Intensive newborn care, organized in neonatal intensive care units (NICUs) and staffed by neonatologists, is highly effective at improving neonatal outcomes.1 Forty years ago, NICUs were available only in tertiary and large maternity hospitals. The scarcity of NICUs at most birth hospitals stimulated regionalization efforts led by national perinatal stakeholders,2 emphasizing early identification of maternal/fetal risk, referral of pregnant women for specialized antenatal care, and, when newborn illness is anticipated, delivery at a hospital with a NICU of appropriate capabilities.3 Regionalized perinatal systems have also promoted the development of systems for newborn transport to hospitals with more advanced care.

The success of perinatal regionalization remains mixed.4 Growth in the number and size of NICUs has been robust, outpacing changes in birth volumes,5,6 NICU admission rates have increased, particularly in newborns with lower overall risk (eg, higher gestational age and birth weight).7 Published estimates of US newborns admitted to a NICU range from 9% to 12.2% depending on the birth population and the definition of a NICU7–9 with a median birth weight of >2500 g (ie, not low birth weight).7 While perinatal regionalization requires cooperation between providers, the economic benefits of NICUs to hospitals (eg, maternity hospital prestige, revenue) have resulted in competitive markets for maternity services. One result is falling NICU patient volumes, and care has shifted to lower level NICUs (ie, Level II). This change is particularly concerning for newborns born very preterm (<28 wk gestation), where delivery in higher volume hospitals with Level III/IV units is associated with lower mortality and morbidity. State regulation of perinatal care systems remains weak, and the US lacks central planning of the clinician workforce and hospital investment found in most other high-income countries.10

Largely ignored by stakeholder perinatal improvement and regionalization initiatives4 is the fact that NICU capacity—the number of NICU beds and neonatologists per live birth—varies substantially across regions.11,12 In the early days of NICU care, this variation reflected the relative scarcity of NICU services, but more recently, attention has turned to possible regions of oversupply. Twenty years ago, greater per capita supply of neonatologists was associated with lower neonatal mortality in higher risk newborns, suggesting possible underuse of intensive care in some regions.13 However, in newborn groups with lower risk more recent studies have observed an association between regional or hospital-level NICU bed supply and admission rates and length of stay without evidence of patient benefit.6,14 From these and other studies, there are concerns that NICU care may now be overused for lower acuity illness.15,16 Supporting this idea is the success of some health systems of implementing care models that reduce NICU admissions.9

Given the persistently poor overall US perinatal outcomes,17 effective and efficient NICU care depends, in part, on capacity growth tracking regional newborn needs. In this study, we investigate the regional patterns of growth in neonatologists and NICU beds over 26 years, 1991–2017, using publicly available datasets. We then quantify the associations between capacity and perinatal risk (ie, health risk of mother, fetal, and newborn) in 1991 and regional growth in capacity to determine if capacity grew in response to greater need.

METHODS Data Sources and Study Population

We conducted a population-based longitudinal ecological study using 3 data sources: (1) total NICU bed location from American Hospital Association hospital surveys; (2) clinically active neonatologists and neonatal fellows location from American Medical Association Masterfiles (American Medical Association, Chicago, IL); and (3) CDC Linked Birth-Death data files, with maternal and newborn characteristics across health service regions. All live infants born in the United States in the years 1991 (n=4,103,528) and 2017 (n=3,849,644) weighing ≥400 g (a commonly used threshold of viability) were included in the analysis. Study years constitute a convenience sample bound by data available from previous studies and the most recent year available to the investigators.

Neonatal Intensive Care Regions

The units of analysis, neonatal intensive care regions (NICRs), are 246 market-based regions of tertiary newborn care delineated using small area analysis18 (Supplemental Fig. 1, Supplemental Digital Content 1, https://links.lww.com/MLR/C676). Initially defined in the late 1990s19 and revalidated in 2019,6 NICRs are based on travel patterns from county of maternal residence to county of birth, irrespective of state lines, for mothers with low–birth weight newborns. These regions have a high concordance of maternal residence, all and low–birth weight births, and neonatal deaths, indicating that they represent geographic markets (ie, service areas) for NICU care. Each live birth was assigned to the NICR of the maternal residence.

Maternal-Infant Characteristics

The 4 primary perinatal risk factors used in analyses were maternal-reported race, educational attainment, and low and very low infant birth weight.20 Lower birth weight is strongly associated with health outcomes and a direct measure of the need for advanced newborn care. Maternal low educational attainment is an indicator of socioeconomic status and an indirect indicator of newborn health risk. At a population level, maternal self-identification as Black indicates a risk of systemic racism and personal discrimination strongly associated with poorer health in reproductive years, difficulty in accessing good health care, and worse maternal and fetal/newborn outcomes.21,22 The expectation is that these measures indicate an increased population newborn health risk, and hence an overall higher need for NICU care and capacity. Six additional maternal and newborn characteristics that are associated with neonatal outcomes are included in secondary analyses (Supplemental Table 1, Supplemental Digital Content 1, https://links.lww.com/MLR/C676). It should be noted that the relationships of these characteristics with outcomes are largely mediated by the primary exposures of low and very birth weight. We calculated the regional proportion and percent changes (ie, 2017 minus 1991) in these key characteristics of perinatal risk (Table 1).

TABLE 1 - Maternal/Newborn Characteristics (≥400 g), US Birth Cohort and Across Neonatal Intensive Care Regions Maternal/newborn characteristics 1991 (N=4,103,528) [n (%)] 2017 (N=3,849,644) [n (%)] Absolute difference* (%) Individual level  Low birth weight 289,569 (7.1) 315,993 (8.2) 1.1  Very low birth weight 50,321 (1.2) 51,224 (1.3) 0.1  Less than high school education 926,572 (22.6) 504,857 (13.1) −9.5  Mothers who identify as Black 680,339 (16.6) 656,166 (17.0) 0.4 Neonatal intensive care regions (N=246)† Median (%) IQR‡ CV§ Median (%) IQR CV  Low birth weight 6.8 5.8–8.0 22 8.3 7.4–9.1 17  Very low birth weight 1.1 1.0–1.4 30 1.3 1.1–1.5 25  Less than high school education 21.2 25.1–27.1 47 12.8 10.2–15.4 32  Mothers who identify as Black 10.7 3.6–25.0 97 13.4 5.7–23.4 82

*All differences of proportions were tested using the Fisher exact test (P<0.001).

†All 246 regions are reported and use for analysis except when suppression was necessary for 2017 percent very low birth weight (n=1), extremely low birth weight (n=6), and maternal black (n=1).

‡IQR=interquartile range: 25th percentile, 75th percentile.

§Coefficient of variation (CV)=SD/mean×100.

Measures of Capacity

We calculated 2 measures of NICU bed capacity per 1000 live births (Table 2): (1) the number of total NICU beds; and (2) the “adjusted” number of neonatologists, by adding the number of clinically active neonatologists to 0.35 times the number of neonatology fellows to account for their lower clinical productivity.13 We also recalculated the capacity measures using the number of low and very low–birth weight newborns as denominators, newborn populations that are important target populations of NICU care. Change in capacity was the 2017 minus 1991 difference of the measures.

TABLE 2 - Growth in Neonatal Intensive Care Unit (NICU) Capacity Across Neonatal Intensive Care Regions (N=246) 1991 2017 Change 1991–2017 All measures per 1000 live births (LB) Overall Median CV* Overall Median CV Overall Median IQR† CV Overall change‡ (%) Median change (%) NICU beds per LB 5.4 4.0 55 7.6 7.3 51 2.4 2.2 0.3, 4.1 167 42 80 NICU beds per low–birth weight (<2500 g) LB 76.1 58.7 52 93.9 89.7 44 16.8 15.9 −8.9, 38.5 346 23 53 NICU beds per very low–birth weight (<1500 g) LB 448.8 344.9 55 665.1 584.7 60 208.6 135.7 −41.9, 265.8 206 48 70 Adjusted neonatologists per LB 0.4 0.4 66 1.3 1.0 63 0.9 0.6 0.4, 1.0 92 200 172 Adjusted neonatologists per low–birth weight (<2500 g) LB 6.3 6.0 63 15.4 12.7 60 9.1 7.0 3.3, 11.5 99 144 129 Adjusted neonatologists per very low birth weight (<1500 g) LB 37.4 35.1 67 95.0 82.4 64 57.6 43.3 20.8, 78.6 103 154 137

*CV=coefficient of variation (SD/mean×100).

†IQR=interquartile range: 25th percentile, 75th percentile.

‡Change in overall capacity tested with paired Student t test; all P<0.001 except for change in intensive NICU beds per low–birth weight live birth which was 0.003.

Statistical Analysis

The magnitude of variation across neonatal intensive care regions was estimated with interquartile ranges (25th percentile, 75th percentile) and coefficients of variation (CVs; SD/mean×100). Change in overall US newborn risk proportions during the study period were tested with the Fisher exact test and changes in capacity with the paired Student t test. After assessing the normality of variable distributions, the associations (ie, coefficient of determination or R2) of perinatal risk and capacity were initially estimated with bivariate linear regression. Then all primary risk factors were modeled simultaneously, and finally the full set of risk factors. In sensitivity testing, we substituted clinically active neonatologists (ie, postfellowship) and added analytic weights to linear models using live births per NICR. A few study areas had a very low number of births that required suppression of the measures and exclusion from analyses (Table 1). We used Stata 16.1 for all analyses. All statistical tests were 2-sided with <0.05 as the threshold of significance. “Statistical significance” does not indicate that the magnitude of an association is clinically or policy relevant. The institutional review board of Dartmouth College exempted the study from human subjects review.

RESULTS Perinatal Risk

The percentages of low–birth weight and very low–birth weight newborns and of mothers who identified as Black increased over the study period (absolute change 1.1, 0.1%, and 0.4%, respectively), while infants born to mothers with less than a high school education decreased by 9.5% (all P<0.001) (Table 1). Most of these risk factors varied moderately across regions in 1991 and 2017 (CV <50) except for the proportion of women who identified as Black, which exhibited greater variation (CV in 1991: 97; CV in 2017: 82).

Growth in Regional Neonatal Intensive Care Unit Capacity

Between 1991 and 2017, there were increases in all measures of overall NICU capacity per 1000 live births across every category of newborn birth weight (Table 2, Supplemental Fig. 2, Supplemental Digital Content 1, https://links.lww.com/MLR/C676). The highest change occurred in the total live birth population, with an increase of 42% in NICU beds and 200% in adjusted neonatologists (all P<0.01). In all, 236 NICRs experienced an increase in the number of adjusted neonatologists per live birth (median increase 0.6 neonatologists per 1000 live births), and 195 NICRs experienced an increase in NICU beds (median increase 2.2 beds per 1000 live births). There was a weak inverse relationship between 1991 NICU beds and growth in supply with an of R2 of 0.10 (P<0.01) indicating that 10% of the regional growth in beds could be explained by lower bed capacity in 1991 (Table 3, Supplemental Fig. 2, Supplemental Digital Content 1, https://links.lww.com/MLR/C676). This, however, was not true for the numbers of adjusted neonatologists (R2=0.00; P>0.05) In both time periods the regional variation was moderate, with the CV ranging from 44 to 67; however, variation in change in regional capacity was high for adjusted neonatologists (CV 92–103) and extremely high for both intensive and total NICU beds (CV 167–568).23

TABLE 3 - Association of Regional NICU Capacity With Perinatal Risk and Change in Perinatal Risk, Neonatal Intensive Care Regions (n=246), 1991 and 2017 Total NICU beds Adjusted neonatologists Perinatal risk measures Coefficient of determination (R 2) Regional NICU capacity with proportion of mothers/births with  Low birth weight   1991 0.13** 0.02*   2017 0.07** 0.00  Very low birth weight   1991 0.11** 0.02*   2017 0.07** 0.01  Lower educational attainment   1991 0.03** 0.01   2017 0.01 0.06**  Who identify as Black   1991 0.07** 0.01   2017 0.05** 0.02* Change (1991–2017) in regional NICU capacity with  1991 NICU beds or neonatologist per live birth 0.10** 0.00  1991 proportion of mothers/births   Low birth weight 0 0.01   Very low birth weight 0 0.02*   Lower educational attainment 0.00 0.01   Who identify as Black 0.00 0.02*  Change in proportion of mothers/births with   Low–birth weight newborns 0.00 0.01   Very low–birth weight newborns 0.03** 0.01   With lower educational attainment 0 0   Who identify as Black 0 0.02*

Coefficients of determination from bivariate linear models after assessment for the assumption of normal distribution.

NICU indicates neonatal intensive care unit.

*P<0.05.

**P<0.01; the direction of associations with P<0.05 was positive. Region suppression was necessary for percent very low birth weight (n=1), extremely low birth weight (n=6), and maternal Black (n=1).

Relationship Between Capacity and Perinatal Risk, 1991 and 2017

There was no meaningful (ie, clinical or policy relevant) association between perinatal risk and adjusted neonatologists in 1991 and 2017 (Table 3). There was a weak relationship in 1991 (R2=0.12; P<0.001) between regional NICU bed capacity and low–birth weight births (Fig. 1), which decreased by 2017 (R2=0.07; P<0.05). This means that only 12% of the regional variance of NICU bed capacity in 1991 could be explained by perinatal risk as measured by low birth weight, declining to 7% in 2017. No other meaningful associations in regional perinatal risk and NICU beds were observed.

FIGURE 1:

Association of neonatal intensive care unit (NICU) beds per 1000 live births with percent of low–birth weight live births, 1991 and 2017, neonatal intensive care regions (n=246). One 2017 observation with an extreme value (42 beds per 1000 live births) was omitted from the figure.

Relationship Between Growth in Capacity and Growth in Perinatal Risk

There were no meaningful associations (R2 for all correlations <0.03) between regions of higher perinatal risk, such as percent of low–birth weight live births, and change in measures of capacity (Table 3; Fig. 2A). Similarly, there were no meaningful associations between regional change in perinatal risk and change in measures of capacity (Table 3; Fig. 2B).

FIGURE 2:

Association of change (1991–2017) in total neonatal intensive care unit (NICU) beds per 1000 live births with percent low birth weight in 1991 (A) and with change in percent low birth weight (B), neonatal intensive care regions (n=246).

Secondary Analyses

Six additional indicators of perinatal risk were also evaluated (proportion with low Apgar score, maternal tobacco use, diabetes, age less than 18 y, hypertension, and multiple births) (Supplemental Table 1, Supplemental Digital Content 1, https://links.lww.com/MLR/C676). These analyses failed to find meaningful positive associations with indicators of perinatal risk and capacity.

Sensitivity Analysis

Neither the substitution of clinically active neonatologists nor weighted bivariate linear models altered the study findings (not shown). Associations were low or absent in linear models simultaneously including primary risk factors and supplemental risk factors (Supplemental Tables 2A–2J, Supplemental Digital Content 1, https://links.lww.com/MLR/C676).

DISCUSSION

In this study, we report a weak regional relationship between NICU capacity and population indicators of the need for advanced newborn care during a period when capacity markedly increased. The growth rates of both beds and neonatologists varied highly across regions, although almost all regions experienced an increase in neonatologist capacity and a majority showed increases in NICU beds. Some of the growth in NICU beds could be explained by higher rates of survival in 2017, but requiring only 14% of the beds added during the study (Supplemental Table 3, Supplemental Digital Content 1, https://links.lww.com/MLR/C676). The weak association between regional NICU capacity and perinatal risk factors in 1991 was even lower in 2017. While the expansion of NICU capacity occurred in regions with lower capacity in 1991, these were not NICRs with greater perinatal risk, nor did expansion occur at a higher rate where risk increased, the regions which might have most benefited from greater capacity. The disassociation of regional NICU capacity from patient needs identifies a type of “unwarranted variation” in health care capacity.24

This study adds to the limited existing literature on the distribution and growth of advanced-care nurseries. A cross-sectional study using 1996–1999 population-based data reported low correlations between regional NICU capacity and low birth weight.6 In a study of increasing supply of NICUs during the period 2002–2013, average travel times did not decrease for rural mothers and newborns to NICUs,25 although a study from the period 2003–2019 reported a 21% decrease in distance from all children’s residences to the nearest neonatologist.26 Importantly, there is no evidence that higher NICU availability in recent years has led to newborn benefit at a population level, possibly because the higher use of NICUs has occurred primarily in newborns of lower acuity.7,16 In contrast, there is an extensive literature on adult health care capacity and its consequences, demonstrating that an irrational distribution of clinicians and capital investments (eg, hospitals, intensive care units, imaging, procedure suites) is common,27,28 and is likely to be a persistent driver of shortcomings in quality and outcomes and higher growth in costs.28,29 One striking contrast is that, almost without exception, regional risk for adults is derived from medical claims files, such as Medicare, while this study uses information collected from mother/newborns unrelated to health care utilization.30

Reports to date on the consequences of current patterns of NICU capacity should raise concerns. Federal and state initiatives to improve the regionalization of perinatal care, for example, are affected by the distribution and growth of NICU capacity.4,31,32 In California, growth in the number of mid-level units during the years 1993–2000 appears to have caused deegionalization of newborn care, increasing the chance that very low–birth weight newborns are admitted to lower level units.33 Greater availability of NICU care also leads to greater competition for a relatively fixed patient population, resulting in lower unit volumes, which have been associated with poorer outcomes in very preterm newborns.34 Regionalization requires cooperation and coordination among hospitals, while hospitals are financially rewarded for expanding capacity, delivering higher levels of service, and competing for deliveries. In addition to the possible disruptive effect on federal and state regionalization initiatives, persistent regional variation of NICU capacity may also be a primary cause of overuse.6,14,35

The cause of the general (ie, not specifically perinatal) maldistribution in health care capacity has generally been attributed to financial incentives that promote hospital investments in more affluent communities with better-insured populations36–38 and the preferences of clinicians to live and practice in places with higher income potential and better partner employment opportunities and education for family members.39 The regulatory and professional response to market forces that influence capacity has been weak. While most countries allow clinicians and health care facilities some autonomy in service location and capital investment, the US reliance on lightly regulated market forces is found in few other advanced economies.40

The most direct state regulation of perinatal care hospital capacity is the designation of hospital nursery/NICU levels in 28 states.41 Following the American Academy of Pediatrics (AAP) guidelines,42 Level I indicates the capability to care for normal newborns or stabilize unexpected illness before transport; Level II units provide intermediate care of mildly ill or slightly premature newborns; Level III units provide critical care for very ill newborns, including those born extremely premature; and Level IV units provide the most advanced medical and surgical care. Even in the states with designation regulations, definitions often deviate from the guidelines,43 many states rely on hospital self-reporting, and hospital websites commonly mischaracterize both NICU levels and the unit’s capabilities.41

A second type of regulation of hospital capital investments that sometimes affects NICU investment is state-level certificate of need (CON) programs. The application and effectiveness of this mechanism, however, varies and has been weakened or repealed by some states.44 Within the 30 states that have retained CON rules for NICUs, the regulations are seen as moderately effective in slowing new NICU building.45 Whether CON states have a better regional alignment of capacity and perinatal risk has not been studied. Given that facility costs represent a high proportion of total newborn care expenditure,46 bed supply is a likely determinant of costs overall and by region. At the same time, under-capacity persists in some regions of very high need.47 The possible value of greater public oversight in NICU size and location in the interest of better population outcomes and lower costs requires further exploration.

The growth and distribution of neonatologists is central to NICU capacity, given that they are required to lead NICU care. Pediatric workforce policy advocacy, however, has primarily been concerned with addressing current or possible future subspecialty shortages without drawing distinctions between the differing need for specific subspecialties.48 In this literature, the need for higher numbers of specific subspecialties is poorly measured, leading to a general call for more training. Despite a stable or declining number of births in the United States, there is a robust expansion in the number of neonatal fellows filling all available training positions.49 There has been no recent discussion regarding possible oversupply in the number of neonatal clinicians.32,48 In general, public policy input into physician training rates by specialty—pediatric or adult—remains very limited in the United States.50

The problem of unwarranted variation in capacity may be resistant to front-line (ie, clinical) improvement efforts. Hospital-level initiatives can shift the care of one category of low-acuity newborns, such as late preterm newborns, from NICUs to maternity-nursery units.9 An important area of future research is to determine if NICU bed occupancy remains constant, and if so, are the beds filled with other low-acuity newborns?

An additional approach to encourage better capacity decisions by providers is to rationalize financing and service delivery.51 High-margin fee-for-service payments for NICU care, either paid according to Diagnostic Related Groups or per diem, in the absence of professional consensus about the benefits of NICU care for the most common newborn occupying a bed, those of low illness acuity, appears to have funded unchecked NICU growth.35 One economic analysis found that the effect of an open NICU bed on the likelihood of admission is highest when the financial incentives are more significant.15 These incentives continue and may lead to further growth in NICU utilization and capacity in some markets. Not surprisingly, while policy initiatives that address market failure in newborn care at a population level have potential in theory, these face resistance from the hospital and specialty communities.

Limitations

Several limitations of this study merit discussion. This study uses data from 2 time periods, and while we cannot assert that the findings apply to all of the intervening years, 2 cross-sectional studies support the lack of association during 1995–1996 and 2012–2013.6,52 Measuring NICU beds and neonatal clinicians at a national scale inevitably leads to measurement errors. NICU beds and neonatologists are, however, very specific resources and previous studies that have cross-checked the sources used in this study have been reassuring.13 The absence of data about neonatal advanced practice nurses and physician assistants will lead to underestimation of NICU clinicians, particularly in 2017. It should be noted that neonatal nurse supplies also vary across states, but their alignment with neonatologist supply is not known.53 We are unable from our data to identify different levels of NICUs. Such information would provide a fuller view of regional capacity. Unfortunately, even among the 22 states that regulate NICU levels, the criteria vary.43,54 We used the Linked Birth/Death data files for the numbers and characteristics of births, but in a few low–birth volume regions, data suppression was necessary. In addition, there is the possibility of inaccurate reporting of maternal and newborn characteristics. In particular, as an assessment of the need for NICU care, gestational age is preferred over birth weight. However, in the Linked Birth/Death file, gestational age is reported only by week, and a less accurate measure was used in 1991 birth certificates. Measurement of patient need for medical care using administrative data is generally imperfect,30 but the US natality file provides the most complete and comprehensive information about perinatal health risk for any large population. The selected variables include indicators of socioeconomic and maternal and neonatal health status. The variables used in this project exceed the information which is usually considered necessary for high discrimination of the need for advanced newborn care.13 The fact that none of the variables are associated with changes in bed capacity strengthens the study’s conclusions. Finally, the absence of analyses to identify factors that lead to differential regional growth in capacity may be viewed as a study weakness. While this approach may be valuable in identifying modifiable factors, the time required for any change in capacity growth rates to affect population-level capacity is exceedingly long. Not surprisingly, public efforts to close hospital beds, even when there is good evidence for excess, have been strenuously resisted.55 The intent of our narrow examination is to measure the magnitude of the mismatch between supply and need to stimulate research into the patient consequences and possible remedies in the financing, health system design, and provider and public transparency of patterns of care.

CONCLUSIONS

NICU capacity grew robustly over the past two and a half decades, while the association between regional capacity and perinatal risk weakened. There was no meaningful association between the change in regional capacity with either the 1991 regions of greater perinatal risk or lower supply, or with regions of increased perinatal risk during the study period. The disconnect of NICU capacity from the need for advanced care raises questio