INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Keywords: epidemiology; mortality; mechanical ventilation; neonatal intensive care; ECMO; variation

Neonatal respiratory failure is a serious clinical problem (1) associated with high morbidity, mortality, and cost (4). The major risk factor is low birth weight (7, 8), a factor more prevalent among the poor, the uninsured, and minorities (9). The mainstay of management is supportive care with mechanical ventilation and high concentrations of inspired oxygen. Important adjunctive therapies include high-frequency ventilation, surfactant therapy, inhaled nitric oxide, and extracorporeal membrane oxygenation (ECMO) (3, 8, 13). However, not all institutions that deliver babies provide these therapies.

Optimal management, therefore, relies on a regionalized care system consisting of a select number of centers capable of providing advanced care, attempts to ensure high-risk pregnancies are delivered at or close to these advanced centers, and considerable use of interhospital transfer of sick neonates born at lower level centers. Regionalized care was promoted in the 1970s but the extent to which it was integrated into practice is unclear. There is also concern that recent market forces are fueling a trend to less regionalization with duplication of services by rival health care systems in lucrative markets and potentially decreased provision of services in other areas (11, 12, 16, 17).

Despite considerable pathophysiological and clinical research, there is a paucity of information on either the epidemiology of neonatal respiratory failure or current patterns of care. We therefore conducted a study of the incidence, cost, and outcome of neonatal respiratory failure in two states, including assessment of patient location, transfer rates, and ECMO use. We explored variation in incidence and outcome by demographic characteristics and generated national estimates of the number of cases and total costs of care.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We constructed our study cohort from the 1994 California and New York state hospital discharge databases, which include a record of every hospitalization at nonfederal hospitals, including children's hospitals. These states were selected because of the size, quality, and accessibility of their data. The information extracted included demographic characteristics (e.g., age at admission, sex, race [non-Hispanic black, non-Hispanic white, Hispanic origin, and other] and birth weight); origin of admission (out-born or in-born); payer; case-mix based on the ICD-9-CM codes for principal diagnosis, up to 24 secondary diagnoses and 21 procedures; discharge destination (home, skilled care, another acute care facility, or death); length of stay; and total hospital charges.

To define neonatal respiratory failure, we selected all hospital discharges for patients  28 d old on admission who underwent mechanical ventilation (ICD-9-CM procedure codes 96.70, 96.71, and 96.72), excluding noninvasive continuous positive airway pressure. Although we would have liked to determine which babies were ventilated in more than one institution, this was not possible because of limitations with unique identifiers for newborns. Instead, we studied hospitalizations that included mechanical ventilation. However, this leads to multiple counting of babies ventilated at more than one institution. Babies who were ventilated and then transferred to another facility within 24 h of birth were the most obvious source of double counting. Therefore, for our main analyses, we excluded hospitalizations of  1 d where the baby was discharged alive (the subsequent hospitalization at another hospital was still included). The number of hospital discharges with mechanical ventilation was stratified by birth weight and expressed as episodes/1,000 live births/year within sex, race, and birth weight class. The birth weight classes were < 1500 g, very low birth weight (VLBW); 1,500-2,499 g, low birth weight (LBW); and  2500 g, normal birth weight (NBW).

We used ICD-9-CM diagnosis and procedure codes to determine different patient groups, including those who received ECMO, underwent major surgical procedures, had congenital abnormalities (grouped as minor or major cardiac, neurological, or other), and sustained bacterial or fungal infection, septicemia, or major hemorrhage. A full list of the ICD-9-CM classification scheme is available from the authors.

We classified hospitals by the provision of ECMO (ECMO centers) and by the provision of ECMO and/or major pediatric cardiac or neurological surgery (higher level hospitals). We calculated hospital costs by multiplying hospital charges by hospital-specific cost-to-charge ratios (average ratio = 0.45 in California and 0.63 in New York) generated from the 1994 HCFA Provider Specific File (HCFA, Washington, DC; www.hcfa.gov).

We obtained 1994-1998 natality statistics from the National Center for Health Statistics to verify state birth rates and calculate age, sex, and race-adjusted national estimates (18, 19). Since 1994, there have been on average 4 million births per annum in the United States, with a variance of  0.2% between 1994 and 1998, both overall and within weight class. We express national estimates of the total costs of care adjusted to 1999 US$ using the consumer price index.

Data Analyses

We compared incidence, morbidity, mortality, and resource use within and across demographic and weight class subgroups using the Student's t test and Mann-Whitney Test for continuous data and the chi-square or Fisher's exact test for categorical data as appropriate. We conducted multivariate logistic regression analyses to test whether variables were independently associated with mortality, treating weight in 250 g increments. We assessed regression model performance by likelihood R² (a measure of the variability accounted for in the model), Hosmer and Lemeshow C-statistic (20) (a measure of goodness of fit), and the area under the ROC curve (21) (a measure of discrimination). We present continuous variables as means, medians, and standard deviations, categorical variables as counts and rates, and OR with 95% confidence intervals.

We explored differences in mortality and transfer rates for lower level hospitals in California, based on whether they were near or far from higher level centers (where near was defined arbitrarily as  20 miles and reexamined at alternative cut-offs) and included hospital admissions with length of stay (LOS)  1 d. We restricted these analyses to California because the number of births > 20 miles from a higher level center in New York was too small. For analyses of transfer rates and regionalization, we included hospitalizations of < 1 d duration.

We generated national estimates by multiplying the observed race, weight class, and sex-specific incidence rates by the corresponding number of births nationwide. Database manipulations were conducted in Visual Foxpro (Microsoft, Redmond, WA; www.microsoft.com). Statistical analyses and geographic computations were performed in Datadesk (Data Description Inc., Ithaca, NY; www.datadesk.com) and SPSS (SPSS Inc., Chicago, IL; www.spss.com).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Incidence and Outcome

There were 16,405 hospitalizations for neonates requiring mechanical ventilation. Of these, 1399 were transferred to another acute care facility within 24 h and were therefore excluded from our main analyses. Descriptive characteristics of the final sample (n = 15,006) are in Table 1. The median age on admission for the final sample was 1 d of age and 93.1% were admitted in the first 2 d of life. The overall mechanical ventilation rate was 18/1,000 live births but this varied more than 100-fold by weight with the highest rate in those weighing 700-800 g at birth (Figure 1). Although LBW and VLBW infants contributed the most cases, one-third were of normal weight. Boys consistently had higher rates, especially in larger babies for whom rates were 50% higher. Although there were significant differences in mortality by sex within weight class, these differences were explained by differences in weight distribution within weight class and the distribution of major anomalies (data not shown).

View larger version (53K): [in this window] [in a new window]   Figure 1.   Distribution of total births and neonatal mechanical ventilation by birth weight. The primary y axis is the mechanical ventilation rate per 1,000 live births within a given weight category. The secondary y axis is the percentage of all births and of all mechanical ventilation hospitalizations. Data are from New York for 1994. Higher mechanical ventilation rates are seen in smaller babies. However, because most babies are > 2,500 g, a large proportion of ventilated babies are of normal weight.

There were several important racial differences. Most notably, the ventilation rate in non-Hispanic blacks was almost double that of non-Hispanic whites. This was not only because of differences in ventilation rates within weight class but also because non-Hispanic blacks were much more likely to be VLBW, the weight class with the highest ventilation rate (2.9% of all non-Hispanic blacks were VLBW compared with 1.1% of all non-Hispanic whites, p < 0.0001). Mortality was also almost 50% higher in non-Hispanic blacks than in non-Hispanic whites.

Clinical characteristics are summarized in Table 2. Infection and septicemia were more common in lower weight classes. Pneumonia was the most common source of infection (28% of cases with infection). Congenital abnormalities were common but most were minor cardiac conditions with hospital mortality rates similar to those without such conditions. Not surprisingly, those with major congenital abnormalities had significantly higher mortality rates than others (63.7 versus 2.7 deaths per 1,000 live births, respectively, p < 0.0001). Congenital anomalies were 10 to 30 times more common in ventilated patients than in nonventilated patients (data not shown). Major bleeding, including intracranial hemorrhage, was common, especially in the VLBW class, and associated with high mortality.

The best fitting multivariate mortality regression model accounted for most of the variability in the data (likelihood R²: 0.802) with moderate goodness of fit (C statistic = 57.2 with 8 degrees of freedom, p < 0.001) and discrimination (ROC = 0.702). The major risk factors for death were lower birth weights (e.g., OR: 12.43 [6.55-23.52] for birth weight < 500 g in comparison with birth weight = 1000-2500 g), major hemorrhage (OR: 2.46 [2.05-2.94] compared with no hemorrhage), and major anomaly (OR: 2.44 [2.16-2.75] compared with no anomaly). Minority race was also associated with increased risk although the risk associated with non-Hispanic blacks was not significant (compared with non-Hispanic whites, ORs were 1.43 [1.12-1.82] for Hispanics, 1.34 [0.90-1.65] for non-Hispanic blacks, and 1.46 [1.15-1.86] for other minorities). Sex was not significant after adjustment for these factors.

Costs of Care

Average LOS and costs are shown in Table 3. In general, neonates requiring ventilation incur long hospitalizations with high cost, with smaller babies staying longer and costing more. LOS in survivors was 2 mo, 3 wk, and 2 wk in the VLBW, LBW, and NBW classes, respectively. Those discharged to another acute care facility generally consumed half to two-thirds of the costs of those discharged elsewhere, but these cost estimates do not include the subsequent costs at the other acute care facility. Nonsurvivors were quite different, with much lower costs and very short median LOS, regardless of weight class. There were no differences in payor mix between those transferred and those not transferred. Medicaid was the principal payer in 49.2% of all cases and did not vary widely by weight class (range: 47-51.6%).

The small proportion of neonates who required surgery were significantly more expensive. LOS was double that of nonsurgical LBW and NBW neonates though similar in VLBW neonates, and costs were at least double in all weight classes with a mean LOS and cost of 34.9 d and $83,300. Those requiring ECMO (n = 178, of whom 10 were < 2,500 g) were more expensive with a mean cost of $89,800.

Patterns of Care

During the study period, there were 482 hospitals in California and New York with 10 or more births. There were 245 hospitals that ventilated neonates, 21 of which were children's hospitals and had no deliveries. Of the 245, 187 (76.3%) provided mechanical ventilation for > 24 h to  10 patients per year. Of the 187, 37 were higher level hospitals and 15 were ECMO centers. The median number of ECMO procedures per ECMO center was 12 with 6 (40%) centers performing  5 per year. Not surprisingly, the in-born rate was much lower in higher level than lower level hospitals (15.9 versus 35.7%, p < 0.0001), especially for NBW infants (8.8 versus 34.8%, p < 0.0001). Even after excluding surgical cases and cases requiring ECMO, average costs of care and LOS were greater at higher level hospitals (mean cost: $55,711 versus $44,461, p < 0.0001; mean LOS: 32.4 versus 30.5, p = 0.002). These trends were apparent in all weight classes but widest in larger neonates. These differences were likely due to a higher severity of illness at higher level hospitals as reflected by the lower in-born rate and a higher rate of infection (20.0 versus 15.6%, p < 0.0001), major congenital abnormalities (8.4 versus 6.2%, p < 0.0001), and hospital mortality (13.9 versus 9.7%, p < 0.0001). Those who died at lower level centers were similar to those who died at higher level centers with respect to weight and had similar or lower rates of major congenital anomalies, infection, septicemia, or major hemorrhage (Table 4).

The location of the hospitals and their distribution of cases for California are shown in Figure 2. Although most cases were within 20 miles of higher level hospitals, 32.0% were >20 miles and 21.8% were >40 miles. We explored the relationship between transfer to acute care hospital rates, hospital mortality rates, and distance from a higher level center for all neonatal hospitalizations with mechanical ventilation (n = 16,405). Transfer rates varied widely (range: 0-100%) but were higher among hospitals located further from higher level centers (25 versus 21.3% for hospitals >20 miles versus hospitals 20 miles from higher level centers, p < 0.001). Distant hospitals were more likely to transfer patients early as evidenced by a higher proportion of transfers on Day 1 (67.8 versus 52.3% of all transfers, p < 0.0001). Despite higher transfer rates, more distant hospitals had higher mortality rates (9.0 versus 7.0%, p < 0.002) than hospitals within 20 miles of higher level centers. These trends were unaffected by different distance cut-offs >20 miles.

View larger version (13K): [in this window] [in a new window]   Figure 2.   Geographic distribution of neonatal mechanical ventilation in California. Data are shown for all hospitals (n = 130) that ventilated 10 or more neonates in 1994 and all cases within those hospitals (n = 10,893). All facilities have Level III/ IV NICUs. Higher level facilities are those that provide ECMO, neonatal neurosurgical services, or neonatal cardiac surgery services. The map shows the distribution of centers geographically. The graph shows the distribution of cases by distance from a higher level center (expressed as the percentage of cases  x miles from a higher level center). Thus, the percentage is 100% at x = 0 miles because all cases are  0 miles from the nearest higher level center whereas 32% of all babies are located  20 miles from the nearest higher level center.

National Estimates

We estimate there are 79,400 hospitalizations in the United States each year for neonates undergoing mechanical ventilation, 8,500 of whom die. Cases are distributed evenly across the three weight classes (VLBW25,600; LBW26,400; NBW23,400) but half the deaths are in the VLBW class (VLBW4,400; LBW1,700; NBW2,400). Importantly, almost 40% (n = 3,300) of deaths occur in cases that never receive care at higher level centers. The annual cost of hospital care is $4.4 billion (VLBW$2.5 billion; LBW$999 million; NBW$891 million).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Several findings emerge from this study. First, the size of the problem is considerable. Our estimate of 80,000 cases of neonatal mechanical ventilation per annum represents 2% of all live births in the United States, is one-fifth the size of the number of cases of adult mechanical ventilation (22), and is similar to (23) or higher than (24, 25) estimates of the number of cases of adult acute respiratory distress syndrome. The total cost of $4.4 billion is also high, especially considering our estimate does not include the high postdischarge costs (4). In comparison, asthma has total annual costs (including both direct and indirect health-related costs) of $5.8 billion (26). That half the costs were incurred in the care of babies insured under state Medicaid programs has important implications for the financing of these programs and for hospital reimbursement. The 8500 deaths are one-third of all neonatal deaths in the United States. Thus, neonatal respiratory failure, often "below the radar screen" when considering priorities for improved quality of care and cost control initiatives, is a much larger public health problem than previously anticipated.

Our data quantitating the strong relationship between birth weight and the need for ventilation may have practical implications for decisions regarding the appropriate sites for delivery and neonatal care and the need for maternal or neonatal transport. For example, at birth weights  1,500 g, fewer than 10% of neonates will need mechanical ventilation. Providing an estimate of risk for neonatal respiratory failure based on expected or actual birth weight allows physicians and parents to make more informed, and perhaps more timely, choices regarding site of care. Although premature babies are at higher risk for neonatal respiratory failure, NBW babies still accounted for a third of cases (2). New therapies for neonatal respiratory failure, such as surfactant, have mainly been studied in smaller babies (8) yet are used increasingly in the management of term, NBW babies (14) despite a different case mix, a lack of efficacy data, and possible harm (5, 6, 27).

Our data raise concerns about the effectiveness of regionalization. The mortality rates at lower level centers were not trivial. Indeed, almost half of all deaths occur outside higher level centers. Furthermore, a large proportion of these babies who died at lower level centers did so without major contraindications to more intensive therapies. The higher mortality at more remote lower level centers suggests these centers have a sicker case mix, are less likely to transfer more complex cases, or provide worse care. Prior work from South Carolina found more remote centers cared for a higher proportion of babies born to young, unmarried mothers, suggesting a sicker case mix (11). The geographic overlap of ECMO centers, with their rather low ECMO procedure rates, is also of concern, especially considering the Extracorporeal Life Support Organization recommendation that a minimum of 12 ECMO procedures per annum is required for competence (30). Taken together, these findings support the contention that there may be duplication of services in major metropolitan areas and a lack of services located in more remote regions.

Our study also clarifies some risk factors for mortality. Sex differences in the severity of lung disease have been reported in premature neonates (8, 31). Our data suggest these differences are due to the distribution of other risk factors and do not represent an independent effect of sex. The fact that some minority groups have a higher neonatal mortality rate is known and has been attributed to a higher incidence of premature births. Our study suggests there may be additional factors as minority race remained a significant risk factor for hospital mortality even after adjusting for birth weight and other patient characteristics. To improve interracial differences, we should focus not only on prevention of prematurity but also assess the extent to which other differences in the provision of care exist between racial groups.

There are limitations to our study. There are no prospective observational cohorts of neonatal mechanical ventilation in the United States from which population-based estimates can be calculated. Therefore, we had to rely on administrative data, the potential drawbacks of which are well documented (32, 33). To minimize inaccuracies, we relied principally on codes and fields that are likely to be accurate. Neonatal respiratory failure was diagnosed on the basis of mechanical ventilation procedure codes. Major procedures are generally coded accurately because they are easily audited and often tied to reimbursement. This approach is used in other areas of medicine (22, 34). We also chose states with well-established data collection procedures, comprehensive auditing, and rigorous reporting. Both New York and California have active quality improvement initiatives based on their state hospital discharge data.

Although our study provides insight into patterns of care, we were unable to link records across hospital admissions and therefore could not follow each child's course from hospital to hospital. Such information would be valuable but is complicated both because of patient confidentiality and because social security numbers, the usual basis for patient linkage, are not assigned immediately. Also, the distance of a hospital from an ECMO or higher level center is a crude and incomplete measure of "remoteness." Further work on the effectiveness of regionalization requires more comprehensive measures of remoteness, as well as consideration of other patient, organizational, and environmental characteristics and evaluation in other states.

Sick neonates, once stabilized, may be transferred back from a regional center to a local hospital. In such instances, we do not include the additional costs of care back at that local hospital. We also do not include postdischarge costs of care. Thus, our estimates of the costs of care, which are already considerable, are likely underestimates.

In summary, we found that neonatal respiratory failure is an expensive and frequently lethal condition that accounts for a large portion of neonatal deaths in the United States with costs similar to those of more prevalent conditions. Low birth weight is the dominant risk factor, but weight is normal in one-third. Although there is evidence of regionalized care, it is perhaps less than optimal. Further research into the effectiveness of therapies in NBW babies, the optimal distribution of NICU services, methods for determining the appropriate site of care, and thresholds for maternal and neonatal transport may lead to improved cost and quality of care.