Comparing Trauma and Injury Severity Score, New Injury Severity Score, and Revised Trauma Score for Trauma Mortality Prediction: A Meta-Analysis Bendix Samarta Witarto1a, Visuddho1, David Nugraha1 1Undergraduate Program, Faculty of Medicine, Universitas Airlangga abendixsw@gmail.com Bendix Samarta Witarto - Visuddho - David Nugraha Table 1. Characteristics of the Included Studies. INTRODUCTION Trauma is one of the world's public health problems affecting people of all ages. Trauma, with its varying degrees of severity, has become one of the leading cause of mortality all around the world, especially in the low and middle income countries.1 Early warning system of medical services plays a significant role in determining the appropriate intervention to reduce unnecessary complications and in order to ensure the right patients’ triage.2 These decisions have a notable impact on mortality outcome of trauma patients. In fact, trauma scores are usually used for the classification of trauma severity and prognostication.3 The proper consideration of the identified prognostic factors and good outcome predictive properties are crucial for a scoring system. There are several scoring systems have been developed and used worldwide, including the anatomicalbased (New Injury Severity Score - NISS), the physiological-based (Revised Trauma Score - RTS), and the combined anatomic-physiological-based scoring system (Trauma and Injury Severity Score - TRISS).4 Nevertheless, the most accurate scoring system remains an elusive target and to the extend of our knowledge, there is no systematic review that analyses each scoring system in depth. Therefore, this study will quantitatively compare which scoring system has the best predictive accuracy for mortality in trauma patients. Author, Year Study Location Javali et al., India 20196 Eryilmaz et al., Turkey 20097 Arikan et al., Total Patients TRISS NISS RTS Patients Characteristics Mean ± SD (M/F) (Trauma and Injury Severity Score) (New Injury Severity Score) (Revised Trauma Score) Study Design Age (Survivor/NonMechanism of % AUC AUC % AUC AUC % AUC AUC % Prev Cut-off % Sn % Sp Cut-off % Sn % Sp Cut-off % Sn % Sp Survivor) Trauma or Injury (95% CI) p- value (95% CI) p- value (95% CI) p- value 200 Prospective (148/52) 66.4 ± 6.9 Blunt trauma 100 ≤ 91.6 97.2 97.06 87.95 < 0.0001 > 17.0 97 91.18 93.37 < 0.0001 ≤ 7.11 94.7 97.06 80.12 < 0.0001 Cohort (166/34) 87 Retrospective Blunt trauma (fall (64/23) 26.9 ± 9.1 100 73.5 99.1 88.9 98.7 0.01 31.5 91.5 100.0 69.2 0.001 1.04 N/A 100.0 5.1 < 0.001 Cross-sectional from height) (78/9) Retrospective Cohort Turkey 20208 Germany Lefering, Retrospective Austria Cohort 20093 Switzerland Tan et al., Malaysia 20179 Retrospective Cohort Shotgun wound Stab wound Other sharp materials injuries 77 (62/15) (56/21) 28.7 ± 10.7 1206 (892/314) (1006/200) 38.2 ± N/A 2208 (1957/251) (1970/238) Blunt trauma 36.0 ± 16.0 Penetrating trauma Blast trauma Blunt trauma Penetrating trauma 47 41 7.6 94.3 95.2 92.2 < 0.001 19.0 95 5 0.5 86.2 (83.0–89.0) 41.0 96.0 N/A 90.5 9.2 0.2 < 0.96 N/A 89.6 85.7 74.0 < 0.001 6.27 49.0 80.0 49.0 (77.0–84.0) 90.0 N/A > 24.0 87.8 86.6 (86.4–89.2) 74.3 N/A 91.5 90.5 92.9 < 0.001 4.10 77.0 44.0 (73.0–80.0) 89.0 N/A < 7.81 80.2 72.4 (78.5–81.8) 82.8 N/A 12 81.2 94.14 48.90 (79.5–82.8) Sensitivity (95% CI) Javali et al. 2019 Eryilmaz et al. 2009 Arikan et al. 2020 Lefering 2009 Tan et al. 2017 0,97 0,89 0,95 0,41 0,94 (0,85 - 1,00) (0,52 - 1,00) (0,76 - 1,00) (0,34 - 0,48) (0,90 - 0,97) Sensitivity 1 SROC Curve Sy AU SE Q SE 0,9 0,8 0,7 0 0,2 0,4 0,6 Sensitivity 0,8 Pooled Sensitivity = 0,73 (0,69 to 0,77) Chi-square = 183,94; df = 4 (p = 0,0000) 1 Inconsistency (I-square) = 97,8 % 0,6 0,5 Specificity (95% CI) Sensitivity 1 METHODS 0,88 (0,82 - 0,92) SROC Curve Javali et al. 2019 Eryilmaz et al. 2009 Arikan et al. 2020 Lefering 2009 Tan et al. 2017 0,99 0,93 0,96 0,49 (0,93 - 1,00) (0,83 - 0,98) (0,95 - 0,97) (0,47 - 0,51) 0,9 0,4 0,3 Symmetric SROC AUC = 0,9542 SE(AUC) = 0,0299 Q* = 0,8962 SE(Q*) = 0,0416 0,2 0,1 Pooled Specificity = 0,67 (0,66 to 0,69) Identification A systematic data searching based Chi-square = 914,11; df = 4 (p = 0,0000) Records identified through 0,80,4 database searching 0 0,2 0,6 0,8 1 Inconsistency (I-square) = 99,6 % on the PRISMA statement (Figure (n = 261) Specificity PubMed (n = 23) 1) was conducted in PubMed, Figure 3. Pooled Sensitivity, Specificity, and AUC of Trauma and Injury Severity Score (TRISS) for Trauma Mortality Prediction. Additional records Scopus (n = 26) identified through ProQuest (n = 106) 0,7 Sensitivity (95% CI) Scopus, ProQuest, ScienceDirect, other sources ScienceDirect (n = 93) (n = 0) CINAHL Plus in EBSCOhost (n = 13) Javali et al. 2019 0,91 (0,76 - 0,98) and CINAHL Plus (EBSCOhost), up Eryilmaz et al. 2009 1,00 (0,66 - 1,00) Arikan et al. 2020 0,86 (0,64 - 0,97) to March 25, 2021. Boolean 0,6 Records after duplicates removed Lefering 2009 0,49 (0,42 - 0,56) (n = 213) Tan et al. 2017 0,87 (0,82 - 0,91) operators were used to broaden and narrow the search result using Records excluded 0,5 Records screened (n = 152) (n =213) the keywords: "TRISS", "NISS", Inappropriate title (n = 79) Pooled Sensitivity = 0,72 (0,68 to 0,76) Inappropriate abstract (n = 63) Chi-square = 91,66; df = 4 (p = 0,0000) "RTS", "mortality", and its No full-text available (n = 10) 0 0,2 0,4 0,6 0,8 1 Inconsistency (I-square) = 95,6 % Articles assessed for 0,4 Sensitivity synonyms. Risk of bias assessment eligibility Specificity (95% CI) Articles excluded, with reasons (n = 61) (n = 56) of the studies was conducted using Javali et al. 2019 0,93 (0,88 - 0,97) Sensitivity SROC Curve Inappropriate population Eryilmaz et al. 2009 0,69 (0,58 - 0,79) 1 (n = 0) QUADAS-2 Tool. Statistical analyses 0,3 Arikan et al. 2020 0,73 (0,60 - 0,84) Studies included in Unmeasured indicator (n = 13) Lefering 2009 0,90 (0,88 - 0,92) qualitative synthesis No outcome of interest (n = 9) were performed using the software Tan et al. 2017 0,74 (0,72 - 0,76) (n = 5) Non-English or Indonesian Symmetric SROC 5 full-text (n = 4) Meta-DiSc version 1.4. Z-test was 0,9 AUC = 0,8730 Conference abstracts or books 0,2 SE(AUC) = 0,0113 (n = 10) Studies included in adopted to compare the pooled Review (n = 13) Q* = 0,8034 quantitative synthesis Pooled Specificity = 0,80 (0,79 to 0,81) Editorial (n = 7) (meta-analysis) AUCs (AUSROCs) of TRISS, NISS, SE(Q*) = 0,0113 Chi-square = 140,69; df = 4 (p = 0,0000) 0,8 (n = 5) 0 0,2 0,10,4 0,6 0,8 1 Inconsistency (I-square) = 97,2 % and RTS with p < 0.05 indicating a Specificity Figure 1. PRISMA Flowchart. statistical significance. Figure 4. Pooled Sensitivity, Specificity, and AUC of New Injury Severity Score (NISS) for Trauma Mortality Prediction. 0,7 Sensitivity (95% CI) 0 Inclusion criteria: Exclusion criteria: 0 0,2Javali et al. 2019 0,4 0,97 (0,85 - 1,00)0,6 0,8 1 1. Observational study 1. Inappropriate population, indicators, 1-specificity Eryilmaz et al. 2009 1,00 (0,66 - 1,00) Arikan et al. 2020 0,90 (0,70 - 0,99) 0,6 2. Hospitalized patients due to any or outcome Lefering 2009 0,44 (0,37 - 0,51) Tan et al. 2017 0,72 (0,66 - 0,78) types of trauma 2. Non-extractable data 3. TRISS, NISS, RTS as indicators 3. Not written in English or Indonesian 0,5 with cut-off, AUC, sensitivity, 4. No full-text available Pooled Sensitivity = 0,64 (0,60 to 0,68) Chi-square = 78,72; df = 4 (p = 0,0000) and specificity values 5. Review, report, editorial, conference 0 Inconsistency (I-square) = 94,9 % 0,2 0,4 0,6 0,8 1 0,4Sensitivity 4. Mortality as outcome abstracts or books Specificity (95% CI) 0 0 0,2 0,4 0,6 0,8 1 1-specificity Sensitivity 1 SROC Curve Screening 0,9 0,8 Sy AU SE Q SE 0,7 Eligibility 0,6 0,5 0,4 0,3 Included 0,2 0,1 0 0 0,2 0,4 0,6 0,8 1 1-specificity Sensitivity 1 SROC Curve 0,9 0,8 0,7 0,6 0,5 Table 1 provides a summary of the studies included in this meta-analysis. The five studies included 3,778 trauma patients in total (82.7% were male) with the mean age of 38. Four studies3,6,8,9 were cohort and one by Eryilmaz et al.7 was a cross-sectional study. Mechanism of trauma varied between studies with blunt trauma being the most prevalent (90.8%). The risk of bias assessment summary Figure 2. QUADAS-2 Summary is provided in Figure 2. The pooled sensitivity and specificity of TRISS to predict mortality (Figure 3) were 73% (95% CI = 69%–77%) and 67% (95% CI = 66%–69%), respectively. The pooled results for NISS (Figure 4) were 72% (95% CI = 68%–76%) for sensitivity and 80% (95% CI = 79%–81%) for specificity, whereas the pooled results for RTS (Figure 5) were 64% (95% CI = 60%–68%) for sensitivity and 83% (95% CI = 82%– 84%) for specificity. The pooled AUC of TRISS to predict mortality (Figure 3) (AUSROC = 95.4%) was significantly higher compared to NISS (Figure 4) (AUSROC = 87.3%; p = 0.01) and RTS (Figure 5) (AUSROC = 86.1%; p = 0.005). However, statistical significance was not observed when comparing the pooled AUCs of NISS and RTS (p = 0.54). References : 1. World Health Organization. The top 10 causes of death - Factsheet. WHO reports. 2018;(May 2018):6–13. 2. Galvagno SM, Massey M, Bouzat P, Vesselinov R, Levy MJ, Millin MG, et al. Correlation Between the Revised Trauma Score and Injury Severity Score: Implications for Prehospital Trauma Triage. Prehospital Emerg Care. 2019;23(2):263–70. 3. Lefering R. Development and validation of the Revised injury severity classification score for severely injured patients. Eur J Trauma Emerg Surg. 2009;35(5):437–47. 4. Orhon R, Eren ŞH, Karadayi Ş, Korkmaz I, Coşkun A, Eren M, et al. Comparison of trauma scores for predicting mortality and morbidity on trauma patients. Ulus Travma ve Acil Cerrahi Derg. 2014;20(4):258–64. Javali et al. 2019 Eryilmaz et al. 2009 Arikan et al. 2020 Lefering 2009 Tan et al. 2017 Sensitivity 0,3 1 RESULT 5. 6. 7. 8. 0,80 (0,73 - 0,86) SROC Curve 0,05 0,93 0,89 0,83 (0,01 - 0,13) (0,83 - 0,98) (0,87 - 0,91) (0,81 - 0,84) 0,2 0,9 0 0,2 0,4 0,3 Symmetric SROC AUC = 0,8614 SE(AUC) = 0,0152 Q* = 0,7921 SE(Q*) = 0,0147 0,2 0,8 0,1 0,4 0,6 Specificity 0,8 Pooled Specificity = 0,83 (0,82 to 0,84) Chi-square = 266,17; df = 4 (p = 0,0000) 1 Inconsistency (I-square) = 98,5 % 0,1 0 0 0,2 0,4 0,6 0,8 1 1-specificity Figure 5. Pooled Sensitivity, Specificity, and AUC of Revised Trauma Score (RTS) for Trauma Mortality Prediction. 00,7 0 0,2 0,4 DISCUSSION AND LIMITATION 1-specificity 0,6 0,8 1 Our study 0,6 showed a better accuracy for TRISS than NISS and RTS, as indicated by the pooled AUC, whereas the pooled specificity of TRISS was found to be lower than NISS and RTS. However, according to the Cochrane 0,5 Handbook10, estimating the pooled sensitivity and specificity may not be the best way to compare accuracies since cut-off values were different between studies. Therefore, estimating and comparing0,4 the summary ROCs (AUSROCs), as we did in our study, are more recommended and will provide more informative approach.10 TRISS also evaluates both anatomical and physiological factors in patients, while NISS and RTS assess only one of either. Scores which include both factors are useful in patient’s prognosis 0,3 in any type of trauma.11 There are several limitations in our study. First, the number of included studies is still limited. Second, there were differences regarding the types of trauma among the patients. 0,2 CONCLUSION TRISS has the 0,1 highest and better accuracy for predicting mortality in trauma patients compared to NISS and RTS, which suggests the utility of TRISS in clinical settings. Further studies are warranted to confirm our 29 findings and determine the best scoring system for patients based on their specific type of trauma. 0 0 0,2 0,4 0,8 1 population. Burn Trauma. 2017;5:1–6. 9. Tan0,6 JH, Tan HCL, Noh NAM, Mohamad Y, Alwi RI. Validation of the trauma mortality prediction scores from a Malaysian 10. Macaskill P, Gatsonis C, Deeks JJ, Harbord RM, Takwoingi Y. Chapter 10: Analysing and Presenting Results. In: Deeks JJ, Bossuyt PM, Gatsonis C (editors), Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy Version 1.0. The Cochrane Collaboration, 2010. Available from: http://srdta.cochrane.org/. 1-specificity 11. Beuran M, Negoi I, Pǎun S, Runcanu A, Gaspar B, Vartic M. Trauma Scores: A review of the literature. Chir. 2012;109(3):291–7. Zamora J, Abraira V, Muriel A, Khan K, Coomarasamy A. Meta-DiSc: A software for meta-analysis of test accuracy data. BMC Med Res Methodol. 2006;6:1–12. Javali RH, Krishnamoorthy, Patil A, Srinivasarangan M, Suraj, Sriharsha. Comparison of injury severity score, new injury severity score, revised trauma score and trauma and injury severity score for mortality prediction in elderly trauma patients. Indian J Crit Care Med. 2019;23(2):73–7. Eryilmaz M, Durusu M, Menteş Ö, Özer T, Killç S, Ersoy G, et al. Comparison of trauma scores for adults who fell from height as survival predictivity. Turkish J Med Sci. 2009;39(2):247–52. Arikan AA, Selçuk E, Bayraktar FA. Predicting outcomes of penetrating cardiovascular injuries at a rural center by different scoring systems. Brazilian J Cardiovasc Surg. 2020;35(2):198–205. Sy AU SE Q SE