Birth certificate replacement

The advent and diffusion of neonatal intensive care services during the late 1960s and 1970s resulted in a marked reduction in neonatal mortality. (1) To facilitate the availability of neonatal care to all high-risk infants and to use the country's neonatal resources efficiently, the Committee on Perinatal Health recommended a regionalized system of neonatal intensive care in 1976. These recommendations included the referral of high-risk mothers and infants to a hospital with a regional neonatal intensive care unit (NICU). Specifically, hospitals with no or intermediate NICUs were expected to refer all infants that weigh 2000 g or less to a regional NICU. (2) This standard was based on expert opinion, and the only population-based studies that support this standard are based on the limited data available from birth certificates. Because failure to control for important risk factors that are not available from birth certificate data can bias studies of neonatal mortality in either direction, this issue needs additional study.

During the late 1980s and 1990s, technology and clinical expertise disseminated outside the tertiary centers, resulting in a proliferation of intermediate NICUs and an expansion of their scope of practice. There has been an increase in the rate of high-risk infants who are born outside tertiary centers (3-8) and a decrease in referrals to regional NICUs. (3)

In a previous study that examined high-risk births of all weights in California during 1990, we showed that there is an advantage of delivery in a hospital with a high-volume regional NICU. (9) It is unclear whether this advantage is limited to certain birth weight (BW) groups. It is possible that because of the development of new technologies (eg, surfactant replacement therapy) and the diffusion of technology and clinical expertise that the 2000-g threshold set in the 1976 recommendation is too conservative and that some uncomplicated births below this weight limit can safely occur outside the tertiary care centers. The objective of this study was to examine the impact of the level of care provided at the hospital of birth on survival of infants of <2000 g over different BW strata.

METHODS

The 1992 and 1993 California birth-infant death cohort file linked to the California Office of Statewide Health Planning and Development (OSHPD) discharge abstracts for 1992 and 1993 were used for this study. (10) These linkages were only for singleton births that occurred in nonfederal hospitals in California in 1992 and 1993; the linking could not distinguish among multiple births, and federal hospitals do not file discharge abstracts with the state. Ninety-nine percent of the maternal and singleton infant discharge abstracts were successfully linked with the birth certificates. These data were also successfully linked to the infant's discharge abstract from the receiving hospital for 99% of the infants who were transferred to another hospital. The advantages of these linked data include having information on the complete neonatal course of hospitalization for infants who are transferred and the diagnosis information from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes recorded on the hospital discharge abstracts. Although there are limitations to the ICD-9-CM code data, specifically a lack of severity information, validation studies have found a very high rate of coding accuracy. (11)

The focus of this study was restricted to singleton infants who had a BW of <2000 g. Because the current standard in most California hospitals is that all infants with a BW of <2000 g be admitted to some level of neonatal intensive care (unless they die before they can be admitted), we did not impose any additional criteria for inclusion in this study. Infants with BW of <500 g were excluded to be consistent with previous research and because the overwhelming majority of these infants died before discharge from the hospital, most in the first day of life. A total of 16 732 singleton infants were born in 1992 or 1993 with a BW of <2000 g and had the discharge record linked to the birth certificate.

We used the 1993 California Association of Neonatologists' Survey of Directors of Neonatology (12) to assign the level of care provided by each hospital. Any uncertainty about appropriate classification was resolved by direct communication with the senior staff of the unit. Each hospital was assigned to 1 of the 4 categories used by the State of California. (13) The criteria for level of care are as follows:

1. No NICU: Cared only for healthy neonates and those with minor medical problems.

2. Intermediate NICU: Cared for moderately sick infants but did not regularly provide assisted ventilation for more than 4 hours.

3. Community NICU: Provided long-term ventilatory support but not all other specialized services normally provided by regional NICUs.

4. Regional NICU: Provided a full range of specialized neonatal intensive care, including pediatric subspecialty consultants and surgery.

In addition to being based on the state criteria, these levels of care are very similar to those used by the Vermont-Oxford Network.

The OSHPD hospital data were used to determine total NICU patient days and to calculate average NICU daily census. These are regulatory data that all nonfederal hospitals in California must file annually with OSHPD. Following Phibbs et al, (9) we classified community NICUs as small when they had an average census of <15 patient and large otherwise. Intermediate NICUs were not sorted by patient census as all of these units had an average patient census of <13, almost all of which had an average census of <10, and we had previously identified a mortality threshold at an average census of 15 patients. All of the regional NICUs had an average census of at least 15 patients. The 15 patient census cutoff from Phibbs et al was based on an empirical volume-outcome relationship. (9)

To risk-adjust hospital mortality rates, we repeated the general approach of the logistic regression analysis that we used previously to model neonatal death using only data for infants with a BW of <2000 g. (9) The model controls for BW, gender, maternal demographic factors, and many clinical diagnoses. For infants who were transferred, diagnoses made at the referral centers were mapped back to the hospital of birth. The main difference in approach is that to prevent possible bias, we restricted the clinical variables considered for inclusion to those used by Phibbs et al (9) that were present at birth. Only the clinical variables that were associated with mortality at the 5% significance level were retained in the multivariate model. Patient demographic and insurance variables, and hospital characteristics were retained in the models regardless of statistical significance. To allow for gender differences in BW-specific mortality, BW was specified separately for each gender. The model was also estimated with a variable added for neonatal transfer to examine the association between transfers and mortality.

The dependent variable was neonatal-related mortality, which we defined as neonatal mortality (28 days) or any other death that occurred in the first year of life if the infant remained hospitalized continuously since birth. These later deaths were included to prevent bias from the exclusion of cases where care in the NICU prolonged death past an age of 28 days. We tested the sensitivity of the analysis to the inclusion both of the late deaths and of out of hospital neonatal deaths. Since both of these types of deaths were not very common, their inclusion or exclusion had very little effect on the results; they were included for completeness.

To examine the effects of NICU patient volume and level of care at the hospital of birth, we added variables for each level of care and NICU census at the hospital of birth to the regression model (regional NICU was the reference category). These models were used to calculate relative odd ratios (OR) and 95% confidence intervals (CI) for mortality at each hospital level, controlling for the risk profile of the infant using the Stata statistical program (Stata Corp, College Station, TX). The standard errors of the hospital level variables were corrected for within-hospital correlation using the "cluster" option in Stata. The models were estimated for infants with BW of <2000 g, <1500 g, and <1250 g to determine whether there is a BW limit below which NICU level at the hospital of birth has a significant effect on mortality. To determine whether there were risk differences for moderately low BW infants, we also estimated models for the 1250 g to 1999 g and 1500 g to 1999 g intervals.

RESULTS