To compare risk-adjusted changes in outcomes of preterm infants <29 weeks gestation born in 1996 to 1997 with those born in 2006 to 2007.
Observational retrospective comparison of data from 15 units that participated in the Canadian Neonatal Network during 1996 to 1997 and 2006 to 2007 was performed. Rates of mortality and common neonatal morbidities were compared after adjustment for confounders.
Data on 1897 infants in 1996 to 1997 and 1866 infants in 2006 to 2007 were analyzed. A higher proportion of patients in the later cohort received antenatal steroids and had lower acuity of illness on admission. Unadjusted analyses revealed reduction in mortality (unadjusted odds ratio (UAOR): 0.83, 95% confidence interval (CI): 0.63, 0.98), severe retinopathy (UAOR: 0.68, 95% CI: 0.50 to 0.92), but increase in bronchopulmonary dysplasia (UAOR: 1.61, 95% CI: 1.39 to 1.86) and patent ductus arteriosus (UAOR: 1.22, 95% CI: 1.07 to 1.39). Adjusted analyses revealed increases in the later cohort for bronchopulmonary dysplasia (adjusted odds ratio (AOR): 1.88, 95% CI: 1.60 to 2.20) and severe neurological injury (AOR: 1.49, 95% CI: 1.22 to 1.80). However, the ascertainment methods for neurological findings and ductus arteriosus differed between the two time periods.
Improvements in prenatal care has resulted in improvement in the quality of care, as reflected by reduced severity of illness and mortality. However, after adjustment of prenatal factors, no improvement in any of the outcomes was observed and on the contrary bronchopulmonary dysplasia increased. There is need for identification and application of postnatal strategies to improve outcomes of extreme preterm infants.
Neonatal intensive care has evolved remarkably over the last few decades. The perception is that there has been a significant reduction in mortality and morbidities for vulnerable preterm infants as a result of the advent of surfactant, administration of antenatal corticosteroids to mothers with imminent preterm births, recent changes in the respiratory management (including high frequency ventilation, use of nitric oxide, tolerance for higher carbon dioxide levels and lower oxygen saturation) and improved nutritional regimens preventing catabolism and supporting anabolism. These strategies have been tested in experimental intervention and some of these strategies have been adopted into practice to a greater or lesser extent, and at variable times across Canadian neonatal intensive care units (NICUs).1, 2 Not only in Canada but also nearly all industrialized countries have undergone major health policy changes directed at improving the value of health-care delivery in all fields including neonatology. Collaborative quality improvement initiatives in the last decade between centers, regions and countries have often been implemented with an expectation of beneficial effects. However, it remains unclear whether or not such advances have resulted in better outcomes for preterm neonates <29 weeks gestational age (GA). Moreover, as the technology grows more complex, there is an ongoing concern that high mortality at extremely low GA will be replaced by survival with severe morbidity. Reports from other networks or collaborative efforts have compared outcomes over time. Some reports have indicated that improvements are at best marginal and some studies provide evidence of worsening outcomes.3, 4, 5
The Canadian Neonatal Network (CNN) is a unique resource that was developed to evaluate changes and trends in neonatal outcomes in the setting of a publically funded health-care system. The network was established in 1996. The network produced its first report after data were collected from 16 centers for an initial period of 22 months.2 In this study, our aim was to compare the neonatal outcomes of preterm infants of <29 weeks GA from all the centers that participated in data collection during both 1996 to1997 and 2006 to 2007 and to assess changes over time. The purpose of this study was twofold: one to assess improvements in the quality of care over the decade (by performing unadjusted analyses) and second was to assess reasons to explain the differences in outcomes (by performing adjusted analyses).
Data from eligible NICU admissions from participating sites are collected and submitted after institutional approval (either from a local Research Ethics Board or an institutional quality improvement process). The CNN maintains an established national database for the purpose of evaluation of outcomes, benchmarking and quality improvement. An initial 2 years endeavor (1996 to 1997) has been extended to continue as an ongoing national data collection. In 2006 to 2007, 22 of 27 tertiary NICUs in Canada were participants in the Network, each of which had trained research personnel who abstracted data from patient charts. Details of data collection and data management have been published elsewhere2 and specifications for data collection have remained consistent over 10 years. Of the 16 centers that participated in 1996 to 1997, one center did not participate in data collection during 2006 to 2007, so their data were excluded.2 Data on infants born at <29 weeks GA admitted to the remaining 15 participating centers from January 1996 to October 1997 and from January 2006 to October 2007 were analyzed. Infants with lethal congenital anomalies, those who were born at <23 weeks GA or those who were moribund on admission (that is, a physician, in consultation with the parents, had made an explicit decision not to provide life support at the time of the NICU admission; n=24 in earlier cohort and 22 in the later cohort without any significant difference at any GA in completed weeks) were excluded from the analyses. It is possible that neonates considered moribund in the first cohort may have been resuscitated in the later cohort; however, there was no difference in the number.
The infants were divided into two groups:
Group 1: Preterm infants admitted during 1996 to1997 to 15 participating centers in the CNN.
Group 2: Preterm infants admitted during 2006 to 2007 to the same 15 participating centers in the CNN.
Study variables were defined according to the CNN manual. Gestational age (GA) was defined as the best estimate based on early prenatal ultrasound, obstetric examination and obstetric history, followed by pediatric estimate unless the postnatal pediatric estimate of gestation differed from the obstetric estimate by more than 2 weeks. In that case, the pediatric estimate was used instead. Small for gestational age was defined as birth weight less than the tenth percentile for GA and sex.6 Score of Neonatal Acute Physiology (SNAP) was calculated to assess severity of illness on admission.7 Neonates were considered to have received a complete course of antenatal steroid if two doses were given to the mother between 2 and 7 days prior to birth. Bronchopulmonary dysplasia (BPD) was defined as the need for any form of respiratory support (oxygen or positive pressure support) at 36 weeks post-menstrual age8 or at the time of discharge to level two centers. Intraventricular hemorrhage (IVH) was defined according to the criteria of Papile et al.9 from the most severe findings on head ultrasound during the infant's stay. Periventricular echogenicity or leukomalacia were reported based on ultrasound findings. However, for these latter morbidities, the method for detection and description of findings changed between 1996 to 1997 and 2006 to 2007. For 1996 to 1997, IVH was reported using a grading system; however, in 2006 to 2007 intracranial pathology was reported descriptively and there was an improved detection of parenchymal lesions due to an improvement in the available technology (higher resolution ultrasound scanners). Retinopathy of prematurity (ROP) was classified according to the international classification.10 Necrotizing enterocolitis (NEC) was defined according to Bell's criteria (stage 2 or higher), and was classified as medical NEC (clinical symptoms and signs plus evidence of pneumatosis on abdominal X-ray in patients who were treated medically) or surgical NEC (laparotomy in addition to medical treatment).11 Patent ductus arteriosus (PDA) was diagnosed clinically, with or without echocardiography. The frequency of echocardiography and thus detection rate varied between the two eras, and no restriction was applied for the diagnosis in either era as far as method of detection is concerned.
Death in the NICU, BPD, severe IVH (grade 3 IVH or periventricular leukomalacia), severe (>stage 2 in either eye) ROP, PDA and NEC were compared between the two groups.
Univariate analyses (χ2-test for categorical variables and Student's t-test for continuous variables) were performed to compare the characteristics of the study populations and to explore the association between population characteristics and clinical outcomes. Both adjusted and unadjusted analyses were performed and reported. One purpose of the study was to investigate improvement in quality of care over the decade and unadjusted analyses were justified. However, the other purpose was to explain the reasons for differences in outcomes and thus, multivariate logistic regression analyses were used to compare neonatal outcomes of infants in the two groups after adjustment for variables identified with P0.10 in univariate analyses. Because of debate surrounding whether to adjust for perinatal risks and process variables, we did not use random effects model for analysis. The statistical software package SAS (Cary, NC, USA) was used for data analyses. A P-value of <0.05 was considered significant. Model appropriateness was assessed using Hosmer–Lemeshow statistics.
A total of 1897 infants in the 1996 to 1997 cohort and 1866 patients in the 2006 to 2007 cohort were eligible for analyses. Univariate comparison (Table 1) revealed that a higher proportion of infants born in 2006 to 2007 compared with 1996 to 1997 were delivered via cesarean section, were born to hypertensive mothers, received a partial or complete course of antenatal steroids and had lower SNAP scores. There was no difference in the number of infants with nosocomial infection (31.9 vs 31.8%; P=0.92). Unadjusted comparisons of outcomes revealed significantly lower odds of mortality and severe ROP, but higher odds of grade 3 and 4 IVH/periventricular leukomalacia among the 2006 to 2007 cohort compared with the 1996 to 1997 cohort. Logistic regression analyses (Table 2) controlling for GA, sex, antenatal steroid use, delivery type, outborn, maternal hypertension and SNAP score revealed no significant differences in mortality, severe ROP or NEC between the two groups; however, there was an increase in BPD. The rates of grade 3 and 4 IVH/periventricular leukomalacia and PDA were increased in patients in the 2006 to 2007 group. However, the criteria for classification of neurological injury on cranial imaging findings and methods and frequency of detection of PDA differed between the two eras. In order to understand difference in competing outcomes, rates of BPD or death were calculated, which were not different (adjusted odds ratio 0.92, 95% confidence interval: 0.79, 1.08) between two cohorts.
In this large cohort study of preterm infants <29 weeks GA, we identified that BPD and PDA increased, whereas mortality and severe ROP reduced in preterm infants born at <29 weeks GA over time in unadjusted analyses. When adjusted for perinatal risks and illness severity on admission, no difference in mortality or severe ROP was identified; however, BPD remained significantly higher in the later cohort. No difference in composite outcome of death or BPD reflect that the increase was due to an increase in BPD. Differences in the odds of neurological imaging findings and PDA were also observed, but are possibly attributable to differences in the rates of detection and classification. Analyses of baseline characteristics revealed higher rates of maternal hypertensive disorders, increased use of cesarean section, antenatal steroids, and improved condition and stabilization after birth of infants in the later cohort. Differences in results between unadjusted and adjusted analyses could possibly be explained by the improvements in obstetric care provision, prenatal interventions and increased use of antenatal steroid. This was obvious because there was reduced severity of illness at admission (SNAP score). These results indicate that improvement in mortality in the later cohort may be explained by higher antenatal steroid usage and lower severity of illness in the later cohort. These could be reflective of an ‘improved package of care’ provided to women in the later cohort; however, definitive remarks regarding this could only be made with detailed data on prenatal and intrapartum care provision in both time periods.
Advent of surfactant and recommendations regarding antenatal corticosteroids in the early 1990s led to a reduction in mortality of preterm infants.12 However, the impact of more recent advances, such as high frequency ventilation, permissive hypercapnea, lower target oxygen saturation thresholds, nitric oxide and increased use of non-invasive ventilation etc., is unclear. We do not have detailed data to evaluate this impact in our data set. Horbar et al.4 evaluated outcomes of infants admitted to hospitals participating in the Vermont Oxford Network and reported that the rates of pneumothorax, mortality and severe IVH were reduced from 1991 to 1995, but there was no change in the mortality and IVH rates between 1995 and 2000 and there was an increase in pneumothorax during the later period. Paul et al.13 in a single center study reported an increasing severity of illness, as measured by SNAP score and thyroxine level, from 1993 to 2002 among very low birth weight infants. There was an associated increase in the combined outcome of death and severe IVH over this time period (18% in 1993 to 1996 cohort compared with 26% in 1999 to 2002 cohort).
Fanaroff et al.3 reported data from participating centers in the National Institute of Child Health & Human Development (NICHD) Neonatal Research Network for very low birth weight infants during three eras: 1987 to 1988, 1993 to 1994 and 1999 to 2000. There was a reduction in mortality (23, 17 and 14%, respectively, over the three time periods), no change in the rate of NEC (6, 6 and 7%, respectively) and increases in the rates of sepsis (17, 21 and 22%, respectively) and PDA (10, 31 and 22%, respectively). The rate of severe IVH was reduced in 1993 to 1994 compared with 1987 to 1988 (from 18 to 11%), but there was no change between 1993 to 1994 and 1999 to 2000 (11 to 12%). Recent data from the NICHD Neonatal Research Network also indicated improved survival, but higher rates of morbidities among very low birth weight neonates in 2003 to 2007.5
Payne et al.14 reported outcomes of 9 years of a quality improvement project to reduce lung injury from eight centers in the United States for very low birth weight infants admitted between 1998 and 2006. There was a significant shift in the delivery room practices toward non-invasive support; however, their BPD-free survival remained unchanged (adjusted odds ratio: 0.9; 95% confidence interval: 0.7 to 1.1) and the BPD rate increased (adjusted odds ratio: 1.3; 95% confidence interval: 1.1 to 1.6). Doyle et al.15 reported that quality-adjusted survival of preterm infants <28 weeks GA in Victoria, Australia, improved from 42% in a 1991 to 1992 cohort to 55% in 1997, but remained unchanged at 53% in 2005.
In a nationwide cohort of extremely low birth weight infants from Finland,16 there was no improvement in mortality between 1996 to 97 and 1999 to 2000, and there was an increase in rates of respiratory distress syndrome, sepsis and IVH in the later period. Thus, it is clear that some of the observations made over the period from 1995 to 2007 in different settings, including ours, show a concerning lack of improvement during this period. By contrast, Zeitlin et al.17 reported improvements in mortality and morbidities in cohort of 2003 compared with a 1998 cohort following regionalization of neonatal care in the Parisian region of France. Itabashi et al.18 from Japan reported an improvement in mortality at each GA <29 weeks over a 15-year period from 1990; however, their overall survival of 83% among extremely low birth weight infants (which mostly included infants <29 weeks GA) is lower than that reported in our study.
Our findings, and those of other authors, raise two main types of questions: process of care-related and population-related issues.
Process of care provision-related issues:
Improvements in prenatal care provision resulted in improvement in unadjusted rates of mortality. It could be argued whether or not it is appropriate to adjust for confounders in such comparison. We believe it was appropriate to explain the reasons for improvement; however, if we are purely looking at change in the quality of care (outcomes) then unadjusted results are appropriate.
It is possible that an increase in BPD reflect a reduction in the use of post-natal systemic steroids. However, this change has not been observed in previous analyses by our network and others, despite a reduction in the use of postnatal steroids.19 Use of postnatal steroids may reduce BPD, but may affect neurodevelopment acting as a dual-edge sword. We do not have data on neurodevelopment from either cohort to compare. It could also be due to change in the oxygen saturation targets that may have occurred over the decade. However, to our knowledge, recent trend in majority of units is to accept lower saturation targets rather than higher.
These results can be thought as if we have reached the threshold of neonatal intensive care provision. However this hypothesis is not supported by data from France17 and mortality data from Japan18, which confirmed that further improvements in survival is possible at lower GA.
One can argue that this could be a reflection of a distinct gap between the results of research findings and the real life scenario when they are applied en masse (effectiveness vs efficacy)? A number of interventions employed today have undergone rigorous evaluation; however, it has always been a concern that when research findings are applied to the general population the effects are significantly reduced because research participants are a selected group of mothers/infants with potentially favorable characteristics.20
Finally, several of the strategies employed today, such as permissive hypercapnea,21 permissive acidosis, permissive hypotension,22 and treatment of hypotension have not undergone rigorous evaluation. These evolving practices in neonatal intensive care may confer no benefit, or may even be harmful. Recently, results from the SUPPORT trial of two oxygen saturation targets revealed increased mortality in the group assigned to lower saturation targets, cautioning stricter evaluation of practices prior to adoption in clinical practice.23
Population related issues:
Over the decade we have noticed a reduction in the stillbirth rate and an increase in the preterm birth rate.24 It is possible that the characteristics of the population have changed sufficiently to offset improvements. However, the overall numbers of admissions to the 15 centers were similar in the two cohorts and additionally there was no difference in individual GA group as well. During both study period, majority of preterm infants were cared by the same units in a geographic region and no regionalization related changes were identified. Additionally, to avoid residual confounding we adjusted for outborn status in multivariate analyses even though in univariate analyses it was borderline significant. Additionally, time of transfer of outborn infants is not an issue in the situation of <29 weeks GA infants, as they are transferred to tertiary NICU in the region from all peripheral centers as soon as possible.
With an increase in the use of assisted reproduction it could be argued that there were more multiples in the later cohort; however, we identified no significant difference in the number of multiples between two cohorts.
Finally, despite the above speculation, no clear mechanisms have been identified as explaining the increase in some adverse outcomes over a decade in our population. These questions should, therefore, provoke further dialogue in the wider neonatal community. Our results and those from many other networks are concerning to global neonatal community and challenges clinicians and researchers to identify, test and implement approaches that can lead to better outcomes.
The strengths of our study include the large sample size of preterm neonates <29 weeks GA, a geographically representative cohort, adjustments for perinatal risks and severity of illness at the time of admission and comparison of outcomes in the same NICUs over a 10-year period. Weaknesses of our study include the lack of data describing differences between the practices at various NICUs (such as hypothermia on admission, oxygen saturation target, primary modes of respiratory support, nutrition, feeding practices, vitamin A administration, surfactant administration criteria, number of infants breast-feeding at discharge etc.), the culture of change at these participating units and the actual practice changes that may have taken place over time in these units. It is also possible that outcomes may have improved at some of the units and deteriorated at others; however, low statistical power for comparison of individual units precluded such analyses. Center variability is an issue even in large randomized controlled trials. Selection bias poses a continual threat in such retrospective comparison; however, similarity in baseline criteria, relatively similar number of patients and similar number of exclusion due to moribund status provide certain reassurance on this bias. These results may not be applicable to other units of other networks, as patient characteristics and practices in units vary.
In conclusion, for preterm infants born at <29 weeks gestation admitted to 15 Canadian NICUs during 2006 to 2007 compared the same NICUs during 1996 to 1997, there was no change in mortality, severe ROP and NEC, but there was an increase in BPD when compared after controlling for confounders. These results are concerning and challenge investigators from other networks to check the validity of these findings and to identify clinically applicable strategies for improvement in outcomes of preterm neonates <29 weeks gestation.
This study was supported by Grant MOP-53115 from the Canadian Institutes of Health Research. Additional funding was provided by participating hospitals. We thank Woojin Yoon for statistical support of this project. Canadian Neonatal Network coordinating center (Mother-Infant Care Research Center) is supported by Ministry of Health and Long-term Care, Ontario, Canada. Funding agencies had no role in design, collection, analyses or interpretation of results of this study.
Source of funding: Please see Appendix.
Site Investigators of the Canadian Neonatal Network
Shoo K Lee (Director, Canadian Neonatal Network); Prakesh S Shah (Associate Director, Canadian Neonatal Network); Wayne Andrews (Janeway Children's Health and Rehabilitation Centre, St John's, NL); Keith Barrington (St Justine's Hospital, Montreal, QC); Wendy Yee (Foothills Medical Centre, Calgary, AB); Barbara Bullied (Everett Chalmers Hospital, Fredericton, NB); Rody Canning (Moncton Hospital, Moncton, NB); Gerarda Cronin (St Boniface General Hospital, Winnipeg, MB); Kimberly Dow (Kingston General Hospital, Kingston, ON); Michael Dunn (Sunnybrook Health Sciences Centre, Toronto, ON); Adele Harrison (Victoria General Hospital, Victoria, BC); Andrew James (Sick Kids Hospital, Toronto, ON); Zarin Kalapesi (Regina General Hospital, Regina, SK); Lajos Kovacs (Jewish General Hospital, Montreal, QC); Orlando da Silva (St Joseph's Health Centre; London, ON); Douglas D McMillan (IWK Health Centre, Halifax, NS); Prakesh Shah (Mount Sinai Hospital, Toronto, ON); Cecil Ojah (St John Regional Hospital, St John, NB); Abraham Peliowski/Khalid Aziz (Royal Alexandra Hospital, Edmonton, AB); Bruno Piedboeuf (Centre Hospitalier Universitaire de Quebec, Sainte Foy, QC); Patricia Riley (Montreal Children's Hospital, Montreal, QC); Daniel Faucher (Royal Victoria Hospital, Montreal, QC); Nicole Rouvinez-Bouali (Children's Hospital of Eastern Ontario, Ottawa, ON); Koravangattu Sankaran (Royal University Hospital, Saskatoon, SK); Mary Seshia (Health Sciences Centre, Winnipeg, MB); Sandesh Shivananda (Hamilton Health Sciences Centre, Hamilton, ON); Zenon Cieslak (Royal Columbian Hospital, New Westminster, BC); Anne Synnes (Children's and Women's Health Centre of British Columbia, Vancouver, BC); and Herve Walti (Centre Hospitalier Universitaire de Sherbrooke, Fleurimont, QC).
About this article
Contribution of authors
Prakesh S Shah: initiated concept, developed draft, analyzed data and wrote the manuscript; Koravangattu Sankaran: participated in design, review of protocol and manuscript writing; Khalid Aziz: participated in design, concept, and revision and editing of the manuscript; Alexander C Allen: participated in design, interpretation and revision of the manuscript; Mary Seshia: initiated concept, participated in design, contributed in writing and editing of the manuscript; Arne Ohlsson: initiated concept, participated in design, contributed in writing and editing of the manuscript; and Shoo K Lee: participated in the design of study, review of results and manuscript preparation.
Mortality Rate-Dependent Variations in the Timing and Causes of Death in Extremely Preterm Infants Born at 23–24 Weeks’ Gestation*
Pediatric Critical Care Medicine (2019)