ACP American College of Physicians - Internal Medicine - Doctors for Adults

Effective Clinical Practice


The Uneven Landscape of Newborn Intensive Care Services: Variation in the Neonatology Workforce

Effective Clinical Practice, July/August 2001

David C. Goodman, Elliott S. Fisher, George A. Little, Thérese A. Stukel, Chiang-hua Chang

For author affiliations, current addresses, and contributions, see end of text.

See related editorial

Context. In the past 30 years, the number of neonatologists has increased while total births have remained nearly constant. It is not known how equitably this expanded workforce is distributed.

Objective. To determine the geographic distribution of neonatologists in the United States.

Data Sources. 1996 American Medical Association physician masterfiles; 1999 survey of all U.S. neonatal intensive care units; 1995 American Hospital Association hospital survey; and 1995 U.S. vital records.

Measures. The number of neonatologists and neonatal mid-level providers per live birth within 246 market-based regions.

Results. The neonatology workforce varied substantially across neonatal intensive care regions. The number of neonatologists per 10,000 live births ranged from 1.2 to 25.6 with an interquintile range of 3.5 to 8.5. The weakly positive correlation between neonatologists and neonatal mid-level providers per live birth is not consistent with substitution of neonatal mid-level providers for neonatologists (Spearman rank-correlation coefficient, 0.17; P < 0.01). There was no difference in the percentage of neonatal fellows in the lowest and highest workforce quintile (14% vs. 16%) or in the percentage of neonatologists engaged predominantly in research, teaching, or administration (14% in lowest and highest quintiles).

Conclusions. The regional supply of neonatologists varies dramatically and cannot be explained by the substitution of neonatal mid-level providers or by the presence of academic medical centers. Further research is warranted to understand whether neonatal intensive care resources are located in accordance with risk and whether more resources improve newborn outcomes.

Take Home Points

The development of neonatal intensive care has been one of the most important recent advances in pediatrics. Neonatal intensive care, in both its technologies and regional organization, (1, 2) has been shown to be highly effective in reducing mortality and improving long-term outcomes for premature and ill newborns. (3-5) This success has led to a dramatic increase in the number of neonatologists and intensive care units in the past 30 years. For example, the average number of extremely low birthweight infants per neonatologist decreased from 39 in 1981 to 11 in 1996, and more than 60% of neonatologists now provide care to normal newborns. (6, 7) Although neonatal intensive care was once primarily the purview of academic medical centers, units are now commonly located in community hospitals providing obstetric care.

Although the growth in capacity is well documented, (6, 8-10) little is known about the regional distribution of the neonatal workforce. Most health care resources, such as hospital beds and physicians, have been shown to vary widely, and this variation can have important implications for utilization, costs, and patient outcomes. The one previous study done in 1983 examined regional differences in the neonatal workforce and found a 1.7 fold difference in the number of neonatologists per birth across the nine U.S. Census regions. (8) These regions, however, are very large and have diverse perinatal care systems.

To provide a more detailed and current view of the geographic distribution of the neonatal workforce, we defined regions on the basis of travel patterns of mothers and low birthweight newborns. We then determined the per capita numbers of neonatologists and neonatal mid-level providers in each region.


Defining Neonatal Intensive Care Regions

Using traditional methods of small area analysis, (11, 12) we defined health services areas called neonatal intensive care regions (NICRs). First, we used 1995 American Hospital Association data (13) to identify counties with at least 20 births. Counties of maternal residence were then assigned to counties with obstetric services based on the highest proportion (plurality) of travel from county of residence to county of birth as indicated in birth records from the 1995 Linked Birth/Death Data Set. (14) These county assignments resulted in 1616 newborn service areas. Newborn service areas were considered as candidate NICRs if they had a neonatal intensive care unit (NICU) in the 1995 American Hospital Association file and in a 1999 survey conducted by the American Academy of Pediatrics Section on Perinatal Pediatrics.

In the next step, newborn service areas of maternal residence for low birthweight infants, defined as those weighing less than 2500 grams, were assigned to NICRs based on 1) the plurality of travel from newborn service area of maternal residence to newborn service area of birth and 2) the degree to which neonatal births and deaths occurred within the same NICR as maternal residence. There was minimal border crossing of mothers and low birthweight infants in the defined regions. Thus, we used 246 NICRs for all analyses. The criteria for reassignment of border crossing and NICR characteristics are described in the Appendix.

Measuring the Workforce


The source of neonatologist counts was the January 1, 1996, American Medical Association masterfiles, which are a census of U.S. physicians irrespective of Association membership. In this census, 3174 physicians designated themselves as neonatologists. To estimate the clinically active neonatology workforce, we excluded neonatologists who devoted most of their professional time to medical teaching (n =97), administration (n =100), or research (n =232) and those who worked fewer than 20 hours per week (n =118), leaving 2627 neonatologists. Clinical neonatology fellows (n =377) were included but were assigned a weight of 0.35 to account for their lower clinical productivity. (15, 16) The median number of neonatologists per NICR was 6, with a range of 1 to 110 (interquartile range, 3 to 12).

Per capita measures were calculated as the number of neonatologists divided by the number of births occurring to mothers residing within the NICR. To further address the concern of border crossing, we created an alternative per capita measure that used NICR of birth as the denominator instead of maternal residence. Our measure was highly correlated with this alternative measure (correlation coefficient, 0.97; P < 0.001). This high correlation suggests minimal border crossing between NICR of maternal residence, NICR of birth, and NICR of very low birthweight death (Appendix).

Mid-Level Providers

To measure mid-level providers, a 1999 survey of NICU directors was conducted in conjunction with the American Academy of Pediatrics Section on Perinatal Pediatrics. Unit directors reported the number of "physician extenders" (neonatal nurse clinicians, neonatal nurse practitioners, and physician assistants). Using a strategy of up to three mailings per unit and telephone calls to nonresponders, the survey had a response rate of 100%.


Variation in Neonatologist Workforce

Neonatologists per capita varied substantially across NICRs (Figure 1). The number of neonatologists per 10,000 live births ranged from 1.2 to 25.6 (interquintile range, 3.5 [20th percentile] to 8.5 [80th percentile]; median 5.8 ). As shown in Table 1, the overall rate of neonatologists varied more than four-fold from the very low quintile of neonatologist supply (2.7 neonatologists per 10,000 births) to the very high quintile (11.6 per 10,000 births).

From the perspective of the physician, the volume of low birthweight deliveries also varied substantially: The number of low birthweight infants per neonatologist in the very high quintile of neonatologist supply was 66 compared with 263 in the very low quintile.

Table 2 shows the NICRs with the 25 highest number of births and the highest number in each state. Regions with a high ratio of local workforce relative to the U.S. average include Washington, DC (2.48), Falls Church, Virginia (1.81), Newark, New Jersey (1.80), and Charleston, South Carolina (1.78). Regions with low ratios include Great Falls, Montana (0.20), Lebanon, New Hampshire (0.37), Omaha, Nebraska (0.38), and St. Louis, Missouri (0.40).

Neonatal Mid-Level Providers

The number of mid-level providers ranged from 0 to 127.4 per 10,000 live births (interquintile range, 1.6 to 14.1; median 5.7). To learn whether neonatologist variation is explained by substitution with mid-level providers, we correlated the supply of mid-level providers with neonatologists. The weakly positive correlation (Spearman rank-correlation coefficient, 0.17; P < 0.01) is not consistent with substitution.

Characteristics of Neonatologists

The characteristics of neonatologists were very similar across quintiles of the neonatologist workforce (Table 3), and the median age differed only slightly across the quintiles. We also looked at two indicators of the presence of an academic medical center in the NICR: the percentage of neonatal fellows and the percentage of neonatologists engaged predominantly in research, teaching, or administration. Little difference was seen in either of these indicators. Thus, regions with a high per capita number of neonatologists did not have a greater presence of academic medical center neonatologists.


The number of neonatologists per newborn differs more than four-fold across market-based regions. This variation represents substantial differences in the availability of an important and costly pediatric specialty directed toward our most vulnerable patients--sick newborns. As a consequence of high neonatologist supply, regions in the "very high" quintile had a particularly low number of low birthweight infants per physician--66 on average during the year compared with 263 in the "very low" quintile. Although caring for premature infants is not the sole activity of neonatologists, it is the area of care for which effectiveness studies are most convincing. (3, 4)

It is possible that the distribution of capacity revealed in our study reflects regional differences in patient need, such as rates of prematurity or congenital anomalies. For example, differences in low birthweight rates are known to vary across states by about two-fold. (17) Smaller areas, such as our NICRs, are likely to have a higher degree of variation in risk. Although this study does not examine regional differences in risk, neonatal intensive care would be a singular medical care resource if it coincided with need, since previous studies have failed to find such an association in the location of hospital beds or the overall numbers of physicians in general. (18)

Could these findings simply reflect measurement error? Not likely. NICRs have minimal border crossing, and in this sense they represent unusually discrete areas of health care resources and populations. Furthermore, we found only weak or nonsignificant correlations between neonatologist supply and measures of NICR border crossing (Spearman rank-correlation coefficient for localization of death vs. number of neonatologists, 0.143, P = 0.02; localization vs. beds 0.045, P > 0.2; see Appendix for further details). This lack of correlation confirms that the observed variation in neonatologists cannot be explained by differences in NICR border crossing. The regions are also large enough that capacity measures are highly precise. Delays in updating the physicians' survey data to their current activities are one potential source of error, but this would be an implausible explanation for the magnitude of variation observed.

This study raises important questions of equity, effectiveness, and efficiency. If neonatal intensive care resources vary, then understanding their relationship to risk is critically relevant to policy. If resources are adjusted for risk, then do we know which adjusted rate is right? That is, which rate provides maximum patient benefits? Or, which achieves a reasonable level of cost-effectiveness? And if it is not the highest rate, then why should these services be supported when the unmet health needs of children are extensive and well documented? (19-21)

Variation in a highly specialized workforce also raises the issue of clinician performance. Low patient volume in high-supply regions may, over time, erode the skills that neonatologists need to care for newborns with complicated illnesses. Although studies have generally shown that low unit volume leads to higher mortality, (22, 23) this area of research is incomplete, and further research is critical as the number of units and neonatologist continues to grow.

Similar questions have been raised many times in response to other studies showing variation in resources. What makes neonatal intensive care unique is the detailed patient-level data collected on every birth through vital records and the frequency of an important and measurable outcome to judge effectiveness--neonatal mortality rates. The questions of regional equity, effectiveness, and efficiency may not be known for neonatal intensive care, but unlike many other types of medical services, they are knowable with current data and methods.

Take Home Points
  • In the past 30 years, the number of neonatologists has increased dramatically, whereas the number of births has remained constant.
  • To learn how this workforce is distributed, we looked at the geographic variation in the supply of neonatologists in 246 hospital market regions.
  • The regional supply of neonatologists varied widely (ranging from 1.2 to 25.6 neonatologists per 10,000 births) as did physician experience with low birthweight babies (ranging from 66 to 263 low birthweight babies per neonatologist).
  • The variation was not explained by the substitution of neonatal mid-level providers or by the presence of academic medical centers.
  • The benefit of a large neonatologist workforce is unknown.


1. Committee on Perinatal Health. (TIOP I) Toward Improving the Outcome of Pregnancy: Recommendations for the Regional Development of Maternal and Perinatal Health Services. White Plains, NY: March of Dimes Birth Defects Foundation; 1976.

2. Little G, Merenstein G. Toward improving the outcomes of pregnancy, 1993: Perinatal regionalization revisited. Pediatrics. 1993;92:611-2.

3. Williams R, Chen P. Identifying the sources of the recent decline in perinatal mortality rates in California. N Engl J Med. 1982;306:207-14.

4. Richardson D, Gray J, Gortmaker S, Goldmann D, Pursley D, McCormick M. Declining severity adjusted mortality: evidence of improving neonatal intensive care. Pediatrics. 1998;102:893-9.

5. Hack M, Klein N, Taylor H. Long-term developmental outcomes of low birth weight infants. The Future of Children. 1995;5:176-96.

6. Goodman D, Little G. General pediatrics, neonatology, and the law of diminishing returns. Pediatrics. 1998;102:396-9.

7. Pollack L, Ratner I, Lund G. United States neonatology practice survey: personnel, practice, hospital, and neonatal intensive care unit characteristics. Pediatrics. 1998;101:398-405.

8. Merenstein G, Rhodes P, Little G. Personnel in neonatal pediatrics: assessment of numbers and distribution. Pediatrics. 1985;76:454-6.

9. Schwartz R. Supply and demand for neonatal intensive care: trends and implications. J Perinatol. 1996;16:483-9.

10. Hall R. Neonatal manpower needs: the writing is on the wall--we should read it and heed it. J Perinatol. 1997;17:423-4.

11. Wennberg J, Gittelsohn A. Small area variations in health care delivery. Science. 1973;182:1102-8.

12. Wennberg J, Cooper M (ed). The Dartmouth Atlas of Health Care in the United States. Chicago, IL: American Hospital Association; 1996.

13. American Hospital Association. Annual Survey of Hospitals Data Base (machine-readable data file). Chicago: American Hospital Association; 1995.

14. National Center for Health Statistics. Linked Birth/Death Data Set. Atlanta: Centers for Disease Control; 1995.

15. Graduate Medical Education National Advisory Committee. Report of the Graduate Medical Education National Advisory Committee to the Secretary, Department of Health and Human Services, Volume 1. Washington, DC: Office of Graduate Medical Education, Health Resources Administration, Public Health Service, U.S. Dept of Health and Human Services; 1980.

16. Jacobsen S, Rimm A. The projected physician surplus reevaluated. Health Aff (Millwood). 1987;6:48-56.

17. The Annie E. Casey Foundation. Kids Count Data Book. Baltimore, MD: The Annie E. Casey Foundation; 1999.

18. Wennberg J, Cooper M (ed). The Dartmouth Atlas of Health Care--1998. 2nd ed. Chicago: American Hospital Association; 1997.

19. Starfield B. Childhood morbidity: comparisons, clusters, and trends. Pediatrics. 1991;88:519-26.

20. Newacheck P, Stoddard J, Hughes D, Pearl M. Health insurance and access to primary care for children. N Engl J Med. 1998;338:513-9.

21. Halfon N, Inkelas M, DuPlessis H, Newacheck PW. Challenges in securing access to care for children. Health Aff (Millwood). 1999;18:48-63.

22. Phibbs C, Bronstein J, Buxton E, Phibbs R. The effects of patient volume and level of care at the hospital of birth on neonatal mortality. JAMA. 1996;276:1054-9.

23. Horbar J, Badger G, Lewit E, Rogowski J, Shiono P. Hospital and patient characteristics associated with variation in 28-day mortality rates for very low birthweight infants. Pediatrics. 1997;99:149-56.

Grant Support

Supported by a grant from the Robert Wood Johnson Foundation.


David C. Goodman, MD, MS, 7251 Strasenburgh Hall, Dartmouth Medical School, Hanover, NH, 03755; telephone: 603-650-1982; fax: 603-650-1225; e-mail:


The accuracy of the definition of health service area for measuring medical resources depends, in part, on the extent to which the resources "match" the resident population. In other words, in areas with minimal border crossing of patients to nonlocal providers, the per capita resource rates will reflect accurate numerators (the resources) and denominators (the populations). Health service areas with populations that extensively seek care from providers in other areas lead to spurious per capita rates without adjustment for this border crossing.

We developed three indices of border crossing (localization indices A, B, and C) that examine how maternal and newborn populations "localize" to their own area's intensive care providers (Appendix Figure 1). Although vital records do not have an explicit measure of place of care, three fields can be used to infer travel for high-risk obstetric and neonatal services for low birthweight infants: place of maternal residence, place of birth, and place of neonatal death.

Localization index A was the proportion of very low birthweight infants born to mothers residing in a given neonatal intensive care region (NICR) and whose births occurred within the same NICR. Mothers in premature labor are transported whenever possible to a hospital with perinatal services and a neonatal intensive care unit. Similarly, localization index B was the proportion of very low birthweight neonatal deaths occurring within a given NICR in infants who were born within the same NICR. Finally, localization index C was the proportion of very low birthweight neonatal deaths of infants occurring in an NICR in which the mother resides. Since most neonatal deaths occur within a neonatal intensive care unit, high death localization indicates that care is generally occurring within the "home" NICR regardless of whether home is defined as the NICR of birth or maternal residence. Localization indices for all 246 NICRs are available from the authors.

As described in the Methods, we initially assigned newborn service areas (NSAs) to NICRs on the basis of plurality of travel from NSA of maternal residence to NSA of birth for low birthweight infants. To achieve a high degree of localization of newborns to the NICR of their medical care, we reassigned NICRs with either a localization index A of less than 0.5 for all births or a localization index B or C of less than 0.5 for very low birthweight infants.

The Appendix Table shows the characteristics of the NICRs. The localization indices show that NICR populations cross borders less often than those of other accepted health service areas (18) and are particularly suited for unbiased measures of perinatal resources.