Pediatric Severe Sepsis: Current Trends and Outcomes From the Pediatric Health Information Systems Database

Amanda Ruth, MD; Courtney E. McCracken, PhD; James D. Fortenberry, MD, MCCM; Matthew Hall, PhD; Harold K. Simon, MD, MBA; Kiran B. Hebbar, MD, FCCM

Disclosures

Pediatr Crit Care Med. 2014;15(9):828-838. 

In This Article

Materials and Methods

Data Collection

This study was an observational cohort review of a prospectively collected administrative database. The study was approved by institutional review boards from CHA and Children's Healthcare of Atlanta. Requirement for informed consent was waived. All patient-related data were de-identified prior to review and enrollment.

Data for this multicenter cohort were obtained from the PHIS database, maintained by CHA, a national collaborative representing more than 220 children's hospitals across the United States. CHA maintains a registry of demographic, outcome, and resource utilization data from 43 freestanding tertiary care children's hospitals (Supplemental Table 1, Supplemental Digital Content 1, http://links.lww.com/PCC/A116). Hospitals participating in the database are located in noncompeting markets of 27 states plus the District of Columbia and account for 15% of all pediatric hospitalizations in the United States. These hospitals provide discharge data, including patient demographics, diagnoses, and procedures. Billing data include medications, radiologic imaging studies, laboratory tests, and supplies charged to each patient. Data are de-identified prior to inclusion in the database. CHA (Overland Park, KS) and participating hospitals jointly assure data integrity and quality as previously described.[19,20]

Data from all 43 participating CHA hospitals were used in our analysis. All children from birth to 19 years admitted to a PICU from January 2004 to December 2012 were identified. PICU admission was determined by a specific database flag. Patients formerly admitted to a neonatal ICU (NICU) and discharged were included in review. Patients admitted to both a NICU and PICU during the same hospitalization were excluded from the cohort since PHIS flags did not distinguish in which unit a severe sepsis episode occurred. Thus 2,269 NICU/PICU patients with PSS were excluded.

For the purpose of reviewing trends in prevalence rates and mortality, we identified a subgroup of 33 children's hospitals that were CHA members and contributed continuous data from the entire 2004–2012 time span for trend analysis.

PICU patients were defined as having PSS if they demonstrated any of the below criteria:

  1. ICD-9 code for severe sepsis (995.92)

  2. ICD-9 code for septic shock (785.52)

  3. An ICD-9 code of infection plus at least one ICD-9 code of organ dysfunction (modified Angus criteria).[15]

To account for the evolution of sepsis billing codes over the past 15 years, we utilized a modified approach to the Angus criteria based on an updated set of ICD-9 codes for severe sepsis and septic shock, as defined in 2012 by Weiss et al[5] (modified Angus criteria) (Supplemental Table 2, Supplemental Digital Content 1, http://links.lww.com/PCC/A116). In the Weiss modification, patients with International Classification of Diseases, 9th Revision, Clinical Modification codes specific for septicemia, sepsis with acute organ dysfunction, and septic shock were additionally included.

Underlying patient disease comorbidities were determined using the definition of a Pediatric Complex Chronic Condition[21] (Supplemental Table 3, Supplemental Digital Content 1, http://links.lww.com/PCC/A116). Length of ICU and hospital stays were calculated and reported as median with interquartile range (IQR) being 25th and 75th percentiles. Site of infection was classified based on grouped ICD-9 codes reflective of a specific source of sepsis (e.g., pneumonia, meningitis, and bacteremia); selected organisms were classified based on ICD-9 codes, but culture results were not specifically available for this determination. Hospitalization charges were supplied by CHA as the total amount charged to the patient by individual hospitals. Cost was estimated by multiplying the total hospital charge by the hospital-specific ratio of cost-to-charge. All reported cost figures were adjusted for inflation and standardized to the year 2012 using U.S. Bureau of Labor Statistics published data for medical cost inflation.[22]

Statistical Analysis

Statistical analyses were performed using SAS 9.3 (Cary, NC). Statistical significance was assessed at the 0.05 level unless otherwise noted. Descriptive statistics were calculated for all variables of interest. Chi-square tests were used to compare categorical variables, and two-sample t tests or Wilcoxon rank-sum tests were used to compare continuous variables between two groups or years. The Cochran-Armitage test for trend was used to identify trends in prevalence rates and mortality rates over time. Although it was possible for a patient to have multiple PICU admissions for severe sepsis over the 9-year cohort, such patients could not be uniquely identified in the PHIS database. Thus, multiple PICU admissions were treated as independent for the purposes of statistical analyses. A hierarchal logistic model was used to identify characteristics of patients with severe sepsis that were associated with an increased risk of mortality while adjusting for PSS cases clustered within hospitals. Initially univariate analysis was used to identify a subset of variables associated with mortality. Variables found to be significant at the 0.15 level in univariate analysis were eligible for inclusion in the final model. To obtain the final model, we performed a stepwise backward elimination procedure in which all candidate predictors were initially included in the model. Variables not significant at the 0.05 level were then systematically removed, provided that they did significantly change the overall model fit when removed. A repeated-measures analysis of variance model was used to assess for changes in the estimated cost of PSS over time while adjusting for the correlation among costs from the same hospital. To evaluate potential impact of length of stay (LOS) and mortality on individual hospital costs, we stratified our cost analysis based on LOS and mortality. Eight separate groups were created using mortality status (survived and died) and four quartiles for LOS: 0–8, 9–17, 18–36, and more than 36 days. The relationship between hospital and cost was then modeled separately for each of the eight groups. Correlations among continuous variables were assessed using Spearman rank-order correlation coefficient. Hospitalization cost was adjusted to the year 2012 as described above.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....