Objectives. Reference data are lacking on the frequency and duration of assisted ventilation in neonates. This information is essential for determining resource needs and planning clinical trials. As mortality becomes uncommon, ventilator utilization is increasingly used as a measure for assessing therapeutic effect and quality of care in intensive care medicine. Valid comparisons require adjustments for differences in a patient's baseline risk for assisted ventilation and prolonged ventilator support. The aims of this study were to determine the frequency and length of ventilation (LOV) in preterm and term infants and to develop models for predicting the need for assisted ventilation and length of ventilator support.
Methods. We performed a retrospective, population-based cohort study of 77 576 inborn live births at 6 Northern California hospitals with level 3 intensive care nurseries in a group-model managed care organization. The gestational age-specific frequency and duration of assisted ventilation among surviving infants was determined. Multivariable regression was performed to determine predictors for assisted ventilation and LOV.
Results. Of 77 576 inborn live births in the study, 11 199 required admission to the neonatal intensive care unit and of these, 1928 survivors required ventilator support. The proportion of infants requiring assisted ventilation and the median LOV decreased markedly with increasing gestational age. In addition to gestational age, admission illness severity, 5-minute Apgar scores, presence of anomalies, male sex, and white race were important predictors for the need for assisted ventilation. The ability of the models to predict need for ventilation was high, and significantly better than birth weight alone with an area under the receiver operating characteristic curve of .90 versus .70 for preterm infants, and .88 versus .50 for term infants. For preterm infants, gestational age, admission illness severity, oxygenation index, anomalies, and small-for-gestational age status were significant predictors for LOV, accounting for 60% of the variance in the length of assisted ventilation. For term infants, oxygenation index and anomalies were significant predictors but only accounted for 29% of the variance.
Conclusions. Considerable variation exists in the utilization of ventilator support among infants of closely related gestational age. In addition, a number of medical risk factors influence the need for, and length of, assisted ventilation. These models explain much of the variance in LOV among preterm infants but explain substantially less among term infants. neonatal intensive care, assisted ventilation, Score for Neonatal Acute Physiology, resource consumption, prematurity.
- LOV =
- length of ventilation •
- NICU =
- neonatal intensive care unit •
- KPMCP =
- Kaiser Permanente Medical Care Program •
- NCPAP =
- nasal continuous positive airway pressure •
- NMDS =
- Kaiser Permanente Neonatal Minimum Data Set •
- SNAP =
- Score for Neonatal Acute Physiology •
- SGA =
- small-for-gestational age •
- LGA =
- large-for-gestational age •
- OI =
- oxygenation index •
- ROC =
- receiver operating characteristic
Assisted ventilation is one of the costliest procedures in neonatal intensive care. It requires extra nursing staff, respiratory care technicians, and significant capital investment. Ventilation accounts for 29% to 37% of the total cost variance in extremely low birth weight infants.1 ,2 Despite its economic and clinical impact, our knowledge regarding utilization of assisted ventilation is scant. Few population-based data are available on the frequency and length of neonatal assisted ventilation.3 ,4 Only the study by Synnes et al4 provided information on both rates and length of assisted ventilation in neonates for different gestational age groups. That study focused only on extremely premature infants born before the introduction of exogenous surfactant therapy.
Because neonatal mortality is now very low except among the extremely immature,5 ,6 morbidity data are needed for monitoring quality of care, assessing internursery differences in outcomes and for resource allocation. Clinical care would benefit from prior identification of infants at high risk for prolonged ventilation. More refined means to predict length of ventilation (LOV) would be helpful in a variety of ways, such as the selection of patients for novel therapies for chronic lung disease, deciding when to use existing treatments such as tracheostomy or steroid therapy, and counseling families. Assisted ventilation, which is easily quantifiable, commonly used and associated with measurable changes in patient health, meets the criteria for a good indicator for measuring quality of care.7 Length of assisted ventilation is being utilized as an outcome measure to compare care among centers8 and to evaluate the cost-effectiveness of therapeutic interventions.9
The aims of this study were to: 1) determine the frequency and duration of assisted ventilation in a birth cohort in a community setting, and 2) develop models for predicting the need for assisted ventilation and length of ventilator support using variables available on the day of admission to the neonatal intensive care unit (NICU). The setting was the Kaiser Permanente Medical Care Program (KPMCP), a group-model managed care organization.
The KPMCP covers ∼30% of the population of Northern California. It is demographically similar to the general California population, although the very poor and very wealthy are underrepresented.10 The study sites consisted of all 6 KPMCP Northern California level 3 NICUs. Board-certified neonatologists staff these 6 units. Infants requiring a few forms of subspecialized care (eg, complex cardiac surgery and extracorporeal membrane oxygenation) are transported to university medical centers but return to KPMCP facilities before discharge home. Surfactant therapy was in wide use in all facilities by 1991. The preferential use of nasal continuous positive airway pressure (NCPAP) over intubation and mechanical ventilation in KPMCP facilities occurred in the early 1990s after an observational study showed a decrease in chronic lung disease in centers where NCPAP was commonly used.11
The methods we used for data collection and electronic linkage have been previously reported.12–16 Two primary data sources were used in this study. The KPMCP hospitalization database provided population-based data on the number of inborn live births for each gestational age. Patient specific data on morbidity and therapy for all infants admitted to the 6 study NICUs were obtained from the Kaiser Permanente Neonatal Minimum Data Set (NMDS) database.12 NMDS research assistants also track infants while they are receiving specialized care outside the KMPCP.
Data quality in the NMDS is maintained by formal training for research assistants, detailed protocols, and electronic checks on data entry that do not permit out of range data. We also performed secondary audits on data from all infants who were LOV outliers in their gestational age group. We defined as outliers, infants with values greater than the upper quartile plus 1.5 times the interquartile range.
We developed a retrospective cohort of inborn live births greater than 22 weeks of gestation who were born at a study facility between August 1, 1992 and November 30, 1997. For regression analysis, only survivors were included, because infants who die are often those from whom life support is withheld or withdrawn. Because strictly defined criteria for discontinuing life support do not exist, inclusion of nonsurvivors in the analysis could significantly distort results.17
Because this computer database was phased in between July 1992 and January 1995, the birth cohort was selected to correspond to the data of NMDS. Currently, the NMDS database captures 95% to 100% of level 3 NICU admissions in a region of ∼2.9 million members and 32 000 deliveries each year.12
The Kaiser Permanente Institutional Review Board for the Protection of Human Subjects approved this project.
For the purposes of this investigation, assisted ventilation included intermittent mandatory ventilation, synchronized intermittent mandatory ventilation, high-frequency oscillatory ventilation, as well as NCPAP. Only infants who were mechanically ventilated in the NICU were included in the numerator for gestational age-specific rates of assisted ventilation. Infants who were briefly ventilated and died before NICU admission were not included, because we were interested in resource utilization in the NICU.
LOV was recorded in days before 1997 and subsequently in hours. For consistency, we used calculated LOV in days with part of a day counting as a whole.
Gestational age for the birth cohort was determined from either obstetrical data or by the Ballard test.18 For all infants admitted to the NICU, the Ballard test was used, rounding down to the nearest gestational age in weeks.
Gestational age is linearly related to LOV until 32 weeks at which time it plateaus. To adhere to the linearity assumption of linear regression, we needed to create a new variable time-to-grow, that is, the length of time an infant requires to reach 32 weeks. For infants with gestational age equal or greater than 32 weeks, this time is zero.
Preterm was defined as infants <37 weeks of gestational age.
Term was defined as infants ≥37 weeks of gestational age.
Score for Neonatal Acute Physiology (SNAP)
SNAP was used to define the admission illness severity. SNAP scores the worst physiologic derangement in each organ system in the first 24 hours. It has been prospectively validated in multiple NICUs and found to be highly predictive of neonatal mortality and highly correlated with other indicators of illness severity.19
Small-for-Gestational Age (SGA)
Infants whose weight was less than the 10th percentile for infants of the same gestational age and sex in a large Canadian population based database were defined as SGA.20 Because of lack of data, however, gender-specific differences were not incorporated into these definitions for infants <25 weeks.21
Large-for-Gestational Age (LGA)
Infants whose weight was more than the 90th percentile for infants of the same gestational age and sex were defined as LGA.20
Our definition for major congenital and chromosomal anomalies is provided in the Appendix.
Oxygenation Index (OI)
The OI is calculated by multiplying the mean airway pressure by the fraction of inspired oxygen and dividing by the arterial partial pressure of oxygen. This number is multiplied by 100 to obtain the OI. We used the worst OI in the first 24 hours in the NICU as a measure of respiratory failure.
All statistical analyses were performed using SAS (SAS Institute, Cary, NC).22 Univariate analyses on categorical data were performed using the 2-tailed χ2 test. Comparisons between non-normally distributed continuous variables were made using the Kruskal-Wallis test.
Multivariable analyses were performed separately on preterm and term infants. Only limited research is available on factors that affect need for ventilator support or LOV. Therefore, we examined predictor variables that have been shown to affect related outcomes (eg, mortality, length of stay, total costs, and development of chronic lung disease19 23–29). Because birth weight and gestational age are highly correlated (R = .76 in our dataset), only 1 should be included. We elected to include gestational age, which is more closely linked to lung maturation than is birth weight. Infants with missing data on any of the predictor variables were excluded from analysis.
We selected the following factors for entry into our model for need for ventilator support: gestational age, 24-hour SNAP, 5-minute Apgar score (<6 or ≥6), SGA status, presence of major anomaly, gender, and race (dichotomized as white vs non-white). We assessed the effect of practice variation over time by including year of birth (1992–1995 or 1996–1997) as a variable. The models for term infants also included LGA status. Gestational age was used as a continuous variable in the models for preterm infants. All surviving infants <27 weeks' gestation required ventilatory support. Therefore, we included only infants between 27 and 36 weeks in this model. For term infants, gestational age was entered as multiple dichotomous variables using 40 weeks of gestation as a reference.
Stepwise, forward, and backward selection methods were applied using inclusion of P ≤ .15 and exclusion ofP ≤ .05 and all resulted in the same variable selection. The significance of the contribution of any variable to the logistic model was determined using the likelihood ratio test. The discrimination of the models was evaluated using the c-statistic that is equivalent to the area under the receiver operating characteristic (ROC) curve. Calibration of the model was evaluated by the Hosmer-Lemeshow χ2 goodness-of-fit, a measure of the ability of a model to perform well in each decile of risk.
LOV was also analyzed separately for preterm and term infants. In this model, all preterm infants between 23 and 36 weeks of gestation were included. The analyses were performed using stepwise linear regression using an inclusion of P < .15 and exclusion ofP < .05. Model selection was also tested using forward and backward methods and the same predictors were included. For modeling, LOV was transformed into its natural logarithm. This transformation is commonly used for positively skewed continuous variables, such as LOV and length of hospital stay.30 ,31Analysis of the residuals was performed to ensure that they were normally distributed. The same variables used in the need for ventilator support models were included with the following modifications. The variable time-to-grow was used instead of gestational age in the model for preterm infants. In addition, the worst OI in the first 24 hours of admission was entered into the models.
During the study, there were 77 576 inborn live births older than 22 weeks of gestation at the 6 facilities. Of these births, 11 199 newborns were admitted to the NICU, of whom 10 995 survived to hospital discharge. Approximately 18% (1928) of these surviving infants required assisted ventilation in the NICU. Table 1 shows the characteristics of the patients admitted to the NICU.
Data on surfactant and antenatal steroid use were not collected before 1997. More recent data from 1997–1998, however, suggest that antenatal steroid use for high-risk pregnancies in the 6 study centers was high and relatively uniform. Overall, 84.3% of infants <33 weeks of gestation were born to mothers who had received antenatal steroids. In the same period, 84.5% of all intubated infants <33 weeks of gestation, and 47.7% of intubated 33- to 36-week-old infants were treated with surfactant therapy.
Utilization of Assisted Ventilation
Figures 1 and2 summarize gestational age-specific rates of assisted ventilation and LOV. All survivors <27 weeks of gestation required assisted ventilation, after which there was a steady decline in the proportion of infants ventilated until term. In contrast, median LOV declined steadily with each additional week of gestation only until 32 weeks. After 32 weeks, there was no statistical difference in LOV (P =. 839; Fig 2). We performed a sensitivity analysis to assess the effect of including nonsurvivors. This had very limited impact on the results of all infants except those <27 weeks of gestation. Among these extremely premature infants, there was a lower rate of assisted ventilation and shorter median LOV. This change was not unexpected because many of these infants were judged to be nonviable, and many of those who were ventilated died early.
In 1997, the NMDS database began capturing different forms of assisted ventilation separately, permitting us to assess the proportion of infants who were intubated and mechanically ventilated, who received NCPAP, or both. The type of assisted ventilation provided varied with gestational age. The use of NCPAP alone or in addition to mechanical ventilation was greatest among the most immature infants. The percentage of infants receiving only mechanical ventilation increased steadily from 18.5% among the youngest preterm infants to 60.9% among term infants (Fig 3).
Barotrauma, defined as radiologically confirmed pneumothorax, pneumomediastinum, pneumopericardium, pneumoperitoneum, or pulmonary interstitial emphysema occurred in 7.9% of all ventilated preterm infants and 15% of ventilated term infants.
Extreme prematurity had a tremendous impact on ventilator resource consumption. Although infants <33 weeks of gestation accounted for only 1.9% of total live births and 12% of NICU admissions, they accounted for the vast majority (73%) of days of assisted ventilation.
Models for Need for Ventilator Support
Table 2 shows results of logistic regression analysis of predictors for assisted ventilation in preterm infants between 27 and 36 weeks of gestation. As expected, gestational age strongly influences the need for assisted ventilation. However, initial severity of illness (SNAP), clinical status at delivery (Apgar scores), and the presence of major anomalies were important independent predictors. Male gender and white race also increased the odds of ventilation in this model. There was a decrease in the proportion of infants ventilated in more recent years. We looked for an interaction between gestational age and SNAP scores and found none. The overall ability of the model to predict need for assisted ventilation was good, with an area under the ROC curve = .90, which was much higher than a model using gestational age alone (.76) or birth weight alone (.70). The goodness-of-fit test result indicates that the predicted and observed number of ventilated infants at each decile of risk were not statistically different (P = .44). Antenatal steroid use was not available on all infants, so it could not be entered into this model. To see how greatly antenatal steroid use would affect our results, we performed a subanalysis using only 1997 data. The addition of antenatal steroid use did not alter the analyses of 483 preterm infants born in 1997.
Gestational age had a limited effect among term and postterm infants. Only infants of 37 weeks of gestation had increased odds of requiring ventilation compared with 40-week gestation infants (Table 3). The effect of postmaturity, which was associated with increased odds of ventilation (odds ratio: 1.3;P =. 02) but was no longer significant after adjusting for SNAP and Apgar scores. As with preterm infants, severity of illness, initial clinical presentation, and presence of anomalies were also strong predictors of the need for assisted ventilation. In fact, SNAP alone had an area under the ROC curve of .82. In this cohort, more recent admissions also had decreased adjusted odds for being ventilated. This model discriminated well between infants who required assisted ventilation and those who did not, (ROC area: .88) compared with birth weight alone (ROC area: .50). Although the difference between expected and observed number of ventilated infants was small (Fig 4), the goodness-of-fit statistic was significant because of the large sample size (P =. 0004). The model's lack of fit was primarily attributable to overprediction of the need for ventilation among infants at low risk and underprediction in infants at high risk for respiratory failure.
Models for LOV
Immaturity (expressed as time-to-grow to 32 weeks) was the principal determinant of LOV among preterm infants (Table 4) explaining 50% of the variance (model R2). Incorporating other variables (OI, anomalies, SNAP, and SGA status) elevated the model R2 from .50 to .60. This was superior to birth weight alone (model R2 = .37). Race, gender, and Apgar scores, which were predictors for the need for ventilator support, did not significantly affect LOV even when SNAP was removed from the model. More recent year of admission was not associated with shorter LOV.
The ability to predict LOV for 178 term and postterm infants was limited (model R2 =. 29). The OI (partial R2 = . 27) and the presence of an anomaly (partial R2 = .02) were the only variables significantly related to LOV.
This population-based analysis of neonatal assisted ventilation provides information on frequency and duration of ventilation that can be used to determine resource needs and in the development of clinical trials. Few studies are available for comparison. Synnes4reported on the experience of a large tertiary care center in the mid-1980s. That study was restricted to infants <29 weeks of gestation born before the introduction of exogenous surfactant therapy and, therefore, does not reflect current neonatal practice or the full spectrum of infants cared for in the NICU. Their study and ours both revealed a wide variation in both rate and LOV between infants of closely related gestational age. This finding highlights the importance of analyzing LOV by small categories rather than the more commonly used broad birth weight categories (eg, very low birth weight and intermediate low birth weight). Interestingly, despite the fact that the infants in the study by Synnes et al4 did not receive surfactant, the median LOV for surviving infants <29 weeks is similar to ours (28 vs 29 days). Among infants who were resuscitated in the delivery room and admitted to the NICU, there was a trend toward improved survival in our group compared with the group of Synnes et al, 76.3% versus 71.0% (P =. 054), respectively, and decreased incidence of barotrauma 14.7% versus 35% (P <. 001). This suggests that widespread surfactant use has decreased morbidity and mortality without affecting the LOV required by surviving extremely premature infants. Although other patient-related or institutional factors such as increased use of antenatal steroids may be partly responsible for these findings, several other studies have also noted improved survival without a reduction in ventilation resource use since the introduction of surfactant therapy.9 32–34 The relatively unchanged use of assisted ventilation combined with improved survival has translated into continued disproportionate resource use by infants in the youngest gestational age group.3 ,33
We also found that LOV is not normally distributed but rather extremely positively skewed. Thus, although the majority of infants in each gestational age group require short courses of assisted ventilation, a few infants required prolonged assistance and tend to inflate the mean LOV. Therefore, the median LOV is a better measure of the central tendency of the population than the mean and nonparametric methods or variable transformation are preferable for data analysis. Unfortunately, LOV is often treated as if it had a normal distribution, which could lead to misleading results.8 ,9 ,33
We noted great variation in the mode of ventilator support received by infants of different gestational age categories. This is not surprising considering the underlying respiratory pathology leading to a requirement for assisted ventilation in the 3 groups. Among the more immature infants, more chronic problems such as resolving respiratory distress syndrome and apnea are more prevalent, thus there is greater use of NCPAP in combination with intermittent mandatory ventilation. In term infants, however, conditions that do not respond as well to NCPAP (eg, meconium aspiration, sepsis, and persistent pulmonary hypertension) are more prevalent.
In our models, many factors besides gestational age influenced the need for assisted ventilation. The most important of these was illness severity as measured by the SNAP. This relationship was particularly strong among term infants where SNAP alone had a large area under the ROC curve. Low 5-minute Apgar scores were positively associated with the need for assisted ventilation independent of SNAP. Similar findings have been reported for neonatal mortality where Apgar scores and SNAP independently affect outcome.35 White race and male gender, which have been linked to increased mortality and risk of chronic lung disease by some25 ,26 ,36 ,37 but not all28 ,38 investigators, were associated with the need for assisted ventilation in preterm infants in our study.
There was no association between SGA status and the need for assisted ventilation among preterm infants in this population. There is discrepancy in the current literature about the relationship between SGA status and respiratory failure. Some studies, like ours, have found no association,39 ,40 whereas others have noted an increased need for assisted ventilation in preterm SGA infants.41–43 We also failed to detect an effect of SGA status among term and postterm infants. This is consistent with recent studies, which have found that suboptimal fetal growth is not associated with increased morbidity or mortality in term infants44 except perhaps in the most severely growth retarded infants.43
We found that the odds of receiving assisted ventilation are lower in more recent years for both preterm and term infants. Although changes in prenatal care are probably partly responsible, this decrease may also reflect a trend toward more selective intubation. This trend has been noted by others45 and is likely attributable to increased tolerance for higher carbon dioxide levels and recognition of the potential adverse consequences of intubation and mechanical ventilation.
The ability of our model to explain the observed variation in LOV among preterm infants was good. We are aware of only 1 study that also attempted to predict the duration of assisted ventilation.46 In their homogeneous population of 69 very low birth weight infants in a tertiary care center, 44% of the variance in days of ventilation could be explained. The improved explanatory power of our model may be attributable to the addition of variables for overall illness severity and SGA status. In addition, we used the worst OI instead of peak inspiratory pressure at 12 hours of life as an indicator for severity of respiratory illness. The OI, which incorporates an infant's response to a given level of ventilator support, may be a better measure of pulmonary status than peak inspiratory pressure.
In our model, SGA status was weakly associated with increasing LOV. This is similar to the results of other researchers39 ,40who have noted that although growth retardation does not influence the need for assisted ventilation, it does seem to be associated with increased LOV in preterm infants. One reason why SGA infants may require longer courses of assisted ventilation is that they have an increased incidence of complications, such as clinically significant patent ductus arteriosus, sepsis,39 ,41 and slower growth. Most importantly, our study adds to the increasing body of literature refuting the belief that SGA infants have accelerated lung maturation and less pulmonary morbidity.
We did not find that the Apgar score, gender, or race were significantly associated with LOV. This suggests that although they influence the need for assisted ventilation in the early perinatal period, they have may have less impact on longer-term morbidity.
Our ability to model LOV for term infants was very limited. This may be partly attributable to our small sample size, but we suspect that other factors are involved. In contrast to premature infants, in whom respiratory failure is predominantly attributable to lung immaturity, term infants experience respiratory failure because of a variety of causes. Inclusion of diagnostic variables might, therefore, improve the amount of explained variance. Attempts to predict duration of mechanical ventilation in adult populations have shown the importance of diagnostic groups, with admitting diagnosis accounting for ∼26.6% of the variance in LOV.47 Because standardized definitions for respiratory diagnoses do not exist, and because using such definitions would result in small cells, we are not yet in a position to test this hypothesis. We also suspect that inclusion of measures of pulmonary status at a later time point in the patient's hospital course would also improve the predictive ability of the model. This enhancement has been shown in models for chronic lung disease.27 ,28 However, our focus was on variables available very early in the NICU course.
Several points need to be considered when interpreting the results of our models. In developing a model of the need for ventilator support, we studied only the population of infants admitted to the NICU, not all live-born infants. As all infants <34 weeks are routinely admitted to the NICU in the KMPCP, this potential selection bias should not greatly affect the model for preterm infants. For term infants, however, it is possible that we missed or underestimated the impact of important variables that influence the need for ventilatory support.
In choosing predictor variables for our models, we aimed to select risk factors that could be easily identified on admission to the NICU and would not be influenced by the treatment the patient received. Although most of our predictors meet these requirements, SNAP, which is assigned 24 hours after admission, can be influenced by treatment. Currently, all severity of illness scores have this limitation.29 ,35 ,48 ,49 Because there is a wide variation in the severity of illness among infants cared for at different units, we believe that inclusion of an illness severity score is important. Development of new scoring systems, such as the recently derived 12-hour SNAP, may help in reducing the bias of treatment on severity indices.50 Finally, these models need to be validated using other cohorts before they can be used to compare outcomes between institutions or between treatment modalities.
Our study demonstrates that several patient- related risk factors besides gestational age should be considered to compare appropriately the utilization of ventilation among centers. We were able to explain much of the variance in LOV for preterm infants, but were limited for term infants. Future studies including information on diagnosis and respiratory status at the end of the first 24 hours of admission may improve the ability to predict duration of ventilation.
This study was supported by The Permanente Medical Group, Inc, and by Kaiser Foundation Health Plan, Inc.
Funding for the Kaiser Permanente Neonatal Minimum Data Set (Grant 115-6137) database was provided by Kaiser Foundation Health Plan, Inc.
We thank Drs Joseph V. Selby, Douglas K. Richardson, David Glidden, Laurel Habel, and Margaret Walker for reviewing the manuscript.
- Received June 11, 1999.
- Accepted November 15, 1999.
Reprint requests to (G.J.E.) Kaiser Permanente Medical Care Program, Division of Research, Perinatal Research Unit, 3505 Broadway Ave, Room 718, Oakland, CA 94611. E-mail:
- Lichtig LK,
- Knauf RA,
- Bartoletti A,
- et al.
- Guyer B,
- MacDorman MF,
- Martin JA,
- Peters KD,
- Strobino DM
- Avery ME,
- Tooley WH,
- Keller JB,
- et al.
- Escobar GJ,
- Fischer A,
- Kremers R,
- Usatin MS,
- Macedo AM,
- Gardner MN
- Escobar GJ,
- Fischer A,
- Li DK,
- Kremers R,
- Armstrong MA
- Meadow W,
- Reimshisel T,
- Lantos J
- Richardson DK,
- Gray JE,
- McCormick MC,
- Workman K,
- Goldmann DA
- ↵Arbuckle TE, Sherman GJ. An analysis of birth weight by gestational age in Canada. Can Med Assoc J. 1989;140:157–60:165. Comments
- ↵SAS Institute. Statistical Analysis Software. 6th ed. Carey, NC: SAS Institute; 1989
- Phibbs CS,
- Williams RL,
- Phibbs RH
- Powell PJ,
- Powell CV,
- Hollis S,
- Robinson MJ
- Sinkin RA,
- Cox C,
- Phelps DL
- Horbar JD,
- Wright EC,
- Onstad L
- Richardson DK,
- Phibbs CS,
- Gray JE,
- McCormick MC,
- Workman-Daniels K,
- Goldmann DA
- Horbar JD,
- Badger GJ,
- Lewit EM,
- Rogowski J,
- Shiono PH
- ↵Bardin C, Zelkowitz P, Papageorgiou A. Outcome of small-for-gestational age and appropriate-for-gestational age infants born before 27 weeks of gestation. Pediatrics. 1997;100(2). URL: http://www.pediatrics.org/cgi/content/full/100/2/e4
- Tyson JE,
- Kennedy K,
- Broyles S,
- Rosenfeld CR
- Poets CF,
- Sens B
- Copyright © 2000 American Academy of Pediatrics