handle shifts and distortions in time, it might still group time-series data with different overall trends or behaviors if their local alignments produce similar DTW distances.

Our cross-correlation analysis provides valuable insights into the temporal dynamics of BPA and DEHP exposure in neonates undergoing cardiac surgery. The strong correlation observed at lag 0 for BPA suggests that the exposure and subsequent detection of BPA metabolites in urine occur almost simultaneously with the use of medical devices. This finding indicates that BPA leaches from medical devices and is rapidly absorbed and excreted by neonates. In contrast, the peak correlation for $\sum \text{DEHP}$ metabolites at lag $-1$ suggests a slight delay in the absorption and excretion of $\sum \text{DEHP}$ metabolites compared to BPA. This delay could be attributed to differences in the chemical properties of DEHP, the types of medical devices used, or the metabolic pathways involved in processing DEHP. Our mixed-effect model further elucidates the relationship between medical treatments, device use, and urinary concentration of BPA and $\sum \text{DEHP}$. The results indicate the urinary BPA concentration increased significantly with each one-unit increase in ECMO treatment and the use of Foley catheters. Specifically, BPA concentrations rose by $51.5\text{\hspace{0.17em}ng}/\mathrm{mL}$ with ECMO treatment and by $6.5\text{\hspace{0.17em}ng}/\mathrm{mL}$ with Foley catheter use. Similarly, $\sum \text{DEHP}$ concentrations increased by $13.4\text{\hspace{0.17em}mM}$ with ECMO treatment and by $6.15\text{\hspace{0.17em}mM}$ with Foley catheter use. Interestingly, when adjusting the medical treatment, the intensity score of medical device use did not show a significant association with urinary BPA and DEHP for either the LME model with or without lag effect. This implies that while ECMO and Foley catheters are major contributors to exposure, their effects might overlap with other variables (i.e., multicollinearity), complicating the analysis. In addition, because the urine passes through the Foley catheter, it could be a source of contamination. Additionally, the pharmacokinetics of endocrine disrupting chemicals (EDCs) in infants are influenced by the infant’s immature metabolic and excretory systems. Absorption of EDCs can occur rapidly, often within hours of exposure, but metabolism and excretion rates are slower in infants than in adults due to underdeveloped liver and kidney functions. For instance, while the half-life of BPA in adults is $\sim 6$ hours, it can be longer in infants.^41,42 DEHP and its metabolites also exhibit varying half-lives, typically ranging from 4 to 24 h in adults, with potentially extended half-lives in infants.^7,28 These potential delays in metabolism and excretion further highlight the complexity of assessing exposure in neonates. Overall, our findings highlight the critical role of medical devices and procedures in the exposure to BPA and DEHP in neonates undergoing cardiac surgery.

There are several limitations to this study. First, the small number of participants limits our ability to draw strong inferences generalizable to other populations. Nevertheless, this is the first study to demonstrate that increasing the use of medical devices and specific medical operations is directly and positively related to neonatal exposure to BPA and DEHP in infants with CHD receiving heart surgery. It is also the first study to estimate exposure levels to DBP and BzBP in infants with CHD undergoing heart surgery. Second, we were not able to relate the exposure in this infant population to adults exposed to ECMO or CPB. There is a paucity of data evaluating plasticizer exposure in adults exposed to ECMO and CPB.^36,43,44 The existing data confirm that there is a significant exposure. We are unaware of any study similar to ours evaluating the longitudinal exposure throughout the hospitalization or the relationship to medical device usage in adults exposed to ECMO and CPB with cardiac surgery. Therefore, there is lack of data to make a direct comparison between infants and adults.

Third, in our clustering analysis, we were able to correlate the cluster of time-course TK profiles with the number of medical devices used, but the results were unable to distinguish exposure to BPA or DEHP from specific medical device use, and the information on which devices contain BPA and DEHP is unclear, limiting our ability to examine further the association between observed groups with high BPA and DEHP concentrations in specific product use groups. In particular, the use of a foley catheter was associated with greater BPA and DEHP metabolite concentrations. Some of the Foley catheters used contain DEHP and BPA. The catheters are in contact with the urethral mucosa and are a likely source of patient exposure. However, because the urine passes through the catheter, it could be a source of contamination. Because control samples were not obtained from the Foley catheter as part of the urine collection system, we cannot rule out potential contamination. For example, participants 16 and 18, who did not undergo ECMO or CPB still exhibited low-to-moderate exposures. While they did not undergo bypass or ECMO, the number of devices containing DEHP/BPA used postprocedure was comparable to those used in infants who had bypass procedures at both facilities. This makes it unclear what their exposure sources were. In contrast, the fluctuating levels of BPA and DEHP metabolites observed in participant 9, who had continuous ECMO exposure, suggest several potential factors that could contribute to the observed peaks and valleys despite continuous ECMO use. The metabolic rate and excretion patterns of the infant can vary and, thus, influence metabolite levels. Changes or replacements in ECMO circuit components, periodic medical interventions such as blood transfusions, and the concurrent use of other medical devices may also contribute to fluctuating exposure levels. The handling and sampling variability, including the timing of urine sample collection in relation to medical procedures and the infant’s hydration status, could further impact the concentration of metabolites measured. These factors collectively highlight the complexity of interpreting exposure levels and suggest that continuous ECMO use does not necessarily lead to continual high exposure levels due to other factors. We made extensive efforts to identify potential sources of exposure by contacting manufacturers and reviewing package inserts for all devices used. Despite these efforts, we cannot be certain that we have identified all potential sources of exposure.

Fourth, differences in urinary MECPP and other non-DEHP metabolites (e.g., MBP, MBzP, and MEP) were observed between institutions, but these differences may be largely attributable to variations in medical treatment rather than institutional factors alone. Since two out of three infants from MUSC did not undergo ECMO or CPB, it is possible that treatment-related factors, such as device exposure and medical interventions, contributed more to the observed variability than institutional differences. The clustering analysis for DEHP exposure was estimated based on the sum of DEHP metabolites and only three participants from MUSC, thus the interinstitutional differences might be minimized. For the cluster analysis for urinary non-DEHP metabolites, we only used the participants from CHOP to avoid unnecessary uncertainty. Additional TK profiles of the study participants from MUSC are needed to further characterize the potential variations and uncertainties between institutions. Fifth, while our current models do not include additional covariates beyond those directly related to medical device use and treatments, we recognize that other factors, such as demographic variables, health status, and environmental exposure, could potentially influence urinary BPA and phthalate levels. Future analyses, if data are available, will need to consider a broader range of covariates to provide a more comprehensive understanding of these associations. Finally, there is a lack of computational models to estimate daily exposures of BPA and DEHP and its metabolites based on urinary concentrations and to assess potential health risk for infants due to exposure to BPA and DEHP and its metabolites. In this regard, it has been shown that physiologically based pharmacokinetic (PBPK) models can be used to predict human external exposure based on urinary metabolite concentrations (i.e., reverse dosimetry analysis) and assess potential health risk due to environmental chemical exposure. PBPK models for BPA and DEHP have been reported mainly in children and adults.^41,45 Additional studies should build upon the knowledge provided here and extrapolate existing PBPK models from children/adults to infants. The new data from this study provides a basis for additional PBPK analyses in the future. Additionally, this study reports elevated phthalate exposure in infants undergoing invasive medical procedures, with short peaks with CPB, longer peaks with ECMO, and a continuous lower-level exposure in all infants when these therapies are not in use. Further work needs to estimate the potential risk due to DEHP and BPA exposures from these medical devices in those infant patients and to determine if the peak, duration, or patterns of exposure are associated with adverse clinical outcomes.

Conclusions

The present study reports original time-dependent data on urinary concentrations of BPA and phthalate metabolites in infants with CHD undergoing heart surgery. We found elevated phthalate exposure in neonates with CHD undergoing invasive medical procedures, with short peaks after CPB, longer peaks with ECMO, and a continuous lower-level exposure in all infants when these therapies are not in use. We applied machine learning–based clustering method to categorize TK profiles of urinary BPA, $\sum \text{DEHP}$, and other non-DEHP metabolites and analyzed the relationship with the use of medical devices and medical procedures such as ECMO and CPB. These results provide strong evidence of the role of medical devices and medical procedures (e.g., ECMO) play in BPA and DEHP exposure levels of infant patients while in critical care settings, findings that were not seen in non-DEHP phthalate exposure. Additionally, cross-correlation analysis revealed a strong temporal relationship between the number of medical devices used and urinary concentrations of BPA and DEHP metabolites, further confirming the impact of medical device use on exposure levels. These insights highlight the urgent need for developing and utilizing BPA-free and phthalate-free medical devices, especially for sensitive populations, such as infants with CHD who undergo heart surgery and require prolonged hospital stays. From a clinical standpoint, this study also identified significant interinstitutional differences in patterns of exposure to phthalates, which could potentially contribute to known interinstitutional differences in outcomes for patients with CHD.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.