Association of Time of First Corticosteroid Treatment with Bronchopulmonary Dysplasia in Preterm Infants

Abstract

Objective:

To evaluate the association between the time of first systemic corticosteroid initiation and bronchopulmonary dysplasia (BPD) in preterm infants.

Study design:

A multi-center retrospective cohort study from January 2010 to December 2016 using the Children’s Hospitals Neonatal Database and Pediatric Health Information System database was conducted. The study population included preterm infants <32 weeks’ gestation treated with systemic corticosteroids after seven days of age and before 34 weeks’ postmenstrual age. Stepwise multivariable logistic regression was used to assess the association between timing of corticosteroid initiation and the development of Grade 2 or 3 BPD as defined by the 2019 Neonatal Research Network criteria.

Results:

We identified 598 corticosteroid-treated infants (median gestational age 25 weeks, median birth weight 760 grams). Of these, 47% (280/598) were first treated at 8–21 days, 25% (148/598) were first treated at 22–35 days, 14% (86/598) were first treated at 36–49 days, and 14% (84/598) were first treated at > 50 days. Infants first treated at 36–49 days (aOR 2.0, 95% CI 1.1–3.7) and > 50 days (aOR 1.9, 95% CI 1.04–3.3) had higher independent odds of developing Grade 2 or 3 BPD when compared to infants treated at 8–21 days after adjusting for birth characteristics, admission characteristics, center, and co-morbidities.

Conclusions:

Among preterm infants treated with systemic corticosteroids in routine clinical practice, later initiation of treatment was associated with a higher likelihood to develop Grade 2 or 3 BPD when compared to earlier treatment.

INTRODUCTION

Bronchopulmonary dysplasia (BPD) is one of the most common serious morbidities of prematurity, affecting over 40% of infants with birth weight less than 1500 grams.^1,2 Postnatal systemic corticosteroids are among the few medications that reduce the risk of developing BPD in randomized controlled trials.^3–6 Because of safety concerns on the developing brain, current clinical practice typically limits corticosteroid treatment to preterm infants at the highest risk for BPD.^7,8 However, the optimal time to initiate corticosteroid treatment in these high-risk infants remains unknown.

Studies evaluating the relationship between timing of corticosteroid initiation and outcomes are limited. While postnatal corticosteroids are generally not recommended in the first seven days of life due to unfavorable effects on neurodevelopment⁹, it is unclear whether the timing of corticosteroid administration in infants older than seven days alters the likelihood of developing severe BPD.¹⁰ A single-center retrospective study by our group¹¹ found an association between later timing of postnatal corticosteroid treatment (>28 days) and longer duration of mechanical ventilation, but this study was underpowered to detect differences in BPD. A larger study by Harmon et al.¹², which included infants from the Neonatal Research Network between 2006 to 2012, demonstrated that first initiation of corticosteroid treatment ≥ 50 days of life was associated with increased odds of severe BPD. However, the shift towards less restrictive corticosteroid recommendations by the American Academy of Pediatrics (AAP) in 2010⁸ may limit the applicability of Harmon’s study to contemporary practice.

In this study, we sought to validate the findings from our previous single-center study by investigating the association between variation in timing of corticosteroid initiation and BPD in a recent, multi-center cohort of preterm infants treated with corticosteroids. Our hypothesis is that, among preterm infants treated with corticosteroids after seven days of life, later treatment is associated with an increased risk of BPD at 36 weeks’ postmenstrual age (PMA) when compared to earlier treatment.

MATERIALS AND METHODS

Data Sources

We used data from the Children’s Hospitals Neonatal Database (CHND) linked with the Pediatric Health Information System (PHIS) database for this study. The CHND contains data from infants admitted to 34 participating tertiary referral NICUs of the Children’s Hospitals Neonatal Consortium (CHNC)^13–15 – a national collaborative of regional NICUs¹⁶ that seeks to optimize clinical practice through benchmarking and critical appraisal of areas with significant practice variability. CHND data are collected by trained chart abstractors who undergo regular assessments for inter-rater reliability scores.¹⁷ The PHIS database is an administrative database that contains daily, patient-specific data on clinical and resource utilization from more than 44 freestanding, tertiary care US children’s hospitals.¹⁸ The linked dataset combining CHND and PHIS records was de-identified, and our research was approved by the Institutional Review Board of Stanley Manne Research Institute (affiliated with the Ann and Robert H. Lurie Children’s Hospital of Chicago), as well as by the local Institutional Review Board of each participating institution.

Selection of Patients

Preterm infants born < 32 weeks’ gestational age who were admitted to CHNC and PHIS-participating centers at ≤ 14 days of life between January 2010 and December 2016 were eligible for inclusion in the study cohort. Infants who had a length of stay of < 3 days, who died ≤ 14 days of life, who had significant congenital anomalies, or who were transferred to another facility prior to hospital discharge were excluded. We identified infants treated with systemic corticosteroids using PHIS codes for dexamethasone (154035) or hydrocortisone (154063) and with additional validation by corresponding CHND code for systemic corticosteroid treatment. Because of uncertainty in determining timing of first corticosteroid use, we excluded infants who received corticosteroids prior to admission. Infants who received mixed dexamethasone and hydrocortisone were likewise excluded as determination of which corticosteroid came first can be difficult to accurately ascertain from the PHIS database. We also excluded infants who received systemic corticosteroids in the first week of life to decrease confounding from prophylactic strategies of corticosteroids for BPD. As our primary outcome, BPD diagnosis, was assessed at 36 weeks’ PMA, we also excluded infants treated with corticosteroids after 34 weeks’ PMA as we felt these treatments were unlikely to alter/mitigate the diagnosis of BPD.

Exposure

We categorized the primary exposure, the timing of initial systemic corticosteroid treatment, into four groups (8–21 days; 22–35 days; 36–49 days; and ≥ 50 days) based on the postnatal age in days when the first dose of corticosteroids was given. These categories were selected to balance group sample sizes and to approximate the time periods used in the prior study by Harmon et al.¹²

Covariates

We evaluated baseline maternal and infant demographic variables including maternal race, multiplicity of birth, gestational age, small for gestational age (weight <10^th percentile for gestation), sex, and Apgar score at 5 minutes. Variables related to admission to CHNC center including admission by 3 days of life, reason for admission (i.e. prematurity, surgical evaluation, respiratory failure, or other), and mechanical ventilation status on admission were also assessed. Data was also collected on important neonatal co-morbidities including patent ductus arteriosus ligation, necrotizing enterocolitis Bell Stage ≥ 2 or spontaneous intestinal perforation, bloodstream infection, and intraventricular hemorrhage grade 3 or 4.

Outcomes of Interest

The primary outcome was Grade 2 or Grade 3 BPD as defined by the 2019 Neonatal Research Network criteria.¹⁹ In this definition, infants at 36 weeks’ PMA with respiratory support of nasal cannula >2 liters per minute or noninvasive positive airway pressure support were categorized as Grade 2 BPD, while infants supported by invasive mechanical ventilation were categorized as Grade 3 BPD. The secondary outcomes were inpatient mortality, total duration of invasive mechanical ventilation (days) for surviving infants, and composite outcome of death or BPD. Subgroup analysis was performed to explore whether associations between timing of treatment and BPD differed based on type of corticosteroid given (dexamethasone versus hydrocortisone). We also evaluated center variation in timing of steroid initiation and its association with BPD and total ventilator days.

Statistical analysis

Descriptive data were reported as counts with percentages and medians with interquartile ranges, as appropriate. Comparisons between steroid initiation cohorts were performed using the Chi-square test and Fisher’s Exact test for categorical variables or the Kruskal-Wallis and Wilcoxon Rank Sum test for continuous variables. We used stepwise, multivariable logistic regression to estimate the association between the timing of corticosteroid initiation and other individual risk factors with dichotomous outcomes.²⁰ The association of total ventilation days and timing of corticosteroids was analyzed with a generalized linear model for gamma distribution and log link.²¹ For all models, we introduced covariates in stages with birth characteristics (race, multiplicity of birth, gestational age, small for gestational age, sex, and Apgar score <4 at 5 minutes) in the first stage; admission characteristics (admission by 3 days of life, reason for admission, and mechanical ventilation status on admission) in the second stage; and co-morbidities (patent ductus arteriosus ligation, necrotizing enterocolitis or spontaneous intestinal perforation, bloodstream infection, and intraventricular hemorrhage) in the third stage. Factors that remained significant were included in the final stage along with timing of corticosteroid initiation and center as an independent effect. Adjusted odds ratios (aOR) with 95% confidence intervals (CI) were calculated, with infants treated at 8–21 days as the reference group. Stepwise logistic regression was also used to perform the specified subgroup analysis, while Pearson correlation was used to evaluate the association between center rate of late corticosteroid treatment with center proportion of BPD and median ventilator days. All analyses were performed in SAS Enterprise Guide v7.1 (Cary, North Carolina).

RESULTS

We identified 598 steroid-treated infants eligible for analysis after applying study exclusion criteria (Figure 1). Demographic and clinical characteristics of the cohort are described in Table 1. Overall, no significant differences in baseline demographics were noted between the steroid initiation cohorts. The median time for initiation of systemic corticosteroid treatment was 23 days (interquartile range:14–37 days). Hydrocortisone was more commonly initiated at 8–21 days and decreased in frequency with subsequent postnatal age groups (22–35 days, 36–49 days, and ≥ 50 days), whereas dexamethasone was initiated less commonly at 8–21 days but was used more often in subsequent postnatal age groups (Table 1 and Figure 2).

Figure 1. Flow of participants.

Table 1.

Characteristics of preterm infants treated with systemic steroids, overall and by different time periods of first steroid initiation.

	Overall	Time of first steroid initiation				P value
		8–21 days	22–35 days	36–49 days	≥ 50 days
No. of patients	598	280	148	86	84
Gestational age	25 [24, 26]	25 [24, 27]	25 [24, 26]	25 [24, 26]	25 [24, 26]	0.27
Birth weight	760 [650, 928]	770 [662, 950]	750 [650, 930]	745 [650, 910]	769 [620, 895]	0.49
Gender
Female	247 (41.3)	109 (38.9)	71 (48.0)	30 (34.9)	37 (44.1)	0.17
Male	351 (58.7)	171 (61.1)	77 (52.0)	56 (65.1)	47 (56.0)	0.17
Race/ethnicitya
Black	199 (33.3)	90 (32.1)	58 (39.2)	21 (24.4)	30 (35.7)	0.12
Hispanic	83 (13.9)	41 (14.6)	21 (14.2)	12 (14.0)	9 (10.7)	0.88
White	270 (45.2)	131 (46.8)	59 (39.9)	44 (51.2)	36 (42.9)	0.32
Other	33 (5.5)	15 (5.4)	7 (4.7)	6 (7.0)	5 (6.0)	0.88
Small for gestational age	74 (12.4)	34 (12.1)	19 (12.8)	10 (11.6)	11 (13.1)	0.99
Multiple births	147 (24.6)	77 (27.5)	33 (22.3)	15 (17.4)	22 (26.2)	0.24
Antenatal steroidsb	327 (54.7)	149 (53.2)	82 (55.4)	47 (54.7)	49 (58.3)	0.87
Apgar score <4 at 5 minutes	136 (22.7)	70 (25)	28 (18.9)	23 (26.7)	15 (17.9)	0.27
Admitted ≤ 3 days of life	342 (57.2)	151 (53.9)	89 (60.1)	57 (66.3)	45 (53.6)	0.17
Intubated on admission	541 (90.5)	256 (91.4)	134 (90.5)	76 (88.4)	75 (89.3)	0.11
Reason for admission
Preterm birth without other co-morbidities	268 (44.8)	111 (39.6)	70 (47.3)	48 (55.8)	39 (46.4)	0.06
Surgical evaluation	207 (34.6)	105 (37.5)	44 (29.7)	27 (31.4)	31 (36.9)	0.37
Respiratory failure	52 (8.7)	23 (8.21)	19 (12.8)	7 (8.1)	3 (3.6)	0.17
Other	71 (11.9)	41 (14.6)	15 (10.1)	4 (0.05)	11 (13.1)	0.36
Type of steroid
Dexamethasone	284 (47.5)	62 (22.1)	75 (50.7)	68 (79.1)	79 (94.1)	<0.001
Hydrocortisone	314 (52.5)	218 (77.9)	73 (49.3)	18 (20.9)	5 (6.0)	<0.001
Co-morbidities among survivors
Bloodstream infection	192 (32.1)	101 (36.1)	43 (29.1)	23 (26.7)	25 (29.8)	0.26
Necrotizing enterocolitis or spontaneous intestinal perforation	215 (36.0)	108 (38.6)	46 (31.1)	31 (36.1)	30 (35.7)	0.50
Patent ductus arteriosus ligation	163 (27.3)	70 (25.0)	37 (25.0)	27 (31.4)	29 (34.5)	0.25
Severe intraventricular hemorrhage	139 (23.2)	69 (24.6)	32 (21.6)	16 (18.6)	22 (26.2)	0.58

Data are presented as counts (percentages) or median [interquartile range]. P values indicate analysis by Chi-square or Fisher’s Exact test for categorical variables and Kruskal Wallis for continuous variables.

^a13 infants had missing data on race/ethnicity

^b18 infants had missing data on antenatal steroids

Figure 2. Breakdown of corticosteroid use in each steroid initiation group.

Hydrocortisone use (shown in orange) was most common in the 8–21 days group and decreased in frequency with subsequent steroid initiation groups. Dexamethasone (shown in blue) was less commonly used at 8–21 days but was increasingly utilized in subsequent postnatal age groups.

The rate of Grade 2 or 3 BPD differed significantly between the different steroid initiation groups and was highest among infants with corticosteroid initiation at 36–49 days and at 50 days or more. When broken down into individual components, both Grade 2 BPD and Grade 3 BPD remained significantly different between groups (Figure 3). Stepwise logistic regression showed that compared to infants with corticosteroid initiation at 8–21 days, the odds of developing BPD was similar in infants treated at 22–35 days but higher in infants treated at 36–49 days (aOR 2.0, 95% CI 1.1–3.7) and 50 days or more (aOR 1.9, 95% CI 1.04–3.3) (Table 2). Other variables included in the final logistic regression model for BPD were small for gestational age and center. A similar pattern of increased total ventilation days among infants with later steroid initiation (36–49 days and ≥50 days) was found compared to infants treated at 8–21 days (Table 2).

Figure 3. BPD rates between steroid initiation groups.

Stacked column plot comparing BPD rates between different steroid initiation groups broken down by severity. Grade 2 BPD (shown in blue) was lower in the 22–35 days group compared to the 36–49 days group (23% vs 46%, P<0.001), while Grade 3 BPD (shown in orange) was lower in the 8–21 days group compared to the ≥50 days group (13% vs 27%, P=0.004). Bonferonni-corrected alpha level of 0.008 (0.5/6) was used to determine significance.

Table 2.

Adjusted analysis for BPD, total ventilation days, and death. Infants with steroid initiation at 8–21 days were selected as reference group.

Grade 2 or 3 BPD	n/N (%)	Adjusted odds ratio	95% CI
8–21 days	101/224 (45)	Reference	Reference
22–35 days	48/129 (37)	0.9	0.6–1.5
36–49 days	47/81 (58)	2.0	1.1–3.7
≥50 days	52/84 (62)	1.9	1.04–3.3
Total ventilation days	Median (IQR)	Adjusted odds ratio	95% CI
8–21 days	31 (19 – 51)	Reference	Reference
22–35 days	36 (27 – 48)	1.1	0.9 – 1.2
36–49 days	42 (35 – 53)	1.3	1.1 – 1.5
≥50 days	55 (33 – 74)	1.3	1.1 – 1.6
Death	n/N (%)	Adjusted odds ratio	95% CI
8–21 days	56/280 (20)	Reference	Reference
22–35 days	19/148 (13)	0.5	0.3 – 0.9
36–49 days	5/86 (6)	0.3	0.1 – 0.8
≥50 days	0/84 (0)	***	***
Death or BPD	n/N (%)	Adjusted odds ratio	95% CI
8–21 days	157/280 (56)	Reference	Reference
22–35 days	67/148 (45)	0.7	0.4 – 1.1
36–49 days	52/86 (60)	1.4	0.8 – 2.5
≥50 days	52/84 (62)	1.2	0.7 – 2.1

Stepwise multivariable logistic regression was used for BPD and death analysis, while generalized linear model with gamma distribution and log link was used for total ventilation days. All analyses adjusted for birth factors (gestational age, race, small for gestational age, multiple births, Apgar <4 at 5 minutes), admission factors (admission to CHNC center ≤ 3 days of life, reason for admission, intubation status on admission), co-morbidities (bloodstream infection, NEC or SIP, PDA ligation, severe IVH), and center.

Figure 4 shows significant center variation in the timing of corticosteroid initiation, rates of BPD, and total ventilator days. Across the different centers, the proportion of systemic corticosteroid treatment after 35 days of life ranged from 12.5% to 60%; while the proportion of BPD ranged from 14% to 75%, and median ventilator days ranged from 16 days to 56 days. Pearson correlation analysis found a moderate positive correlation between later timing of corticosteroid initiation with BPD (Pearson’s correlation coefficient, 0.50; P=0.03) and with total ventilator days (Pearson’s correlation coefficient, 0.49; P=0.04). Exploratory subgroup analysis did not find significant treatment effect by type of corticosteroid (dexamethasone versus hydrocortisone) on the association between timing of corticosteroid initiation and BPD (Supporting Information E-Table 3).

Figure 4. Comparison of BPD rates, ventilator days, and steroid initiation by center.

Shown are scatter plots of center proportion for steroid initiation after 35 days of life plotted against (A) center proportion of Grade 2 or 3 BPD and (B) center median ventilator days. Each bubble represents a center, and the size of the bubble is representative of the number of infants from that center.

Death before 36 weeks’ PMA was observed in 13.4% (80/598) of the entire cohort. The majority of deaths (70%) occurred among infants first treated with systemic corticosteroids at 8–21 days of life and decreased among infants with later steroid initiation (36–49 days and ≥ 50 days) (Table 2). A comparison of baseline characteristics and cause of death among infants with corticosteroid initiation at 8–21 days compared to infants with corticosteroid initiation ≥ 22 days did not identify differences that could account for this finding (Supporting Information E-Table 4). No significant association was observed between timing of steroid initiation and the composite outcome of death or BPD (Table 2).

DISCUSSION

In our multi-center retrospective study of corticosteroid-treated preterm infants, we found that later initiation of corticosteroid treatment was independently associated with higher odds of Grade 2 or 3 severe BPD. Although the timing of initiation varied widely among the different study centers, a positive correlation was observed between later corticosteroid initiation after 35 days of life and the rate of severe BPD at 36 weeks’ PMA. Our results validate our prior single-center study and extend its generalizability to a contemporary cohort of preterm infants transferred to regional NICUs. Moreover, these results raise the question that earlier initiation of corticosteroid treatment may mitigate the rate of Grade 2 or 3 severe BPD at 36 weeks’ PMA.

The significant center variation in the timing of corticosteroid initiation in our study reflects the lack of consensus surrounding corticosteroid treatment in preterm infants. This is highlighted in the recent network meta-analysis by Ramaswamy et al.²² which identified 14 different regimens of postnatal corticosteroids for BPD. Our finding of a moderate correlation between center rate of later corticosteroid initiation and center rate of severe BPD is interesting. Variation in practices and outcomes can provide insight into potentially better practices, and a collaborative quality improvement initiative that aims to decrease the rate of later corticosteroid treatment as part of a bundle to reduce BPD seems reasonable.^23–25 However, it is important to remember that while such an enterprise can help decrease variation in corticosteroid initiation, a consequent reduction in BPD is not guaranteed. Medical care in the NICU is complex and introducing changes – no matter how well-intentioned – can have little or no effect on patient outcomes; and can even have unexpected adverse effects in some cases.^26–28

The majority of the 21 trials included in the Cochrane meta-analysis of postnatal corticosteroids after seven days randomized infants to corticosteroid treatment or placebo by four weeks of age.¹⁰ Only two trials – Avery et al. ²⁹ and Doyle et al. ³⁰ – allowed randomization beyond 4 to 6 weeks of life. Despite the paucity of evidence to support initiating corticosteroids beyond 4 to 6 weeks of age, such late initiation of treatment is not uncommon in clinical practice.²³ Our study adds to the current literature by providing evidence that later initiation of corticosteroids may not be as effective in decreasing BPD when compared to earlier treatment, perhaps indicating the existence of a relative inflection point when chronic lung injury becomes less reversible.

The study by Harmon et al.¹² (study period: 2006–2012) included the era when corticosteroid use was significantly restricted following publication of the 2002 Joint AAP and Canadian Pediatric Society guidelines limiting treatment to only those infants with exceptional clinical circumstances.^31,32 In contrast, our study period (2010–2016) included infants treated with corticosteroids after the 2010 AAP Revised Statement, which relaxed treatment guidelines and supported selective use for infants at high risk for BPD.⁸ While slightly different exclusion criteria were used, both study cohorts included infants with median gestational age of 25 weeks and had a comparable rate of corticosteroid treatment (ratio of infants treated vs not treated of 12%). When comparing the timing of treatment between our study and Harmon et al., our study revealed a lower rate of later initiation (rate of initiation after 5 weeks: 32% in current study vs 41% in Harmon) and a higher rate of earlier initiation (rate of initiation at 2 to 3 weeks: 43% in current study vs 22% in Harmon). Interestingly when comparing BPD, we found a higher rate of severe BPD in Harmon et al. compared to our study [60% (517/863) vs 48% (248/518)]. While several factors may explain this difference in BPD rates – such as changes in clinical practice over time and slight differences in BPD definitions used – the possibility that the gradual shift towards earlier corticosteroid initiation contributed to decreased rates of severe BPD warrants further investigation.

Because death is an important competing event, composite outcomes that combine death with BPD have traditionally been used as a primary outcome in neonatal studies.^33,34 However, interpretation of composite outcomes can be challenging when individual constituents do not move in line with each other. ³⁵ Such is the case in our study, where infants with steroid initiation at 8–21 days had lower BPD but higher death. While we cannot exclude the possibility that the lower odds of BPD among earlier-treated infants is an underestimation of their true risk for BPD, several other important factors need to be considered. First, the finding of higher deaths occurring during earlier time points is not unexpected as this is consistent with the reported pattern of mortality in very preterm infants.³⁶ Second, the association of higher deaths with earlier treatment was likely related to confounding by indication and survival biases inherent in cohort studies such as ours. Lastly, a causative relationship between earlier treatment initiation and death is unlikely as randomized trials that initiated systemic corticosteroids after 1–3 weeks of age showed no differences in survival compared to placebo^37–40; and the Cochrane meta-analysis of trials with corticosteroid initiation between 7–14 days actually showed an increased rate of survival with corticosteroid treatment.⁴¹

Corticosteroid treatment in preterm infants can be initiated for several reasons including vasopressor-resistant hypotension, airway edema, or BPD prevention or treatment. As treatment rationale is difficult to ascertain retrospectively, the precise clinical indications for corticosteroid treatment in our study are unknown. While knowledge of therapeutic intent is valuable, medications have pharmacologic effects beyond their prescribed use. In the subgroup analysis of the Caffeine for Apnea of Prematurity Trial⁴², the beneficial effects of caffeine on BPD remained significant whether initially prescribed to prevent apnea, treat apnea, or facilitate extubation, suggesting that clinical indication may not have a large impact on the observed effects of medications in certain circumstances.

We cite several limitations. Our study is limited by biases and confounders inherent in a retrospective observational study design. Although we used logistic regression in our analysis, residual confounding factors and biases may still be present. For example, infants selected for earlier corticosteroid treatment were likely sicker compared to later-treated infants despite accounting for differences in baseline demographics, admission characteristics, and co-morbidities. The lack of data on underlying severity of lung injury prior to steroid initiation is an important limitation that hampered our ability to further mitigate selection bias. More robust data collection and, preferably, a prospective study, may better address this and other biases in the future. Our study was also limited by the availability of data recorded in CHND and PHIS databases. We did not have data on corticosteroid dosing or on long-term neurodevelopment. Lastly, CHND and PHIS databases represent infants cared for at regional NICUs. While most of the study cohort was admitted by three days of life, we acknowledge that patients cared for at tertiary centers can have referral biases and may reduce generalizability of our findings to other settings.

In conclusion, we found that in a multicenter cohort of preterm infants managed in the current era of less restrictive postnatal corticosteroids, later initiation of treatment remains associated with higher odds of BPD. Further studies are needed to validate these results in a prospective manner and determine the effect of earlier treatment on BPD and other outcomes such as long-term neurodevelopment.