Background. The outcomes for very low birth weight infants vary among neonatal intensive care units (NICUs), but the reasons for this variation are not well understood. We used the database of a large neonatology research network to determine whether either admission characteristics of the infants or specific characteristics of the units such as annual patient volume and the presence of a pediatric residency program could account for observed differences in neonatal mortality rates among units.
Methods. We studied 7672 infants with birth weights from 501 to 1500 g treated during 1991 and 1992 at 62 NICUs participating in the Vermont Oxford Network Database.
Results. Overall, 14.7% of the study infants died within 28 days of birth (interquartile range 9.9% to 18.1%). The ratio of the number of observed deaths at an NICU to the number of deaths predicted based on the characteristics of infants treated at the NICU (standardized neonatal mortality ratio, [SNMR]) varied significantly among units (range 0 to 1.69, z = 4.24). There was no association between annual patient volume and either mortality rate (r = .17) or SNMR (r = .22). Observed mortality rates (17% vs 13%) and SNMR (1.04 vs .87) were both higher at the 24 hospitals with pediatric residency training programs than at the 38 hospitals without such programs. Hospitals with residency programs had higher average annual patient volumes (104 vs 66). In an analysis simultaneously adjusting for patient characteristics, volume, and presence of a residency program, neither volume (odds ratio [OR] per 10 additional cases treated 1.01, 95% confidence interval [CI], .98 to 1.04) nor presence of a pediatric residency program (OR 1.18, 95% CI, .94 to 1.47) was significantly associated with neonatal mortality risk.
Conclusion. There are differences in neonatal mortality rates among NICUs that cannot be explained by differences in the measured admission characteristics of the infants, suggesting that the effectiveness of medical care varies among units. Neither the annual volume of very low birth weight infants treated in a unit nor the presence of a pediatric residency training program was independently associated with neonatal mortality rates for very low birth weight infants.
- NICU =
- neonatal intensive care unit •
- OR =
- odds ratio •
- CI =
- confidence interval •
- ROC =
- receiver operating characteristic •
- SNMR =
- standardized neonatal mortality ratio •
- CRIB =
- clinical risk index for babies •
- SNAP =
- score for neonatal acute physiology
Medical practices and patient outcomes vary markedly among hospitals for a wide range of clinical conditions including those treated in adult, pediatric, and neonatal intensive care units (NICUs).1-5 Although the sources of this variation are imperfectly understood, characteristics of the patients and the hospitals at which they are treated have both been shown to play a role.6,7 The purpose of the present study is to identify patient and hospital characteristics responsible for differences among NICUs in mortality rates for very low birth weight infants (501 to 1500 g at birth). The study was performed using the database of the Vermont Oxford Network8; it was carried out in two stages. First, we identified patient characteristics associated with neonatal mortality and demonstrated that there is significant unexplained variation in mortality rates among NICUs after adjusting for differences in the admission characteristics of the infants treated in these units. We then evaluated whether the unexplained residual variation in mortality was associated with features of the NICUs themselves.
The specific NICU features of interest are the annual volume (case load) of very low birth weight infants treated at a NICU and the presence of a pediatric residency training program in the hospital. Both of these features have been studied in other clinical settings. Increasing patient volume and the presence of a residency training program have each been associated with improved patient survival for a variety of medical and surgical conditions.7,9-12 In contrast, Pollack et al3 have recently reported that the probability of survival is decreased for infants and children treated in pediatric intensive care units located within teaching hospitals where care is provided by residents in training; they found no association between survival and case load.
The effects of patient volume and presence of a residency training program on outcomes for very low birth weight infants treated in NICUs are unknown. It is important to understand how these factors influence outcomes because the current trend toward deregionalization of neonatal care in the United States will increase the proportion of the very low birth weight infants receiving treatment at smaller, nonteaching units and decrease the proportion of infants receiving care at larger regional academic centers, where residents participate actively in patient care.13
The Vermont Oxford Network is a voluntary collaborative group of neonatologists and other health professionals representing NICUs in North America, Europe, Asia, and Australia. It was established in 1989 to improve the effectiveness and efficiency of neonatal intensive care through a coordinated program of randomized trials, outcomes research, and quality improvement projects.5,8,14 Participating institutions submit data to the network for all inborn infants weighing 501 to 1500 g at birth, including those who died before NICU admission, and for all outborn infants in this birth weight category who are admitted within 28 days of birth. Infants are followed until hospital discharge. Those who are transferred to other hospitals are tracked to determine their survival status and final date of discharge. For the current study we included institutions in the United States that participated in the network during 1991 or 1992. Institutions were excluded if 10% or more of their cases in the database were missing data for variables under investigation. In addition, individual cases with missing data for variables under investigation were excluded.
Mortality was defined as deaths occurring on or before the 28th day after birth. The day of birth was considered as day 1. Patient characteristics were defined according to the Vermont Oxford Trials Network Database Manual of Operations.15Gestational age was defined as the attending neonatologist's best estimate of gestational age based on obstetric history, obstetric examination, prenatal ultrasounds, and postnatal physical examinations. An infant whose birth weight was less than the 10th percentile for gestational age was considered small for gestational age.16Because published percentiles did not include data for extremely premature infants and percentiles for some gestational age categories were based on small sample sizes, we estimated these percentiles using smoothed regression curves based on cubic polynomial regression. The type and severity of all major birth defects reported in the network database were independently reviewed and coded by two neonatologists before data analysis.
Because hospital membership in the network varied from a partial year to 2 full years, the annual volume of very low birth weight infants treated at an institution was estimated using the average time between admissions for such infants. An institution was defined as having a pediatric residency program if it was listed as an accredited program in Pediatrics in the 1991–1992 Graduate Medical Education Directory.17
Statistical tests are adjusted for the fact that infants within each NICU represent a cluster of individuals who are not independent.18-21 The standard errors, confidence intervals [CIs], and associated significance levels for odds ratio [ORs] based on logistic regression have been adjusted using the method of Liang and Zeger.18 These adjusted standard errors and CIs are generally larger than those obtained from methods that assume independence. The point estimate of the OR is unaffected.
Variation in 28-day Mortality
Stepwise logistic regression was used to predict 28-day mortality (F-to-enter, P < .10). Candidate variables for inclusion in the regression model were birth weight, 1-minute Apgar score, race (black/nonblack), gender, mode of delivery (cesarean/vaginal), multiple birth (yes/no), antenatal steroid use (any or none), small size for gestation age (<10 percentile, ≥10th percentile), location of birth (inborn/outborn), prenatal care (yes/no), and major birth defect (yes/no). The performance of the logistic model was evaluated using the Hosmer-Lemeshow goodness-of-fit test22 and area under the receiver operating characteristic (ROC) curve.23 The standard error of the area under the ROC curve was computed based on its relationship to the Mann-WhitneyU statistic.23 Multiple cross-validation was used to estimate the predictive value of the final model based on randomly dividing the total sample into ten equal subsamples, and applying to each tenth the equation estimated from the remaining 90% of cases.24 The square of the simple correlation between predicted probabilities and observed outcomes is presented as a measure of predictive power. Standardized neonatal 28-day mortality ratios (SNMR) were computed for each NICU as the ratio of the observed mortality at the NICU to the expected mortality which was derived by summing the predicted probabilities of mortality for individual infants at the NICU based on the logistic regression model.25 CIs presented for individual hospital SNMRs are unadjusted for the number of hospitals examined. A z-test was used to test whether the variability observed across NICUs with respect to SNMRs was greater than would be expected due to random binomial variability.25
Effects of Volume and Teaching on Mortality
Univariate comparisons of infant characteristics between hospitals with and without pediatric residency programs were performed using Wald tests that adjust for clustering. Nested analysis of variance was used for comparisons of continuous variables. The univariate association between annual volume and patient characteristics was investigated using correlation analyses with individual NICUs as the unit of analysis.
Logistic regression was used to estimate the adjusted ORs and 95% CIs in the multivariate analysis associated with volume and presence of a residency program. Logistic regressions and univariate Wald tests were performed using SAS Procedure IML.26
Sixty-eight institutions in the United States participated in the Vermont Oxford Network in 1991 and/or 1992. Six institutions were excluded from analysis because data were missing in over 10% of cases for one or more of the variables included in the regression models. There were 7846 infants with birth weights 501 to 1500 g cared for at the remaining 62 institutions in the 2 study years; data were complete for the variables under investigation for 7672 infants. Twenty-four (39%) of the institutions had pediatric residency training programs; five (8%) were for-profit institutions. The median estimated annual volume of infants 501 to 1500 g treated at an institution was 76 (interquartile range 47 to 113).
The characteristics of the study infants are shown in Table1. The infants had a mean birth weight of 1059 g and mean gestational age of 28.5 weeks. Overall, 18% of the infants were outborn, 29% were black, 57% were delivered by cesarean section, 23% were the result of a multiple gestation, and 90% received at least one prenatal care visit. Antenatal corticosteroid treatment was given to 25% of the mothers whose infants were in the study population. The frequency of antenatal steroid use varied among the participating institutions (interquartile range 10% to 31%). Fifty-three percent of the infants received surfactant treatment (interquartile range 43% to 62%). Overall, 14.7% of the infants died on or before 28 days after birth (interquartile range 9.9% to 18.1%).
The patient level characteristics meeting the criteria for inclusion in the logistic regression model for predicting the risk of mortality within 28 days of birth are shown in Table 2. Increasing birth weight, higher 1-minute Apgar score, black race, small size for gestational age, antenatal corticosteroid treatment, female gender, cesarean delivery, and prenatal care were associated with a decreased risk of death; multiple gestation and birth defects were associated with an increased risk of death. Location of birth (inborn/outborn) was not significantly associated with 28-day mortality after adjusting for the above variables. In the model, small for gestational age infants have a lower risk of mortality than appropriate for gestational age infants at any given birth weight because they are more mature.
The performance of the logistic regression model based on infant characteristics for predicting death within 28 days of birth was assessed. There was no significant lack-of-fit (Hosmer-Lemeshow goodness-of-fit statistic 10.5, df = 8, P = .23). The area under the ROC curve was .87 (SE = .009). The averageR2 based on the cross-validation procedure was .30.
The logistic regression model was used to calculate the predicted number of deaths at each institution. The ratio of the observed to the predicted number of deaths at each institution, the SNMR, ranged from 0 to 1.69 (Fig 1). The SNMR values of 0 at two sites reflect small sample sizes as indicated by the large CIs of these estimates. The variation in SNMR among institutions was greater than would be expected based on random binomial variation (z = 4.24, P < .001) indicating that significant variation in mortality among NICUs remained after adjusting for the patient characteristics in the model.
Comparisons of patient characteristics at institutions with and without a pediatric residency program are shown in Table 3. The estimated annual volume of infants 501 to 1500 g treated at institutions with a pediatric residency program (104.4 ± 39.3) was higher than that at institutions without such a program (66.3 ± 37.2) (P < .001). A higher percentage of infants treated at hospitals with a pediatric residency program were outborn (29% vs 8%, P < .001), and a lower percentage were born to mothers who had received antenatal corticosteroid treatment (18% vs 33%, P = .02). The lower rate of antenatal corticosteroid treatment at institutions with pediatric residency programs could not be explained by the higher proportion of outborn infants at those institutions as similar differences were observed when analyses were limited to inborn infants.
A higher proportion of infants treated at hospitals with a pediatric residency program died within 28 days of birth (17% vs 13%,P = .002). The mean SNMR at the 24 hospitals with a pediatric residency program was 1.04 (SE = .06) compared with .87 (SE = .05) at the 38 hospitals without such a program (P = .04).
The correlations between annual volume and the various measured infants characteristics are shown in Table 4. In addition, descriptive statistics are displayed for infants within the 62 NICUs broken down by quartiles based on estimated annual volume. The only characteristic significantly associated with NICU volume was the occurrence of major birth defects where a weak positive correlation was observed (r = .28, P = .03). Small NICUs (<47 annual admissions) were observed to have a somewhat lower rate of defects (2.0%) compared with larger NICUs that had rates of approximately 3.0%.
There was no statistically significant association between 28-day mortality rate and annual volume (r = .17,P = .18) (Fig 2). There is a weak positive correlation between SMNR and annual volume (r = .22, P = .08).
The results of the logistic regression including hospital-level characteristics (volume, pediatric residency) and patient level characteristics are shown in Table 5. Neither the presence of a pediatric residency program (OR 1.18, adjusted 95% CI .94 to 1.47) nor the estimated annual volume of patients 501 to 1500 g (OR per 10 additional cases 1.01, adjusted 95% CI .98 to 1.04) were significantly associated with the risk for mortality within 28 days of birth after adjusting for all of the patient level characteristics included in the Table.
We have shown that mortality rates for very low birth weight infants vary among NICUs in the United States and that this variation cannot be explained solely by differences in the characteristics of the patients they treat. Furthermore, we found that neither the annual volume of very low birth weight infants treated at a NICU, nor the presence of a pediatric residency program in the hospital where the unit is located could explain this variation in mortality.
Our results should be interpreted with caution for several reasons. First, our sample was drawn from the database of the Vermont Oxford Network, a voluntary collaboration of health professionals whose goal is to improve the effectiveness and efficiency of neonatal intensive care. It may be that members of the Network are systematically different form nonmembers in ways that affect 28-day mortality. In addition, as compared with the universe of NICUs in the United States, the sample included relatively few very small and very large units, and few major university centers with extensive research programs. Thus, our findings are only suggestive of the possible effects of extremes in annual patient volume or highly sophisticated research and training programs on neonatal mortality among very low birth weight infants.
Williams27 was the first investigator to show that mortality rates for newborn infants varied significantly from hospital to hospital. In a study based on over 3 million infants of all birth weights born in 504 California hospitals between 1960 and 1973, he demonstrated how the standardized mortality ratio could be used to identify variation among institutions. Subsequent investigators have focused on high-risk newborns treated in NICUs. Both mortality and the occurrence of chronic lung disease, a major morbidity of low birth weight infants, have been shown to vary among NICUs even after adjusting for differences in case-mix.28-30 However, these previous studies included small numbers of NICUs, and adjusted only for basic demographic variables such as birth weight, race, and gender. Furthermore, they were performed before the introduction of surfactant therapy which has had a major impact on the outcomes of neonatal intensive care.31,32 The present study has the advantage of using a large database from the current era of neonatal care that was collected prospectively using uniform definitions of data items at all sites. The National Institute of Child Health and Human Development Neonatal Research Network has also reported variation in practices and outcomes at participating NICUs.4,33
Because differences in patient-mix may be an important source of differences in mortality rates among NICUs, we adjusted for a number of patient characteristics in the analysis. The mortality risks associated with these factors were consistent with those found in previous studies.34-36
Except for the 1-minute Apgar score, we did not adjust for physiological measurements or other indicators of disease severity in our analysis. Disease severity scores for infants receiving neonatal intensive care have been recently developed and tested. Tarnow-Mordi and the International Neonatal Network have described the Clinical Risk Index for Babies (CRIB) score for predicting mortality risk for very low birth weight infants.36,37 Richardson and colleagues35,38 have developed the Score for Neonatal Acute Physiology (SNAP), which can be used to predict mortality risk for NICU admissions of any birth weight. We did not use these scores for several reasons. First, the database does not include detailed physiological data. Second, the CRIB and SNAP scores, respectively, use information collected during the 12 or 24 hours after NICU admission. Data collected after an infant is admitted to the NICU reflect the results of treatment as well as the infant's underlying disease severity and excludes those infants dying before this time. The effects of treatment may be particularly important for infants receiving surfactant treatment, which can cause rapid changes in oxygenation and ventilation. We limited our choice of variables for risk adjustment to those whose values are determined before NICU admission. Moreover, a comparison of the performance of our patient level logistic model with models created based on data from the CRIB and SNAP scores indicates that there is no major difference in performance between the model we used and the others. The area under the ROC curve for our model, which uses only admission data, is .87 as compared with .90 for the CRIB score.36 When predictive models based on SNAP and birth weight are tested on infants weighing less than 1500 g, the area under the ROC curve ranges from .73 to .91 depending on birth weight.35 The cross-validated R2 of .30 for our model also compares favorably with the values of .14 to .25 reported for models used to predict hospital associated mortality for adult Medicare patients,24 and with recently reported values for 14 multivariate severity models used to predict hospital death rates for adults with pneumonia.39
We did not find evidence of a relationship between the annual volume of very low birth weight infants treated at a NICU and its mortality rate. We restricted our analysis to very low birth weight infants because of their unique care requirements. Our results may not apply, however, to other populations of infants cared for in NICUs. Furthermore, although neonatal mortality did not vary with annual volume, other neonatal outcomes and morbidities of neonatal intensive care were not evaluated.
The volume-outcome relationship has been evaluated for a number of clinical conditions.10 Several explanations have been offered for the association of increased hospital volume with improved outcomes in those settings where it has been observed. They include learning by doing and maintenance of skills (“practice makes perfect”), economies of scale attributable to increased specialization within a hospital, and selective referral of patients to hospitals with better outcomes. In adult cardiac surgery, where the relationship of volume to mortality has been extensively studied, performing a higher volume of surgery is associated with lower mortality.40,41 The same relationship has recently been reported for pediatric congenital heart surgery.11 In pediatric intensive care, on the other hand, the volume of cases treated does not appear to be associated with the risk of mortality.3 This is consistent with our findings for neonatal intensive care.
After adjusting for differences in both patient characteristics and volume, no statistically significant effect of the presence of a residency program on neonatal mortality rates was found. Although the average SNMR was higher at hospitals with pediatric residency training programs than at those without such programs, the comparison may be confounded by differences in annual patient volume. Because annual volume was higher at hospitals with residency training programs, and because there was also a weak but positive correlation between volume and SNMR, we believe that inferences regarding associations between hospital characteristics and mortality should be based on the regression analysis which simultaneously adjusts for annual volume, presence of a pediatric residency program, and admission characteristics of the patients. In that analysis neither annual volume nor the presence of a pediatric residency program were significantly associated with neonatal mortality.
These results should be interpreted cautiously because hospitals with pediatric residency programs may differ from hospitals without such programs in a variety of ways, some of which we did not measure. These include the availability of subspecialists and advanced technology, and differences in staffing patterns and models for unit coordination and management. Also, although resident and student physicians are likely to play a larger role in patient care at hospitals with accredited pediatric residency programs, even those NICUs without such programs may have trainees participating in patient care. The degree of supervision provided to these trainees may be an important determinant of patient outcomes that should be addressed in future studies. We did not measure differences in staffing or supervision.
Previous studies that have analyzed the impact of a teaching program on patient outcomes have not shown consistent effects. Hartz et al7 found that mortality rates for general hospital patients were lower in hospitals with a teaching program, whereas Pollack et al3 reported that mortality rates for infants and children are higher if they were treated in a pediatric intensive care unit staffed by pediatric residents. Several studies have shown that neonates treated in tertiary level intensive care units or born at hospitals with such units have a lower risk for mortality, but these studies have not specifically looked at the role of teaching or residency training.36,42,43
Phibbs et al,43 in a study of infants born in California during 1990, reported that risk-adjusted neonatal mortality was lower for births that occurred in hospitals with large level III NICUs. Their study design differed from ours in three important ways. Their analysis included infants of all birth weights, used average daily census as the measure of volume, and was based on the hospital of birth rather than the hospital at which intensive care was provided. These differences in design make comparison of the results difficult. However, it is important to recognize that our results do not address the question of how the level of care available at the hospital of birth affects outcome because we did not have data on the characteristics of the hospitals that transferred patients to units in our network.
Antenatal steroids have been documented in numerous randomized trials and observational studies to reduce mortality and morbidity for preterm infants.44-46 It is not clear why hospitals in this study with pediatric residency programs used antenatal steroid therapy less frequently. We did not evaluate the relationship of antenatal steroid therapy with the presence of an obstetrical residency program. The increased 28-day mortality rates at hospitals with pediatric residency programs may be explained in part by their less frequent use of antenatal steroid therapy. This issue merits further investigation.
In summary, we found differences among NICUs in neonatal mortality rates for very low birth weight infants that were not explained by differences in the characteristics of the infants treated. This suggests that the effectiveness of care varies among units. A major goal of the Vermont Oxford Network is to help participating NICUs improve the effectiveness of their care through a coordinated program of randomized trials, outcomes research, and quality improvement projects. In addition, we found that neither the annual volume of very low birth weight infants treated at a NICU, nor the presence of a pediatric residency training program were associated with differences in mortality risk. This finding is of importance because the current trend toward deregionalization of neonatal intensive care will result in a larger proportion of infants receiving their treatment at smaller, nonteaching units. We found no evidence to suggest that this trend will result in higher mortality rates for very low birth weight infants.
The following institutions and investigators participated in this study: Arnot-Ogden Medical Center, Elmira, NY–Willard Helmuth, MD, Robert Balcom, MD; Aultman Hospital, Canton, OH–Martha W. Magoon, MD, Louis J. Heck, MD; Baptist Memorial Hospital–East, Memphis, TN–Esmond L. Arrindell, MBBS, James M. Hamlett, III, MD; Cardinal Glennon Children's Hospital/St Louis University, St Louis, MO–Claire C. Juzwicki, MD, Phyllis Anderson, CRTT; Children's Hospital Medical Center of Akron, Akron, OH–John H. Vollman, MD, Deborah L. Giebner; Columbia Hospital for Women Medical Center, Washington, DC–Kenneth L. Harkavy, MD; Community Hospital of the Roanoke Valley, Roanoke, VA–C. Gilbert Frank, MD, Leah Sinozich; DeVos Children's Hospital, Grand Rapids, MI–Dinah Sutton, RN, Ed Beaumont, MD; Emanuel Children's Hospital, Portland, OR–Patrick K. Lewallen, MD, Karen L. Waske, RN, MN; Fitzgerald Mercy Hospital, Darby, PA–David L. Schutzman, MD, Maria Frankenfield-Chernicoff, RN; Good Samaritan Hospital, Cincinnati, OH–David H. Levine, MD, Horacio S. Falciglia, MD; Good Samaritan Medical Center, West Palm Beach, FL–Setty G. Viralam, MD, Mary Abramson, RN; Greater Baltimore Medical Center, Baltimore, MD–Siew-Jyu Wong, MD, Ambadas Pathak, MD; Henrico Doctors' Hospital, Richmond, VA–Charles R. Frakes, MD; Huntsville Hospital, Huntsville, AL–Meyer E. Dworsky, MD, Linda S. Reynolds, RN, BSN; Lutheran General Children's Hospital, Park Ridge, IL– Bhagya Puppala, MD, H. Mangurten, MD; Marshfield Clinic/St Joseph's Hospital, Marshfield, WI–George J. Hoehn, MD, Edward C. Denny, MD; McKay-Dee Hospital and Medical Center, Ogden, UT–Michael Clark, MD, Kathy Zundel, RN; Mease Hospital, Dunedin, Dunedin, FL–Mary T. Newport, MD, Deborah Rogala, NNP; Medical Center Hospital of Vermont, Burlington, VT–Roger F. Soll, MD, Kathleen Leahy, RN, NNP; Memorial Hospital, Hollywood, FL–Lester McIntyre, MD, Bruce Schulman, MD; Memorial Hospital of South Bend, South Bend, IN–Robert D. White, MD; Memorial Medical Center, Savannah, GA–Linda M. Sacks, MD, Roberta M. Smith, MD; Mercy Hospital and Medical Center, Chicago, IL–Rohitkumar Vasa, MD; Miami Children's Hospital, Miami, FL–Ian P. Jeffries, MB, Mary E. Schwartz, RN, NNP; Miami Valley Hospital, Dayton, OH–Marc R. Belcastro, DO, Sue Mackey, RRT; Mt Sinai Medical Center, Cleveland, OH–Douglas P. Powell, MD, Jeffrey Schwersenski, MD; North Oaks Medical Center, Hammond, LA–Ivan A. Villalta, MD, Marcia Brewton, RNC, NNP; Ochsner Foundation Hospital, New Orleans, LA–Victor E. Lunyong, MD, Harley G. Ginsberg, MD; Parkview Memorial Hospital, Fort Wayne, IN–Pat Carteaux, RN, Joel W. Secrest, MD; Penn State University Children's Hospital, Hershey Medical Center, Hershey, PA–Keith H. Marks, MB, PhD, Kathleen L. Gifford, RN; Plantation General, Boca Raton, FL–M. Stern, MD; Presbyterian Hospital, Albuquerque, NM–Sydney M. Swetnam, MD, Virginia M. Hallinan, MD; Presbyterian/St Luke's Hospital, Denver, CO–Mark S. Brown, MD, Connie Rusk, NNP; Promina Kennestone Hospital, Marietta, GA–Patricia B. Hunt, RNC, MS, CNNP, Vickie P. Fox, RNC, MSN, CNNP; Riverside Hospital, Toledo, OH–Michael D. Shaw, MD, Joan Zolla-Boldt, RN; Sinai Hospital of Baltimore, Baltimore, MD–Mollie Wheatley, RN, CNNP, S. Lee Marban, MD, PhD; Sparrow Hospital/Michigan State University, Lansing, MI–Padmani Karna, MD, Karen Taylor; St Agnes Hospital–Baltimore, MD–Howard J. Birenbaum, MD, Barbara A. Long, RN; St Francis Hospital and Medical Center, Hartford, CT–Hema N. DeSilva, MD; St Francis Medical Center, Peoria, IL–Tim C. Miller, MD, Connie McConnell, RN; St John Hospital and Medical Center, Detroit, MI–Maria L. Duenas, MD, Ali Rabbani, MD; St John's Mercy Medical Center, St Louis, MO–Michael Maurer, Jr., MD, Linda Baker; St Joseph Hospital, Denver, CO–David A. Belenky, MD, Christinia T. Ukrainski, MD; St Joseph's Hospital Health Center, Syracuse, NY–Larry Consenstein, MD, Phyllis Palla, RNC, NP; St Louis Regional Medical Center, St Louis, MO–Corinne Walentik, MD, MPH, P. A. Menon, MD; St Luke's Hospital, Bethlehem, PA–Lloyd Tinianow, MD, Andrew Unger, MD; St Luke's Regional Medical Center, Boise, ID–Matthew S. Sell, MD; St Peter's Medical Center, New Brunswick, NJ–Thomas Hegyi, MD, L. Christmas, RNC; Sunrise Children's Hospital, Las Vegas, NV–Nita Johnson, RN, Jennifer Beecham, RN, CNS; Texas Tech University Health Sciences Center, Amarillo, TX–Mubariz Naqvi, MD, Phillip Platt, RN, NNP; The Children's Hospital of Greenville Hospital System, Greenville, SC–J. Ferlauto, MD, J. A. Wareham, MD; The Children's Hospital, Denver, CO–Adam A. Rosenberg, MD, Camille Shea-McAleavey, RN, MS; The Medical Center of Central Massachusetts-Memorial, Worcester, MA–Francis J. Bednarek, MD, Renee Gosselin, RN; The Toledo Hospital, Toledo, OH–V. Krishnan, MD, MPH; University of California, Davis Medical Center, Sacramento, CA–T. Allen Merritt, MD, Robin White, BA; University of Oklahoma, Tulsa, OK–George P. Giacoia, MD, Kathy Rossman, RN; University of South Florida/Tampa General Hospital, Tampa, FL–R. M. Nelson, Jr, MD, Cathy Groh, RN; University of Tennessee Medical Center at Knoxville, TN–Mark E. Anderson, MD, Tara M. Burnette, MD; University of Texas Medical Branch, Galveston, TX–Michael H. Malloy, MD, Margarette Allen; Wesley Medical Center, Wichita, KS–Paula Delmore, MSN, Barry T. Bloom, MD; West Boca Medical Center, Boca Raton, FL–Luiz A. Grajwer, MD.
Funded by a grant from the Center for the Future of Children of the David and Lucile Packard Foundation.
- Received July 18, 1996.
- Accepted October 17, 1996.
- Address correspondence to: Jeffrey D. Horbar, MD, Department of Pediatrics, University of Vermont College of Medicine, Given Building, Burlington, VT 05405.
Reprint requests to A. Lynn Stillman, Database Manager, Vermont Oxford Network, 444 South Union Street, Suite 1S, Burlington, VT 05401.
↵¶ Participating institutions and investigators are listed in the Appendix.
Dr Horbar is the Executive Director of the Vermont Oxford Network.
- Hack M,
- Horbar JD,
- Malloy MH,
- et al.
- The Investigators of the Vermont-Oxford Trials Network Database Project: the Vermont-Oxford Trials Network: very low birth weight outcomes for 1990
- ↵Luft HS, Garnick DW, Mark DH, McPhee SJ. Hospital Volume, Physician Volume and Patient Outcomes. Assessing the Evidence. Ann Arbor, MI: Health Administration Press; 1990:97–131
- Jenkins KJ,
- Newburger JW,
- Lock JE,
- et al.
- ↵Vermont Oxford Trials Network. Vermont Oxford Trials Network: Database Manual of Operations. Burlington, VT: Vermont Oxford Trials Network; 1991
- Arbuckle TE,
- Sherman GJ
- ↵American Medical Association. Graduate Medical Education Directory, 1991–1992. Chicago, IL: American Medical Association; 1991
- Liang KY,
- Zeger Sl
- Bye BV,
- Riley GF
- Morel JG
- Donner A,
- Klar N
- ↵Hosmer DW, Lemeshow S. Applied Logistic Regression. New York, NY: Wiley; 1989:307
- ↵Ewy W, Flood AB, Brown BW, Scott RW. A methodology for assessing hospital performance. In: Flood AB, Scott WR, eds. Hospital Structure and Performance. Baltimore, MD: Johns Hopkins University; 1987:126–173
- ↵SAS Institute Inc. SAS/IML Software. Usage and Reference Version 6. 1st ed. Cary, NC: SAS Institute Inc; 1989:501
- Avery ME,
- Tooley WH,
- Keller JB,
- et al.
- Horbar JD,
- McAuliffe TL,
- Adler SM,
- et al.
- Horbar JD,
- Wright EC,
- Onstad L,
- et al.
- Schwartz RM,
- Luly AM,
- Scanlon JW,
- Kellog RJ
- Richardson DK,
- Phibbs CS,
- Gray JE,
- et al.
- Richardson DK,
- Gray JE,
- McCormick MC,
- et al.
- ↵Crowley P. Promoting pulmonary maturity. In: Chalmers I, Enkin M, Keirse MJNC, eds. Effective Care in Pregnancy and Childbirth. Oxford, England: Oxford University Press; 1989:746–764
- Copyright © 1997 American Academy of Pediatrics