Objective. Despite intense interest in allocation of resources to neonatal intensive care, no description exists of resource use by the large numbers of newborns admitted for triage, the process of short-term evaluation and management of infants after delivery. This study characterized the triage phase of neonatal care with respect to infant demographics, risk factors for illness, and the course of the hospital admission. We hypothesized that triage infants were responsible for a significant fraction of total intensive care resource utilization, and that patterns of use were predictable.
Design. Cross-sectional cost analysis of prospectively collected data.
Participants. Data were collected prospectively on 2486 inborn infants admitted to two neonatal intensive care units (NICUs) for <24 hours and subsequently discharged to routine care. Over the 11-month study period, these two hospitals delivered 15 097 live births and admitted a further 1837 infants for nontriage NICU care.
Interventions. On a 50% random subsample, we calculated severity of illness using the Score for Neonatal Acute Physiology (SNAP) and applied a NICU resource checklist. Daily NICU workload was estimated according to the number and labor intensity of NICU admissions using Medicus and SNAP. Charges were obtained from patient-level item charge records and converted to costs using Medicare ratios of costs to charges. Length of stay (LOS) and costs for triage were correlated with diagnoses, perinatal descriptors, severity of illness, and markers of concurrent NICU workload using stepwise regression.
Results. Mean birth weight for triage infants was 3367 g (standard deviation, 600 g) and mean gestational age 39.1 weeks (standard deviation, 1.8 weeks). The predominant reasons for evaluation were exclusion of sepsis (34%), birth complications including meconium aspiration, perinatal depression and trauma (24%), and transitional respiratory distress (23%). Severity illness, as measured by SNAP, was minimal, with 70% having scores of 0, indicating no derangement. Only 6% experienced depressed 5-minute Apgar scores (<7), and 80% required no delivery room resuscitation. The most frequent forms of resource use were antibiotic administration (34%), placement of a peripheral intravenous line (40%), cardiac monitoring (53%) and external warming (26%). Median LOS was 102 minutes, corresponding over the study period to 2% of total NICU hours but 7% of NICU days charged. Median cost was $870, with aggregate costs accounting for a total of 9.5% of total NICU costs. In the multivariate model, LOS was increased by respiratory diagnosis or hypoglycemia, severity of illness, lower gestational age, the need for intravenous placement, daytime shift, hospital, and lower acuity of concurrent NICU admissions (R2 = 0.24).
Conclusions. Neonatal triage is a low-acuity but time-intensive process that contributes significantly to total resource use by newborns because of the large numbers of infants involved. Both LOS and costs are affected not only by infant medical characteristics but also by nonmedical markers of unit structure, which may be amenable to change. This source of resource consumption should be recognized in future assessments of costs associated with neonatal intensive care.
Cost studies to date in the field of newborn care have concentrated almost exclusively on global assessments of neonatal intensive care,1–3 particularly that of very low birth weight prematures,4 or have targeted new technological advances such as surfactant5–9 or erythropoietin.10,,11 However, the American Academy of Pediatrics in its Guidelines for Perinatal Care12 recognizes several components of newborn care. The highest profile is the acute subspecialty care of the sickest infants, involving the intensive application of personnel and technology to a relatively limited patient population. Another aspect is that of triage, the evaluation and short-term management of infants after delivery. These infants have defined perinatal risks, transitional conditions, and mild illness, such as jaundice and exclusion of sepsis. The goal is rapid diagnosis and assignment of the infant to an appropriate level of neonatal care (regular nursery, intermediate care, or intensive care) contingent on clinical findings.
In contrast to intensive care services, triage involves high patient volumes and low acuity of illness. Furthermore, little of triage care has a strong base of evidence, potentially leaving more discretion—and thus variability—in diagnosis and management. In these respects, neonatal triage is similar to the evaluation process in emergency medicine, where the walking wounded and worried well are directed to appropriate levels of care. In the adult emergency room setting, several studies have examined the resource implications and reliability of the process.13,,14,15 In contrast, this function has been ignored in the pediatric literature.
As the initial step toward more evidence-based management and more discriminating resource use, this study characterized the triage phase of neonatal care with respect to infant demographics, risk factors for illness, and the course of the hospital admission. We hypothesized that triage infants were responsible for a significant fraction of total intensive care resource utilization, and that patterns of use were predictable.
Study Setting and Subjects
The study involved two neonatal intensive care units (NICUs) located in urban academic hospitals. These units represented a spectrum of patient populations, with Hospital A having level 3 (intensive or subspecialty care) capabilities and 46 NICU and intermediate beds, and Hospital B providing level 2 (intermediate or specialty) care with 12 intermediate beds. At the time of the study, site A was affiliated with high-risk obstetric services while site B transferred mothers with impending high-risk deliveries to site A. The combined deliveries totaled 15 097 live births between November 1, 1989 and October 1, 1990. Both sites were staffed with a neonatal fellow 24 hours per day, and had daytime coverage by attending level neonatologists. Site A also had pediatric residents, who provided most of the triage care.
In both institutions, all infants who had abnormal symptoms requiring blood work or monitoring for risk factors were transferred from the regular nursery or labor and delivery ward to the NICU. In most cases, such transfer occurred after telephone consultation with a physician. However, guidelines existed for automatic transfer for certain conditions, such as maternal insulin-dependent diabetes mellitus.
As it was standard practice at these institutions to evaluate all sick infants within the NICU, triage patients in this study were defined retrospectively as those inborn infants admitted to a NICU for <24 hours before being discharged to regular nursery care. Over the 11-month study period, 2486 infants met these criteria, with an additional 1837 infants being admitted for more prolonged neonatal care. Thus, triage infants accounted for 16% of all births and 57% of all contacts with the neonatology service at the two hospitals. All infants meeting study criteria were included.
DATA COLLECTION AND ANALYSIS
The study made use of data collected prospectively as part of the validation project for the Score for Neonatal Acute Physiology (SNAP).16 These data had been obtained with an a priori intention of examining the triage process and neonatal resource utilization. In addition to demographic, perinatal, and length of stay (LOS) descriptors, admission illness severity was calculated using the SNAP. Resource utilization was tallied using the checklist of treatments included in the Neonatal Therapeutic Intensity Scoring System (NTISS).17 Both scores were collected on a 50% random sample of triage infants and on all NICU admissions. As markers of concurrent unit acuity, the number, illness severity, and estimation of nursing workload (Medicus) were calculated for all other admissions to the unit on the day of triage for a given infant. Diagnoses or conditions were taken from the International Classification of Diseases, 9th Revision coding at discharge, or inferred from information on specific therapies prescribed or resources utilized.
Billings for diagnostic and therapeutic services were obtained from individual item charge records and converted to costs using cost center-specific Medicare ratios of costs to charges. Per diem rates were not converted. Professional and personnel costs were considered to be included in these rates and were not costed separately. All costs are expressed in 1990 US dollars.
All statistical analyses were performed using the Statistical Analysis System (SAS Institute, Cary, NC). Models of resource utilization, as measured by LOS and costs, were generated using multiple linear regression, with a stepwise model selection procedure.
The institutional review boards at both hospitals provided approval for review of medical records.
Demographic Descriptors of Triage Patients
Demographic characteristics closely reflected those of healthy liveborn infants at the two hospitals. Fifty-three percent of infants were male and 52% were white. The mean gestational age was 39.2 weeks (standard deviation, 1.8 weeks) and mean birth weight was 3367 g (standard deviation, 600 g). Low birth weight infants (<2500 g) comprised 9% of the sample, and there were no infants weighing <1500 g. Operative deliveries constituted 33% of triage births. Seventy-two percent of infants were admitted directly from the labor and delivery service, with the remainder being referred from the postpartum floors.
Severity of Illness
Illness severity was minimal in the large majority of patients. Only 6% experienced depressed 5-minute Apgar scores (<7), and 80% required no resuscitation. Of those remaining infants receiving delivery room interventions, 14% required only supplemental oxygen, bag-mask ventilation was administered to 5%, and only 1% required intubation for respiratory symptoms; none received cardiac compressions or resuscitative medications. The SNAP indicated no physiologic derangement in 70% (SNAP = 0), while 94% had scores <3, indicating very mild derangement. No infant had a score >8, which is considered the threshold between mild and moderate illness severity.
The predominant reasons for triage were exclusion of sepsis (34%), birth complications including meconium aspiration, perinatal depression and trauma (24%), and transitional respiratory distress (23%). Less frequently, infants were admitted for hyperbilirubinemia (13%), prematurity and postmaturity (9%), suspicion of congenital anomalies (7%) and small for gestational age status (5%). Exclusion of cardiac disease or hypoglycemia were each rationales in 3% of cases. These categories were not mutually exclusive.
Half of triage infants underwent cardiorespiratory monitoring, while 30% also required oximetry. Only 4% received supplemental oxygen. Intravenous catheter placement corresponded closely to the frequency of antibiotic prescription, in 40% and 34% respectively. Incubator or radiant warming was provided in 26%.
LOS, as depicted in Fig 1, showed a skewed distribution with a median value of 102 minutes. This corresponded to 2% of total neonatal intensive care admission hours accrued at the participating hospitals, or up to 7% of total NICU days if a correction factor of ½-day minimum was applied to take into account the higher workload involved in new admissions. Given the nonnormal distribution, further analyses of LOS were performed after natural log transformation of the variable.
Costs were distributed as shown in Fig 2, with a median value of $870. Costs for triage infants accounted for a total of 9.5% of total NICU costs.
Multiple linear regression models for cost and for log-transformed LOS are reported in Tables 1 and2, respectively. The models show highly significant correlations of both LOS and costs with the predictor variables (P = .0001 for both models) and explain 24% and 31% of the total variance, respectively. The results for costs were robust to log-transformation of that variable.
Variables entering into the equations were of two classes. Medical and resource descriptors, such as the presence of respiratory distress, hypoglycemia, severity of illness, and the need for intravenous placement all tended to increase the summary resource use measures. Similarly, system descriptors, including higher concurrent unit acuity, night shift, and hospital, were associated with decreases in costs or LOS.
The models for cost and LOS contain similar predictors, with minor variations consistent with the difference in dependent variables. For example, the exclusion of sepsis is likely to be a shorter process than monitoring for resolution of hypoglycemia, but might be expected to incur higher costs given the frequent use of antibiotic and intravenous therapies.
Although physician characteristics have been correlated with individual admission behaviors,18 our study is believed to be the first in the neonatal literature to describe the important subset of infants who undergo triage evaluation. As hypothesized, this process is characterized by high volume, comprising 16% of all births and 57% of all infant contacts with newborn specialists in the institutions studied. Triage is also characterized by low illness acuity, with 70% of infants showing no physiologic instability as measured by an objective scoring system, indicating clinicians' response to risk rather than illness. Despite the mild severity of illness and the brief duration of stay, these contacts were resource-intensive, and accounted for 9.5% of total neonatal intensive care costs.
Examination of the assigned and inferred diagnoses sheds some light on the rationale for evaluating such a large population of asymptomatic infants. Fully 40% of infant contacts were dictated by maternal and perinatal risk factors for neonatal sepsis or hypoglycemia.
Regression analysis confirms that, while admission for symptoms had the greatest impact on LOS and costs, an important component of the explained variance was associated with evaluation for sepsis. The sepsis work-up is frequently a source of frustration for practicing pediatricians and anxiety for new parents. Clinical protocols designed to identify perinatal risk and institute early antibiotics19 have never been prospectively tested, nor evaluated for cost-effectiveness. More recently, similar guidelines from the Centers for Disease Control and Prevention20 and the American Academy of Pediatrics21 have provided a rationale for identification and treatment of infants at risk of early onset group B streptococcal sepsis. However, cautious interpretations of risk factors may lead to low specificity and high work-up rates in clinical practice. Furthermore, recent evidence suggests that epidural use in labor may be commonly inducing maternal hyperthermia,22 provoking pediatric staff to respond with sepsis evaluations for maternal fever.
Systemic factors such as hospital of birth or markers of unit acuity and workload also have effects on resource use. Hospital A appeared to process cases 15% faster and infants were ushered through more quickly on days of high NICU acuity. Moreover, similar infants evaluated on the night shift had an average of $118 less in ancillary charges.
Given the subtlety involved in such high-volume, risk-based decisions, it is ironic that most institutions make use of junior house officers to staff the triage service. Resource use by residents may be less discerning than by more experienced clinicians, through increased ordering of tests and longer LOS,23 and some of the variability seen between hospitals and shifts might be attributable to personnel.
These findings have important policy implications. The cost of predischarge newborn care in the United States has been estimated at over $4 billion, as measured in 1989 currency.24Although the cost of neonatal intensive care has been more difficult to characterize, application of average costs reported by the Office of Technology Assessment4 to the estimated 200 000 annual NICU admissions yields a minimum estimate of almost $800 million in 1984 currency. These substantial tallies are compounded by the high proportion of this population whose care is uncompensated or only partially compensated.24 Moreover, increased penetration by managed care organizations and ongoing medical cost inflation have placed severe stress on medical care budgets at all levels.
Given evidence that up to 10% of these costs are attributable to a high-volume, low-acuity process driven by risk factors, tests and short-term observation, and that these costs are significantly affected by hospital infrastructure, it will be important to avoid focusing cost-effectiveness analysis solely on new technological advances. Indeed, such costs might be more easily reduced without affecting outcomes than those associated with proven life-saving therapies during acute intensive care. This is the economic essence of discretionary care.
Variability between institutions might in some circumstances also lead to potential abuse of payment systems, as in the case of hospitals that formally admit these infants as intensive care patients to bill a higher level of care. Given the large numbers of infants involved, such up-coding might be a significant source of inefficiency in resource use.
Before applying these findings however, several limitations of the study methodology must be recognized. First, our definition of triage was retrospective, including only those infants well enough to be released within 24 hours. This underestimates the difficulty of prospectively selecting those safe enough to leave, but should not underestimate the prevalence of triaged infants.
Another salient limitation is that this was a cost analysis only, performed to highlight a previously unrecognized population consuming neonatal care resources. Any adjustments to delivery of medical care must be based on high-quality evidence of effectiveness. Specifically, higher admission rates for monitoring may in fact be morecost-effective if they avert more complicated illnesses associated with higher costs. In most cases such evidence is lacking for the interventions associated with triage, and policy must therefore be made cautiously.
Generalizability of the findings is also affected by the study population. Although we were able to risk-adjust for infant severity of illness, such risk adjustment was not available for most maternal or perinatal characteristics. Thus, it is possible, for example, that the high resource utilization for sepsis evaluation highlighted earlier is a reasonable response to more prevalent risks. Similarly, the larger of the institutions studied had a very active high-risk maternal service, received frequent maternal transports and ran a regional diabetic care program, which may have resulted in the need for more frequent monitoring. Furthermore, the process of transferring all triage infants to a geographically separate area is not universal; it is not clear whether this would decrease costs (in the case of increasing economies of scale, or more efficient processing by a dedicated team) or increase them (through increased testing, triage staff with idle time, etc).
Several factors affecting the cost estimates must also be acknowledged. We had to approximate costs using charges; the necessary conversions reflect to some extent the idiosyncrasies of a particular institution's cost accounting system.25 Furthermore, the true labor input of professionals and other hospital personnel may not be accurately captured in these charges. Finally, only direct medical costs were assessed; other costs important to a societal perspective, such as time lost from work, were not available. These omissions are likely to result in an underestimation of the cost assessments.
In summary, triage constitutes a high-volume, low acuity-process that consumes a significant proportion of the resources devoted to newborn care. This study has documented the discretionary nature of triage care and the impact of nonmedical factors on resource use. In light of these characteristics, research into the effectiveness and costs of alternative strategies of delivery of care to this group have great potential for improving both care and efficiency and should be given higher priority.
This study was funded in part by Agency for Health Care Policy Research Grant R01 HS06123.
We thank Drs Eric Eichenwald and Gabriel Escobar for their thoughtful review of the manuscript, and members of the Departments of Neonatology at the Brigham and Women's, Beth Israel Deaconness, and Children's Hospitals for valuable discussions during the analysis phase.
- Received April 3, 1998.
- Accepted July 29, 1998.
- Address correspondence and reprint requests to Douglas K. Richardson, MD, MBA, Department of Newborn Medicine, Beth-Israel Deaconness Medical Center, 330 Brookline Ave, Boston, MA 02115.
This work was presented in part at the Annual Meeting of the Pediatric Academic Societies, Washington, DC, 1997.
- NICU =
- neonatal intensive care unit •
- SNAP =
- Score for Neonatal Acute Physiology •
- LOS =
- length of stay •
- NTISS =
- Neonatal Therapeutic Intensity Scoring System
- Rogowski J
- Stevenson RC,
- Pharoah PO,
- Stevenson CJ,
- et al.
- ↵US Congress, Office of Technology Assessment. Neonatal Intensive Care for Low Birth Weight Infants: Costs and Effectiveness (Health Technology Case Study 38), OTA-HCS-38. Washington, DC: US Congress, Office of Technology Assessment; 1987
- Tubman TR,
- Halliday HL,
- Normand C
- Mansicalco WM,
- Kendig JW,
- Shapiro DL
- ↵American Academy of Pediatrics. Guidelines for Perinatal Care. 4th ed. Elk Grove Village, IL: American Academy of Pediatrics; 1997
- Weingarten S,
- Ermann B,
- Bolus R,
- et al.
- Richardson DK,
- Gray JE,
- McCormick MC,
- et al.
- Gray JE,
- Richardson DK,
- McCormick MC et al
- ↵Cloherty JP, Stark AR. Manual of Neonatal Care. Boston, MA: Little, Brown & Co; 1980
- ↵Centers for Disease Control and Prevention. Prevention of perinatal group B streptococcal disease: a public health perspective. MMWR. 1996;45(No. RR-7):1–24
- Halsey NA,
- Schuchat A,
- Oh W,
- Baker CJ
- Lieberman E,
- Lang JM,
- Frigoletto F Jr.,
- et al.
- Copyright © 1998 American Academy of Pediatrics