Article

Comparison of modified Kampala trauma score with trauma mortality prediction model and trauma-injury severity score: A National Trauma Data Bank Study

a b s t r a c t

Background: Mortality prediction of trauma patients relies on anatomical, physiological or combined scores. The purpose of this study is to compare the diagnostic accuracy of the modified Kampala Trauma Score (M-KTS) with the Trauma Mortality Prediction Model (TMPM), and Trauma-Injury Severity Score (TRISS) using data from a large dataset from a developed registry, the National Trauma Data Bank .

Methods: Using 2011 and 2012 data from NTDB, patient based trauma scores (M-KTS, TMPM, and TRISS) were calculated and Predictive ability of M-KTS for mortality was compared with other trauma scores using receiver operating characteristics (ROC) curves.

Results: A total of 841 089 patients were included in the study. TRISS outperformed other scores (AUC = 0.922,

%95 CI 0.920-0.924) with M-KTS as the second best score (AUC = 0.901, %95 CI 0.899-0.903) followed by TMPM (AUC = 0.887, 95% CI 0.844-0.889). For blunt trauma, TRISS (AUC = 0.917, 95% CI 0.915-0.919) performed bet- ter than M-KTS (AUC = 0.891, %95 CI 0.889-0.893) and TMPM (AUC = 0.874, 95% CI 0.871-0.877). For penetrat- ing trauma, M-KTS (AUC = 0.956, 95% CI 0.954-0.959) and TMPM (AUC = 0.955, 95% CI 0.951-0.958) had

similar performance after TRISS (AUC = 0.969, 95% CI 0.967-0.971).

Conclusion: M-KTS performed worse than TRISS although its’ main advantage is simple use in resource-limited settings.

(C) 2017

Introduction

Injury severity is one of the key interests in trauma-related research, and mortality is the outcome of these studies. Use of injury severity scor- ing systems is essential in evaluating and benchmarking outcomes and for objective comparison of Trauma systems. Complex formulas requiring computational power may limit their daily clinical benefits. Trauma scores were developed to identify the impact of injury and quantitate the severity of injuries. Trauma registries have been used to develop and evaluate these trauma scores. They include hospital administrative datasets and regional or national trauma registries, as well as the Nation- al Trauma Data Bank (NTDB(R)), the largest trauma registry available.

Trauma scores can be categorized as anatomic, physiologic, and combined scores based on the method of calculation. The most com- monly used trauma score is the Injury Severity Score (ISS). As an

? No funding was provided for this study.

* Corresponding author at: Yenikale Mh. M. Seyfi Eraltay Sk No:20 D:5, Narlidere/ Izmir, 35320, Turkey.

E-mail addresses: [email protected] (S. Akay), [email protected] (A.M. Ozturk), [email protected] (H. Akay).

anatomic scoring system, ISS is composed of the sum the square of the three highest Abbreviated Injury Scale scores for the three most injured ISS regions [1]. Because of ISS’s imprecise ability to perfectly pre- dict mortality, several other anatomically based scoring systems were developed, but they failed to replace ISS [2,3]. To overcome this prob- lem, Glance et al. developed Trauma Mortality Prediction Model (TMPM) for anatomically based trauma assessment, and codes were de- veloped to predict mortality after trauma and published in the Interna- tional Classification of Diseases, Ninth Edition (ICD-9-CM) [4,5].

The Revised Trauma Score is a physiologic scoring system de-

veloped using a logistic regression method. It comprises three physio- logic values [Glasgow Coma Scale score, systolic blood pressure (SBP), and respiratory rate (RR)] [6]. Lack of anatomic scores and under-performance limited its’ common use. The Trauma and Injury Se- verity Score (TRISS) is a combined trauma score based on the use of co- efficients derived from the Major Trauma Outcome Study. It uses the RTS, ISS, and age index [7].

KTS is a combined trauma score and was created by Kobusingye and Lett [8]. The Injury surveillance system in Uganda (ICC) validated KTS for local use and as an alternative to other trauma scores, while re- searchers in several studies recommended it to be used as a predictor

http://dx.doi.org/10.1016/j.ajem.2017.02.035

0735-6757/(C) 2017

of mortality in resource-poor settings [8-10]. KTS is calculated with five categorical components: age, SBP, RR, neurologic status [alert, responsive to verbal stimulus, responsive to painful stimulus, unrespon- sive (AVPU)], and number of Serious injuries.

This study aims to compare the accuracy M-KTS to more established trauma scoring system using a large dataset from a developed world registry. In the present study, we compared M-KTS with more established Trauma scoring systems (TMPM), as well as with a combined score (TRISS), using a database derived from a population compiled in a developed database world registry (NTDB(R)), and we evaluated M-KTS’s Prognostic accuracy. We hypothesized that M-KTS could serve as a valuable predictor of mortality compared with other trauma scores and sought to compare the diagnostic accuracy of M-KTS using the largest available database of trauma patients.

Patients and methods

This study included patients in NTDB(R) registry of 2011 and 2012. Analysis of NTDB(R) was approved by the American College of Surgeons NTDB(R) committee. NTDB(R) is defined as the largest aggregated U.S. trauma registry ever assembled. The dataset included 1 620 156 patients hospitalized after trauma over a 2-year period. Data used in the study and calculation of trauma scores included demographic char- acteristics, type and mechanism of injury (based on ICD-9-CM codes), vital signs, GCS values, AIS and ISS codes, and patient outcomes (defined as survival to hospital discharge). Patients without AIS and ISS codes, those with burns or non-traumatic injuries (e.g., poisoning, drowning, and suffocation), patients with missing or invalid data (data missing on age, gender, outcome), and patients younger than 1 year were ex- cluded. Patients with missing values or with incalculable scores for any of the five categories because of missing components (age, SBP, RR, AIS and ISS scores, and GCS) were excluded. Patients who were dead on arrival to the emergency department (ED), transferred to an- other hospital, or left against medical advice were also excluded. We also included patients treated at trauma centers with a caseload of at least 500 patients per year to robust patient care with trauma registry systems. ICD-9-CM codes used for injury type define injuries as either blunt or penetrating trauma and according to mechanism of injury (pedestrian, bicycle, motorcycle, motor vehicle collision, fall, stabbing, and gunshot wounds). The final dataset included patients with all valid trauma scores and outcomes.

AIS is an anatomic scoring system developed by the Association for

the Advancement of Automotive Medicine to classify and describe inju- ries. Since its introduction in 1969, seven major updates have been pub- lished. Beside its extensive use for classification, AIS severity is assessed by assigning scores ranging from 1 for Minor injury to 6 for maximum (fatal) injury for one of six body regions. Because AIS is a measurement system for single injuries and lacks an aggregation function, Baker et al. developed an AIS-derived ISS for use as an overall score to describe the severity of multiple injuries [11]. ISS comprises the sum of the square of the highest AIS in three regions of the five regions (head-neck, face, tho- rax, abdomen-pelvic contents, extremities-pelvic girdle and external). If any of the scores of the six sub-regions is 6, the ISS score is set to 75. The NTDB(R) contains a precalculated ISS along with ICD-9 codes with AIS 1998 revision (AIS 98). Precalculated ISS scores were supplied by the contributing trauma centers, and they were used in this study. AIS codes were globally mapped to AIS 98. If the hospital did not submit a corresponding AIS code, the ISS was based on the AIS derived using ICDMAP-90 software. Four types of ISS were in the NTDB(R); and the precalculated ISS submitted by the trauma center to NTDB(R) was used in this study. The other three methods of ISS calculation–ISS derived from the AIS score submitted by the hospital, ISS derived from mapping of existing AIS codes to AIS 98, and ISS derived from AIS score calculated

using ISS/AIS mapping–were excluded.

TMPM based on ICD-9 (TMPM-ICD-9) was calculated by using the method described by Glance et al. [4]. The five worst injuries of patients

are taken into consideration, and the Probability of death is calculated with a two-stage approach. STATA routine (Tmpm.ado) was provided by its developers and is available on the Internet. TMPM-ICD-9 matches injuries to one of six precalculated model-averaged regression coeffi- cients (MARC), and probability of death is a product of the cumulative inverse normal function of the sum of the five highest MARC values multiplied by the model coefficients [4]. To achieve consistency of trauma scoring, TMPM-ICD-9 was calculated by using the AIS codes of each trauma patient submitted by the corresponding trauma center.

M-KTS was calculated by using the method described by Kobusingye and used the original formula [8]. We used categorical components for calculating an overall score: age, SBP, respiratory change, neurologic status, and number of serious injuries (AIS score greater than or equal to 3). The neurologic status components of the AVPU score are alert, re- sponsive to verbal stimulus, responsive to painful stimulus, and unre- sponsive; however, since NTDB(R) registry records report patients’ more commonly accepted GCS for neurologic status, we converted NTDB(R) registry GCS records to corresponding AVPU scores using a step- wise method described in a recent study by Weeks et al. [12]. Patient with a motor response score of 6 (obeys commands) and an eye re- sponse score of 4 (spontaneously opens eyes) was considered “alert.” An eye response score of 3 (opens eyes to verbal commands) was con- sidered “responds to verbal stimulus.” A motor response from 2 to 5 (extension to painful stimulus =2, localized painful stimulus =5) or an eye response score of 2 (opens eyes in response to painful stimulus) was considered “responds to painful stimulus.” Motor and eye re- sponses of 1 (no motor or eye response to painful stimulus) were con- sidered “unresponsive.” The total M-KTS score ranges from 0 to 10.

The TRISS is a combined trauma score that uses different coefficients for blunt or penetrating trauma derived from the Major Trauma Outcome Study [7]. It uses RTS, ISS, and age index (which uses catego- rized age values) to predict Ps with the equation Ps = 1/(1 + e-b), where b is calculated using b = b0 + b1 (RTS) + b2 (ISS) + b3 (age index). The RTS was calculated by using the function RTS = (0.9368 x GCS) + (0.7326 x SBP) + (0.2908 x RR) [6].

Probability of death using TRISS, TMPM, and M-KTS was assessed by using the receiver operating characteristic (ROC) method to evaluate how each score discriminates between survivors and non-survivors. An ROC curve plots the false-positive rate (100-specificity) on the x-axis and the true-positive rate (sensitivity) on the y-axis. A random guess would lie on the line drawn from the left lower corner to the right upper corner, and a perfect discrimination passes through the upper left corner [13].

Patients were categorized further by race and/or ethnicity (Cauca- sian, African-American, Asian, American Indian, Pacific Islander, or other) and by insurance status as privately insured (Blue Cross/Blue Shield, other Commercial insurance), publicly insured (Medicaid, Medi- care, other government insurance), uninsured (including self-pay), and other forms. The mechanism of injury was classified as blunt, penetran, or other form according to the International Classification of External Causes of Injury published by the World Health Organization as part of its Family of International Classifications [14]. Mortality was the primary outcome, and patients who had “survival to discharge” were included in this study. Mortality was reported with a separate ROC curve, and Area under curve calculations were performed for the overall population; by injury type (blunt or penetrating); according to ISS severity (minor = 1-8, moderate = 9-15, severe = 16-24, and very severe =25 or higher); and by age group, defined as children (1-17 years), young (18-64 years), and elder (>= 65 years). This severity classification was commonly used in previous studies [5].

Results

After all exclusion criteria were applied, among the 1 620 156 patients reported by 840 trauma centers, 841 089 (51.9%) met the study inclusion criteria, and the total dataset included patients with

Table 1

Patient characteristics

n Percentage

trauma. The mean TMPM, M-KTS, and TRISS scores for blunt trauma pa- tients were 0.05, 8.49, and 0.96, respectively, whereas the correspond- ing scores for penetrating trauma patients were 0.06, 8.76, and 0.95,

Male Gender 529 339 62.93

Age Categories

respectively.

The ability of each trauma scoring method to discriminate survivors

1-17

105 933

12.59

from non-survivors is shown in Table 2. When the scoring methods

18-64

517 773

61.56

were compared among the entire population, TRISS had the highest

AUC and M-KTS had the second highest, followed by TMPM. For age groups, TRISS outperformed other scores, with TMPM outperforming M-KTS. For the injury types (blunt or penetrating), regardless of injury severity, TRISS had a higher AUC for both blunt and penetrating trauma types, but for Blunt injuries, M-KTS had a higher AUC than TMPM. No difference between TMPM and M-KTS was found for Penetrating injuries.

>= 65

Race

217 383

25.85

American Indian

4876

0.58

Asian

13 205

1.57

African American

113 579

13.50

Pacifician

1842

0.22

Caucasian

598 771

71.19

Other

108 816

12.94

Insurance

Commercial

315 614

37.52

Goverment

306 812

36.48

Uninsured

121 138

14.40

Other

50 397

5.99

Unknown

47 128

5.60

Mechanism of Injury

Fall

353 790

42.06

Motor Vehicle Collusion

205 749

24.46

Pedestrian

38 835

4.61

Bicycle

25 528

3.04

Motorcycle

52 171

6.20

Stabbing

33 927

4.03

Gunshot

37 623

4.47

Other

93 466

11.11

Injury Type Blunt

760 075

90.37

Penetrating

81 014

9.63

Injury Severity ISS 1-8

375 296

44.62

ISS 9-15

295 614

35.15

ISS 16-24

107 228

12.75

ISS >= 25

62 951

7.48

For injury severity categorized by ISS ranges, TRISS had better AUCs for minor (ISS 1-8), moderate (ISS 9-15), severe (ISS 16-24) and very severe (ISS >= 25) injuries. For minor, moderate and severe injuries M- KTS predicted mortality better than TMPM while for very severe injuries TMPM had higher AUC than M-KTS.

Discussion

3 051 342 instances of complete injury scoring data and mortality. Some patients were excluded by meeting more than one criterion. Table 1 lists the demographic characteristics of the patients. The mean (SD) age of the population was 45.8 years, and men accounted for 62.9% of the pop- ulation. Among the patients, 4.1% were b 18 years of age, whereas 61.6% were between 18 and 65 years and 25.8% were more than N 65 years of age. Caucasian patients accounted for 71.1% of the population. Unin- sured patients comprised 14.4% of the sample. The most common cause of trauma was falls (42.1% patients), followed by motor vehicle crashes (24.5%) and penetrating injuries (8.5%). The overall mortality rate was 3.0%, and the rate was higher for patients with penetrating in- juries than in those who sustained blunt traumas (4.6% vs 2.8%, respec- tively; p b 0.001).

Among the 25 344 patients who died, 21 595 (85.2%) had experi- enced blunt trauma and 3749 (14.8%) had sustained penetrating

In this study, we compared the discriminative ability of M-KTS with two commonly and effective used measures: anatomic scores (TMPM), and combined score (TRISS). For the entire population, in trauma groups categorized by age, injury type, and injury severity, TRISS outperformed other scores, followed by TMPM and M-KTS. Among all three trauma scores, M-KTS failed to predict mortality better than the TRISS.

As the first and most commonly used trauma score, ISS has been in- cluded in trauma registry systems and employed as a common predictor of mortality, as well as for benchmarking outcomes. Also, as a method used to calculate TRISS, its use had been a milestone in trauma score-re- lated research. Relatively lower predictability compared with trauma scores developed after its creation has made it a less preferred measure- ment tool, leading to development of the New Injury Severity Score [15]. Since the development of the New Injury Severity Score, numerous studies have showed that it outperforms the former ISS but has not re- placed ISS’s popularity.

RTS and TRISS were developed from regression models and were proven to be predictors of mortality. Being a solely physiologic parame- ter-based scoring method, RTS failed to be superior to TRISS, probably due to lack of an anatomic profile of injuries. TMPM, developed using a more complex model of regression analysis, takes account of injury se- verity only by using the anatomic profile of injuries, failing to consider how the trauma patient is reacting to injuries, or in other words “how sick the patient is.” For these reasons, combined trauma scores predict mortality better than anatomically based scores. KTS was developed to

Table 2

Comparison of Area Under Curve (AUC) for the Receiver Operating Characteristics (ROC) for each trauma score

AUC for each trauma score with 95% CI

Population

n

TMPM

TRISS

M-KTS

All

841 089

0.887 (0.884-0.889)

0.922 (0.920-0.924)

0.901 (0.899-0.903)

Age groups

1-17

105 933

0.981 (0.976-0.985)

0.987 (0.983-0.990)

0.978 (0.974-0.981)

18-64

517 773

0.932 (0.929-0.934)

0.938 (0.936-0.941)

0.931 (0.929-0.933)

>= 65

217 383

0.818 (0.813-0.822)

0.839 (0.835-0.844)

0.813 (0.809-0.817)

Injury Type

Blunt

760 075

0.874 (0.871-0.877)

0.917 (0.915-0.919)

0.891 (0.889-0.893)

Penetran

81 014

0.955 (0.951-0.958)

0.969 (0.967-0.971)

0.956 (0.954-0.959)

ISS

ISS 1-8

375 296

0.672 (0.657-0.686)

0.841 (0.831-0.850)

0.804 (0.793-0.814)

ISS 9-15

295 614

0.662 (0.654-0.670)

0.793 (0.786-0.799)

0.762 (0.755-0.769)

ISS 16-24

107 228

0.715 (0.707-0.723)

0.812 (0.806-0.819)

0.800 (0.794-0.807)

ISS >= 25

62 951

0.804 (0.800-0.808)

0.816 (0.813-0.820)

0.791 (0.787-0.795)

be a combined score and was found to perform as the worse performing score in the largest group of patients, between ages 18 and 65 years.

More information, in both volume and scope, of trauma patients helps to improve prediction models. TRISS was developed as a com- bined score using ISS and RTS, comprising physiologic parameters (GCS, RR, and SBP), and it has been shown to be a robust method [7]. An explanation for this outcome is that it takes into account physiologic acuity (“how sick the patient is”) based on the patient’s vital signs, as well as the anatomic traumatic profile based on ISS. Non-inclusion of vital signs (e.g., RTS) with the anatomic profile (e.g., AIS, ISS, or TMPM) or vice versa may explain the underperformance of trauma scoring.

A limitation of this study is that it is based on a retrospective analysis of a trauma registry rather than a Prospective data collection. Data were collected using standards defined by the American College of Surgery. Data included required parameters for trauma scores, whereas cases with missing data required for calculation were excluded from this anal- ysis using listwise deletion. A similar percentage of excluded cases was observed in the study of Turner et al. which authors build and tested the TMPM score [4]. Since NTDB data is entered by trauma registrars and manipulation of data by researchers is not appropriate, such a high per- centage of missing data is unavoidable. The data we used for analysis were derived from selected trauma centers with a volume of at least 500 cases, and missing data were not re-appended. A relatively higher number of excluded cases due to missing cases can be questionable but although given the high number of missing cases and possible cause of missing data are due to missing complete on random, we be- lieve such high percentage of missing data don’t affect the final analysis. Also, patients admitted to non-trauma centers were not included. Using dedicated trauma centers is a robust method and helps to avoid the un- desirable outcome of mortality. Prospective population-based cohort studies that include trauma patients can overcome this problem.

The trauma score of interest in this study, M-KTS, is analogous but not equal to KTS, which the GCS is converted to AVPU score because of lack of AVPU scores in NTDB. AVPU score is another and easier method of evaluating neurological status of Trauma victims not for physicians but for other health professionals. Using M-KTS with GCS component instead of AVPU score is an alternative to KTS but further studies is needed for evaluating variability between KTS and M-KTS.

Although KTS was developed and suggested to be used in resource- limited settings, quantitative definition of “serious injury” is lacking in the original study [8]. Subsequent studies of KTS described serious inju- ries as either AIS score more or equal to than two or three [10,12]. Thus calculation of KTS differs from previous studies. AIS manual published by American Association of Automative Association describes serious injury with a score of three or more [16]. TRISS, which uses AIS and AIS derived, relies on an injury scale of one to six, M-KTS broadly groups anatomic injuries as serious or minor but this too depends on AIS group- ing. A simplified version of AIS lexicon for M-KTS injury severity classi- fication can favor M-KTS. Such M-KTS calculation is compelling without a lexicon which limits its’ use for both attending physicians and triage personnel.

Like ISS and ISS-based TRISS, M-KTS uses the single worst injury and the three most severely injured regions of trauma and does not take into account the second worst injury from the same region or more than one injury from the same region. TMPM was developed to address this issue. It employs a more complex probit regression method, and a computa- tion is needed to calculate the probability of death using either a statis- tical software program or a spreadsheet program. Although M-KTS was suggested for its ability to be used as a triage tool without need for a lexicon or personnel finance and computational skills, unlike ISS, TMPM, RTS, and TRISS, and its’ use a potentially valuable scoring system for EDs is compulsive for previously mentioned reasons. ROC values of M-KTS for the overall population, as well as for age groups and injury

types, were higher than those reported in the literature, but they tended to be lower for injury severity, which limits its utility for injury severity- based analysis and its suitability [12]. Future prospective studies need to address this problem regarding possible use of M-KTS in scoring for ap- propriate classification of injuries, except very severe penetrating inju- ries (ISS of 25 or higher), where RTS is superior. Also inclusion of physiologic parameters to the “best” anatomic score TMPM can achieve a better scoring system.

Trauma scoring systems are used for quality improvement and trau- ma research based studies and still a perfect scoring system does not exist. Distinct source, impact on body, pathophysiology, treatment, acute and long term consequences of trauma are Potential causes of de- velopment of a perfect scoring system. M-KTS is another scoring system developed for this purpose. KTS was developed by Kobusingye et al. to be used for assessment of injury severity by first-line health profes- sionals in low-resource settings, and focused training resulted in high intraclass correlation [8]. Not very ease of calculation and use in patient care are major limiting factors for first-line and ED professionals but suitable for quality improvement and research; M-KTS can succeed achieving this goal.

Acknowledgements

No funding was provided for this study. We would like to thank Dr. Olive C. Kobusingye, Dr. Alan Cook and Dr. Turner Osler for their invaluable help and expertise.

Competing interests.

Authors declare no conflicts of interests.

References

  1. [No authors listed]Rating the severity of tissue damage. I. The abbreviated scale.

    JAMA. 1971;215:277-80

    Osler T, Baker SP, Long W. A modification of the injury severity score that both im- proves accuracy and simplifies scoring. J Trauma 1997;43:922-5.

  2. Osler T, Rutledge R, Deis J, et al. ICISS: an international classification of disease-9 based injury severity score. J Trauma 1996;41:380-6.
  3. Glance LG, Osler TM, Mukamel DB, et al. TMPM-ICD9: a trauma mortality prediction model based on ICD-9-CM codes. Ann Surg 2009;249:1032-9.
  4. Haider AH, Villegas CV, Saleem T, et al. Should the ICD-9 trauma mortality prediction model become the new paradigm for benchmarking trauma outcomes? Trauma Acute Care Surg 2012;72:1695-701.
  5. Champion HR, Sacco WJ, Copes WS, et al. A revision of the trauma score. J Trauma 1989;29:623-9.
  6. Boyd CR, Tolson MA, Copes WS. Evaluating trauma care: the TRISS method. Trauma Score and the Injury Severity Score J Trauma 1987;27:370-8.
  7. Kobusingye OC, Lett RR. Hospital-based trauma registries in Uganda. J Trauma 2000; 48:498-502.
  8. Weeks SR, Juillard CJ, Monono ME, et al. Is the Kampala trauma score an effective predictor of mortality in low-resource settings? A comparison of multiple trauma severity scores. World J Surg 2014;38:1905-11.
  9. Macleod JBA, Kobusingye O, Frost C, et al. A comparison of Kampala trauma score (KTS) with the revised trauma score (RTS), injury severity score (ISS) and TRISS method in a Ugandan trauma registry. European Journal of Trauma 2003;6:392-8.
  10. Baker SP, O’Neill B, Haddon W. et al. The Injury Severity Score: a method for describ- ing patients with multiple injuries and evaluating emergency care J Trauma 1974; 14:187-96.
  11. Weeks SR, Stevens KA, Haider AH, et al. A modified Kampala trauma score (KTS) ef- fectively predicts mortality in trauma patients. Injury 2016;47:125-9.
  12. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristics (ROC) curve. Radiology 1982;143:29-36.
  13. International Classification of External Causes of Injuries (ICECI) Version .12 Webpage: http://www.rivm.nl/who-fic/ICECI/ICECI_1-2_2004 July.pdf Accessed 26 05 2015
  14. Lavoie A, Moore L, LeSage N, et al. The new injury severity score: a more accurate predictor of in-hospital mortality than the injury severity score. J Trauma 2004;56: 1312-20.
  15. Thomas A. Gennarelli, Elaine Wodzin. The Abbreviated Injury Scale 2005. Up- date 2008. American Association for Automotive Medicine (AAAM), Des Plaines, IL 2008.

Leave a Reply

Your email address will not be published. Required fields are marked *