685 - Replacing Capillary Refill Time with High-Risk Comorbidity Improves the qPS4’s Predictive Sensitivity for Phoenix Sepsis and Septic Shock

Publication Number: 3662.685

Nathan Georgette, Boston Children's Hospital, Jamaica Plain, MA, United States; Kenneth Michelson, Lurie Children's Hospital, Chicago, IL, United States; Matthew Eisenberg, Boston Children's Hospital, Boston, MA, United States

Background: The quick Pediatric Septic Shock Screening Score (qPS4) has been previously shown to have only moderate predictive sensitivity for sepsis; moreover, one of its components (capillary refill time) may have limited inter-rater reliability.

Objective: We aimed to determine the predictive characteristics of modifying the qPS4 by replacing capillary refill time with presence of a high-risk co-morbidity.

Design/Methods: We performed a secondary analysis of the validation set from a retrospective cohort study of children aged 1 month to < 18 years presenting to a single ED with suspected infection over a 10-year period. We excluded encounters with sepsis within 1 hour of presentation ("overt sepsis"). The qPS4 has four components (elevated Temperature- and Age-adjusted Mean Shock Index, tachypnea, altered mental status, and capillary refill time >= 3 s). We defined the modified PS4 (mPS4) as the qPS4 where the capillary refill time criterion is replaced by the presence of a high-risk comorbidity (i.e. < 56 days old, Asplenia or sickle cell disease, Bone marrow or solid organ transplant, Central line, Malignancy, Significant CNS/Functional tech dependence, Other immunocompromise). Both the mPS4 and qPS4 were considered positive if >= 2 components were abnormal. The primary outcome was sepsis, secondary outcome was septic shock (both by the Phoenix Criteria and within 24 hours of arrival). The predictive characteristics of the mPS4 and qPS4 were compared; to be considered a true positive, the tool was required to turn positive prior to the outcome criteria being met.

Results: 47,176 encounters were included, of which 6,814 (14.4%) had a high-risk comorbidity. 628 (1.3%) met the sepsis outcome, and 228 (36.3% of sepsis encounters) met the septic shock outcome. For predicting sepsis (Table 1), the mPS4 achieved a sensitivity of 74.7% (95%CI: 71.1-78.0%) and PPV of 7.1% (6.5-7.7%); qPS4 demonstrated a sensitivity of 67.8% (64.0-71.5%) and PPV of 8.1% (7.4-8.8%) (Figure 1A). For predicting septic shock (Table 2), the mPS4 achieved a sensitivity of 90.8% (86.3-94.2%) and PPV of 3.1% (2.7-3.5%); qPS4 demonstrated a sensitivity of 85.5% (81.0-90.1%) and PPV of 3.6% (3.1-4.1%) (Figure 1B).

Conclusion(s): The mPS4, which replaced delayed capillary refill time with presence of a high-risk comorbidity, had improved predictive sensitivity with slightly lower PPVs for Phoenix sepsis and septic shock. If higher predictive sensitivity is desired, or if capillary refill time is not reliably captured, mPS4 may provide a useful alternative to qPS4.

Figure 1
Precision-recall charts of the mPS4 and qPS4 tools for predicting Phoenix sepsis (Panel A) and septic shock (Panel B) in the cohort (N= 47,176). The vertical bars indicate 95% CIs for PPV; the horizontal bars indicate those for sensitivity. Dashed line indicates prevalence of outcome. Abbreviations: PPV, positive predictive value; mPS4, modified Pediatric Sepsis Shock Screening Score; qPS4, quick Pediatric Septic Shock Screening Score.

Table 1
Predictive characteristics for outcome of Phoenix sepsis. Prevalence of outcome = 1.3%. Parentheses enclose the 95% Confidence Intervals. Abbreviations: mPS4, modified Pediatric Sepsis Shock Screening Score; qPS4, quick Pediatric Septic Shock Screening Score; PPV, positive predictive value; NPV, negative predictive value; LR -, negative likelihood ratio; LR +, positive likelihood ratio.

Table 2
Predictive characteristics for outcome of Phoenix septic shock. Prevalence of outcome = 0.5%. Parentheses enclose the 95% Confidence Intervals. Abbreviations: mPS4, modified Pediatric Sepsis Shock Screening Score; qPS4, quick Pediatric Septic Shock Screening Score; PPV, positive predictive value; NPV, negative predictive value; LR -, negative likelihood ratio; LR +, positive likelihood ratio.