Role of early peritoneal dialysis after neonatal arterial switch operation

Abstract

Objective

Peritoneal dialysis (PD) commenced early in the postoperative period has the potential to mitigate the postcardiopulmonary bypass inflammatory response. We evaluated the role of early PD on postoperative outcomes after the arterial switch operation (ASO).

Methods

Newborns (≤30 days, n = 318) undergoing ASO were classified into those who did (early PD, n = 90) or did not (control, n = 228) receive PD within 6 hours of admission to intensive care unit after surgery. Using observational data and imitating a preplanned clinical trial (target trial framework), we evaluated the role of early PD on postoperative outcomes.

Results

Infants in the early PD group had greater serum lactate (median [interquartile range]: 2.6 [2.1, 4.1] vs 2.2 [1.8, 2.9]) and lower central venous saturation (median [interquartile range]: 45.2 [39.3, 51.4] vs 51.3 [42.2, 59.9]) at admission. Early PD was associated with a shorter duration of mechanical ventilation, but this effect was restricted to the subgroup receiving extracorporeal membrane oxygenation (ECMO) in the perioperative period (incidence rate ratio [95% confidence interval]: for early PD/control: 0.28 [0.17-0.47] for those requiring ECMO and 1.14 [0.93-1.39] for those not requiring ECMO, P interaction <.001). Similar results were seen for intensive care unit length of stay.

Conclusions

Early PD after ASO was associated with a reduction in duration of mechanical ventilation and intensive care stay for infants who required ECMO in the perioperative period. Future studies of early PD, ideally clinical trials, in high-risk infants (such as those requiring ECMO after cardiac surgery) will be of benefit to either confirm or refute these findings.

Early PD after ASO improves outcomes for newborns requiring ECMO.

Central Message.

We report the role of early PD (within 6 hours) in a comprehensive cohort of newborns who underwent ASO.

Perspective.

Peritoneal vascularity and its permeability to solutes/proteins vary among humans. Under conditions of elevated cytokine production (prolonged CPB+ECMO) with associated capillary-leak, they “spill-over” into peritoneal cavity, allowing removal through a PD catheter. Having a catheter provides early opportunity to remove cytokines, optimize fluid-balance in high-risk operations, and influence outcomes.

Newborns with transposition of the great arteries (TGA) typically undergo the arterial switch operation (ASO) in the first month of life,^1,2 and current estimates show low postoperative mortality.^1,3 This population, like many others with congenital heart disease, is susceptible to the development of a post-cardiopulmonary (CPB) inflammatory state and ischemia-reperfusion (IR) injury resulting in a low cardiac output syndrome in the early postoperative period. After the ASO, up to 25% develop a low cardiac output,⁴ and this is can associated with significant morbidity. Prevention of this hemodynamic deterioration is important to optimize clinical outcomes.

Observational evidence has suggested that early commencement of peritoneal dialysis (PD) after surgery is associated with improved outcomes.5, 6, 7 A small, single-center study on the use of PD after ASO⁸ has shown benefits such as shorter duration of mechanical ventilation and stay in the intensive care unit (ICU) as well as lower hospital costs. TGA is a relatively homogenous group of heart defects, and infants with this condition undergo an intermediate-risk operation.⁹ Given a high proportion of them experience low cardiac output after ASO,⁴ understanding the role of PD in this targeted population is important. Peritoneal catheters are routinely placed in the operating theater after all cases of ASO in our institution. This allows us to ascertain the role of early PD in this population.

Traditional Observational Study Evidence and Its Problems

Conventional observational studies experience methodologic problems that can produce biased estimates. The main biases are selection bias, sampling bias (attributable to nonparticipation), confounding bias, and immortal time bias (related to improper specification of time-zero).¹⁰ Recent developments in health research methodology have strengthened causal inference by minimizing these biases.^10,11 These methods allow researchers to design a clinical trial to answer the question of interest and aim to emulate components of this target trial protocol using observational data. This improves transparency and replicability and minimizes unnecessary sources of bias.¹⁰

Controlled Trial Evidence and Its Problems

Evidence from conventional clinical trials may be limited in its applicability to day-to-day practice for several reasons. (1) When a treatment is highly controlled in a trial setting (unlike the flexibility inherent in real-world setting), its usefulness and generalizability is reduced. (2) Clinical trials often take years to complete and during this period, major changes occur in the target population; in particular, rates (and number) of primary outcome events decline and, among participants who experience those events, the reasons for the events also change. (3) The requirement for informed consent for complex trial interventions in critical care settings has the potential to alter the types of participants entering trials. Having a critically ill child in the ICU is stressful for families, who may find it difficult to provide consent under these conditions. Consequently, the sickest children, who are likely to benefit from an intervention, are also the most difficult to recruit into trials.

Several of the aforementioned issues with traditional clinical trials can be overcome to some extent by pragmatic clinical trials involving random assignment embedded in everyday practice.¹² In the absence of such pragmatic trials, target trial emulation using observational data can be viewed as pragmatic clinical trials lacking a placebo, blind treatment assignment, and blind outcome ascertainment.¹⁰ We aimed to apply such a methodology to study the role of early PD in newborns undergoing the ASO.

Methods

The project was approved by the institutional ethics board on March 14, 2024 (HREC/101749/RCHM-2023); individual consent was not required.

Aims

Using observational data imitating a preplanned target trial, we assessed the effect of early PD (compared with control) in newborns undergoing ASO. The target trial framework is shown in Table E1.

Eligibility

Newborns (≤30 days) undergoing ASO and admitted to ICU postoperatively from January 2010 to December 2023 were eligible.

Interventions of Interest

Intervention group: early PD, defined as PD commenced within 6 hours of admission to intensive care after surgery. Control group, defined as one or more of the following: PD commenced after 6 hours, PD never commenced, or peritoneal catheter under passive drainage. The target trial framework as defined allows estimation of the per-protocol effect of the intervention.

Outcomes

Postoperative duration of mechanical ventilation (primary), duration of intensive care and hospital stay.

Statistical Analysis

The study was designed to emulate the components of a pragmatic trial using routinely collected data. Balance in study covariates between study groups was checked with the use of inverse probability of treatment weighting (IPTW) method.¹³ Baseline was defined as the time of admission to intensive care after surgery.

First, a logistic regression using the study exposure as outcome (early PD vs control, 1/0) conditional on baseline covariates was performed. This allowed estimation of the predicted probability of choosing early PD for each participant. Second, using IPTW method,¹³ the calculated probability was used to weight each subject in such a way that a weight of 1/predicted probability was allocated for early PD and a weight of 1/(1 – predicted probability) to control subjects, with the aim of achieving balance between the groups. Balance of individual covariates between the groups was assessed by calculation of absolute standardized differences (AbSD) of means and proportions; adequate balance was defined as a maximum AbSD of <0.1 after IPTW¹³ between the study groups. The list of covariates is shown in Table E1.

The causal effect of early PD on primary (duration of mechanical ventilation) and secondary (length of ICU and hospital stay) end points are reported as incidence rate ratios (IRRs) with 95% confidence intervals (CIs). IRRs were calculated using negative binomial regression with the inclusion of IPTWs and a robust variance-estimator. In addition, 3 postbaseline covariates (within 6 hours of ICU admission), known to predict treatment decision, were also balanced using IPTW: highest lactate, lowest venous saturation, and highest vasoactive inotrope score.¹⁴ An additional analysis was also conducted using an alternate definition of early PD (using 6 hours from the end of CPB) and study outcomes. Preplanned subgroup analyses included (1) gestational age into 2 groups (with approximately equal numbers in each group), (2) sex, (3) prenatal versus postnatal diagnosis, (4) requirement for extracorporeal membrane oxygenation (ECMO) before and/or after surgery, (5) CPB duration (3 equal groups), (6) type of operation, ie, ASO versus ASO ± ventricular septal defect closure ± arch repair (7), and admission serum lactate (3 equal groups). Appropriate interaction terms were included in the model to test subgroup effects and model reparameterization was performed to estimate the effect of early PD for each level of the subgroup variable.

Results

Patient Characteristics

During the study period (January 2010 to December 2023), 319 newborns with TGA underwent the ASO within 30 days of life; 318 were included in the study (Figure 1). The median (interquartile range, or IQR) gestation at birth for the full cohort was 39 (38.1, 39.5) weeks and median (IQR) weight at surgery was 3.4 (3, 3.7) kg. In total 226 of 318 (71%) underwent atrial septostomy before ASO (67 of 90 [74%] in the early PD and 150 of 228 [70%] in the control group). Sixty-five (20%) were admitted after surgery with an open sternum. Ninety (28%) were started on early PD and 228 (72%) were control (Table 1). There were 6 deaths in the ICU (6/318, 1.9%) (3 in early PD and 3 in control) and 8 deaths overall in hospital (8/318, 2.5%) (4 in each study group).

Figure 1.

Study flow. PD, Peritoneal dialysis; ICU, intensive care unit.

Table 1.

Patient characteristics

Characteristic	Early PD (n = 90)	Control (n = 228)	Absolute standardized difference
			Pre-IPTW	Post-IPTW
Gestational age at birth, wk	38.4 (37.4, 39.5)	39 (38.2, 39.5)	0.24	0.06
Birth weight, kg	3.35 (2.89, 3.64)	3.31 (2.99, 3.64)	0.15	0.02
Age at surgery, d	8 (6, 12)	8 (6, 11)	0.03	0.01
Weight at surgery, kg	3.4 (2.9, 3.7)	3.4 (3.1, 3.7)	0.19	0.04
Female	28 (31)	78 (34)	0.07	0.04
Prenatal diagnosis	67 (74.4%)	166 (72.8%)	0.04	0.05
Australian state of prenatal care			0.05	0.06
Victoria	63 (70)	132 (58)
Western Australia	9 (10)	34 (15)
South Australia	11 (12)	40 (18)
Other	7 (8)	22 (10)
Preoperative ICU admission	84 (93)	205 (90)	0.12	0.09
Preoperative mechanical ventilation	71 (79)	180 (79)	0.001	0.02
Preoperative ECMO	10 (11)	11 (5)	0.23	0.03
Preoperative creatinine, μmol/L	47 (39, 56)	46 (39, 55)	0.14	0.03
Congenital anomaly/genetic syndrome	8 (9)	12 (5)	0.14	0.04
CPB duration, min	213 (178, 237)	201 (173, 244)	0.08	0.05
Crossclamp duration, min	126 (102, 153)	120 (105, 152)	0.04	0.06
Type of surgery			0.01	0.007
ASO	71 (79)	169 (74)
ASO + VSD repair ± arch repair	19 (21)	59 (26)
Open sternum from operating theater	22 (24)	43 (19)	0.14	0.02
Lactate at ICU admission, mmol/L	2.6 (2.1, 4.1)	2.2 (1.8, 2.9)	0.49	0.003
Venous saturation at ICU admission, %	45.2 (39.3, 51.4)	51.3 (42.2, 59.9)	0.11	0.09
VIS at ICU admission	5 (4, 7)	5 (3, 6)	0.35	0.01
Highest lactate in first 6 h, mmol/L	3.1 (2.5, 4.8)	2.5 (1.9, 3.1)	0.52	0.05
Lowest venous saturation in first 6 h, %	46.9 (39, 54.5)	50 (42.4, 59.5)	0.15	0.06
Highest VIS score in first 6 h	8 (5, 12)	7 (5, 10)	0.30	0.03

Percentages may not add to 100 because of rounding; venous saturation from right internal jugular; data are median (interquartile range) or n (%). PD, Peritoneal dialysis; IPTW, inverse probability of treatment weighting; ICU, intensive care unit; ASO, arterial switch operation; VSD, ventricular septal defect; VIS, vasoactive inotrope score.

Before applying IPTW, variable distribution showed that infants commenced on early PD were born earlier, a greater proportion had received preoperative ECMO (11% vs 5%), had greater lactate, and lower central venous saturation on admission to ICU after surgery.

The AbSDs pre-post IPTW are shown in Table 1 and Figure E1. After IPTW, the AbSD was ≤0.1 for all variables, suggesting a well-balanced cohort. For example, the AbSD of 0.24 pre-IPTW for gestation reduced to 0.06 post-IPTW. The pre-IPTW AbSD was 0.49 for admission lactate, which was reduced to 0.003 post-IPTW.

Figure E1.

Absolute standardized differences before and after IPTW. IPTW, Inverse probability of treatment weighting; ICU, intensive care unit; VIS, vasoactive inotrope score; ECMO, extracorporeal membrane oxygenation; CPB, cardiopulmonary bypass.

Characteristics of PD Use

The median (IQR) time to commence PD in the early PD group was 1.8 (1, 3.9) hours and it was 11.3 (7.8, 18.5) hours in the control group (P < .001) (Table 2). In total, 108 of 228 (47%) participants in the control arm were commenced on PD postoperatively. The median (IQR) duration of PD in the first 24 hours in early PD versus control arms were 17.5 (13, 21) and 0 (0, 8) hours, respectively (P < .001). The distribution of time to PD commencement by time since ICU admission is shown in Figure E2.

Table 2.

Characteristics of PD use in the study groups

Characteristic	Early PD (n = 90)	Control (n = 228)
Time to commence PD after ICU admission, h	1.8 (1, 3.9)	11.3 (7.8, 18.5)∗
PD duration, h, in the first 24 h	17.5 (13, 21)	0 (0, 8)
PD duration, h, after 24 h	9 (0, 19)	0 (0, 12)
Duration, h, of peritoneal free drainage in the first 24 h†	0 (0, 4)	0 (0, 9.5)

Data are reported in table are as median (interquartile range). The mean (standard deviation) duration of PD in the first 24 hours in the early PD and control groups was 16.7 (5.7) and 3.9 (5.9) hours, respectively. The mean (standard deviation) duration of PD after 24 hours in the early PD and control groups was 19.5 (40.4) and 11.2 (25.8) hours, respectively. PD, Peritoneal dialysis; ICU, intensive care unit.

^∗In total, 108 of 228 (47%) in control arm were started on PD.

^†Overall, 141 of 318 (44%) had some period of free drainage in the first 24 hours.

Figure E2.

Time to commence peritoneal dialysis (n = 198 in both study groups combined) from the time of admission to ICU after arterial switch operation. Data reported as median (IQR). Total of 198 (90 in the early PD group and 108 in the control group) were started on PD in the postoperative period. ICU, Intensive care unit; IQR, interquartile range; PD, peritoneal dialysis.

Primary and Secondary End Points

In the adjusted analysis (Table 3), the primary end point of duration of mechanical ventilation was 40% lower in the early PD group than the control group (adjusted IRR, 0.60; 95% CI, 0.42-0.86). A similar reduction was seen in the duration of ICU stay (adjusted IRR, 0.55; 95% CI, 0.37-0.82). The adjusted IRR for the duration of hospital stay was 0.75 (95% CI, 0.55-1.04). A sensitivity analyses using an alternate definition for early PD (6-hour cut-off from time of CPB finish) showed results consistent with the main analysis (Table E2).

Table 3.

Study outcomes and association with early PD

Postoperative outcome	Unadjusted analysis		Adjusted analysis		Early PD/control: Adjusted incidence rate ratio (95% CI)
	Early PD	Control	Early PD	Control
	Mean ± SD		Mean (95% CI)
Length of mechanical ventilation, h	105.3 (85)	121.5 (254.6)	89.4 (75.6-103.1)	148.4 (101.3-195.5)	0.60 (0.42-0.86)
Length of ICU stay, d-	6.2 (4.5)	7.7 (17.1)	5.3 (4.7 to 6.0)	9.7 (6.0-13.3)	0.55 (0.37-0.82)
Length of hospital stay, d-	18.2 (15.5)	17.6 (25.5)	15.9 (13.7 to 18.1)	21 (15.1-27)	0.76 (0.56-1.04)

PD, Peritoneal dialysis; SD, standard deviation; CI, confidence interval; ICU, intensive care unit.

Subgroup Analysis

The effect of treatment with early PD and mechanical ventilation was evaluated across subgroups (Figure 2). Early PD effect was greatest in infants who required perioperative ECMO (adjusted IRR, 0.28; 95% CI, 0.17- 0.47); those who did not require ECMO derived no demonstrable benefit (1.14; 95% CI, 0.93-1.39) (P for interaction < .001). Forty-five infants overall received ECMO (15 in the early PD and 30 in control) (Table E3).

Figure 2.

Subgroup analyses reporting duration of mechanical ventilation (MV). CI, Confidence interval; CPB, cardiopulmonary bypass; ICU, intensive care unit; ASO, arterial switch operation; VSD, ventricular septal defect; ECMO, extracorporeal membrane oxygenation.

Discussion

In this study of 318 newborns undergoing the ASO, early use of PD (within 6 hours) was associated with a reduction in postoperative mechanical ventilation and ICU duration in children receiving perioperative ECMO. For infants who were not on ECMO (the majority of the study cohort), early PD did not confer any major benefit in this study. It has to be acknowledged that most of the participants in this study would have received what could be considered low dose PD (10 mL per kg of 1.5% dextrose solution, hourly cycles), so the role of alternate specifications (shorter cycle durations and/or higher dextrose concentrations) cannot be understood from these results.

Observational evidence has shown that early PD might optimize postoperative outcomes5, 6, 7, 8; most of these studies are also likely to have been limited by sampling bias and selection bias,¹⁵ and findings have to be interpreted with caution. Although the authors of a previous clinical trial studied the role of PD versus furosemide in infants who remained oliguric after cardiac surgery,¹⁶ this current study aims to report the role of early PD (within 6 hours) versus control (which included late PD, no PD, passive peritoneal drainage), so they address different questions. The role of passive peritoneal drainage has been discussed in literature before. A 2014 study in which the authors compared PD with passive peritoneal drainage showed superior performance of PD in terms of fluid balance, clinical outcomes, and cytokine levels,¹⁷ and a recent multicenter study of PD catheter use from Neonatal and Pediatric Heart and Renal Outcomes Network (NEPHRON) investigators show superior fluid drainage by using PD rather than passive drainage.¹⁸

Subgroup analysis in this study shows that early PD might optimize outcomes in the high-risk group (those requiring ECMO, having ASO + ventricular septal defect operation or greater lactate at ICU admission). Essentially, the majority of the neonates (who were non-high risk) did not seem to derive any major benefit. The systemic inflammatory response syndrome (SIRS) associated with CPB and IR injury¹⁹ plays an important role in the postoperative recovery after ASO.⁴ While this SIRS is predictable, peaking within 4 to 6 hours after CPB and reaching baseline between 24 and 48 hours,^20,21 there are exceptions. Infants supported with ECMO have an exaggerated SIRS as a result of the cumulative effects of ECMO and CPB, leading to a widespread activation of the innate and adaptive immune systems.^22,23 Patients with this exaggerated response are more likely to respond to modulating therapies, as is suggested by our findings. PD in addition reduces lung injury by augmenting fluid removal and stabilizes acid-base balance which optimizes conditions for organ recovery after CPB. A study conducted on a mice model of ischemic acute kidney injury revealed that high-dose PD reduces lung injury and inflammation associated with removal of serum interleukin (IL)-6 and peritoneal IL-6.²⁴

CPB and IR injury results in microvascular damage and endothelial-cell injury.^25,26 Neutrophil-mediated endothelial damage resulting from tumor necrosis factor-alpha and IL-1 increases capillary permeability.²⁷ The activated endothelium expresses IL-8, stimulates neutrophils producing additional endothelial damage.²⁷ IR injury caused by return of blood to anoxic/hypoxic regions furthers exacerbates this injury causing cell-to-cell junction dissociation between endothelial cells.²⁵ Importantly, the glycolax (on the intraluminal side of blood vessel), which provides structural support to the endothelial layer, is also damaged, further increasing the capillary leak.²⁸ Under these conditions having a PD catheter might allow easy removal of cytokine molecules.

Peritoneal Vascularity and the Role of Early PD

The degree of peritoneal vascularity and its permeability to solutes and proteins vary widely among humans^29,30; this results in wide-ranging diffusion rates of solutes, waste products, and cytokines across the peritoneum. Cytokine molecules with a large molecular weight³¹ have very low clearance across the peritoneal capillaries under normal conditions.³² In addition to size selectivity, cytokine transport across the peritoneal membrane^32,33 also may depend on a concentration gradient. However, under conditions of elevated production of cytokines (prolonged CPB and ECMO) and the associated capillary leak, they may easily “spill-over” into the peritoneal space, allowing removal through the PD catheter.

There is substantial heterogeneity within the high-risk neonatal cohort undergoing cardiac surgery. Traditional methods of stratifying risk (such as age, risk adjustment for congenital heart surgery [RACHS-1] or Society of Thoracic Surgeons-European Association for Cardio-Thoracic Surgery [STAT] categories, CPB duration, baseline lactate, vasoactive inotrope score, etc)^6,17,34,35 although helpful, are limited in scope. Sophisticated methods that provide real-time risk stratification such as a multiomics offers promise.²³ Such approaches after neonatal cardiac surgery are likely to be useful in characterizing changes across the cytokine storm umbrella and provide guidance in therapy.²³ Precision diagnostics (eg, proteomics) in the postoperative period opens the possibility of targeting specific molecules with directed immune-modulating drugs. Viewed in this context, although PD may modulate SIRS, it is a relatively broad and imprecise therapy.

Among various cytokines involved in the post-CPB inflammatory response, IL-6 has been widely studied.^20,24,36 With elevated IL6, downstream signaling and activation of immune and nonimmune cells occur.^23,37 Three cardiac studies^17,38,39 have reported cytokine measurements with the use of PD catheter; their collective evidence suggests that IL-6 and IL-8 are removed by PD, removal of IL-6 is more efficient than IL-8 and renal clearance of ILs is very minimal. In a mice model of ischemic acute kidney injury,²⁴ use of high-dose PD resulted in removal of IL-6 from serum and showed decrease in lung inflammation without any evidence of alterations in kidney inflammation markers.

Strengths and Limitations

This study included a complete cohort of newborns who underwent ASO. Because our protocol is to insert PD catheters at the end of all ASOs, we did not miss any patient; in this regard, sampling bias, wherein PD catheters are inserted only in a self-selected high risk ASOs was eliminated. Given the main question was to address the role of early PD, all patients who did not satisfy this criterion (no PD use, PD after 6 hours, use of free drainage) were classified as controls; therefore, this can be considered as a pragmatic target trial with clear definitions of intervention arms. We followed this approach in design and analysis^10,40; this to some degree minimizes selection bias and reduces immortal time bias, both major problems in standard observational studies.⁴¹

Although IPTW was used to restore balance, the possibility of residual confounding does exist both while addressing the main study question and within subgroups; in a sense, this possibility exists even in a randomised controlled clinical trial. Although our routine protocol for PD is 10 mL/kg of 1.5% dextrose, hourly cycles, and study infants would have been started on this prescription, it is possible that in some cases dextrose concentrations or cycle duration may have subsequently been altered. While baseline parameters (such as lactate) were used to divide into subgroups, more meaningful subgroups on the basis of cytokine profile would have been ideal, but cytokines are not routinely measured after surgery.

In conclusion, use of early PD was associated with a reduction in mechanical ventilation and intensive care duration post ASO in those who received ECMO. It is likely that the exaggerated inflammatory response inherent in this high-risk group is modulated by the use of early PD. Clinicians in cardiac intensive care may consider the use of early PD in the postoperative period for newborns on ECMO after ASO.

Conflict of Interest Statement

The authors reported no conflicts of interest.

The Journal policy requires editors and reviewers to disclose conflicts of interest and to decline handling or reviewing manuscripts for which they may have a conflict of interest. The editors and reviewers of this article have no conflicts of interest.

Appendix E1. Potential Benefits of Using a Target Trial Protocol

Table E1.

Target-Trial emulation used in the study of PD use after ASO

Protocol component	Target trial specification	Emulation using observational data
Eligibility criteria	Neonates (≤30 d) undergoing ASO and admitted to intensive care unit postoperatively.	Same as for specification. Study baseline: time of admission to intensive care from the operating theater.
Treatment strategies	(1) Intervention: early PD commenced within 6 h of ICU admission. (2) Control: PD commenced after 6 h or PD not commenced at all or peritoneal catheter under passive free drainage.	Same as for specification.
Treatment assignment	Random assignment to intervention or control. Both participants and treating teams will be aware of the assignment.	Eligible participants will be assigned to the strategies with which their data were compatible at the time of admission (eligibility) and as per the treating team. Baseline balance between study groups will be ensured by IPTW of a wide range of baseline covariates. The baseline covariates include prenatal diagnosis, gestational age, weight, sex, weight at surgery, noncardiac congenital anomaly, preoperative mechanical ventilation, preoperative ICU, preoperative ECMO, preoperative creatinine, surgeon, cardiopulmonary bypass duration, crossclamp duration, type of surgery (ASO or ASO + arch repair; ASO ± IAA or HA repair ± VSD repair), cardiac function on echocardiography performed in operating theater (normal, mild, moderate, severe, missing), chest open from operating theater, lactate on admission, venous saturation on admission (from right internal jugular vein), and vasoactive inotrope score on admission. In addition, the following 3 postbaseline covariates (within the first 6 h of ICU admission) will also be balanced using IPTW: greatest lactate, lowest venous saturation and greatest VIS score. We assumed no unmeasured confounding at baseline conditional on measured baseline prognostic factors.
Outcomes	Primary outcome: Duration of mechanical ventilation from admission to intensive care and during their primary intensive care stay. Mechanical ventilation is defined as being endotracheally intubated and ventilated. Secondary outcome(s): Duration of intensive care hospital stay.	Same as for specification.
Follow-up	From time of ICU admission after surgery to discharge (or death) from ICU.	Same as for specification.
Causal contrasts	Which counterfactual contrasts will be estimated using the study data? • Intention-to-treat effect (effect of being assigned to treatment) • Per-protocol effect (effect of receiving treatment as indicated in the protocol)	Observational analogue of the per protocol effect. Valid estimation of the per-protocol effect will be performed by balancing baseline and post baseline prognostic factors within 6 hours of ICU admission (see section notes under treatment assignment). This balancing will be performed using IPTW.
Statistical analysis	Intention-to-treat analysis Per-protocol analysis (requires adjustment for preassignment and postassignment confounders).	Same as per-protocol analysis. Negative binomial regression with robust std errors to calculate incidence rate ratio for duration of mechanical ventilation, intensive care and hospital length of stay. IPTW produced using baseline and post-baseline (within 6 h) covariates associated with treatment decision will be included in the regression model. Missing data were handled by creating a category for missing observations within the variable (in the case of a continuous variable, it was categorized along with the introduction of a category for the missing observations). Appropriate transformation of the variables was performed to ensure adequate balance between study groups. Subgroup analysis: see analysis plan for further details.

The structure of the table of Target-Trial emulation using observational data was taken from Hernan 2021.^E1PD, Peritoneal dialysis; ICU, intensive care unit; IPTW, inverse probability of treatment weighting; ECMO, extracorporeal membrane oxygenation; ASO, arterial switch operation; IAA, interrupted aortic arch; HA, hypoplastic arch; VSD, ventricular septal defect.

Table E2.

Alternate definition of early PD (using 6 hours from the end of cardiopulmonary bypass) and study outcomes

Postoperative outcome	Unadjusted analysis		Adjusted analysis		Early PD/control: Adjusted incidence rate ratio (95% CI)
	Early PD	Control	Early PD	Control
	Mean ± SD		Mean (95% CI)
Length of mechanical ventilation, h	101.2 (85.2)	120.6 (241.5)	90.9 (74.2-107.6)	143.9 (99.9- 187.9)	0.63 (0.44-0.90)
Length of ICU stay, d	6.1 (4.5)	7.6 (16.2)	5.5 (4.6-6.4)	9.3 (5.8-12.8)	0.59 (0.39-0.89)
Length of hospital stay, d	19.8 (17.6)	17.2 (24.3)	19.1 (14.8-23.4)	20.1 (14.9-25.4)	0.95 (0.67-1.34)

PD, Peritoneal dialysis; SD, standard deviation; CI, confidence interval.

Table E3.

Characteristics of infants who received ECMO (n = 45)

Characteristic	Early PD (n = 15)	Control (n = 30)
Time to commence PD after ICU admission, h	1.5 (0.8 – 3.9)	11.8 (6.8 – 23.5)
PD duration, h, in the first 24 h	18 (9 – 19)	0 (0 – 8)
PD duration, h, after 24 h	28 (0 – 70)	19.5 (0 – 58)
Duration of peritoneal free drainage, h, in the first 24 h	0 (0 – 9)	3.5 (0 – 15)
Timing of ECMO commencement
Postoperative ECMO	5 (33)	19 (63)
Pre- and postoperative ECMO	2 (13)	1 (3)
Preoperative ECMO	8 (53)	10 (33)
Duration of mechanical ventilation, h, mean (SD)	164 (96)	455 (598)
Duration of intensive care stay, d, mean (SD)	9.7 (5.8)	28.9 (41.4)
Duration of hospital stay, d, mean (SD)	18.3 (10)	47.5 (61)

Data reported as median (interquartile range) or no (%) unless otherwise stated. PD, Peritoneal dialysis; ECMO, extracorporeal membrane oxygenation; SD, standard deviation.

Table E4.

Study outcomes (excluding those who died in ICU) and association with early PD

Postoperative outcome	Unadjusted analysis		Adjusted analysis		Early PD/control: Adjusted incidence rate ratio (95% CI)
	Early PD	Control	Early PD	Control
	Mean ± SD		Mean (95% CI)
Length of mechanical ventilation, h	100.8 (80)	108.4 (14)	87.6 (74.2-101)	128.4 (89-167.7)	0.68 (0.48-0.96)
Length of ICU stay, d	6 (4.4)	6.9 (15)	5.27 (4.6-5.9)	8.2 (5.3-11.1)	0.64 (0.44-0.93)
Length of hospital stay, d	18.5 (15.7)	16.7 (23.6)	16 (13.8-18.4)	19.3 (14.1-24.4)	0.83 (0.61-1.13)

Total n = 312 after excluding 6 infants who died in ICU. PD, Peritoneal dialysis; CI, confidence interval; SD, standard deviation; ICU, intensive care unit.

(See target trial specification used in the study in Table E1.)

Handling of Selection Bias

Observational studies that did not specify a target trial may have led to optimistic estimates of the benefit of the intervention.^E1 This may have been the case in previously published studies reporting on the role of early peritoneal dialysis (PD).E2, E3, E4 For example in previous studies, early PD was compared with late PD; this meant that participants not started on PD or those on passive peritoneal drainage were not included in the analysis.

Sampling bias due to nonparticipation is a major problem in randomized controlled trials, particularly when unblinded and testing a complex intervention (like PD). Nonparticipation can occur when participants deny consent (a common scenario), when participants require emergency treatment (no time to consent) or other reasons. In all of these situations, it is likely the disease severity (and outcome) may differ between participants and nonparticipants. The study intervention is likely to work within a particular range of disease severity (termed a sweet spot),^E5 and this range might be poorly represented in the recruited population. Sampling bias can produce misleading estimates of a study intervention in the wider population of interest; in essence a major threat to external validity. Target trial emulation in this situation, can eliminate such a bias by including all participants meeting the eligibility criteria.

Handling of Time-Zero

Time-zero is often variably defined in observational studies and this can lead to variable estimates. In this study, time-zero is defined as the time of admission to intensive care after surgery when eligibility is confirmed. Treatment assignment (to early PD or not) will occur within the first 6 hours. Immortal time bias is unlikely in this setting because in the time-interval between intensive care unit (ICU) admissions to 6 hours after ICU admission, no infant is likely to be extubated (primary outcome) discharged to from ICU.

Handling Confounding Bias

The study minimizes confounding bias by including a sufficiently rich confounder adjustment set. A wide range of baseline demographic, clinical and treatment factors which are present at the time to admission to intensive care and have prognostic significance were collected from hospital medical records. These factors were balanced between the treatment assignment groups before proceeding with the analysis of the outcome.

Appendix E1. Supplemental Analysis for Mortality

In an adjusted analysis (inverse probability treatment weighted analysis), there was no evidence of an association between the use of early peritoneal dialysis (PD) and mortality. For death in intensive care unit: Early PD versus control, odds ratio: 0.70 (95% confidence interval, 0.13-3.83). For death in hospital: early PD versus control: odds ratio, 0.64 (95% confidence interval, 0.15-2.78).