NICU indicates neonatal intensive care unit; PDA, patent ductus arteriosus.
Model 1 adjusted for gestational age; model 2, for gestational age, sex, and birthweight z score; model 3, for gestational age, weighted by the inverse of the propensity score; model 4, for gestational age, weighted by the inverse of the propensity score and with analysis accounting for clustering on neonatal units. The position of each square represents the point estimate of the exposure effect. Error bars indicate 95% confidence intervals.
Y-axes shown in blue indicate range from 0% to 20%.
eTable 1. Basal Characteristics in the Overall Cohort and in the Matched Cohort
eTable 2. Relationship Between the Instrumental Variable and Exposition to Early Screening by Echocardiography in the Overall Cohort
eTable 3. Instrumental Variable and Covariates in the Overall Cohort
eTable 4. Ductus Arteriosus Treatment, Death, and Morbidity in the Matched Cohort
eTable 5. Causes of Death in the Matched Cohort
eFigure 1. Relationship Between Unit Patient Volume and Early Echocardiography Screening Strategy
eFigure 2. Propensity Score Distribution in the Nonexposed and Exposed Groups
eFigure 3. Multivariable Analysis of In-hospital Mortality in Survivors at Day 2
eFigure 4. Multivariable Analysis of In-hospital Mortality in Units That Routinely Did or Did Not Perform Early Screening Echocardiography
Rozé J, Cambonie G, Marchand-Martin L, Gournay V, Durrmeyer X, Durox M, Storme L, Porcher R, Ancel P, for the Hemodynamic EPIPAGE 2 Study Group. Association Between Early Screening for Patent Ductus Arteriosus and In-Hospital Mortality Among Extremely Preterm Infants. JAMA. 2015;313(24):2441-2448. doi:10.1001/jama.2015.6734
There is currently no consensus for the screening and treatment of patent ductus arteriosus (PDA) in extremely preterm infants. Less pharmacological closure and more supportive management have been observed without evidence to support these changes.
To evaluate the association between early screening echocardiography for PDA and in-hospital mortality.
Design, Setting, and Participants
Comparison of screened and not screened preterm infants enrolled in the EPIPAGE 2 national prospective population-based cohort study that included all preterm infants born at less than 29 weeks of gestation and hospitalized in 68 neonatal intensive care units in France from April through December 2011. Two main analyses were performed to adjust for potential selection bias, one using propensity score matching and one using neonatal unit preference for early screening echocardiography as an instrumental variable.
Early screening echocardiography before day 3 of life.
Main Outcomes and Measures
The primary outcome was death between day 3 and discharge. The secondary outcomes were major neonatal morbidities (pulmonary hemorrhage, severe bronchopulmonary dysplasia, severe cerebral lesions, and necrotizing enterocolitis).
Among the 1513 preterm infants with data available to determine exposure, 847 were screened for PDA and 666 were not; 605 infants from each group could be paired. Exposed infants were treated for PDA more frequently during their hospitalization than nonexposed infants (55.1% vs 43.1%; odds ratio [OR], 1.62 [95% CI, 1.31 to 2.00]; absolute risk reduction [ARR] in events per 100 infants, −12.0 [95% CI, −17.3 to −6.7). Exposed infants had a lower hospital death rate (14.2% vs 18.5% ; OR, 0.73 [95% CI, 0.54 to 0.98]; ARR, 4.3 [95% CI, 0.3 to 8.3]) and a lower rate of pulmonary hemorrhage (5.6% vs 8.9%; OR, 0.60 [95% CI, 0.38 to 0.95]; ARR, 3.3 [95% CI, 0.4 to 6.3]). No differences in rates of necrotizing enterocolitis, severe bronchopulmonary dysplasia, or severe cerebral lesions were observed. In the overall cohort, instrumental variable analysis yielded an adjusted OR for in-hospital mortality of 0.62 [95% CI, 0.37 to 1.04].
Conclusions and Relevance
In this national population-based cohort of extremely preterm infants, screening echocardiography before day 3 of life was associated with lower in-hospital mortality and likelihood of pulmonary hemorrhage but not with differences in necrotizing enterocolitis, severe bronchopulmonary dysplasia, or severe cerebral lesions. However, results of the instrumental variable analysis leave some ambiguity in the interpretation, and longer-term evaluation is needed to provide clarity.
Closure of the ductus arteriosus is completed in most normal-term infants within 72 hours of birth. However, failure to close is common in extremely premature infants, resulting in a condition called patent ductus arteriosus (PDA). The hemodynamic consequences of PDA are a left-to-right shunt, resulting in high pulmonary blood flow and low systemic output. Clinical problems associated with this shunt are pulmonary hemorrhage, hypotension, intraventricular hemorrhage, necrotizing enterocolitis, and mortality.1,2
Ductal constriction can be induced in preterm infants through medical intervention with cyclooxygenase inhibitors. However, the optimal management of PDA continues to be debated because numerous randomized clinical trials have failed to demonstrate any benefit of treatment on survival and long-term outcome.2,3 However open-label rescue treatments were authorized in all these trials, hindering any conclusions about the effects of untreated PDA, even if spontaneous closure occurred in some extremely preterm infants.4 With a lack of consensus5 reflecting a low level of proof, the management of PDA is highly variable among neonatal intensive care units (NICUs). Some neonatologists have been opting for conservative treatment of PDA, although there is no evidence to support this approach.3 Others perform early screening echocardiography for PDA, which allows the diagnosis and treatment of PDA at a preclinical stage. A recent randomized clinical trial6 based on this strategy showed that early targeted indomethacin treatment reduced the rate of pulmonary hemorrhage.
Observational studies are another way to evaluate a complex strategy.7 EPIPAGE 2 (Etude Epidémiologique sur les Petits Ages Gestationnels 2),8 a national population-based prospective cohort study, provided a unique opportunity to evaluate the association between early screening echocardiography for PDA and in-hospital morbidity and mortality in unselected preterm infants born before 29 weeks of gestational age and hospitalized in unselected NICUs. We hypothesized that early screening echocardiography for PDA would be associated with more frequent PDA treatment and better outcome.
The EPIPAGE 2 study is a national, prospective, population-based cohort study conducted in France in 2011. Recruitment took place at birth in all maternity units in France from April 2011 through December 2011, during an 8-month period for births occurring at 24 to 26 weeks and a 6-month period for births occurring at 27 to 28 weeks. The follow-up of participants is planned up to 2023. For this study, eligible children were those born at 24 to 28 weeks of gestation, admitted to the NICUs, alive at day 3, and with available information regarding whether an echocardiogram was performed, and if so, the timing and indication of the first echocardiographic assessment (screening or because of clinical signs).
Recruitment and data collection occurred only after families had received information and agreed to participate in this cohort by oral informed consent. The study was approved by the National Data Protection Authority, by the consultative committee on the treatment of information on personal health data for research purposes and the committee for the protection of people participating in biomedical research.
Preterm infants were placed into 2 groups based on whether they had undergone early (before day 3) screening (not because of PDA-related clinical signs) echocardiography. Those who had undergone this procedure were classified as exposed to early PDA screening echocardiography. The optimal timing for screening echocardiography is unknown; 3 days was chosen based on a prior trial of PDA treatment and clinical feasibility.9 Infants who had not undergone early screening were classified as nonexposed. All data were prospectively collected during NICU hospitalization. Information was collected about ductus arteriosus diagnosis, including PDA markers from the echocardiography (ductus arteriosus diameter >1.5 mm),10 high-velocity flow in the left pulmonary artery (mean velocity >0.4 m/s or end-diastolic velocity >0.2 m/s),11,12 diastolic aortic flow reversal13 or low superior vena cava flow (<40 mL/kg/min),14 pulsatile ductus shunt flow pattern,15 date of treatment with ibuprofen or surgical ligation, and indication for treatment.
The primary outcome was death between day 3 and discharge. The secondary outcomes were major neonatal morbidities: pulmonary hemorrhage; severe bronchopulmonary dysplasia, defined as a need for oxygen for at least 28 days plus a need for oxygen concentration 30% or greater, mechanical ventilator support, or continuous positive airway pressure at 36 weeks’ postmenstrual age; severe cerebral lesions, defined as intraventricular hemorrhage with ventricular dilation and intraparenchymal hemorrhage; cystic periventricular leukomalacia; and stage II and III necrotizing enterocolitis, according to Bell staging.
We used a propensity score approach to control for observed confounding factors that might influence both group assignment and outcome.16 Propensity score was defined as the infants’ probability of undergoing early screening echocardiography based on their individual observed covariates. Probability was estimated using a logistic regression model with early screening echocardiography as the dependent variable in relation to the following baseline maternal and infant characteristics (eTable 1 in the Supplement): gestational age, sex, birth weight z score based on Olsen curves,17 antenatal corticosteroids, main pregnancy complications, inborn status, mode of delivery, Apgar score, intubation at birth, surfactant therapy, respiratory distress syndrome, and early neonatal infections. Given the variation of early screening echocardiography before day 3 of life across intensive care units (range, 0%-100%), we did not consider the treatment center as a covariate in the propensity score. Instead, we used a proxy for center expertise, defined by summing the number of 24- to 28-week preterm infants spending their first consecutive 48 hours in the center. The proportion of patients with missing data ranged from 0% to 8.5%, exceeding 4% only for Apgar score at 5 minutes. Missing data were treated as a separate category.
The primary analysis was based on propensity matching. We used a 1:1 matching algorithm without replacement to match exposed and nonexposed newborns on gestational age, sex, and propensity score within a caliper of 0.2 standard deviation of the logit of the propensity score.18 Imbalance after matching was checked. The analysis of the matched set was carried out in 2 steps. First, the timing of ductus arteriosus echocardiographic evaluation, echocardiographic parameters, treatment, and in-hospital mortality were compared between exposed and nonexposed infants. Second, odds ratios (ORs) were calculated to quantify the association between early screening and outcomes using logistic regression fit by generalized estimating equations to account for paired data.19 The survivor functions of infants in both groups were estimated by Kaplan-Meier product-limit estimator and compared using Cox models with robust variance to account for correlations within the matched pairs.20
We used an instrumental variable approach as a confirmatory analysis of the principal analysis, since propensity score cannot remove hidden biases caused by unmeasured confounders.21 We used neonatal unit preference for early screening echocardiography as preference-based instrument.22 We estimated the unit effect in a mixed-effects logistic regression model with early screening echocardiography as dependent variable, maternal and infant characteristics as fixed effects, and a random neonatal unit effect. We categorized the estimated random unit effects into quartiles to define our instrumental variable. To assess the validity of the instrumental variable, we confirmed that the instrument was strongly associated with early screening echocardiography (eTable 2 in the Supplement) and examined the covariate balance across the instrument categories (eTable 3 in the Supplement). Instrumental variable analysis was carried out using the 2-stage residual inclusion approach,23 with additional adjustment for gestational age, sex, and birthweight z score. The strength of the instrument was formally assessed by the partial F statistic in the first-stage regression model.24
Six sensitivity analyses were performed. In the first, infants with signs of PDA were excluded from the propensity score analysis to clarify the role of screening if PDA is not suspected. The second used multiple imputation of missing data instead of a separate category.25 Multiple imputation by Monte Carlo Markov chains as implemented in the SAS MI procedure was carried out using all baseline variables of the propensity score model, as well as early screening echocardiography and the outcome, as recommended.26 Twenty-five independent imputed data sets were generated. A propensity score was estimated on each imputed data set and was used to create 2 matched sets, as for the main analysis. Each matched imputed data set was then analyzed and the resulting estimates were pooled according to the Rubin rule.25 Third, in the overall cohort, we used logistic regression models to assess the relationship between early screening echocardiography and the primary outcome after adjustment for gestational age (model 1) and after adjustment for gestational age, sex, and birthweight z score (model 2). We also used generalized estimating equation logistic regression with adjustment for gestational age and propensity score weighting, in which each infant was weighted by the inverse propensity of his or her group (model 3).27 The same analysis was performed taking center into account as an additional unit of clustering (model 4).
Fourth, we performed a negative control analysis using late-onset infections as a negative control outcome to detect uncontrolled confounding.28 Fifth, we analyzed preterm infants hospitalized in units that routinely performed early screening echocardiography, ie, for more than 85% of infants, and in units that performed early screening echocardiography for less than 15% of their patients. Only NICUs with more than 5 infants were enrolled in these analyses. No propensity score–based approach was used because of the limited risk of confounding in neonatal units with a homogeneous strategy, which did not depend on the infants’ characteristics. A sixth analysis was performed using data from preterm infants still alive at day 2 (n = 1538) to ensure that early screening echocardiography was not associated with an increase in the early mortality rate.
For analyses performed on the overall cohort, we used weighted percentages and gestational age adjustments in the multivariable analysis to take into account the differences in the recruitment times for the infants born at 24 to 26 weeks of gestation and those born at 27 to 31 weeks of gestation.
All tests were 2-sided. P values less than .05 were considered signiﬁcant. All analyses were performed using SAS version 9.3 (SAS Institute Inc).
Among the 1633 extremely preterm infants admitted to NICUs during the study period, 70 died within the first 2 days, leaving 1563 alive at day 3. Among these, information about the timing of the echocardiographic assessment and its indication was available for 1513 infants: 847 underwent early screening echocardiography (exposed group) and 666 did not (nonexposed group) ( Figure 1). These infants were hospitalized in 68 NICUs, wherein the percentage of early screening echocardiography varied from 0% to 100%. There was no significant relationship between this percentage and the volume of patients per NICU (Pearson correlation r = 0.06, P = .63; eFigure 1 in the Supplement), proportion of inborns (r = −0.03, P = .84), and proportion of extremely preterm (<27 weeks of gestation) infants (r = −0.05, P = .69) observed among the 61 NICUs with more than 5 infants enrolled. Differences between the exposed and nonexposed infants are presented in Table 1 and in eTable 1 in the Supplement.
Propensity scores were calculated for 1513 extremely preterm infants. Distributions of propensity scores are summarized in eFigure 2 in the Supplement. Propensity scores ranged from 0.17 to 0.92. The area under the receiver operating characteristic curve for the propensity score model was 0.64 (95% CI, 0.62 to 0.66). Among the 1513 extremely preterm infants, 1210 could be matched, with 605 in each group, exposed and nonexposed. The matched groups were found to be well balanced for all recorded baseline variables (Table 1 and eTable 1 in the Supplement).
Exposed infants had an echocardiogram performed at a median of 1 day (interquartile range, 0-2) vs 3.5 days (interquartile range, 0-98) for nonexposed infants. In the nonexposed group, 82% underwent nonscreening echocardiography. Among these, 65% were evaluated at day 3 or later, and 17% were evaluated before day 3 because of clinical signs potentially related to PDA (Table 1). The proportion of infants who received a treatment for PDA at any time during their hospitalization was significantly lower in the nonexposed group than in the exposed group (43.1% vs 55.1%, P < .001 [Table 1]) (odds ratio [OR], 1.62 [95% CI, 1.31 to 2.00]; absolute risk reduction [ARR] in events per 100 infants, −12.0 [95% CI, −17.3 to −6.7]). If only infants for whom echocardiography was performed are considered, the proportion who received a PDA treatment when any marker indicated hemodynamic significance did not differ between the exposed and nonexposed groups, regardless of the marker: 74% vs 75% for ductus arteriosus diameter (P = .70), 83% vs 85% for high-velocity flow in the left pulmonary artery (P = .59), 88% vs 84% for diastolic aortic flow reversal or low superior vena cava flow (P = .52), and 83% vs 78% for ductus low pattern (P = .37), respectively (Table 1).
Exposed infants had a lower death rate (14.2% vs 18.5%; OR, 0.73 [95% CI, 0.54 to 0.98]; ARR in events per 100 infants, 4.3 [95% CI, 0.3 to 8.3]) (Figure 2). The number of infants needed to be screened to prevent 1 death was 23. The difference in in-hospital mortality was mainly accounted for by infants who were not treated for PDA (13.9% vs 20.6%; OR, 0.62 [95% CI, 0.40 to 0.96]; ARR, 6.7 [95% CI, 4.0 to 9.6]) (Table 2). The results remained unaltered after excluding infants with clinical signs of PDA (OR, 0.62 [95% CI, 0.43 to 0.89]). Analysis using multiple imputation was consistent (OR, 0.74 [95% CI, 0.55 to 1.00]). In the extremely premature infants (24-26 weeks of gestation), the in-hospital death rate was 22.0% in the exposed group vs 28.4% in the nonexposed group (OR, 0.71 [95% CI, 0.48 to 1.06]; ARR, 6.3 [95% CI, −1.0 to 13.6]). In infants born at 27 to 28 weeks of gestation, death rates were 8.0% and 10.7%, respectively (OR, 0.73 [95% CI, 0.44 to 1.21]; ARR, 2.7 [95% CI, −1.7 to 6.7]). Figure 3 shows Kaplan-Meier survival curves of nonexposed and exposed groups in the matched cohort: the difference in survival accumulated gradually.
In addition, exposed infants had a lower rate of pulmonary hemorrhage than nonexposed infants (OR, 0.60 [95% CI, 0.38 to 0.94]). No significant differences were observed for necrotizing enterocolitis, severe bronchopulmonary dysplasia, and severe cerebral lesions (Table 2, eTable 4 in the Supplement). The causes of death are reported in eTable 5 in the Supplement. Death attributable to pulmonary hemorrhage was significantly lower in the exposed group (0/86 [0%] vs 7/112 [6.3%], P = .03).
The instrumental variable (ie, neonatal unit preference) was associated with early screening echocardiography rate (eTable 2 in the Supplement) but not with antenatal use of corticosteroids, inborn status, or late-onset infection rates, 3 quality markers usually related to outcome (eTable 3 in the Supplement). The partial F statistic for the instrumental variable in the first-stage model was 601(1,1509). Using this instrumental variable approach, the adjusted OR for the association between early screening echocardiography and mortality was not statistically significant (0.62 [95% CI, 0.37 to 1.04]).
Sensitivity analyses in the overall cohort were consistent with the primary analysis based on propensity matching; nevertheless, the association was not statistically significant in models 2 and 4 (Figure 2).
The negative control analysis did not show any significant association between early screening echocardiography and late-onset infection after adjustment for propensity score in the overall cohort (adjusted OR, 1.1 [95% CI, 0.9 to 1.4]). Analysis performed in preterm infants alive at day 2 (eFigure 3 in the Supplement) showed results similar to those of the primary analysis. Among preterm infants alive at day 3 and admitted into the 21 NICUs that performed early screening echocardiography for either more than 85% or less than 15% of their patients, the association was statistically significant (adjusted OR, 0.52 [95% CI, 0.31 to 0.86]) (eFigure 4 in the Supplement).
In this study, early screening echocardiography was associated with lower rates of in-hospital mortality and pulmonary hemorrhage, a well-recognized life-threatening complication of PDA.13 The higher mortality in the nonexposed group was mainly observed in untreated infants, who were in higher proportion in that group.
These findings are consistent with previous studies that reported a relationship between PDA and increased mortality in very premature infants.2,29,30 Animal experiments have shown a causal link between PDA and morbidity.31 Our survival curves reveal that the increase in mortality in the nonexposed group occurred at the end of the first postnatal week, consistent with the usual delay observed before PDA becomes clinically symptomatic. Moreover, the decrease in pulmonary hemorrhage rate is in agreement with the results of a recent randomized therapeutic trial based on echocardiographic screening.6 Nevertheless, the significant decrease in mortality attributable to pulmonary hemorrhage only partially explains the difference in in-hospital mortality between the 2 groups, since multiple organ failure with or without proven sepsis and respiratory failure also accounted for excessive mortality in the nonexposed group.
Randomized clinical trials are the standard strategy to assess medical interventions. However, neither individual randomized clinical trials nor meta-analyses have demonstrated any benefit of PDA treatment on survival and long-term outcomes of patients.2,3,32 Unfortunately, this lack of benefit can be misinterpreted,3 because most randomized clinical trials focused on the timing of early treatment, rather than on the risks of PDA closure vs prolonged patency. Indeed, open-label rescue treatment was given to 15% to 40% of patients in the placebo groups.6,32,33 This backup treatment strategy, which is difficult to control for in randomized clinical trials, may induce biases and hinder any definitive conclusion about PDA treatment.3,5
As in all observational studies, the main limitation of our study is uncontrolled confounding. We used multiple statistical approaches to reduce biases as much as possible. To control for the indication bias inherent in this type of study, we performed a propensity score analysis and made a rigorous adjustment for confounding factors, minimizing the likelihood of incorrectly attributing the association to early echocardiographic screening. At the individual level, performing an echocardiographic evaluation because of clinical signs of PDA, instead of screening echocardiography, classified an infant in the nonexposed group. This could have caused a bias toward classifying the sickest infants in the nonexposed group. Indeed, the presence of PDA in a premature infant is a risk factor for unfavorable outcomes. Nevertheless, the difference in in-hospital mortality between the 2 groups was not apparent in the subgroup of preterm infants who displayed clinical signs of PDA, underwent echocardiography, and were subsequently treated. On the contrary, increased mortality was observed in the infants from the nonexposed group in whom PDA was never assessed or treated. At the NICU level, the complementary analysis of infants hospitalized in the subgroup of neonatal units with a highly consistent screening strategy (either >85% or <15% of patients exposed) confirmed the relationship between early screening echocardiography and decreased mortality. Although the instrumental variable was not statistically significant and thus not completely confirmatory, this analysis provided a point estimate close to that from the primary analysis using propensity score matching and could still be considered supportive because of similar effect size and because the confidence interval just crossed 1. Moreover, we did not observe any relationship between the characteristics of a NICU and its PDA screening strategy, suggesting the lack of performance bias. The reasons for PDA screening seem to be based on medical opinion.34
The strengths of the EPIPAGE 2 study include its population-based cohort design and prospective enrollment of all infants born prematurely in France in 2011. Furthermore, the data on hemodynamic status were prospectively collected with a detailed standardized questionnaire and standardized definitions. Therefore, comprehensive and accurate information was available on our variable of interest. The originality of this study consists in evaluating the association between early screening echocardiography, rather than early treatment, and in-hospital mortality. Neonatologists who favor conservative management of PDA usually do not actively screen infants. Therefore, we chose a marker that would reflect the overall opinion about PDA.
In this national population-based cohort of extremely preterm infants, screening echocardiography for PDA before day 3 of life was associated with lower in-hospital mortality and likelihood of pulmonary hemorrhage but not with differences in necrotizing enterocolitis, severe bronchopulmonary dysplasia, or severe cerebral lesions. However, results of the instrumental variable analysis leave some ambiguity in the interpretation, and longer-term evaluation is needed to provide clarity.
Corresponding Author: Jean-Christophe Rozé, MD, PhD, Department of Neonatal Medicine, University Hospital of Nantes, Nantes University, 36 Boulevard Jean Monnet, 44093 Nantes, France (firstname.lastname@example.org).
Author Contributions: Drs Rozé and Ancel had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Rozé, Cambonie, Durrmeyer, Porcher, Ancel.
Acquisition, analysis, or interpretation of data: Rozé, Marchand-Martin, Gournay, Durrmeyer, Durox, Storme, Porcher, Ancel.
Drafting of the manuscript: Rozé, Cambonie, Marchand-Martin, Gournay, Durrmeyer, Storme, Ancel.
Critical revision of the manuscript for important intellectual content: Rozé, Marchand-Martin, Durrmeyer, Durox, Porcher, Ancel.
Statistical analysis: Rozé, Marchand-Martin, Porcher, Ancel.
Obtained funding: Ancel.
Administrative, technical, or material support: Gournay, Durox.
Study supervision: Ancel.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Group Information: Members of the Hemodynamic EPIPAGE 2 (Etude EPIdémiologique sur les Petits Ages GEstationnels) Study Group: Gilles Cambonie, MD, PhD (Department of Neonatal Medicine, Montpellier University Hospital, Montpellier, France); Jean-Christophe Rozé, MD, PhD (Department of Neonatal Medicine, Nantes University Hospital, Nantes, France); Pierre-Yves Ancel, MD, PhD (INSERM, U1153, Obstetrical, Perinatal and Pediatric Epidemiology Team, Epidemiology and Biostatistics Sorbonne, Paris, France); Laetitia Marchand-Martin, MS (INSERM, U1153, Obstetrical, Perinatal and Pediatric Epidemiology Team, Epidemiology and Biostatistics Sorbonne, Paris, France); Mélanie Durox, MSc (INSERM, U1153, Obstetrical, Perinatal and Pediatric Epidemiology Team, Epidemiology and Biostatistics Sorbonne, Paris, France); Veronique Gournay, MD, PhD (Pediatric Cardiology Unit, Nantes University Hospital, Nantes, France); Xavier Durrmeyer, MD, PhD (Department of Neonatal Medicine, Intercomunal hospital, Creteil, France); Laurent Storme, MD, PhD (Department of Neonatal Medicine, Lille University Hospital, Lille, France); Raphael Porcher, PhD (INSERM, U1153, METHODS Team, Epidemiology and Statistics Sorbonne Paris Cité Research Center, Paris, France); Patrice Morville, MD (Department of Neonatal Medicine, Reims University Hospital, Reims, France); Olivier Brissaud, MD (Department of Neonatal Medicine, Bordeaux University Hospital, Bordeaux, France); Patrick Truffert, MD, PhD (Department of Neonatal Medicine, Lille University Hospital, Lille, France); Antoine Bouissou, MD (Department of Neonatal Medicine, Tours University Hospital, Tours, France); Isabelle Ligi, MD, PhD (Department of Neonatal Medicine, Marseille University Hospital, Marseille, France); Marie Odile Marcoux, MD (Department of Neonatal Medicine, Toulouse University Hospital, Toulouse, France); Fabrice Cneude, MD (Department of Neonatal Medicine, Grenoble University Hospital, Grenoble, France); Geraldine Gascoin, MD, PhD (Department of Neonatal Medicine, Angers University Hospital); Gerard Thiriez, MD, PhD (Department of Neonatal Medicine, Besançon University Hospital, Besançon, France); Hugues Patural, MD, PhD (Department of Neonatal Medicine, St Etienne University Hospital, St Etienne, France); Doriane Madeleneau, MD (Department of Neonatal Medicine, Cochin University Hospital, Paris, France); Antoine Burguet, MD, PhD (Department of Neonatal Medicine, Dijon University Hospital,Dijon, France); Patrick Pladys, MD, PhD (Department of Neonatal Medicine, Rennes University Hospital, Rennes, France).
Funding/Support: This study was supported by the French Institute of Public Health Research/Institute of Public Health and its partners the French Health Ministry, the National Institute of Health and Medical Research, the National Institute of Cancer, and the National Solidarity Fund for Autonomy; grant ANR-11-EQPX-0038 from the National Research Agency through the French Equipex Program of Investments in the Future; and the PremUp Foundation.
Role of the Funders/Sponsors: The funders/sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.
Additional Contributions: We gratefully acknowledge Philippe Ravaud, MD, PhD, Director of the Epidemiology and Statistics Sorbonne Paris Cité Research Center, Paris Descartes University, France, for methodological guidance and review of the manuscript. Dr Ravaud received no compensation for his contributions.