OBJECTIVE. Our goal was to describe patient and hospital characteristics associated with in-hospital mortality, length of stay, and charges for critically ill children with severe sepsis.
METHODS. Our study consisted of a retrospective study of children 0 to 19 years of age hospitalized with severe sepsis using the 2003 Kids' Inpatient Database. We generated national estimates of rates of hospitalization and then compared in-hospital mortality, length of stay, and total charges according to patient and hospital characteristics using multivariable regression methods. Severity of illness was measured by using all-patient refined diagnosis-related group severity of illness classification into minor, moderate, major, and extreme severity.
RESULTS. There were an estimated 21448 hospitalizations for severe pediatric sepsis nationally in 2003. The in-hospital mortality rate was 4.2%. Comorbid illness was present in 34% of hospitalized children. Most (70%) of the extremely ill children were admitted to children's hospitals. Length of stay was longer among patients with higher illness severity and nonsurvivors compared with survivors (13.5 vs 8.5 days). Hospitalizations at urban or children's hospitals were also associated with longer length of stay than nonchildren's or rural hospitals, respectively. Higher charges were associated with higher illness severity, and nonsurvivors had 2.5-fold higher total charges than survivors. Also, higher charges were observed among hospitalizations in urban or children's hospitals. In multivariable regression analysis, multiple comorbid illnesses, multiple organ dysfunction, and greater severity of illness were associated with higher odds of mortality and longer length of stay. Higher hospital charges and longer length of stay were observed among transfer hospitalizations and among hospitalizations to children's hospitals and nonchildren's teaching hospitals compared with hospitals, which had neither children's nor teaching status.
CONCLUSIONS. Mortality from severe pediatric sepsis is associated with patient illness severity, comorbid illness, and multiple organ dysfunction. Many characteristics are associated with resource consumption, including type of hospital, source of admission, and illness severity.
Pediatric sepsis that is associated with organ dysfunction, termed severe sepsis, is associated with higher mortality compared with sepsis without organ dysfunction.1 Severe sepsis is a major cause of child mortality and morbidity in the United States and is also associated with high annual costs of >2 billion dollars.1,2 Although organ dysfunction progresses over time in the disease state, both in terms of severity and the number of organs involved, timely resuscitation of septic shock has been associated with improved outcomes assessed by reversal of organ dysfunction and improved survival.3
Although extensive research and clinical efforts are continually being made to reduce the morbidity and mortality associated with severe pediatric sepsis, no published information exists about potential variation that might occur in outcomes and resource utilization across different acute care settings. Also, the clinical course, sequelae, and outcomes of severe pediatric sepsis may be significantly impacted by the performance of time-sensitive and resource-dependent maneuvers, such as fluid resuscitation,3,4 interhospital transfer, subspecialist referral, and intensive care.5 The availability of, and access to, such key interventions might vary among different care settings, thereby providing an opportunity to investigate the association between various hospital characteristics and the outcomes and resource utilization for severe pediatric sepsis, while accounting for patient characteristics previously associated with such outcomes. Such information may be useful to clinicians and health care policy makers and highlight opportunities to alleviate illness burden.
Our study was conducted to describe hospital and patient characteristics that are independently associated with hospital mortality, length of stay (LOS), and charges for children hospitalized with severe sepsis, and to test the hypothesis that these measures would vary among different types of hospitals.
We conducted a retrospective study of hospitalized children with sepsis, 0 to 19 years old, who also had organ dysfunction involving at least 1 organ system. Our data source was the 2003 Kids' Inpatient Database (KID). The KID was developed by the Agency for Healthcare Research and Quality (AHRQ) and includes nearly 3 million pediatric discharge records obtained from 3438 hospitals in 36 states.6 The KID is the only national, all-payer database of hospitalizations for children. The database contains 80% of the normal nonnewborn discharges from these states and is nationally representative with the inclusion of discharge weights in analyses. Information on patient demographics, hospital characteristics, and diagnosis codes is included for each hospitalization.
Study Sample and Variable Identification
Children with a primary or secondary diagnosis of sepsis associated with organ dysfunction were identified by using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes applying methodology described in previous studies.2,7,8 Severe sepsis was subsequently identified as the occurrence of sepsis and dysfunction of ≥1 organ system. Organ dysfunction was measured both according to the actual organ systems involved and the number of organ systems involved, because previous literature suggest worse outcomes with an increase in the number of organ systems involved in the disease state.1,2,8 Patient characteristics other than organ dysfunction were age, gender, presence, and count of comorbid illness, and severity of illness using the all-patient refined diagnosis-related group (APRDRG) classification within the KID. The categories of APRDRG severity of illness were minor, moderate, major, and extreme loss of function, hereafter referred to as illness severity. The APRDRG severity classification is a proprietary, validated, and extensively used measure of illness severity that uses patient discharge data including principal and secondary diagnoses, procedures, and demographic information to assign patients to subclasses of illness severity.9 Because of small sample size for hospitalizations with minor illness severity, the categories for minor and moderate illness severity were combined for the analyses. Comorbidities were identified by using ICD-9-CM codes, applying methodology described in the literature.10 Comorbid illness was incorporated into the analyses because previous studies have associated the presence of comorbid illness with mortality2 and cost of hospitalization.11
Hospital characteristics were total bed size (large, medium, and small) defined by hospital location and teaching status, the type of hospital (children's hospital status and teaching status: children's teaching, children's nonteaching, nonchildren's teaching, and nonchildren's nonteaching hospitals), source of admission (emergency department [ED], clinic, other hospitals), and location (urban or rural) as reported by publications available from AHRQ.6
Children's hospitals were identified by the American Hospital Association Annual Survey of Hospitals and information from the National Association of Children's Hospitals and Related Institutions. Teaching hospital status was determined by whether a hospital had an American Medical Association-approved residency program, was a member of the Council of Teaching Hospitals, or had a ratio of full-time equivalent interns and residents to beds of 0.25 or higher.6 Because of small sample size for admissions (77 admissions) to the nonteaching children's hospitals, children's teaching and nonteaching hospitals were combined for the analyses. The institutional review board of the University of Michigan Medical School approved the study.
To identify factors associated with in-hospital mortality, LOS, and total charges, we initially identified the number of hospitalizations for severe sepsis in the KID and, subsequently, the frequency distribution of these discharges was obtained. Thereafter, rates of in-hospital mortality, and the frequency distribution of LOS and charges (with associated 95% confidence intervals [CIs]) for the hospitalizations was described according to both patient and hospital characteristics. The number of hospitalizations in our results was unweighted, although all effect estimates and accompanying 95% CIs were calculated by using sample weights to account for the complex survey design.
All estimates used the survey commands in Stata for Windows (Stata Corp, College Station, TX), which accounted for the complex survey design. To allow generation of stable estimates, cell frequencies that were too small (<70) for precise estimation were suppressed in the report as recommended by AHRQ. In these analyses, multivariable logistic regression, negative binomial regression, and multiple linear regression models for complex survey data were fit to assess differences in mortality, LOS, and total charges, respectively. Variance estimates accounted for clustering of data at the hospital level by using hospital identifiers as the primary sampling units.
There were 12604 hospitalizations for severe sepsis in the database, representing 21448 hospitalizations nationally in 2003. Over half (53%) of all hospitalized patients were male, and 58% were infants and children <5 years of age. Another peak in the frequency of hospitalizations (18%) was in children 15 to 19 years of age. Over half (59%) of hospitalizations were related to major or extreme illness severity (Table 1). For 58% of the hospitalizations, admission occurred via the ED. Approximately half (51%) of all cases were admitted into large-sized hospitals. The majority (98%) of all hospitalizations were in urban hospitals, and half of all hospitalizations were to children's (teaching and nonteaching) hospitals.
The overall in-hospital mortality rate was 4.2%, and comorbid illness was present in 34% of all hospitalized children. Cardiac dysfunction occurred in 46.4% of the hospitalized children. Other dysfunctional organ systems were, in decreasing order of frequency: respiratory (39.2%), hematologic (19.5%), neurologic (8.1%), renal (7%), and hepatic (0.7%). The overall mean LOS was 8.7 days (95% CI: 8.1–9.3 days), and the mean total charge per hospitalization was $ 47126 (95% CI: $42159–$52097).
In bivariate analysis, fatal outcomes were associated with specific patient and hospital characteristics (Table 1). Nearly half of all fatalities occurred among children <5 years of age, and 25% occurred among children 15 to 19 years of age. There was no gender difference in mortality; however, patients with comorbid illness or extreme illness severity had higher mortality rates. Similarly, higher mortality rates were observed among hospitalizations via the ED, to an urban hospital, or to children's hospitals.
Analysis of illness severity revealed that 40% of hospitalizations to children's hospitals had extreme illness severity, much higher than in nonchildren's hospitals with teaching or nonteaching status (Table 2). Likewise, 70% of all hospitalized children with extreme illness severity were admitted into children's hospitals compared with 18% and 12% to nonchildren's teaching and nonchildren's nonteaching hospitals, respectively. Of note, half of all transfer hospitalized children had extreme illness severity, much higher than for those with any other source of admission (Table 2).
LOS was 5 days longer among nonsurvivors and was notably longer among patients with major or extreme illness severity compared with minor/moderate severity (Table 3). Hospitalizations into urban hospitals or children's hospitals were associated with longer LOS compared with rural or nonchildren's hospitals.
Total hospital charges were 2.5-fold higher among nonsurvivors when compared with survivors (Table 4). Higher charges were also associated with higher illness severity, and among hospitalizations into urban hospitals or children's hospitals, compared with rural or nonchildren's hospitals, respectively.
Multivariable logistic regression analysis revealed that the number of comorbid illnesses (odds ratio [OR]: 1.54; 95% CI: 1.38–1.73), number of dysfunctional organ systems (OR: 2.25; 95% CI: 2.02–2.51), and severity of illness were associated with mortality (Table 5). There was no statistically significant difference in risk-adjusted mortality according to the type of hospital (children's/teaching hospital status), despite hospitalization of children with the highest illness severity within children's hospitals.
In multivariable negative binomial regression analysis, longer LOS was associated with multiple comorbid illnesses, multiple organ dysfunction, and higher illness severity (Table 6). Hospitalizations from either the clinic or as transfer from other hospitals were associated with higher LOS compared with ED admissions. Hospitalization into children's hospitals or nonchildren's teaching hospitals had longer LOS compared with hospitals with neither children's nor teaching status, after controlling for patient factors.
Hospital charges for children with severe sepsis varied by patient and hospital characteristics. Multivariable linear regression analysis revealed higher charges among patients who were admitted on transfer from other hospitals (Table 7). Higher charges were also observed among hospitalizations in children's hospitals and nonchildren's teaching hospitals compared with hospitals with neither children's nor teaching status. Likewise, urban hospitals had higher charges compared with rural hospitals. Higher charges were also observed with increasing illness severity, multiple comorbid illnesses or multiple organ dysfunction.
This is the first study, to our knowledge, to describe an association between characteristics of the hospitals where care is provided to children with severe sepsis and in-hospital mortality, LOS, and charges. We found higher charges and longer length of stay for children admitted to children's hospitals and urban hospitals compared with admission to nonchildren's or rural hospitals, respectively. Of note, although most of the children with extreme illness severity were admitted to children's hospitals, there was no statistically significant difference in risk-adjusted mortality between these hospitalizations and those within nonchildren's hospitals. This study also corroborates previous studies that have described a significant association between the presence of comorbid illness and multiple organ dysfunction with mortality and LOS for children hospitalized with severe sepsis.1,2,7,8
Our study findings highlight significant variation in illness burden and resource utilization according to the type of hospital and suggest a practice of patient triage according to illness severity. This results in funneling of the most severely ill patients to children's hospitals where specialized care might be more readily available. On a similar note, patients whose care involved transfer consumed more resources, both in terms of longer LOS and higher total charges, when compared with children admitted directly from the ED. This observation corroborates previous findings among critically ill adults that higher resource utilization occurred among certain patients who underwent interhospital transfer.12,13
Children's hospitals, regardless of their teaching status, along with nonchildren's teaching hospitals, had significantly longer LOS of hospitalizations when compared with hospitals without teaching or children's hospital status. A previous study highlighted variation in LOS by hospital type among hospitalizations for common pediatric conditions, without clearly defined reasons for such observation.14 A significant limitation of that study was the inability to adjust for illness severity. However, in our study, similar findings were observed after rigorous adjustment for illness severity by using available methods. We speculate that this differential LOS and resource consumption according to the type of hospital for severe sepsis hospitalizations might be attributable to extensive in-hospital rehabilitation and multidisciplinary care for the most severely ill patients selectively admitted to children's hospitals. Outside illness severity, however, multiple patient characteristics, practice patterns, or institutional factors may cause the wide differences across hospitals in LOS.15 Additional study of differences in LOS among patients with severe sepsis according to the type of hospital might help elucidate whether these or other factors are germane to the critically ill pediatric population with severe sepsis.
A previous study of hospitalizations for common pediatric conditions16 reported significant differences in charges by hospital type, similar to our study. It has been speculated that subspecialty care and use of advanced technology within children's hospitals might explain some of the differential charges reported. Furthermore, previous studies have reported higher charges among adult hospitalizations in tertiary hospitals,17 suggesting elevated hospital charges might occur with specialization of medical care.
Fatal outcomes for children with severe sepsis were associated more strongly with patient characteristics than hospital characteristics. Historically, a significant limitation of the use of administrative databases for outcomes research pertained to the inability to control for patient severity of illness, leading to the use of various methods over time.18,19 In our study, however, available methodology was rigorously applied to account for comorbid illness and multiplicity of organ dysfunction as markers of illness severity. In addition, the use of APRDRG severity provided an opportunity for categorization of patients according to level of illness severity.
The finding of an association between multiple organ dysfunction and death implies an urgent need for concerted clinical and educational efforts within clinical care settings aimed at limiting progression of illness severity and organ dysfunction in children with severe sepsis. This finding indirectly corroborates previous studies3,4 that emphasize the need for early resuscitation to prevent or limit organ dysfunction as a major mortality risk-reduction strategy in severe pediatric sepsis.
Resource utilization, measured by LOS and hospital charges, was associated with illness severity and varied significantly by hospital characteristics, including type of hospital and geographical location. Efforts to alleviate this resource burden, particularly among children's hospitals and hospitals located in urban regions, will need to develop better understanding of circumstances surrounding interhospital transfers, timing of resuscitative care, and other health system processes that might be incorporated into the care of children with severe sepsis.
The findings of our study should be interpreted in light of certain limitations. The KID is a database of administrative discharge data without clinical information beyond what can be captured in ICD-9-CM diagnosis and procedure codes. Therefore, it was not possible to study the clinical course for each patient, the need for and receipt of various therapeutic interventions, and how clinical care was coordinated at the patient and hospital level. An additional limitation is that the identification of cases of severe sepsis was ascertained via ICD-9-CM diagnosis codes and is susceptible to inaccuracies of detection and attribution that may have biased our findings.
Also, although the care of patients with severe sepsis might be improved by early resuscitation in the community before transfer to referral hospitals,3 determination of the characteristics of the referring hospitals and the circumstances surrounding transfer among the transfer hospitalizations was not possible in our study. This limited the ability to investigate any delay in resuscitative care, which might impact clinical outcomes and resource utilization at the receiving hospital. The characterization of types of hospitals within the database did not permit identification of which hospitals were freestanding children's hospitals, which are likely to be highly specialized with extensive resource capability. The KID contains nonclinical administrative data collected during patient hospitalizations and has no follow-up or longitudinal information on patients after hospital discharge. This limitation did not permit determination of patient morbidity and functional status.
Mortality from severe pediatric sepsis was influenced largely by patient characteristics including comorbid illness and multiple organ system dysfunction. Resource consumption, assessed by LOS and hospital charges, was influenced both by these patient characteristics, and multiple hospital characteristics, such as type and location of hospital and source of admission.
Efforts to curb mortality from severe pediatric sepsis will need to address patient illness severity and explore mechanisms to stem the progression of multiorgan system dysfunction. The success of such mortality-reduction strategies and others aimed at alleviating both the overall and differential resource burden associated with severe pediatric sepsis will hinge on improved understanding of the role of patient triage, interhospital transfer, hospital characteristics, and other yet unmeasured factors on clinical outcomes and resource utilization within various acute care settings.
- Accepted November 7, 2006.
- Address correspondence to Folafoluwa O. Odetola, MD, MPH, 6C07, 300 North Ingalls St, Ann Arbor, MI 48109. E-mail:
The authors have indicated they have no financial relationships relevant to this article to disclose.
- ↵Han YY, Carcillo JA, Dragotta MA, et al. Early reversal of pediatric-neonatal septic shock by community physicians is associated with improved outcome. Pediatrics.2003;112 :793– 799
- ↵Agency for Healthcare Research and Quality. Healthcare Cost and Utilization Project (HCUP) Kids' Inpatient Database (KID) 2003: Description of Data Elements. Rockville, MD. Agency for Healthcare Research and Quality; 2005
- ↵Muldoon JH. Structure and performance of different DRG classification systems for neonatal medicine. Pediatrics.1999;103 :302– 318
- ↵Srivastava R, Homer CJ. Length of stay for common pediatric conditions: teaching versus nonteaching hospitals. Pediatrics.2003;112 :278– 281
- ↵Merenstein D, Egleston B, Diener-West M. Lengths of stay and costs associated with children's hospitals. Pediatrics.2005;115 :839– 844
- ↵Iezzoni LI, Shwartz M, Ash AS, Mackiernan YD. Predicting in-hospital mortality for stroke patients: results differ across severity-measurement methods. Med Decis Making.1996;16 :348– 356
- Copyright © 2007 by the American Academy of Pediatrics