Clinical efficacy of different supplemental doses of vitamins A and D in preventing bronchopulmonary dysplasia in preterm infants: A network meta-analysis based on randomized controlled trials

Abstract

Background:

Bronchopulmonary dysplasia (BPD) is a prevalent and severe chronic respiratory condition in preterm infants, with vitamin deficiency recognized as a contributing factor. Although vitamins A and D are known to play protective roles in lung development, the optimal supplementation doses for BPD prevention remain unclear.

Methods:

Search PubMed, Ovid, the Cochrane Library, Web of Science, CNKI, and Wanfang from database inception to November 30, 2024, identifying randomized controlled trials that investigated the role of vitamins A and D in the prevention of BPD in preterm infants. Data were extracted for network meta-analysis. Patient demographic data, the incidence of BPD, mortality, and mechanical ventilation duration were analyzed.

Results:

In our analysis, encompassing 20 studies with 4357 patients, we observed that high-dose vitamin D (HDVD, ≥800 IU/d) demonstrated the most notable reduction in the incidence of BPD; low-dose vitamin A (LDVA, <3330 IU/d) exhibited the lowest mortality; high-dose vitamin A (≥3330 IU/d) had the shortest mechanical ventilation duration.

Conclusion:

Current scholarly literature suggests that HDVD (≥800 IU/d) supplementation may be the most effective regimen for preventing BPD in preterm infants, followed by LDVA (<3330 IU/d). No statistically significant differences in mortality were observed among any of the supplementation strategies or placebo. High-dose vitamin A (≥3330 IU/d) was associated with a shorter duration of mechanical ventilation. Consequently, to prevent BPD in preterm infants, supplementation with HDVD (≥800 IU/d) and LDVA (<3330 IU/d) may be considered.

1. Introduction

Bronchopulmonary dysplasia (BPD) predominantly affects very low birth weight (VLBW) and extremely preterm infants, and is associated with a poor prognosis. It can result in persistent respiratory and neurological impairments, including pulmonary hypertension, aberrant pulmonary vascular development, and neurodevelopmental disorders, among others.^[1–3] BPD is the most prevalent and intricate chronic respiratory condition in preterm infants.^[4] Studies report that the incidence of BPD ranges from 10% to 40%,^[5] and the mortality rate among BPD infants during the perinatal period in developing countries is approximately 4-fold that of developed countries.^[6] Consequently, early identification and preventative measures for BPD in preterm infants are of significant value.

The primary pathogenic mechanisms underlying BPD likely involve continuous chronic injury to the immature lungs of preterm infants that lack pulmonary surfactant production, along with abnormal repair processes following such injury, due to adverse factors such as hyperoxia and inflammation.^[7] BPD’s etiology is multifactorial, encompassing genetics, low birth weight, prematurity, sepsis, excessive oxygen exposure in the neonatal intensive care unit (NICU), postnatal malnutrition, and low antioxidant levels, with vitamin deficiency being a notable factor.^[8] Vitamins are essential nutrients in humans, critical for regulating immune function, promoting cellular differentiation and proliferation, maintaining the integrity of respiratory epithelial cells, and facilitating alveolar and vascular development.^[9,10] Vitamin transport mechanisms via the placenta are incomplete in preterm fetuses,^[9] increasing the risk of vitamin deficiency.^[11] This deficiency may result from insufficient placental transfer of vitamins due to preterm birth, limited liver storage, inadequate vitamin supply, and poor postnatal vitamin intake, especially in extremely low birth weight and VLBW infants.^[12] Research has established that deficiencies in vitamin A (VA) or vitamin D (VD) are associated with various pulmonary diseases, including asthma and airway obstruction.^[8]

VA preserves the integrity of pulmonary epithelial cells and fosters the synthesis of surfactant-associated proteins.^[13] VA is acknowledged as a safe and efficacious agent for preventing BPD. However, dosing guidelines for VA supplementation differ among institutions and regions; the standard National Institute of Child Health and Human Development protocol is 5000 IU per dose, administered 3 times weekly for 4 weeks,^[14] while European pediatric guidelines propose a dosage range of 1332 to 3330 IU/kg/d for newborns.^[15] A meta-analysis by Rakshasbhuvankar et al suggests that the benefits of VA supplementation in reducing BPD are confined to intakes below 1500 IU/kg/d.^[16] VD is known to stimulate alveolar and vascular development, modulate fibroblast proliferation and differentiation, expedite lung repair post-injury, and regulate immune responses to mitigate inflammatory reactions.^[17,18] Although numerous studies explore the relationship between VD supplementation and BPD, the efficacy of VD in preventing BPD remains a subject of debate. No literature has yet identified an optimal dosage of VD supplementation for BPD prevention.

A recent network meta-analysis by Zhang et al evaluated the efficacy of VA, VD, and vitamin E supplementation for BPD prevention in preterm infants, comparing 9 different dosing regimens across the 3 vitamins. The authors concluded that moderate-dose VD (400–800 IU/d) and low-dose VA (<1332 IU/kg/d) were associated with reduced BPD incidence. Although informative, that analysis has several notable limitations. First, its literature search was conducted up to August 2023, leaving a gap of approximately 15 months into which newer trials may have been published. Second, its Bayesian framework and 9-group classification – which included vitamin E – make direct comparisons with the simpler 2-level dose categorizations commonly used in clinical practice less straightforward. Third, the analysis did not yield statistically significant findings for several secondary outcomes, including mechanical ventilation duration, suggesting that further investigation with an updated dataset is warranted.

The present study updates and extends the evidence in several important ways. First, we conducted a more recent literature search up to November 30, 2024, incorporating newly published randomized controlled trials (RCTs). Second, we employed a frequentist network meta-analysis approach with a pragmatic 2-tier dose classification for VA (<3330 IU/d vs ≥3330 IU/d) and VD (<800 IU/d vs ≥800 IU/d), based on current European neonatal/pediatric guidelines. Accordingly, this study aims to comprehensively evaluate the comparative efficacy of different doses of VA and VD supplementation for preventing BPD in preterm infants, with particular attention to BPD incidence, mortality, and mechanical ventilation duration.

2. Materials and methods

This study adheres to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 statement,^[19,20] ensuring a structured methodology and reporting format, and A Measurement Tool to Assess systematic Reviews 2 guidelines.^[21]

2.1. Data sources

Two independent investigators (the first and second authors) performed an exhaustive literature search, with any disagreements resolved by a third researcher (the corresponding author). This process included the systematic evaluation of titles, abstracts, and full manuscripts to assess study eligibility.

We performed a comprehensive database search of PubMed, Embase, Ovid, the Cochrane Library, Web of Science, and Wanfang from database inception to November 30, 2024. RCTs that investigated the role of VA and VD in the prevention of BPD in preterm infants were included in this analysis. The search incorporated the following indexed terms: “bronchopulmonary dysplasia,” “BPD,” “chronic lung disease,” “lung injury,” “vitamins,” “provitamins,” “vitamin A,” “VA,” “retinol,” “retinol palmitate,” “retinoic,” “tretinoin,” “alitretinoin,” “retinoids,” “vitamin D,” “VD,” “cholecalciferol,” “Ergocalciferols,” “random,” and “randomized.” In addition, Google Scholar was also queried to identify additional pertinent literature. The reference lists of identified reports were meticulously reviewed for any further relevant studies. Our meta-analysis included studies without language limitations, with the search strategy detailed in Table 1 (illustrated with PubMed).

Table 1.

Search strategy on PubMed.

#1	Search: (((bronchopulmonary dysplasia) OR (BPD)) OR (chronic lung disease)) OR (lung injury) Sort by: Most Recent
#2	Search: Vitamins or Provitamins or vitamin A or VA or retinol or retinols or retinol palmitate or retinoic or tretinoin or alitretinoin or retinoids or vitamin D or VD or cholecalciferol or Ergocalciferols Sort by: Most Recent
#3	Search: Randomized controlled trial Sort by: Most Recent
#4	Search: (((((bronchopulmonary dysplasia) OR (BPD)) OR (chronic lung disease)) OR (lung injury)) AND (Vitamins or Provitamins or vitamin A or VA or retinol or retinols or retinol palmitate or retinoic or tretinoin or alitretinoin or retinoids or vitamin D or VD or cholecalciferol or Ergocalciferols)) AND (Randomized controlled trial) Sort by: Most Recent

2.2. Eligibility criteria

The inclusion criteria for this study encompass the following: participants: preterm infants with a gestational age of <37 weeks, without restrictions on race, age, gender, or disease duration; study types: RCTs; interventions: the experimental group administered a specific dosage of VA or VD monotherapy; and outcomes: incidence of BPD, mortality rate, and mechanical ventilation duration.

The exclusion criteria were as follows: case reports, systematic reviews, meta-analyses, editorials, letters to the editor, protocols, biomechanical analyses, animal and cadaveric experiments, and so on; trials where vitamins are not used as the sole intervention; studies with incomplete data; and duplicate or similarly themed publications by the same author in various journals.

Based on the intervention measures from the included RCTs and a review of existing literature on vitamin supplementation in preterm infants, the interventions were categorized into 4 groups for comparative analysis. Given the heterogeneity in dosing units across studies (some reported IU/d, others IU/kg/d, and some used fixed regimens regardless of weight) and the limited number of studies available for each specific dose, a formal dose–response analysis was not feasible. Therefore, we adopted a pragmatic categorical approach informed by existing neonatal/pediatric guidelines.

For VA, the categorization was based on the European neonatal/pediatric guidelines, which recommend 1332 to 3330 IU/kg/d for infants weighing over 1 kg. For a standard 1 kg infant, this range corresponds to 1332 to 3330 IU/d. Accordingly, we defined high-dose vitamin A (HDVA) as a daily intake of ≥3330 IU/d and low-dose vitamin A (LDVA) as a daily intake of <3330 IU/d. For studies reporting weight-adjusted doses, the total daily intake was calculated based on the reported mean body weight of the study population.

For VD, the categorization was based on the Central European neonatal/pediatric guidelines, which recommend 400 to 800 IU/d starting from birth. We defined high-dose vitamin D (HDVD) as a daily intake of ≥800 IU/d and low-dose vitamin D (LDVD) as a daily intake of <800 IU/d. This threshold (800 IU/d) has been used as a comparator in multiple prior RCTs investigating VD supplementation in preterm infants.

2.3. Data extraction

Two investigators independently conducted the literature screening according to predefined inclusion and exclusion criteria, utilizing a uniform custom-designed extraction form to gather basic information from each registered study. Discrepancies were addressed through consensus discussion or arbitration by a third researcher. The information extracted encompassed the following: essential details of the included studies, including the first author, publication year, and country; methodologies and critical components of the risk of bias assessment in RCTs; specifics of the interventions, such as timing, administration route, dosage, and treatment duration; and pertinent outcomes: incidence of BPD, mortality, and mechanical ventilation duration. In instances where standard deviations (SDs) were not reported, they were estimated using Cochrane Handbook-specified methods:

1. In the absence of directly reported group SDs, the formula from section 6.5.2.2 of the Cochrane Handbook was applied to derive group mean SDs from the standard error of the mean or 95% confidence intervals (CIs).

2. When data were presented as medians and interquartile ranges, the conversion to mean ± SD was performed using the Wan et al formula detailed in section 6.5.2.5 of the Cochrane Handbook:

$${\displaystyle \begin{array}{lll}& [\mathrm{S}\mathrm{D}=\mathrm{S}\mathrm{E}\mathrm{M}\times \sqrt{n}\mathrm{o}\mathrm{r}& \mathrm{S}\mathrm{D}=\sqrt{n}\times (\mathrm{u}\mathrm{p}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{l}\mathrm{i}\mathrm{m}\mathrm{i}\mathrm{t}-\mathrm{l}\mathrm{o}\mathrm{w}\mathrm{e}\mathrm{r}\mathrm{l}\mathrm{i}\mathrm{m}\mathrm{i}\mathrm{t})/3.92]\end{array}}$$

1. For unreported values, the Follmann et al formula was employed to estimate the SDs of changes from baseline, assuming a correlation coefficient of 0.50 between baseline and post-intervention values (Cochrane Handbook sections 6.5.2.8, 2):

$${\displaystyle \begin{array}{lll}& [S{D}_{E,change}& =\sqrt{S{D}_{E,baseline}^2+S{D}_{E,final}^2-(2\times 0.50\times S{D}_{E,baseline}\times S{D}_{E,final})}]\end{array}}$$

2.4. Quality assessment

The quality of included RCTs was assessed using the Cochrane Risk of Bias tool. Two investigators (the first and second authors) evaluated the risk of bias in accordance with the Cochrane Handbook criteria, which include randomization, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, completeness of outcome data, selective reporting, and other biases. Each domain is classified as unclear risk, low risk, or high risk.

2.5. Statistical analysis

To conduct a comprehensive network meta-analysis, we utilized the statistical software packages “Network” and “mvmeta” within STATA 17.0 software (StataCorp LLC). Dichotomous variables, specifically the incidence of BPD and mortality, were analyzed using relative risk (RR) with corresponding 95% CI. Meanwhile, continuous variable, including mechanical ventilation duration, was analyzed using weighted mean differences with corresponding 95% CI. When the 95% CI of the RR or weighted mean difference contained the value 1, the comparison was considered statistically nonsignificant.

For direct comparisons, a conventional meta-analysis was conducted to aggregate the results using random-effects models, serving as sensitivity analyses. The network meta-analysis employed a frequentist approach with a random-effects model to estimate both direct and indirect comparisons. The primary objective of the network meta-analysis was to assess whether any of the comparator interventions demonstrated superiority. To evaluate potential inconsistencies between indirect and direct comparisons, we employed global inconsistency, local inconsistency (using a node-splitting approach), and loop inconsistency. Statistical significance for global inconsistency was determined using P values, with P > .05 indicating no significant global inconsistency. Local inconsistency was assessed through node-splitting analysis, and P > .05 indicated no significant local inconsistency. Heterogeneity within each closed loop was estimated using the inconsistency factor (IF), with a 95% CI (IF) value of zero signifying no statistical significance. In each prespecified outcome, a global network diagram was employed to illustrate direct comparisons between interventions. The size of the nodes in the diagram corresponded to the number of participants receiving each treatment. Treatments subject to direct comparisons were linked by lines, and the thickness of these lines was proportional to the number of trials evaluating the specific comparison.

Within the “Results” section, the ranking probability of each intervention was presented using a cumulative probability ranking graph. The graph incorporated the Surface Under the Cumulative Ranking Curve (SUCRA) value, serving as an index that summarized the cumulative ranking probability. The SUCRA value ranged between 0% and 100%, where a larger SUCRA value indicated a higher ranking for the intervention, typically reflecting a more favorable or less favorable effect. All intervention measures were ranked based on their respective SUCRA values or the area under the curve, resulting in a comprehensive ranking of the interventions.

To assess the potential for publication bias, the comparison-adjusted funnel plot was utilized. This analysis aimed to determine whether there was evidence of a small sample effect or publication bias within the intervention network.

3. Results

3.1. Search results

Initially, a total of 1352 studies were identified from PubMed (n = 536), Ovid (n = 45), Web of Science (n = 366), the Cochrane Library (n = 1), CNKI (n = 113), and Wanfang (n = 291). Utilizing the “Find Duplicates” feature in EndNote software, 277 duplicate studies were excluded. Following a comprehensive screening of titles and abstracts, 984 nonrelevant studies were excluded. The full texts of the remaining 91 studies were then retrieved. In total, 20 RCTs,^[22–41] encompassing 4357 patients, met the criteria for inclusion in this network meta-analysis. The study selection process is illustrated in Figure 1, and the baseline characteristics of the included studies are detailed in Table 2.

Figure 1.

Flow diagram of the study selection process.

Table 2.

Baseline characteristics of the included studies.

Author	Country	Study design	Group	No	Gestational age (wk)	Birth weight (g)	Outcome
Jon E 1999	USA	RCT	VA 5000 IU Placebo	405 402	26.8 ± 1.9 26.7 ± 1.7	770 ± 135 769 ± 138	①②
Ravishankar 2003	USA	RCT	VA 1500–3000 IU Placebo	22 18	<32 wk <32 wk	500–1500 500–1500	①②
Basu 2019	India	RCT	VA 10,000 IU Placebo	98 98	30.9 ± 2.9 30.7 ± 2.7	1185 ± 194 1163 ± 181	①②③
Meyer 2024	Germany	RCT	VA 5000 IU Placebo	449 466	<32 wk <32 wk	400–1000 400–1000	①②③
Matthew A 2021	USA	RCT	VA 5000 IU Placebo	405 402	<28 wk <28 wk	400–1000 400–1000	①②③
Prem Fort 2016	USA	RCT	VD 200 IU VD 800 IU Placebo	34 30 36	25.3 ± 1.4 25.1 ± 1.6 25.5 ± 1.4	744 ± 210 796 ± 228 774 ± 210	①②③
Kiatchoosakun 2014	Thailand	RCT	VA 5000 IU Placebo	40 40	29.0 + 1.7 28.9 + 1.9	1152.8 + 203.6 1123.1 + 218.1	①②③
Rakshasbhuvankar 2021	Australia	RCT	VA 5000 IU Placebo	33 33	25.5 + 1.55 25.8 + 1.48	810 + 200 877 + 251	①
Ge 2022	China	RCT	VD 800 IU Placebo	57 55	29.2 ± 1.4 29.8 ± 1.1	1240 + 150 1290 + 170	①③
Aristizaba 2022	USA	RCT	VD 200 IU VD 800 IU Placebo	19 23 31	26 ± 1 25 ± 1 25 ± 1	846 ± 154 839 ± 224 773 ± 206	①
Abhijeet 2021	Australia	RCT	VA 5000 IU Placebo	94 94	25.8 ± 1.4 26.0 ± 1.3	853 ± 201 852 ± 211	①②③
Sun 2019	China	RCT	VA 1500 IU Placebo	132 130	26.8 ± 1.9 27.1 ± 2.0	981.8 ± 233.6 983.9 ± 216.9	①②③
Pearson 1992	USA	RCT	VA 2000 IU Placebo	27 22	27 ± 1 27 ± 1	885 ± 118 887 ± 102	①②
Wardle 2001	UK	RCT	VA 5000 IU Placebo	77 77	25–27 25–27	710–890 662–880	①②
Shenai 1987	USA	RCT	VA 2000 IU Placebo	20 20	27.9 ± 1.3 28.0 ± 1.3	1006 ± 163 976 ± 162	①②③
Tang 2016	China	RCT	VA 10,000 IU Placebo	32 32	29.98 ± 2.3 30.35 ± 2.0	1251.62 ± 181.5 1269.56 ± 148.32	①②
Liu 2023	China	RCT	VD 400 IU VD 800 IU	49 49	29.64 ± 1.68 29.58 ± 1.49	1278.26 ± 132.77 1245.69 ± 144.73	①③
Huang 2024	China	RCT	VD 400 IU VD 800 IU Placebo	20 20 20	30.04 ± 2.53 30.2 ± 2.81 30.12 ± 2.76	1200 ± 390 1240 ± 370 1220 ± 410	①③
Nie 2022	China	RCT	VD 400 IU VD 800 IU	73 87	30.43 ± 5.31 30.25 ± 4.67	1469.88 ± 221.74 1478.24 ± 203.71	①
Bai 2020	China	RCT	VD 800 IU Placebo	43 43	34.79 ± 1.02 34.86 ± 1.13	2059.74 ± 218.59 2137.42 ± 365.58	①③

① The incidence of BPD, ② mortality, ③ mechanical ventilation duration.

RCT = randomized controlled trial, VA = vitamin A, VD = vitamin D.

3.2. Risk of bias and quality assessment

The quality assessment of the included RCTs was conducted using the Cochrane Collaboration’s “Risk of Bias” tool. The risk of bias assessment for the included studies is presented in Table 3.

Table 3.

Risk of bias of the included randomized controlled trials.

Study	Sequence generation	Allocation concealment	Blinding	Completeness of data	Selective reporting bias	Other bias
Jon E 1999	A randomization list	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Ravishankar 2003	Stratification by weight	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Basu 2019	Random permuted blocks	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Meyer 2024	A randomization list	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Matthew A 2021	Stratified by center and birth weight group	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Prem Fort 2016	Computer-generated	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Kiatchoosakun 2014	A randomization list	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Rakshasbhuvankar 2021	Computer-generated	Unclear	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Ge 2022	Random number table	Unclear	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Aristizaba 2022	Unclear	Unclear	Unclear	Low risk	Low risk	Low risk
Abhijeet 2021	Computer-generated	In identical amber-colored containers	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Sun 2019	A blocked randomization method	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Pearson 1992	Stratified into 4 groups according to birth weight and gender:	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Wardle 2001	A computerized random number generator	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Shenai 1987	Unclear	Sealed envelopes	Double-blind (participant and therapist)	Low risk	Low risk	Low risk
Tang 2016	Unclear	Unclear	Single-blind (participant)	Low risk	Low risk	Low risk
Liu 2023	A random number	Unclear	Single-blind (participant)	Low risk	Low risk	Low risk
Huang 2024	A random number	Unclear	Single-blind (participant)	Low risk	Low risk	Low risk
Nie 2022	Unclear	Unclear	Single-blind (participant)	Low risk	Low risk	Low risk
Bai 2020	Unclear	Unclear	Single-blind (participant)	Low risk	Low risk	Low risk

3.3. Evidence network

This study encompassed 5 interventions, including HDVA, LDVA, HDVD, LDVD, and placebo. Figure 2 visually delineates the evidence network, with straight lines representing direct comparisons between pairs of interventions. Interventions lacking direct connections are compared indirectly via the network meta-analysis. The width of the lines reflects the number of trials, while the size of the nodes corresponds to the total sample size across multiple treatments.

Figure 2.

Network analysis of eligible comparison for (A) the incidence of BPD, (B) mortality, and (C) mechanical ventilation duration. The size of each node represents the number of participants, while the thickness of the line represents the number of studies directly comparing the 2 interventions. BPD = bronchopulmonary dysplasia, HDVA = high-dose vitamin A, HDVD = high-dose vitamin D, LDVA = low-dose vitamin A, LDVD = low-dose vitamin D.

3.4. Inconsistency test

Figure 3 illustrates an inconsistency plot assessing heterogeneity within the closed loops of our network meta-analysis, exemplified by the incidence of BPD. The plot includes one closed loop with an IF of 0.88, and the 95% CI spans from 0.00 to 1.94, encompassing 0, which suggests data consistency. Due to the structure of the evidence network, only this single closed loop was available for inconsistency assessment; a more comprehensive evaluation (e.g., global inconsistency test or node-splitting analysis) was not feasible. Therefore, the possibility of undetected inconsistency cannot be fully excluded.

Figure 3.

Inconsistency plot of eligible comparison for the incidence of BPD. BPD = bronchopulmonary dysplasia, HDVA = high-dose vitamin A, HDVD = high-dose vitamin D, LDVA = low-dose vitamin A, LDVD = low-dose vitamin D.

3.5. Results of network meta-analysis

3.5.1. Incidence of BPD

A total of 20 studies with 4357 patients reported the incidence of BPD involving interventions of HDVA, LDVA, HDVD, LDVD, and Placebo. The network meta-analysis revealed that compared with HDVA, both HDVD (RR = 0.36, 95% CI: 0.22–0.60) and LDVA (RR = 0.54, 95% CI: 0.32–0.89) were associated with a significantly lower incidence of BPD; compared with HDVD, the placebo group had a significantly higher incidence of BPD (RR = 3.23, 95% CI: 2.02–5.16), as did the LDVD group (RR = 1.69, 95% CI: 1.02–2.80); compared with LDVD, the placebo group also had a significantly higher incidence of BPD (RR = 1.91, 95% CI: 1.08–3.36); and no other comparisons yielded statistically significant differences (Fig. 4A).

Figure 4.

Forest plots for (A) the incidence of BPD, (B) mortality, and (C) mechanical ventilation duration. BPD = bronchopulmonary dysplasia, CI = confidence interval, HDVA = high-dose vitamin A, HDVD = high-dose vitamin D, LDVA = low-dose vitamin A, LDVD = low-dose vitamin D.

A ranking graph illustrating the distribution of probabilities for the incidence of BPD is presented in Figure 5A. The SUCRA rankings for the incidence of BPD were as follows: HDVD (3.7%) < LDVA (31%) < LDVD < (42.1%) < HDVA (74.2%) < placebo (98.9%), which suggests that HDVD is associated with the highest probability of reducing the incidence of BPD, whereas placebo is associated with the lowest probability. Therefore, the efficacy of reducing the incidence of BPD was ranked from best to worst as follows: HDVD, LDVA, LDVD, HDVA, and placebo.

Figure 5.

SUCRA for (A) the incidence of BPD, (B) mortality, and (C) mechanical ventilation duration. BPD = bronchopulmonary dysplasia, HDVA = high-dose vitamin A, HDVD = high-dose vitamin D, LDVA = low-dose vitamin A, LDVD = low-dose vitamin D, SUCRA = surface under the cumulative ranking.

3.5.2. Incidence of mortality

Twelve articles with 3662 patients reported mortality involving interventions of HDVA, LDVA, HDVD, LDVD, and placebo. The results showed that there was no statistical significance between the interventions (Fig. 4B).

A ranking graph illustrating the distribution of probabilities for mortality is presented in Figure 5B. The SUCRA rankings for mortality were as follows: LDVA (22.6%) < HDVD (26%) < HDVA (53.9%) < placebo (63.4%) < LDVD (84%), which suggests that LDVA is associated with the highest probability of reducing mortality, whereas LDVD is associated with the lowest probability. Therefore, the efficacy of reducing mortality was ranked from best to worst as follows: LDVA, HDVD, HDVA, placebo, and LDVD. However, because none of the pairwise comparisons reached statistical significance, these SUCRA rankings are purely exploratory and should not be interpreted as evidence of differential treatment effects.

3.5.3. Mechanical ventilation duration

Only 8 studies with 1559 patients reported mechanical ventilation duration involving interventions of HDVA, LDVA, HDVD, LDVD, and placebo. Compared with HDVA, the placebo group had a significantly longer duration of mechanical ventilation (MD = 1.93 × 10⁶, 95% CI: 4.89–7.61 × 10¹¹), indicating that HDVA was associated with a shorter ventilation duration. No other comparisons yielded statistically significant differences (Fig. 4C).

A ranking graph illustrating the distribution of probabilities for mechanical ventilation duration is presented in Figure 5C. The SUCRA rankings for mechanical ventilation duration were as follows: HDVA (15.9%) < LDVA (20.1%) < HDVD < (61.6%) < LDVD (69.7%) < placebo (82.8%), which suggests that HDVA is associated with the highest probability of shortening mechanical ventilation duration, whereas placebo is associated with the lowest probability. Therefore, the efficacy of shortening mechanical ventilation duration was ranked from best to worst as follows: HDVA, LDVA, HDVD, LDVD, and placebo.

3.6. Publication bias

Based on the outcomes observed for the incidence of BPD, mortality, and mechanical ventilation duration, adjusted funnel plots were utilized to assess publication bias and potential small-study effects in the network meta-analysis. The analysis demonstrated an even distribution of data points within the funnel plot and a balanced pattern on either side. Moreover, the regression line is nearly parallel to the x-axis, suggesting a negligible chance of publication bias or small-study effects (Fig. 6).

Figure 6.

Funnel plots of (A) the incidence of BPD, (B) mortality, and (C) mechanical ventilation duration. BPD = bronchopulmonary dysplasia, HDVA = high-dose vitamin A, HDVD = high-dose vitamin D, LDVA = low-dose vitamin A, LDVD = low-dose vitamin D.

4. Discussion

Identifying the most effective dosing regimen of VA or VD for preventing BPD in preterm infants is a matter of ongoing debate. While numerous studies have reported on the efficacy of VA or VD in preventing BPD with positive results, there is a dearth of meta-analyses on the optimal dosing. Our network meta-analysis fills this void by providing a comprehensive analysis of the efficacy of various VA and VD dosages in preventing BPD, including HDVA (≥3330 IU/d), LDVA (<3330 IU/d), HDVD (≥800 IU/d), and LDVD (<800 IU/d). As far as we are aware, this is the first systematic review to compare these dosages for BPD prevention in preterm infants, and we believe our results will guide clinical decision-making toward the most effective treatment strategies.

In this study, incorporating 22 RCTs, the analysis suggests that HDVD (≥800 IU/d) is potentially the most effective supplementation strategy for preventing BPD in preterm infants, with LDVA (<3330 IU/d) as the next most effective. LDVA (<3330 IU/d) was linked to reduced mortality rates in preterm infants, while LDVD (<800 IU/d) supplementation significantly increased the risk of mortality, and HDVA was associated with shorter durations of mechanical ventilation.

VD facilitates fetal pulmonary development and maturation, and VD deficiency correlates with the incidence of respiratory diseases in infants, particularly extremely preterm infants.^[42] Park et al^[43] previously meta-analyzed the relationship between VD levels and BPD, finding an association between neonatal BPD and low VD levels along with prenatal VD insufficiency. In preterm infants, high oxygen exposure in the NICU leads to oxidative stress and O2 free radical release, inducing chronic pathological processes in the lungs and promoting BPD. This can be mitigated by antioxidants, anti-fibrotics, and anti-inflammatory hormones, such as VD.^[8,44] VD reduces inflammatory cytokines like interleukin-1β and interferon-γ, which increase under hyperoxic conditions.^[45] Furthermore, it elevates vascular endothelial growth factor and vascular endothelial growth factor receptor 2 levels, protecting against alveolar simplification caused by hyperoxia.^[46] Ge et al^[36] demonstrated that VD supplementation significantly suppresses serum inflammatory factors in preterm infants, including CRP, interleukin-6, and tumor necrosis factor-α. Compared with the control group, VD supplementation also lowered PaCO₂ levels and raised PaO₂ levels, indicating improved gas exchange capacity. Elfaragy et al^[47] reported significantly reduced levels of Krebs von den Lungen-6 and urinary β2-microglobulin (2 important predictive markers of BPD) in newborns supplemented with VD. Czech-Kowalska et al^[48] suggest that preterm infants are at higher risk of VD deficiency, and early high-dose VD supplementation helps maintain adequate VD levels. Fort et al^[28] divided subjects into 3 groups based on VD supplementation dosages of 200 IU/d, 800 IU/d, and placebo, indicating that 800 IU/d is the optimal dosage for achieving 25(OH) VD levels within the ideal range of 20 to 60 ng/mL. This aligns with our findings that preterm infants supplemented with VD at ≥800 IU/d had the lowest incidence of BPD and significantly reduced mortality rates. Based on these conclusions, this study posits that HDVD (≥800 IU/d) is the most effective regimen for preventing BPD in preterm infants.

VA is the sole nutritional therapy endorsed for the prevention and treatment of BPD in preterm infants^[49]; it is essential for normal respiratory, visual, cardiovascular, immune, and gastrointestinal function,^[50] and is indispensable for normal growth and development.^[11] Extremely preterm infants are born with low levels of VA, placing them at high risk for VA deficiency. However, the optimal VA supplementation for these demographic remains unknown, and despite evidence of benefits, early oral VA supplementation is not widely practiced in preterm infants.^[11,12] A trial administering 5000 IU of VA intramuscularly 3 times a week for 4 weeks found that this dosage could reduce the risk of VA deficiency and chronic lung disease in low birth weight infants,^[24] but this study was published in 1999. Rakshasbhuvankar et al^[16] published a meta-analysis on VA supplementation in extremely preterm infants, suggesting that the benefits of VA supplementation for reducing BPD in preterm infants may be limited to LDVA (intake of 1500 IU/kg/d), and that oral VA supplements are as effective as other routes of administration. This study suggests that VA supplementation doses <3330 IU/kg/d help reduce the incidence of BPD and mortality, results that are roughly consistent with theirs, and also confirm that lower doses of VA are more beneficial in preventing BPD in preterm infants. This study, based on European guidelines, divided VA doses into high and low doses at 3330 IU/kg/d, with each group including separate RCTs. In addition, this paper included more relevant studies, more comprehensively covering VA supplementation regimens for the prevention of BPD.

It is also worth considering whether gestational age or birth weight modifies the observed effects. Most included studies enrolled VLBW or extremely low birth weight infants, and the findings may not be directly generalizable to more mature preterm infants. Due to the limited number of studies, we were unable to perform subgroup analyses by gestational age or birth weight. Future individual patient data meta-analyses are needed to explore whether the effects of vitamin supplementation on BPD prevention differ across gestational age or birth weight strata.

Several important factors should be considered before translating these findings into clinical practice. First, while no serious adverse events were reported in the included trials, the safety profile of high-dose vitamin supplementation – particularly long-term safety – remains to be established. Second, the cost-effectiveness of routine VA or VD supplementation for BPD prevention may vary across healthcare settings, especially for intramuscular VA, which requires syringes and trained personnel. Oral formulations may be more feasible in resource-limited settings, but their bioavailability in preterm infants requires further study. Third, implementation feasibility differs widely; while VD supplementation is relatively simple and low-cost, HDVA protocols used in some trials (e.g., intramuscular injections 3 times weekly for 4 weeks) may be challenging to implement in NICUs with limited staffing or resources. Therefore, our findings should be considered hypothesis-generating, and local context should guide any clinical decision-making.

Limitations of this study include the following: first, and most importantly, our categorization of VA and VD supplementation doses into high-dose and low-dose groups was pragmatic rather than pharmacokinetically or pharmacodynamically derived. The chosen thresholds (3330 IU/d for VA and 800 IU/d for VD) were based on existing European neonatal/pediatric guidelines rather than on established therapeutic ranges, toxicity thresholds, or physiological saturation points. This arbitrary dichotomization may have introduced misclassification bias, particularly for studies with doses near the cutoff values. Furthermore, this approach does not capture potential nonlinear dose–response relationships, such as threshold effects or U-shaped curves, which may be clinically important. The categorization also did not fully account for weight-based dosing, as some studies reported fixed doses regardless of infant weight. Consequently, our findings should be interpreted as exploratory, and the optimal dosing of VA and VD for BPD prevention remains to be determined through well-designed dose–response trials or individual patient data meta-analyses. Second, substantial clinical and methodological heterogeneity exists across the included trials. The studies spanned nearly 4 decades (1987–2024), during which neonatal intensive care practices – including respiratory support strategies, surfactant administration protocols, caffeine use, and nutritional support – have evolved considerably. More specifically, the diagnostic criteria for BPD varied across studies: some used oxygen requirement at 28 days of life, others used oxygen requirement at 36 weeks postmenstrual age, and more recent studies adopted the National Institutes of Health severity-based classification. Mortality assessment periods also varied, with some studies reporting in-hospital mortality and others reporting mortality at specific postnatal time points. For mechanical ventilation duration, some studies reported total days, while others reported hours or minutes, and the criteria for extubation or weaning may have differed. In addition, concomitant use of therapies known to affect BPD risk was not uniformly reported. Due to the limited number of included studies, we were unable to perform subgroup analyses or meta-regression to explore these sources of heterogeneity. Third, our inconsistency assessment was limited to a single closed loop due to the structure of the evidence network; a more comprehensive evaluation (e.g., global inconsistency test or node-splitting analysis) was not feasible, and the possibility of undetected inconsistency cannot be excluded. Fourth, the methods, routes, and durations of vitamin supplementation varied among studies, which could be more thoroughly investigated in future research. Fifth, the study included only 20 RCTs, and after categorization, some groups had a small number of cases, leading to a limited sample size. In addition, while SUCRA rankings were used to suggest treatment hierarchies, these represent probabilities rather than definitive evidence. The rankings should be interpreted with caution, particularly when pairwise comparisons show no statistically significant differences, as was the case for mortality. This study did not employ a formal GRADE framework to assess the certainty of evidence. Given the multiple limitations discussed above – particularly the pragmatic dose categorization, clinical heterogeneity, and limited inconsistency testing – the overall certainty of evidence for most outcomes is low to very low. Consequently, our findings should be considered exploratory, and no strong clinical recommendations can be made based on this analysis alone. Well-designed dose–response trials or individual patient data meta-analyses are needed to confirm these findings and guide clinical practice.

Collectively, the findings suggest that HDVD supplementation (≥800 IU/d) may be the most potent regimen for preventing BPD in preterm infants, followed by LDVA (<3330 IU/d). LDVA supplementation (<3330 IU/d) is associated with a decrease in mortality among preterm infants, while LDVD supplementation can significantly elevate the risk of mortality, and HDVA can reduce the duration of mechanical ventilation. Consequently, to prevent BPD in preterm infants, it is recommended to supplement with HDVD (≥800 IU/d) and LDVA (<3330 IU/kg/d), and to mitigate the risk of neonatal death, LDVD (<800 IU/d) should be avoided. This study warrants further validation through extensive multicenter, large-sample studies.

Author contributions

Conceptualization: Qingfeng Yu, Xiao Chen, Jing Chen.

Data curation: Qingfeng Yu, Xiao Chen, Jing Chen, Lan Peng, Kui Luo.

Formal analysis: Qingfeng Yu, Xiao Chen, Jing Chen.

Investigation: Qingfeng Yu, Xiao Chen, Jing Chen, Lan Peng, Kui Luo.

Methodology: Qingfeng Yu, Xiao Chen, Jing Chen.

Project administration: Qingfeng Yu, Xiao Chen, Jing Chen, Lan Peng, Kui Luo.

Supervision: Qingfeng Yu, Xiao Chen, Jing Chen.

Validation: Qingfeng Yu, Xiao Chen, Jing Chen, Lan Peng, Kui Luo.

Visualization: Qingfeng Yu, Xiao Chen, Jing Chen, Lan Peng, Kui Luo.

Resources: Xiao Chen, Jing Chen.

Software: Xiao Chen, Jing Chen.

Funding acquisition: Jing Chen.

Writing – original draft: Qingfeng Yu, Xiao Chen, Jing Chen.

Writing – review & editing: Qingfeng Yu, Xiao Chen, Jing Chen.