Objective. Yale-New Haven Hospital (Y-NHH) has maintained the longest running, single-center longitudinal database of neonatal sepsis, started in 1928. The objective of this study was to update this database with review of neonatal sepsis cases at Y-NHH to identify longitudinal trends in demographics, pathogens, and outcome.
Methods. Records of infants with positive blood cultures obtained while they were inpatients in the NICU at Y-NHH from 1989 to 2003 were reviewed retrospectively. Records of infants who were ≤30 days of age, had positive blood cultures, and were hospitalized at Y-NHH outside the NICU from the same period were also reviewed, and all findings were compared with 60 years of preexisting data.
Results. A total of 862 organisms were identified in 755 episodes of sepsis from 647 infants. The percentage of cases of early-onset sepsis decreased and late-onset sepsis increased compared with the previous 10-year study period. A marked increase in cases as a result of commensal species was observed, particularly in preterm infants who had indwelling central vascular catheters, were receiving parenteral nutrition, and required prolonged mechanical ventilation. The overall percentage of sepsis caused by group B streptococcus and Escherichia coli decreased. No episodes of sepsis from Streptococcus pneumoniae or S pyogenes, common in the early years of the survey, were observed. The sepsis-related mortality rate steadily decreased, from 87% in 1928 to 3% in 2003.
Conclusions. The demographics, pathogens, and outcome associated with neonatal sepsis continue to change. The increase in late-onset sepsis in preterm infants who required prolonged intensive care indicates that strategies to prevent infection are urgently needed for this population of infants.
Sepsis is a significant cause of morbidity and mortality in the newborn, particularly in preterm, low birth weight infants.1,2 Despite advances in neonatal care, overall case-fatality rates from sepsis range from 2% to as high as 50%.3 Analysis of longitudinal trends assist in the formulation of strategies to treat and prevent neonatal sepsis,4–7 with the ultimate goal of decreasing morbidity and mortality. Yale-New Haven Hospital (Y-NHH) has maintained the longest running, single-center database of neonatal sepsis, started in 1928.8–12
The last published survey in this series covered the period from January 1, 1979, to December 31, 1988.12 During that period, an increase in sepsis caused by commensal species, organisms that are present as normal skin and intestinal flora such as coagulase-negative Staphylococcus (CONS), Staphylococcus aureus, viridans streptococci, and species of Candida, was observed. In addition, an increase in the rate of sepsis in the NICU at Y-NHH in the population of infants who were older than 30 days was also observed. We speculated that the rise of a more premature patient population with longer duration of NICU hospitalization and an increase in the placement of central vascular catheters might be responsible for the increase in infections from commensal organisms.12 This report details the trends in neonatal sepsis at Y-NHH in the 15-year period from 1989 to 2003 and compares recent trends with those observed over the previous 60 years in an effort to identify potential changes in the cause, presentation, associated risk factors, and outcomes associated with neonatal sepsis.
The medical records of all infants and newborns with positive blood cultures obtained at any time while infants were inpatients in the NICU at Y-NHH from January 1, 1989, to December 31, 2003, as identified by microbiology laboratory culture logs, were reviewed. Positive blood cultures that were from the same period and obtained from any non-NICU patient up to and including 30 days of age hospitalized or seen in the pediatric emergency department, pediatric ward, PICU, or ambulatory clinic at Y-NHH were also reviewed. In accordance with criteria established in the last study period, this approach sought to identify all positive neonatal blood cultures (eg, those obtained in infants ≤30 days of life) in addition to all NICU cultures in an effort to identify trends not only within the patient population but within our ICU as well.12 This study was approved by the Human Investigation Committee of the Yale University School of Medicine.
Using criteria similar to those described in the last publication,12 we divided microbial organisms into 2 categories: traditional neonatal pathogens and commensal species.
Traditional Neonatal Pathogens
These include species of both bacteria and fungi that are known to be common causes of neonatal sepsis. In particular, group B and D β-hemolytic streptococci, Listeriamonocytogenes, yeast, Escherichia coli, and all other aerobic and anaerobic Gram-negative rod species compose this group.
These include species of bacteria and fungi that are present as normal skin and intestinal flora in the newborn patient population. They include Staphylococcus species, viridans Streptococci, Gram-positive rods other than Listeria, Gram-negative cocci other than Neisseriameningitidis and N gonorrhoeae, Gram-positive anaerobes, and species of Candida. Blood cultures that yielded organisms that were believed to be contaminants, including Corynebacterium and nonspeciated Gram-positive rods, were eliminated from review.
Sepsis was defined as a blood culture that yielded a traditional neonatal pathogen or a commensal species. Cultures that yielded CONS, which composed the majority of commensal species–positive cultures, were reviewed using modified specific criteria of the Centers for Disease Control and Prevention13 (Table 1). Multiple positive blood cultures that were from a single patient and yielded the same species with identical antibiotic susceptibility patterns were considered as a single episode. Cases of sepsis were classified according to the infant's age when the positive blood culture was obtained as early onset (≤4 days of life), late onset (5–30 days), and late, late onset (>30 days).14 Patients were classified as inborn or transported. The term “transported” was used to include those who were admitted to the Y-NHH NICU from outside facilities, as well as non-NICU patients from other locations at Y-NHH.
Risk Factors, Signs, and Symptoms
Preterm labor was defined as onset of labor before 37 weeks' gestation. Maternal fever was defined as core temperature of >38°C before delivery. Prolonged rupture of membranes was defined as rupture of membranes ≥18 hours before delivery.
The presence of a central vascular catheter was included only when placed before the onset of sepsis and in place at the time of the positive blood culture. Vascular catheters included umbilical venous and arterial catheters, percutaneous central venous catheters, Arrow, and Broviac Silastic catheters. Surgery as a potential risk factor for sepsis was included only when the procedure occurred ≤7 days before the onset of the positive blood culture.
Apnea and bradycardia were included as clinical signs of sepsis only when the episodes were new in onset or more frequent, prolonged, or severe than previously observed. Hypothermia was defined as a core body temperature of <36.5°C. Fever was defined as a core body temperature of >38°C. Hyperglycemia was defined as blood glucose of >140 mg/dL that did not coincide with an increase in the glucose infusion rate or volume of enteral intake. Hypoglycemia was defined as blood glucose of <40 mg/dL that did not coincide with a reduction in the glucose infusion rate or volume of enteral intake. The absolute neutrophil count was obtained at the same time as the positive blood culture and was defined as the sum of the segmented and band forms (or the automated neutrophil count) divided by 100 and then multiplied by the total white blood cell count.
Death was considered related to infection when it occurred within 7 days of the positive blood culture or when clinical signs and symptoms of sepsis were documented as the direct cause of death. Mortality was calculated with the numerator representing the number of episodes of sepsis with death and the denominator as the total number of episodes of sepsis. In the case in which an infant had multiple episodes of sepsis, only the last 1 was included in the calculation of mortality and previous episodes were recorded as sepsis with survival.
Blood cultures were assessed during this period using a fluorescent detection system for the presence of CO2 (Bactec 960 or 660; Becton Dickinson, Piscataway, NJ), a method that has been used at our institution for >20 years.
The SPSS 12.0 statistical software package (SPSS Inc, Chicago, IL) was used for data analyses. The Spearman rank correlation (rs) test was performed using the 2-tailed t test to reject the null hypothesis that rs = 0. This nonparametric test was chosen over the parametric Pearson correlation because of its robustness in analyzing fewer data points. A multivariate logistic-regression model was created to analyze the effects of several covariates on the binary outcome “sepsis secondary to commensal species.” P < .05 was considered statistically significant.
Major Demographics of the Y-NHH NICU, 1989–2003
Dividing the 15-year study period into three 5-year blocks, the mean numbers of live births were 5063 (1989–1993), 4773 (1994–1998), and 4728 (1999–2003). The mean numbers of long-term (>12 hours) admissions to the NICU during the study period were 777 (1989–1993), 824 (1994–1998), and 850 (1999–2003). The inborn admission rate of extremely low birth weight (ELBW) infants (birth weight <1000 g) was 14.2 per 1000 live births per year from 1989 to 1993, 15.1 per 1000 live births per year from 1994 to 1998, and 14.4 per 1000 live births per year from 1999 to 2003. The admission rate for newborns with birth weight <500 g was 2.9 per 1000 live births per year from 1989 to 1993, 1.9 per 1000 live births per year from 1994 to 1998, and 1.9 per 1000 live births per year from 1999 to 2003. Thus, during the study period, there was little variation in the numbers of live births or ELBW infants who were admitted to the NICU and a slight increase in total number of NICU admissions per year.
A total of 1809 blood cultures from 939 patients were identified. The medical records of all 939 patients were reviewed, and 862 organisms were identified in 755 episodes of sepsis (there were 82 episodes of polymicrobial sepsis that contained 2 or more organisms) from 647 infants that met our definition for sepsis. Of the 755 episodes of sepsis, 27 episodes from 27 neonates who were admitted to the PICU or pediatric wards or seen as ambulatory patients were combined with 208 episodes from 173 neonates who were transported to our NICU for data analysis (n = 235 episodes).
The entire population studied had a male:female ratio of 1.2:1, a mean gestational age of 31 ± 12 weeks, and mean birth weight of 1760 ± 2315 g. The median time to onset of sepsis was 19 days (range: 1–157 days). The overall mortality in the population being studied was 22%, and overall mortality as a result of sepsis was ∼12%.
Of the 755 episodes of sepsis, 108 (14%) were in inborn neonates who were ≤4 days of age, 276 (37%) were obtained from inborn neonates who were 5 to 30 days of age, and 136 (18%) were from inborn neonates who were >30 days of age. The inborn sepsis rate for the 15-year period was 7.1 cases per 1000 live births.
Table 2 shows the number of isolates of each species for both inborn and transported cases. The most common organism identified was CONS (29%) followed by E coli (12%), group B streptococcus (GBS; 10%), and S aureus (8%). Selected clinical and demographic data are presented in Table 3.
The effect of birth weight on the incidence of sepsis was analyzed in neonates who were born at Y-NHH (Table 4). A total of 141 cases of bacteremia (348 cases per 1000 live births) occurred in the population with birth weight of <750 g, with the highest proportion of infections occurring at 5 to 30 days of life. Infection rate decreased with increasing birth weight. In the population of neonates with birth weight ≥2000 g, the sepsis rate was only 2 per 1000 live births, with the largest percentage of cases occurring at 0 through 4 days of life.
Presence of apnea and bradycardia and alterations in core temperature and blood glucose levels at the time of positive blood culture were assessed (Table 5) . In early-onset sepsis (EOS) group, hypoglycemia and hypothermia were the most common findings. In late-onset sepsis (LOS) and late, late-onset sepsis, hypothermia and hyperglycemia, and hypothermia and apnea were the most common findings, respectively.
This group included 105 inborn neonates with mean gestational age of 34 ± 6 weeks and birth weight of 2388 ± 1257 g. The median onset of infection was 1 day (range: 1–4 days). GBS was the predominant organism cultured (47%), followed by E coli (23%), Staphylococcus species (13%), and aerobic Gram-negative rods other than E coli (8%). The maternal antenatal histories of these patients were reviewed in an attempt to identify possible risk factors for EOS. Forty-three percent of mothers had preterm labor, 26% had fever before delivery, 46% had prolonged rupture of membranes, and 20% had documented or suspected chorioamnionitis. Forty-three percent received systemic antibiotics before delivery, and 26% received antenatal steroids.
The majority of infants in this EOS group were >30 weeks' gestation (68%) and >1500 g birth weight (65%). Ten percent had a central vascular catheter at the time of infection, and only 1% had had surgery within 1 week of the onset of symptoms. A commensal species was identified in 14% of cases. The overall mortality was ∼18%, with an overall mortality as a result of sepsis of 11%.
This group included 239 inborn neonates with onset of infection between 5 and 30 days of life. The mean gestational age and birth weight were 30 ± 5 weeks and 1475 ± 1043 g, respectively. The majority of these infants had a gestational age <30 weeks (62%) and a birth weight < 1500 g (68%). Median onset of infection was 17 days (range: 5–30). The predominant organism cultured was CONS (39%), followed by E coli (9%) and Candida albicans (8%). Aerobic Gram-negative rods other than E coli accounted for 20% of organisms cultured. Fifty-seven percent of the organisms cultured were commensal species, similar to that reported by the National Institute of Child Health and Human Development Neonatal Research Network.15
More patients in the LOS group than in the EOS group had an indwelling central vascular catheter at the time of infection (78% vs 10%; P < .0001) and had had a surgical procedure before infection (8% vs 1%; P < .0001). In addition, a larger percentage required mechanical ventilation (49% vs 34%; P = .029). The overall mortality rate for this group was 21%, with a mortality sepsis rate of ∼11%.
Late, Late-Onset Group
This group included 103 inborn neonates with mean gestational age and birth weight of 29 ± 5 weeks and 1243 ± 860 g, respectively. The majority of neonates were <30 weeks' gestational age (67%) with birth weight <1500 g (70%), mirroring the well-known high risk for sepsis in preterm infants.16 Onset of infection occurred at a median of 42 days of age (range: 31–152). Like the LOS group, the predominant organism cultured was CONS (25%), followed by S aureus (12%) and Enterococcus faecalis (11%). Gram-negative aerobic rods other than E coli accounted for 25% of organisms cultured, whereas E coli accounted for 7% of organisms cultured. Forty-seven percent of organisms cultured were commensal species.
A very high percentage of infants had an indwelling central vascular catheter at the time of infection (79%). Seven percent had surgery within 7 days of onset of infection, and 53% were receiving mechanical ventilation at the time the positive blood culture was obtained. The overall mortality rate for this group was 29%, with an overall mortality rate as a result of sepsis of 15%.
This group included 200 infants with mean gestational age and birth weight of 33 ± 6 weeks and 2037 ± 1148 g, respectively. The majority were >30 weeks' gestational age (61%) and 1500 g birth weight (77%). Typically, patients are transported to the NICU at Y-NHH for management of surgical, cardiac, and genetic conditions and for management of extreme prematurity. Onset of infection occurred at a median of 22 days of age (range: 1–157). CONS (29%), E coli (15%), S aureus, and E faecalis (9% each) were the predominant organisms cultured. Gram-negative rods excluding E coli were responsible for 21% of infections in this group.
The transported group had a high percentage of central vascular catheters (67%), infection with a commensal species (45%), and the highest percentage with surgery before infection (12%). The overall mortality rate for this group was 23%, with an overall mortality as a result of sepsis of 12%.
The sepsis-related mortality rate has steadily decreased over the past 75 years, from 87% in 1928 to 3% in 2003. The overall mortality rate as a result of sepsis over the 15-year study period was 12%. In newborns with birth weight <1000 and <500 g, mortality rates as a result of sepsis were 17% and 22%, respectively. In the entire population studied, those who were infected with organisms that were identified as commensal species had a mortality rate as a result of sepsis of 9%, whereas those with organisms that were identified as traditional neonatal pathogens had a mortality rate as a result of sepsis of 12%. More specific, the mortality rate from GBS infection was 5%, from S aureus was 13%, from CONS was 5%, from E coli was 13%, and from all other Gram-negative aerobic rods was 20%. In the category of aerobic Gram-negative rods, infection with Pseudomonas aeruginosa had the highest associated mortality at ∼46%, in agreement with previous reports.1,17
Figure 1A shows the overall trend in inborn cases of EOS per 1000 live births per year for all organisms and for GBS, E coli, and commensal species. Statistical analysis revealed a significant increase in the number of early-onset cases of infection with a commensal species from 1979 to 2003 (rs = 0.43, P = .03). During the recent 15-year period, however, this trend is not observed (rs = -0.06, P = .85). In addition, there is a trend toward a decrease in the number of cases of GBS during the past 25 years (rs = -0.37, P = .07) but not during the past 15 (rs = -0.07, P = .81). Also of note is that the overall number of cases of EOS has not undergone a significant change in our nursery during the past 25 (rs = -0.32, P = .13) or 15 years (rs = -0.25, P = .38).
Figure 1B shows the overall trends in inborn cases of LOS per 1000 live births per year for all organisms and for commensal species. The total number of cases of LOS has increased significantly on a yearly basis from 1979 to 2003 (rs = 0.86, P < .01) as has the number of commensal species–related cases of LOS (rs = 0.84, P < .01). During the current 15-year study period, the overall number of LOS cases has increased significantly (rs = 0.60, P = .02), whereas those caused by commensal species have not (rs = 0.35, P = .21).
Figure 1C shows the overall trends in inborn cases of late, late-onset sepsis per 1000 live births per year for all organisms and for commensal species. There is a statistically significant increase in both the total number of cases of sepsis (rs = 0.65, P < .01) and the number of cases caused by commensal species (rs = 0.69, P < .01) from 1979 to 2003. During the current 15-year study period, there is a statistically significant increase by year only in the number of total cases of late, late-onset sepsis (rs = 0.75, P < .01).
The inborn mortality rate associated with sepsis at Y-NHH since the onset of the collection and analysis of newborn blood cultures in 1928 is shown in Fig 2. This graph uses the midpoints of each 5-year period or the midpoint of each study in the series to identify changes in the percentage of mortality attributable to sepsis during a 75-year period.12 A statistically significant decline in mortality rate from sepsis during the 75-year period from 87% in 1928 to 3% in 2003 (rs = -.99, P < .01) was observed. Since the last published review in 1988, the mortality rate caused by sepsis has declined from ∼15% to 3%.
The overall change in percentages of organisms identified from 1928 to 2003 is shown in Table 6. A significant rise in the percentage of commensal species cultured is observed as is a decline in the percentages of GBS and E coli cultured.
The predominance of commensal species cannot be related directly to a rise in the number of admissions or the number of patients who had a birth weight <1000 g and were admitted to the Y-NHH NICU. Therefore, a multivariate logistic-regression model (n = 755) was created to identify potential risk factors for hospital-acquired infection in the NICU population at Y-NHH. The model included the covariates birth weight, gestational age, gender, presence of patent ductus arteriosus, diagnosis of intraventricular hemorrhage, presence and duration of a central vascular catheter, presence and duration of mechanical ventilation, use of total parenteral nutrition, intravenous intralipids and H2 blockers at the time of infection, surgery within 7 days of infection, and the total number of days of antibiotic therapy before the most recent episode of infection. This analysis identified the presence of a central vascular catheter (P < .01), prolonged mechanical ventilation (defined as >14 days; P = .02), and the use of total parenteral nutrition at the time of infection (P < .01) as significant risk factors for sepsis with a commensal species.
Comparison of the past 15 years of neonatal sepsis at Y-NHH with the previous 60 years demonstrates continuing changes in the pathogens that cause neonatal sepsis. In early surveys, S pneumoniae and group A streptococci were significant causes of neonatal sepsis, composing almost half of the cases from 1933 to 1943. Since that time, the proportion of sepsis from these organisms has declined steadily. No episodes of S pneumoniae or group A streptococcal sepsis were identified in the last 15-year period.
The decline in the percentage of GBS sepsis is the first observed in this longitudinal series since 1966. A national decline in EOS caused by GBS has been described.18–20 For the last 5 years at Y-NHH, the absolute number of EOS cases caused by GBS has remained fixed between 1 and 3 per year (data not shown). These findings mirror similar trends observed in Connecticut in the late 1990s.21 The lack of significant change during the last 5 years may reflect small sample size and/or the inclusion of preterm infants who were not affected by intrapartum GBS testing and intrapartum antibiotic prophylaxis (IAP).
The percentage of cases of sepsis caused by E coli has decreased steadily since the 1958–1965 period. Similar to GBS, the absolute number of EOS sepsis cases from E coli has remained stable between 0 and 2 over the past 5 years (data not shown). These data, based on small numbers, are similar to other reports that have documented no increased risk for non-GBS infections among traditional neonatal pathogens in the era of maternal IAP21,22 and are in contrast to other reports that have observed an increase in the number of cases of LOS or late, late-onset sepsis caused by E coli.20,23–26
A striking observation in the cause of sepsis is the continued increase in commensal species at Y-NHH, first observed in the previous review period and rising since 1979. Commensal species now compose 44% of all organisms that are responsible for sepsis in the NICU. Episodes of sepsis caused by CONS (8–29%), S aureus (3–8%), and Candida species (1–8%) all increased during this study period compared with the previous period. This trend is not unique to Y-NHH. The predominance of CONS infections, particularly in relation to LOS, was reported previously.1,15,27–29 Prematurity, prolonged intravascular access and mechanical ventilation, the use of intravenous intralipids, and the duration of total parenteral nutrition are among the proposed risk factors associated with NICU-acquired CONS infections.15,16,30–34 The need for mechanical ventilation and total parenteral nutrition administered via central vascular catheters, necessary measures for survival of many NICU patients, weighed against the marked increase in infection risk from these interventions continues to be a dilemma. Strategies to prevent LOS are urgently needed for this population of infants.
The rate of sepsis-related mortality continues to decrease, although we did not identify one specific cause. The definition of death as a result of sepsis used in this study, a positive blood culture obtained within 7 days of death or clinical evidence of sepsis as the cause of death, may overestimate the effect of sepsis as the primary cause of death. Without extensive postmortem examination of all patients, however, cause of death in a neonate is often difficult to substantiate.
The decline in the incidence and virulence of EOS may also contribute to the decrease in mortality. Previous studies revealed that EOS is associated with the highest risk for death, and historically the species that are responsible for EOS carry the highest mortality rate.35,36 From 1979 to 1988, EOS composed 46% of all inborn cases of sepsis. However, during the past 15 years, that percentage has dropped to 21%, with LOS now composing the majority of cases (53%). Currently, overall mortality from EOS (10.2%) does not differ significantly from LOS (9.1%; P = .76). This trend may be related to changes in prenatal care, including screening and IAP for GBS.
The relative rise in cases of LOS and late, late-onset sepsis from commensal species, which have a lower mortality rate than certain species (eg, E coli) that have dominated neonatal infections in the past, may contribute to decreased sepsis-related mortality. In the last study period, the overall percentage of infections related to a commensal species was ∼15%, with the majority of infections caused by traditional neonatal pathogens GBS (37%) and E coli (20%). Commensal species now account for 46% of inborn infections, whereas the percentage of GBS and E coli has fallen to 12% and 11%, respectively (Table 6). Finally, the decrease in mortality from sepsis may reflect the overall decline in neonatal mortality, which has been attributed to advances in perinatal care.36
Supported in part by National Institute of Child Health and Human Development Training Grant T32 HD 07094 (M.J.B.).
- Accepted May 2, 2005.
- Address correspondence to Patrick G. Gallagher, MD, Department of Pediatrics, Yale University School of Medicine, 333 Cedar St, PO Box 208064, New Haven, CT 06520-8064. E-mail:
No conflict of interest declared.
Dr Raskind's current affiliation is Cleveland Clinic Foundation, Cleveland, Ohio.
- ↵Klein JO. Bacterial sepsis and meningitis. In: Remington JS, Klein JO, eds. Infectious Diseases of the Fetus and Newborn Infant. Philadelphia, PA: WB Saunders; 2001:944
- Leibovitz E, Flidel-Rimon O, Juster-Reicher A, et al. Sepsis at a neonatal intensive care unit: a four-year retrospective study (1989–1992). Neonatal Intensive Care.1998:44– 48
- Nyhan WL, Fousek MD. Septicemia of the newborn. Pediatrics.1958;22 :268– 278
- ↵Stoll BJ, Hansen N, Fanaroff AA, et al. Late-onset sepsis in very low birth weight neonates: the experience of the NICHD Neonatal Research Network. Pediatrics.2002;110 :285– 291
- ↵Karlowicz MG, Buescher ES, Surka AE. Fulminant late-onset sepsis in a neonatal intensive care unit, 1988–1997, and the impact of avoiding empiric vancomycin therapy. Pediatrics.2000;106 :1387– 1390
- ↵Baltimore RS, Huie SM, Meek JI, Schuchat A, O'Brien KL. Early-onset neonatal sepsis in the era of group B streptococcal prevention. Pediatrics.2001;108 :1094– 1098
- ↵Gray JE, Richardson DK, McCormick MC, Goldmann DA. Coagulase-negative staphylococcal bacteremia among very low birth weight infants: relation to admission illness severity, resource use, and outcome. Pediatrics.1995;95 :225– 230
- Gaynes RP, Edwards JR, Jarvis WR, Culver DH, Tolson JS, Martone WJ. Nosocomial infections among neonates in high-risk nurseries in the United States. National Nosocomial Infections Surveillance System. Pediatrics.1996;98 :357– 361
- ↵Baltimore RS. Perinatal bacterial and fungal infections. In: Jenson HB, Baltimore RS, eds. Pediatric Infectious Diseases. Philadelphia, PA: WB Saunders; 2002:1119– 1134
- ↵Stoll BJ, Holman RC, Schuchat A. Decline in sepsis-associated neonatal and infant deaths in the United States, 1979 through 1994. Pediatrics.1998;102 (2). Available at: www.pediatrics.org/cgi/content/full/102/2/e18
- Copyright © 2005 by the American Academy of Pediatrics