The aim of the present study was a comparison of prognostic accuracy assessment results for scores in different ways on the example of validity analysis of the prognostic model ISS-RTS-TRISS for assessing the severity of the condition in children with trauma. The prospective study was conducted using clinical and physiological data collected at the admission and during the first 24 hours of hospitalization from 414 children with trauma. We had three groups of patients, common group children with traumatic brain injuries, and two groups of patients were divided into the two following ways: 141 (34%) patients with isolated traumatic brain injuries and 273 (66%) patients with combined injuries with traumatic brain injuries The validity and prognostic accuracy of prognostic scores were assessed by determining there discrimination and calibration ability. Analysis of the discrimination ability of score was carried out by assessment the areas under the ROC curves. For analysis of the calibration ability of score was used three methods of the Hosmer-Lemeshow test (H-criterion, C-criterion in two variants). The ISS-RTS-TRISS score showed significantly outstanding predictive accuracy in studied groups (AUROC >=0.9) .However, estimation the calibration ability of score using the C-criterion, by dividing into groups with the same number of cases of patients (recommended for abnormal distribution) did not show an unambiguous results. It was shown that to obtain an unambiguous correct result, it was necessary to use the C-criterion method using the division of cases into groups with the same number of patients with the lethal outcome. When using this method, satisfactory results of study of calibration ability were shown for all the studied groups (p<0.05).
Keywords: Prognostic score; prognostic accuracy of score; score ISS-RTS-TRISS; the Hosmer-Lemeshow test; validity of prognostic scores.