Applying a discriminant model based on combination of pediatric critical illness score,D-dimer and procalcitonin in early predicting the severity of pediatric infection

Abstract

Objective To establish a discriminant method based on clinical and laboratory data and common examinations for early predicting the severity of pediatric infection. Methods Consecutive hospitalized patients diagnosed as septic shock were included who were admitted between June 2014 and May 2015 retrospectively.Gender (male-female ratio: 1.251.00)and age (1 month to 6 years old) were matched in all of 18 patients with septic shock, and 27 patients diagnosed as systemic inflammatory response syndrome (SIRS), sepsis and severe sepsis on admission were included respectively in order of sequential admission number during the same period.Additional 36 gender- and age-matched children with common infection (non-SIRS)were enrolled as controls.The clinical and laboratory examination data of all the included patients were collected and then the pediatric critical illness scores (PCIS)were made according to the worst condition within 24 hours of hospitalization.The parameters correlated with the severity of infection were evaluated by rank correlation and Logistic regression analysis.The discriminant models were established based on κth-nearest-neighbor analysis and evaluated with clinical diagnosis by interrater agreement test. Results Except for platelet count, the other indexes including PCIS, neutrophil count, C-reactive protein, procalcitonin (PCT), international normalized ratio of prothrombin time, activated partial thromboplastin time, thrombin time, fibrinogen, fibrin/fibrinogen degradation product (FDP)and D-dimer (D-D)all had differences among groups with varying infection severity (all P<0.001). The Spearman's coefficient ρ of PCIS, PCT, D-D and FDP correlated to infection severity were -0.837, 0.680, 0.679 and 0.648, respectively (all P<0.001). Multivariate cumulative odds Logistic regression analysis showed PCIS, D-D and PCT were related to infection severity (all P<0.05). The total error rate of discriminant models based on 3-index combination (Mahalanobis transformation, k=2)was 0.091 that was lower than any models based on 2-index combination or single-index.Using the discriminant model based on three-index combination, the infection severity of 26 patients admitted during June 2015 were predicted with a high interrater agreement (weighted Kappa coefficient: 0.670, P<0.001)compared to clinical diagnosis. Conclusion The discriminant model based on combination of PCIS, D-D and PCT could assist predicting the severity of pediatric infection earlier. Key words: Infection; Discriminant analysis; Pediatric critical illness score; D-dimer; Procalcitonin