Abstract
Infectious complications are major contributors to morbidity and mortality in children with leukemia; however, clinical trial reporting is incomplete (Miller 2016 J Clin Oncol). Using administrative data to identify SS events could improve estimates of incidence and clinical impact. However, accurate SS determination is challenging, and the 2005 International Pediatric Sepsis Consensus Conference (IPSCC) definition is often modified for individual studies. ICD-9 codes, assigned at discharge, do not permit precise timing estimates. We hypothesized PHIS resource utilization (RU) codes could identify vasopressor exposures as proxies for IPSCC-defined SS events. We present the operating characteristics of our methods and use the approach to estimate SS incidence for a national cohort.
We previously identified and validated a longitudinal cohort of children aged 0-21 with newly diagnosed ALL using PHIS billing and diagnosis codes (Fisher 2014 Med Care), expanded through 12/31/2013. PHIS RU data were used to identify distinct vasopressor exposure patterns hypothesized to represent true SS (Figure 1). Epinephrine alone with asparaginase within ± 3 days was considered anaphylaxis rather than SS. We captured vasopressor patterns occurring ≥7 days from first chemotherapy until the first of: 30 days prior to relapse, 10 days prior to stem cell transplantation, or 3 years from diagnosis.
We reviewed charts of all patients with ALL diagnosed from 2004 to 2013 at the Children's Hospital of Philadelphia (CHOP) to establish the SS gold-standard. We assumed true SS events would be associated with blood cultures. Each blood culture triggered a chart review and was classified by IPSCC sepsis criteria. A PHIS RU-defined vasopressor exposure had to be within ±14 days of the chart-defined SS to be considered a true positive.
We predicted a patient may meet criteria for a PHIS vasopressor exposure if vasopressors were ordered to the bedside for impending SS but ultimately not administered. For CHOP patients with available electronic medication administration record (MAR) data (1/1/2011-12/31/2013), we assessed whether using vasopressor orders marked as "given" or "given or held at bedside" resulted in different operating characteristics relative to gold standard IPSCC SS events. We calculated sensitivity and positive predictive value (PPV) of PHIS RU exposure patterns for IPSCC SS, then used them to estimate SS incidence for the entire PHIS cohort and for certain risk group subsets.
We identified 360 patients common to both PHIS and CHOP chart review cohorts. Chart review revealed 38 events meeting IPSCC SS criteria. PHIS RU data identified vasopressor exposure patterns in 35 of these 38 SS events within ±14 days (sensitivity 92%). PHIS RU data identified an additional 43 vasopressor patterns that did not correlate with a chart review IPSCC SS event (PPV: 35/78, 45%). When considering chart-defined sepsis event of any severity (ie, not restricted to SS), the PPV for PHIS-defined vasopressor events increased to 88%. MAR data were available for 149 patients to delineate whether ordered vasopressors were given or held. Among these 149 patients, 10 SS events met IPSCC criteria by chart review. Restricting RU-identified vasopressor exposure events to patients who actually received vasopressors resulted in sensitivity of 70% and PPV 77%; including orders where vasopressors were given or held, the sensitivity increased to 100% and PPV decreased to 55%. Using PHIS RU vasopressor patterns, we estimated an incidence of 12.6% (95% CI 12.0 - 13.3 %) of patients ever experiencing SS in the PHIS ALL cohort. Infants, children aged ≥10 years, with public insurance, or receiving daunorubicin all had significantly higher SS rates than the cohort baseline (Table 1).
PHIS RU-defined vasopressor exposure patterns have a high sensitivity for IPSCC-defined SS but a low PPV. The PPV may be the result of including patients with vasopressors ordered but not given; however, IPSCC-defined SS excludes children with fluid-responsive shock. The low PPV suggests this RU-defined approach results in "overcapture" of SS, but the gold standard definition may be too restrictive. Vasopressor exposure patterns from RU data may identify patients with SS or impending SS that can augment incomplete clinical trial data collection. PHIS RU SS incidence estimates for a national pediatric ALL cohort exceed 12%, with even higher rates in high-risk subgroups.
Fisher:Merck: Research Funding; Pfizer: Research Funding.
This feature is available to Subscribers Only
Sign In or Create an Account Close Modal