Individualizing Care in Severe Bronchopulmonary Dysplasia: A Series of N-of-1 Trials Comparing Transpyloric and Gastric Feeding

Abstract

Objective:

Compare rates of hypoxemia during transpyloric and gastric feedings in very preterm infants with severe bronchopulmonary dysplasia.

Design:

N-of-1 multiple crossover trial with individual patient and pooled data analyses.

Setting:

Level IV intensive care nursery.

Patients:

Infants receiving positive airway pressure between 36-55 weeks postmenstrual age were enrolled 12/2014-7/2016.

Intervention:

N-of-1 trial consisting of 2 blocks, each with a 4-day gastric and 4-day transpyloric feeding period assigned in random order.

Main Outcome Measures:

The primary outcome was the frequency of daily intermittent hypoxemic events (SpO₂ ≤80% lasting 10-180s). Secondary outcomes included the daily proportion of time with an SpO₂ ≤80% and mean daily fraction of inspired oxygen.

Results:

Of 15 infants, 13 completed the trial and 2 stopped early for transient worsening in respiratory status during gastric feedings. In the intention-to-treat analyses, transpyloric feedings resulted in increased rates of intermittent hypoxemia in 5 infants, greater time per day in hypoxemia in 3 infants, and more supplemental oxygen use in 3 infants. One infant received more supplemental oxygen during gastric feedings. The remaining study outcomes were similar between the feeding routes in all other infants. Pooling all data, transpyloric feedings resulted in a higher frequency of intermittent hypoxemic events (median 7.5/day [IQR 1-23.5] vs. 3/day [1-11]; adjusted incidence rate ratio 1.8, 95% CI 1.3-2.5) and a greater proportion of daily hypoxemia time (median 0.8% [IQR 0.1-2.3] vs. 0.4% [0.07-1.8]; adjusted mean difference 1.6, 95% CI 1.1-2.5).

Conclusions:

Transpyloric compared to gastric feedings modestly increased rates of hypoxemia among study participants.

Introduction

Bronchopulmonary dysplasia (BPD) predisposes very preterm infants to severe, enduring cognitive and physical impairments.^1–4 Gastroesophageal reflux (GER) and pulmonary aspiration may contribute to these adverse sequelae by inducing lung inflammation, bronchospasm, and recurrent hypoxemia.^5–13 Transpyloric feedings reduce aerodigestive sequelae of GER in older children and adults.^14,15 However, the safety and efficacy of this feeding practice in infants with BPD is uncertain. The limited data from randomized trials conducted in preterm newborns show that early use of transpyloric compared to gastric feedings does not prevent aspiration pneumonia and may increase mortality risk.¹⁶ In contrast, recent observational studies found that transpyloric feedings were associated with less frequent apnea and bradycardia in preterm infants and a lower risk of death or BPD.^17–19 These data also suggest that any potential benefit with transpyloric feedings may vary between infants.¹⁸

Because of the heterogeneity of GER in preterm infants^20,21 and potential intra-patient variability in the response to transpyloric feedings,¹⁸ individualized feeding route selection in infants with BPD may be optimal. N-of-1, or single patient trials, are a methodologically robust means to select a treatment strategy in individual patients when the preferred therapy may vary between subjects.^22–24 Data from N-of-1 trials can also be pooled to estimate overall treatment effects. We report a series of N-of-1 trials comparing the safety and efficacy of gastric and transpyloric feedings in infants with severe BPD at the individual and group levels.

Methods

Study Infants

Infants born <32 weeks’ gestation with severe BPD²⁵ who were receiving positive airway pressure and full enteral nutrition at 36^0/7 to 55^6/7 weeks postmenstrual age were eligible. Infants with clinical signs attributed to gastroesophageal reflux (GER) by their medical team (e.g. frequent regurgitation, suspected aspiration) were targeted for enrollment. No infants underwent diagnostic evaluation for GER, as this is not part of our routine practice. Infants with recent, significant changes to their medical care (e.g. change in mode of ventilation, new medication), a gastric fundoplication, or severe congenital malformation were excluded. Informed consent was obtained from a parent or guardian of each infant. The Institutional Review Board at the Children’s Hospital of Philadelphia approved this study. The trial was registered at clinicaltrials.gov ( NCT 02142621).

N-of-1 Trial Design

Each N-of-1 trial lasted 16 days and consisted of two blocks. Each block consisted of a 4-day gastric and 4-day transpyloric feeding period assigned in random order using a computer-generated sequence. A “washout period” was not utilized between feeding periods owing to the need to provide enteral nutrition throughout the study.

Intervention

All feedings were administered using 6f CORFLO nasogastric/nasointestinal feeding tubes (Halyard Health, Alpharetta, GA) placed by nursing staff. Gastric tube position was confirmed by aspirate pH testing (acceptable pH≤5) or abdominal radiograph if the pH result was inconclusive. Transpyloric tubes were placed in the second portion of the duodenum and confirmed by radiograph. Feeding route allocation could not be blinded as clinicians needed to monitor the feeding tube location. Gastric feedings were run continuously for the first 24 hours of each assigned period. Thereafter, clinicians were allowed to condense feedings. Transpyloric feedings were run continuously throughout the study. All other care decisions were at the discretion of the medical team. However, clinicians were asked to avoid making non-urgent changes in medical management during the trial. Following each trial, the study team discussed the results with the medical team and provided a summary report (online supplementary Figure 1). The treating physician selected the post-study feeding route.

Outcomes

The primary outcome was the daily number of intermittent hypoxemic events defined as a peripheral oxygen saturation (SpO₂) ≤80% lasting between 10 seconds and 3 minutes.^6,7 The secondary outcomes were the proportion of time per day with an SpO₂ ≤80%, the daily number of prolonged hypoxemic events (SpO₂ ≤80% lasting ≥60 seconds)¹³, and the daily time-averaged fraction of inspired oxygen (FiO₂). All SpO₂ data were recorded using a Masimo Rad-8 pulse oximeter (Masimo, Irvine, CA) set to 2-second averaging and sampling rates.²⁶

Occurrence of the following pre-specified adverse events were recorded: (1) absolute increase in FiO₂ >20% above baseline for ≥8 hours, (2) need for non-elective endotracheal intubation; (3) aspiration pneumonia defined as an observed regurgitation event with a new infiltrate on chest radiograph, increased respiratory support needs, and antibiotic therapy; (4) bag-valve ventilation for an acute hypoxemic event; (5) water loss stools with associated electrolyte abnormalities; (6) necrotizing enterocolitis or intestinal perforation; (7) and death. An independent data safety monitor reviewed the study data at half enrollment and adjudicated all adverse events.

Oximeter Data and Statistical Analysis

SpO₂ data were downloaded from the study oximeters every 72 hours. Prior to any statistical analyses, oximeter values with a 0 pulse rate or a “Low Signal IQ” exception message were removed. The primary analyses utilized an intention-to-treat approach. Pre-specified as-treated analyses were also performed. Mixed effects negative binomial regression models were used to assess the relationship between the feeding route and rates of intermittent and prolonged hypoxemic events in the individual and pooled data. The daily proportion of time spent hypoxemic and the mean daily time-weighted FiO₂ were log transformed and assessed using mixed effects linear regression models. A dichotomous variable distinguishing the first and second treatment blocks was included in all models to adjust for temporal variation in disease severity. The data for one infant with zero hypoxemia time observed on 2 days were replaced with the lowest percentage of hypoxemia time in the data set (0.003%) to generate analyzable log transformed results for those days. To account for within subject correlation of repeated outcome measures, a random effect for each patient with an unstructured covariance matrix and robust variance estimates was used in the mixed-effects models for the pooled analyses.

We assessed infant-level heterogeneity of treatment effects for the primary study outcome in two ways. Firstly, we tested the statistical significance of a treatment-by-participant interaction term added to the pooled-data regression model. Secondly, we tested the equivalence of the infant-level treatment effect estimates using Cochran’s chi-square and quantified the proportion of variation due to infant-level heterogeneity with the I² statistic in a fixed effect inverse variance model.^27,28 All statistical testing was conducted using Stata/SE, version 15.1 (StataCorp, College Station, TX).

Sample Size Determination

The sample size calculation for the pooled analysis utilized the following parameters: 2 paired crossovers, a within subject outcome correlation of 0.7, and an intra-subject variance in the frequency of intermittent hypoxemic events of 84.^7,29 Based on these assumptions, 15 infants provides 90% power to detect a 13% relative reduction in the daily number of intermittent hypoxemic events between the two feeding routes (see additional description of the sample size calculation in the online supplement).

Results

Study Participants

A total of 15 infants were enrolled between December 2014 and July 2016 (Figure 1). The majority were male, receiving non-invasive positive airway pressure, and clinically prescribed transpyloric feedings at enrollment (Table 1). Five infants (27%) received a proton pump inhibitor or histamine-2 antagonist during their N-of-1 trial. Pulse oximetry data were available for 77% of study period after excluding low perfusion readings and times when the oximeter was not attached to the infant. The proportion of analyzable data was similar between the feeding routes.

Figure 1.

Study flow diagram

Table 1.

Characteristics of the 15 study participants

Gestational age, wk - median (IQR)	26.0 (24.6-27.9)
Birth weight, g – median (IQR)	750 (640-930)
Male, n (%)	12 (80%)
Outborn, n (%)	14 (93)
Postmenstrual age at enrollment, wk – median (IQR)	47.3 (39.3-53.1)
Weight at enrollment, g – median (IQR)	4400 (2923-5260)
Feeding route immediately prior to enrollment, (%)
Gastric	5 (33)
Transpyloric	10 (67)
Respiratory support at enrollment, (%)
High flow nasal cannula	6 (40)
Nasal CPAP	6 (40)
Mechanical ventilation	3 (20)
Gastric acid suppression medication use at enrollment, n (%)	5 (27)

Individual Patient Analyses

Five infants (20%) (no. 4, 6, 7, 13, and 14) experienced a significantly higher frequency of intermittent hypoxemic events per day while receiving transpyloric compared to gastric feedings (Figure 2). Rates of intermittent hypoxemia were similar between the two feeding routes in the remaining 10 infants. There was evidence of infant-level heterogeneity of treatment effects for this outcome (treatment-by-participant interaction term p<0.001; I²=49.8, p=0.015). Transpyloric feedings were also associated with a higher frequency of prolonged hypoxemic events in two infants (no. 4, 13) and a greater proportion of time per day with an SpO₂ ≤80% in 3 infants (no. 2, 6, 13) (online supplementary Table 1). Three infants (no. 10, 12, 15) received more supplemental oxygen during transpyloric feedings; one (no. 5) received more during gastric feedings.

Figure 2. Primary outcome results for the individual patient and pooled analyses (shaded box).

Incidence rate ratios (IRR) calculated using mixed effects negative binomial regression. A dichotomous variable distinguishing treatment blocks 1 and 2 was included in each model to account for temporal variation in disease severity over time. IRR values indicate the average relative increase (IRR >1) or decrease (IRR<1) in the daily number of intermittent hypoxemic events with transpyloric compared to gastric feedings. Stars (*) show the approximate timepoints these infants were removed from the intervention portion of the trial and transpyloric feedings were administered thereafter. All study participants received gastric feedings run continuously when this feeding route was assigned except for infants no 12 and 14 who received bolus gastric feedings after the first 24 hours of each gastric feeding period. CI, confidence interval.

Adverse Events and As-Treated Analyses

No pre-defined adverse events were observed. However, the trial intervention was stopped early in two infants (no. 4 and 9) for clinical worsening during gastric feedings. Infant no. 4 received transpyloric feedings prior to enrollment. His oxygen requirement acutely worsened after approximately 48 hours of initial gastric feedings. This change was attributed to a clinically suspected post-emesis aspiration event that did not meet the pre-defined stopping rule for aspiration pneumonia. Gastric feedings were subsequently discontinued and transpyloric feedings resumed. The intention-to-treat analysis for this subject showed a higher frequency of intermittent hypoxemic events during transpyloric feedings (Figure 2). The as-treated analysis showed no difference between the feeding routes (IRR 3.0, 95% CI 0.81-11.4). Infant no. 9 was removed from the trial during the second gastric feeding period following an acute rise in FiO₂. The independent data safety monitor determined this clinical change most likely resulted from a concomitant decrease in the ventilator mean airway pressure. After a brief period without enteral nutrition, the medical team chose to administer transpyloric feedings. Rates of intermittent hypoxemia in this infant were similar between the two feeding routes in the intention-to-treat and as-treated analyses.

Post-Trial Feeding Route and Discharge Status

Four of the 5 infants whose intention-to-treat analyses showed greater intermittent hypoxemia during transpyloric feedings were prescribed gastric feeding by their medical team following the N-of-1 trial (Table 2). This was a change from clinically prescribed, pre-trial transpyloric feedings in 2 of the 4 infants. Of the 10 infants with equivocal trial results for the primary outcome, 6 were prescribed post-trial gastric feedings (5 of whom received pre-trial transpyloric feedings) and 4 were prescribed transpyloric feedings (2 of whom received pre-trial transpyloric feedings). Rates of gastric fundoplication or jejunal tube feedings at discharge were similar among the infants whose N-of-1 trials favored gastric feedings and those with equivocal results (60% vs. 50%; p=1.0).

Table 2 –

Individual patient feeding route pre and post N-of-1 trial

Infant	N-of-1 trial primary outcome result	Pre-trial Feeding route	Post-trial Feeding route	Feeding route at NICU discharge	Respiratory support at NICU discharge
1	Equivocal	Transpyloric	Gastric	Nasogastric tube	Room air
2	Equivocal	Transpyloric	Gastric	Oral	Room air
3	Equivocal	Transpyloric	Gastric	GT/Fundoplication	Nasal cannula
4	Gastric Bettera	Transpyloric	Transpyloric	Jejunal tube	Tracheostomy
5	Equivocal	Gastric	Gastric	Nasogastric tube	Nasal cannula
6	Gastric Better	Gastric	Gastric	GT/Fundoplication	Tracheostomy
7	Gastric Better	Transpyloric	Gastric	Gastrostomy tube	Nasal cannula
8	Equivocal	Transpyloric	Gastric	GT/Fundoplication	Nasal cannula
9	Equivocal	Transpyloric	Transpyloric	GT/Fundoplication	Tracheostomy
10	Equivocal	Transpyloric	Transpyloric	Care Withdrawn	Care Withdrawnb
11	Equivocal	Gastric	Transpyloric	GT/Fundoplication	Nasal cannula
12	Equivocal	Gastric	Transpyloric	Jejunal tube	Nasal cannula
13	Gastric Better	Transpyloric	Gastric	GT/Fundoplication	Nasal cannula
14	Gastric Better	Gastric	Gastric	Oral	Room air
15	Equivocal	Transpyloric	Gastric	Oral	Nasal cannula

^aAs-treated analysis showed equivocal results

^bMedical care was withdrawn long after the N-of-1 trial in this infant owing to significant unremitting cardiopulmonary disease

Abbreviations: GT, gastrostomy tube; NICU, neonatal intensive care unit

Pooled Data Analyses

In the pooled analyses, transpyloric compared to gastric feedings significantly increased the frequency of intermittent hypoxemic events and the proportion of time per day with an SpO₂ ≤80% (Figure 2, Table 3). Stated as time, the differences in SpO₂ levels equated to an average of 3 additional minutes (95% CI 0.3 to 8 minutes) of hypoxemia per day during transpyloric feedings. The feeding route was not associated with differences in FiO₂ (Table 3). Results of the pooled as-treated analyses were similar to the intention-to-treat analyses (data not shown).

Table 3.

Pooled study results

Outcome	Gastric feedings Median (IQR)	Transpyloric feedings Median (IQR)	IRR for transpyloric feedings (95% CI)a
Intermittent hypoxemic events (no. per day)	3 (1-11)	7.5 (1-23.5)	1.8 (1.3, 2.5)
Prolonged hypoxemic events (no. per day)	0 (0-1)	0 (0-2.5)	1.9 (1.1, 3.2)
			Mean change with transpyloric feedings (95% CI)b
Proportion of time with SpO2 ≤ 80% per day (%)	0.4 (0.07-1.8)	0.8 (0.1-2.3)	1.6 (1.1, 2.5)
Mean time-weighted daily FiO2 (%)	29.2 (23.7-34.5)	29.0 (24.0-35.9)	1.0 (1.0, 1.0)

^aMixed effects negative binomial regression models were used to assess the relationship between the feeding route and rates of intermittent and prolonged hypoxemic events. IRR values indicate the average relative increase in the daily number of intermittent or prolonged hypoxemic events with transpyloric compared to gastric feedings.

^bThe daily proportion of time spent hypoxemic and the mean daily time-weighted FiO2 were log transformed and assessed using mixed effects linear regression models. Values shown for these outcomes are back-transformed (exponentiated) model parameter estimates. They indicate the average relative increase (for values >1) in the proportion of time in hypoxemia per day and the time-weighted daily FiO₂ exposure during transpyloric compared to gastric feedings.

A random effect from each individual patient and an unstructured covariance matrix with robust variance estimates were used in all models. A dichotomous variable distinguishing the first and second treatment blocks was included in the models to account for temporal variation in disease severity over time. Use of these model parameters account for the differences in the ratios of the median values for the study outcomes and the estimates of treatment effects shown in the table.

Abbreviations: CI, confidence interval; FiO₂, fraction of inspired oxygen; IQR, interquartile range; IRR, incidence rate ratio; SpO₂ peripheral capillary oxygen saturation level

Discussion

We conducted a series N-of-1 trials to compare the safety and efficacy of transpyloric to gastric feedings in individual very preterm infants with severe BPD. Transpyloric feedings resulted in a modest, but statistically significant increase in the rates intermittent hypoxemia in one third of study participants. No infants experienced more hypoxemia during gastric feedings. When pooling the data for the entire study cohort, transpyloric feedings resulted in greater hypoxemia and no difference in the prescribed FiO₂.

We compared transpyloric and gastric feedings using N-of-1 trials rather than a parallel-arm group trial to enable real-time, individualized therapy selection. N-of-1 trials are a preferred option to personalize treatment choices when the following conditions are met: a relatively stable chronic illness, a therapy with heterogeneous treatment effects, availability of a clinically relevant short-term endpoint, a treatment effect that is expected to persist with continued use of the therapy, and a lack of adequate evidence from standard group trials establishing a preferred treatment.²² Transpyloric feeding in severe BPD meets these requirements. Severe BPD is a chronic illness marked by prolonged periods with “stable” physiology. GER with pulmonary aspiration likely exacerbates respiratory disease in some but not all infants with severe BPD.³⁰ The utility of transpyloric feedings may vary among infants.¹⁸ Hypoxemia is easily measured in real-time and observational data suggest that transpyloric feeding may reduce episodic apnea and bradycardia within 72 hours in preterm infants.¹⁷ Pathologic GER persists throughout the first year of life in some infants with severe BPD, suggesting control of GER will provide stable treatment effects over time.³¹ Finally, there are no proven safe and effective treatments for GER in preterm infants.³²

Our study findings raise concern that transpyloric feedings may carry potential for harm in some infants with severe BPD. The etiology of the apparent adverse effects is unclear. Data from chronically ill young children suggest that transpyloric feeding does not eliminate GER.³³ Moreover, gastric feeding is an important buffer of gastric pH in infants and may act to reduce the acidity and irritability of esophageal refluxate.³⁴ It is possible that GER occurring during transpyloric compared to gastric feedings has a higher concentration of gastric acid and injurious enzymes. Transpyloric feeding is also not physiologic. Rapid entry of fortified, high osmolarity nutrition into the jejunum can induce bloating and dumping. We did not observe significant diarrhea or electrolyte abnormalities during transpyloric feedings, but it is possible that more subtle effects on intestinal function could impact patient comfort and clinical stability. Although the existing neonatal trial data primarily include younger infants, these studies showed increased gastro-intestinal disturbances necessitating temporary cessation of enteral nutrition among infants randomized to receive transpyloric feedings.¹⁶ Variability in any of these multiple physiologic processes could explain why some but not all study participants experienced more hypoxemia during transpyloric compared to gastric feedings.

We acknowledge several study limitations. Firstly, the trials were conducted in a small and select cohort of infants. Although our findings suggest clinicians should carefully consider the potential risks of transpyloric feedings in infants with severe BPD, larger studies are needed to fully characterize the safety and efficacy of this feeding method in preterm infants. Secondly, the study endpoints focused on oxygen saturation stability and not the frequency or severity of GER or pulmonary aspiration. Continuous esophageal pH-multiple impedance monitoring may help evaluate the utility of transpyloric feeding. However, this diagnostic tool is not commonly used in our unit as it does not differentiate benign GER from esophageal reflux that may contribute to pulmonary aspiration and exacerbation of BPD. Thirdly, we did not include a “washout period” between feedings routes as it was necessary to maintain enteral feedings throughout the study. Whether the effects of one feeding route carried over to the other is unknown. To monitor the feeding tube location, investigators and clinicians were also not masked to the assigned feeding route. Lastly, 4-day trial periods may not provide sufficient time to fully characterize the risks and benefits of transpyloric feedings. The study duration in N-of-1 trials must be sufficiently long to provide meaningful information about a therapy but short enough to allow adequate post-trial time to positively influence patient care. The 16-day trial used in this study was chosen to balance these considerations. Future studies should explore the effects of transpyloric feedings administered over longer periods and in infants with objectively diagnosed GER.

In conclusion, this series of N-of-1 trials found that transpyloric feedings, on average, led to greater oxygen saturation instability among infants with severe BPD. Subsequent studies should investigate the physiologic effects of transpyloric feedings in preterm infants and further evaluate the safety and efficacy of this feeding strategy.

What is known about this topic?

• Gastroesophageal reflux and pulmonary aspiration may contribute to hypoxemia in some very preterm infants with severe bronchopulmonary dysplasia (BPD).

• Transpyloric feeding reduces the risk of aspiration pneumonia among adults receiving invasive mechanical ventilation; the utility in infants with severe BPD is uncertain and may vary between infants.

• N-of-1, or single-patient, randomized crossover trials are a methodologically robust means to estimate treatment effects within individual patients and to personalize care.

What this study adds?

• Transpyloric feeding increased rates of hypoxemia among some very preterm infants with severe BPD; no infants demonstrated less hypoxemia with transpyloric feedings.

• N-of-1 trials are feasible to conduct in the intensive care nursery and may help individualize care in infants with severe BPD.