Prediction of Trauma Mortality Incorporating Pre-injury Comorbidities Into Existing Mortality Scoring Indices

Abstract

Objectives

The purpose of the study is to develop a comprehensive risk score of mortality in trauma victims that can predict the in-hospital mortality better than trauma injury severity score (TRISS) and A Severity Characterization of Trauma (ASCOT) score.

Methods

All hospitalized trauma patients, between the ages of 16 and 89 years old were included in the study. The National Trauma Quality Improvement Program (TQIP) database of the calendar year 2011-2016 was accessed for the development of a traum mortality scoring system (TMS). The prediction of mortality was tested by creating a receiver operating characteristics (ROC) curve and an area under the curve (AUC). ROCs and AUCs of TMS were compared with TRISS and ASCOT score.

Results

The AUC of TMS (0.892, 95% CI: 0.888-0.896) was better than TRISS (0.864, 95% CI: 0.859-0.869, P <0.0001) and ASCOT (0.841, 95% CI: 0.835-0.846, P <0.0001), respectively, in blunt injury. Similarly, TMS prediction (AUC: 0.949, 95% CI: 0.940-0.958) was better in penetrating injury when compared with TRISS (0.942, 95% CI: 0.934-0.951, P = 0.030) and ASCOT (0.924, 95% CI: 0.912-0.936, P <0.0001), respectively.

Conclusion

TMS can predict the in-hospital mortality better than TRISS and ASCOT.

Keywords

multiple trauma risk score mortality

Introduction

Trauma is the leading cause of death and disability among young adults.¹ Globally 5.8 million people die due to injury each year representing about 10% of all deaths.² In the year 2013, the cost associated with the care of injured patients was $671 billion in the United States alone.³ Almost one third of that cost was associated with caring of fatal injuries.³ The development of trauma system and specialized care at trauma centers has reduced the incidence of mortality.^4,5 Early identification of high-risk patients, proper resource allocations, and close monitoring can reduce the overall mortality and morbidity.⁶ Based on the mortality risk of the patient, early palliative care consult can be initiated in some cases.⁷

Different scoring system and predictor models for mortality/survivability of trauma patients have been reported in the literature. The two most widely cited trauma scores, revised trauma score (RTS) and trauma injury severity score (TRISS), were developed a few decades ago.^8,9 About 10 years later after developing RTS, the same group came up with another score, A Severity Characterization of Trauma (ASCOT) score that showed a better prediction of mortality when compared to TRISS score in blunt injury.¹⁰ These scores include some physiological parameters, anatomical site injury, and age factors and are mostly used for triaging the patients at the scene and transferring to trauma center vs non-trauma center. Sartorius et al¹¹ in 2010 developed another scoring system, known as MGAP score, to predict the in-hospital mortality by adding mechanism of injury to the previously reported scoring system component, age, Glasgow coma scale [GCS], and blood pressure.¹¹ A modified version of the MGAP score, known as GAP, was created that accounts for GCS, age, blood pressure, and has shown better accuracy of predicting in-hospital mortality.¹² All previously reported trauma mortality score systems lack the comorbidities in their predictor models,^8-13 although comorbidities are one of the factors affecting trauma mortality.¹⁴ Excluding the pre-existing condition from the model may have underestimated the in-hospital mortality.

To address this issue, we designed the study to identify the risk factors of in-hospital mortality using National Trauma Quality Improvement Program (TQIP) database, including patient’s demography, mechanism of injury, injury severity, systolic blood pressure (SBP), and comorbidities. For ease of calculating the predictability of mortality in trauma patients for point of care physician, we have created a Trauma Mortality Score (TMS) calculator that will be available to all providers through a web-based link. Our hypothesis is TMS will predict the in-hospital mortality better than TRISS and ASCOT.

Methods

Data Source

The TQIP database of the calendar years 2011-2016 was accessed for the study. The database is one of the largest trauma quality program database in the United States. The database is maintained by the American College of Surgeon (ACS). Individual trauma centers’ performance, mortality and morbidity, is compared by the rest of the trauma centers in the country. The results of the performance are shared with the individual center twice a year.

Patient’s Characteristics

All trauma patients, between the ages of 16 and 89 years old, who sustained injury with an injury severity score (ISS) ≥1 and admitted to the hospital, were included in the study. Other characteristics included in the study are sex, race, mechanism of injury, initial SBP mmHg, hypotension, defined as (SBP<90 mmHg), injury severity score (ISS) and Glasgow coma scale (GCS) and patients’ comorbidities, chronic alcoholism, ascites, coagulopathy, use of chemotherapy, congestive heart failure (CHF), chronic renal failure (CRF), on dialysis, stroke, diabetes mellitus (DM), disseminated cancer, functional dependency (from completely dependency to complete independency), myocardial infarction (MI) within last month, peripheral vascular disease (PVD), hypertension (HTN) on medication, obesity, chronic pulmonary obstructive disease (COPD), use of steroid, cirrhosis, and dementia.

Outcomes

The primary endpoint of the study was to identify the risk of in-hospital mortality, validate the results, and develop a simple scoring system for the point of care physicians for better resource utilization.

Statistics

Patient demographic information was summarized using summary statistics (median with interquartile range (IQR) [first quartile – third quartile] for continuous variables, and frequency and percentage for categorical variables). To compare the groups, the Wilcoxon rank-sum test was used for continuous variables, and the chi-square test was used for the categorical variables. The normality of data was tested using the Anderson-Darling test. Initially, all patients’ characteristics mentioned above were used for performing the risk assessment. A multiple logistic regression model was used to assess the chance of having mortality with those specified variables. The backward elimination approach was used for the final selection of the model at significance level of .05. The receiver operating characteristics (ROC) curve was constructed, and the corresponding area-under-the curve (AUC) was calculated. The parameter estimates from the fitted model were summarized using β coefficient estimates and 95% confidence intervals (CIs), and 95% CIs for the odds ratio (OR) as measures of precisions. For the purpose of developing and validating the risk assessment model, we divided the data into two data sets: the training data set, which contains 80% of the randomly selected data, and the testing data set, which contains the other 20% of the data. The 2-sided P-value was reported for each test. A trauma mortality scoring system (TMS) has been created from the β coefficient of the selected variables for the point of care physicians. The ROC and AUC of TMS were compared with TRISS and ASCOT scores for blunt and penetrating injury using the DeLong’s test.^15-18 The TMS calculator has been developed from the beta regression coefficients. The sum of the coefficients was inputted into a formula, which was then converted into a percentage to output the risk of mortality following a traumatic injury. A P-value of <.05 is considered an indication of statistical significance. Statistical analysis was performed using the R language.¹⁹

Results

Univariate Analysis

There were 767,115 patients qualify for the study. Of that total, 34,320 (4.5%) patients died. The patients who died were significantly older, predominantly male and Caucasian. The mortality group suffered from higher severity of injury and presented with severely compromised neurological status. Higher comorbid conditions were found in patients who died compared to who survived (Table 1).

Table 1.

Univariate Analysis of Trauma Patients Who Died and Who Survived.

Variable	Value	All Patients (n = 767,115)	Survived (n = 732,795)	Died (n = 34,320)	P-Value
Age (years)	Median [Q1-Q3]	59 [ 40-75 ]	58 [ 39-74 ]	68 [ 52-80 ]	<.001
Age category (years)
	16-44	228,611 (29.8)	222,515 (30.4)	6,096 (17.8)	<.001
	45-54	103,705 (13.5)	100,025 (13.6)	3,680 (10.7)
	55-64	122,891 (16)	117,688 (16.1)	5,203 (15.2)
	65-74	115,967 (15.1)	109,772 (15)	6,195 (18.1)
	75-89	195,941 (25.5)	182,795 (24.9)	13,146 (38.3)
Sex					<.001
	Female	292,576 (38.1)	281,446 (38.4)	11,130 (32.4)
	Male	474,539 (61.9)	451,349 (61.6)	23,190 (67.6)
Race					<.001
	American Indian	6,995 (.9)	6,724 (.9)	271 (.8)
	Asian	12,812 (1.7)	12,045 (1.6)	767 (2.2)
	Black or African American	96,291 (12.6)	92,952 (12.7)	3,339 (9.7)
	Native Hawaiian or other Pacific Islander	1,553 (.2)	1,492 (.2)	61 (.2)
	Other race	50,168 (6.5)	48,031 (6.6)	2,137 (6.2)
	White	599,296 (78.1)	571,551 (78)	27,745 (80.8)
ISS	Median [Q1-Q3]	10 [ 9-17 ]	10 [ 9-17 ]	25 [ 16-29 ]	<.001
ISS category					<.001
	1-14	500,185 (65.2)	492,744 (67.2)	7,441 (21.7)
	15-24	172,744 (22.5)	166,059 (22.7)	6,685 (19.5)
	25-34	76,166 (9.9)	61,087 (8.3)	15,079 (43.9)
	35-45	14,407 (1.9)	10,712 (1.5)	3,695 (10.8)
	46-75	3,613 (.5)	2,193 (.3)	1,420 (4.1)
Mechanism					<.001
	All—others	2,769 (.4)	2,648 (.4)	121 (.4)
	Bicycle hit by	14,907 (1.9)	14,519 (2)	388 (1.1)
	fall	384,504 (50.1)	365,656 (49.9)	18,848 (54.9)
	GSW	35,333 (4.6)	32,384 (4.4)	2,949 (8.6)
	Machinery	4,127 (.5)	4,019 (.5)	108 (.3)
	Motorcycle	45,957 (6)	44,299 (6)	1,658 (4.8)
	MVT	161,523 (21.1)	155,258 (21.2)	6,265 (18.3)
	Other transport	69,521 (9.1)	67,865 (9.3)	1,656 (4.8)
	Pedestrian hit by	28,905 (3.8)	26,836 (3.7)	2,069 (6)
	SW	19,569 (2.6)	19,311 (2.6)	258 (.8)
GCS	Median [Q1-Q3]	15 [ 15-15 ]	15 [ 15-15 ]	7 [ 3-15 ]	<.001
GCS category					<.001
	03-08	61,083 (8)	42,908 (5.9)	18,175 (53)
	09-12	22,264 (2.9)	19,843 (2.7)	2,421 (7.1)
	13-15	683,768 (89.1)	670,044 (91.4)	13,724 (40)
Hypotension	No	745,154 (97.1)	714,830 (97.5)	30,324 (88.4)	<.001
	Yes	21,961 (2.9)	17,965 (2.5)	3,996 (11.6)
Comorbidities
Chronic alcoholism	No	670,662 (87.4)	640,795 (87.4)	29,867 (87)	.022
	Yes	96,453 (12.6)	92,000 (12.6)	4,453 (13)
Ascites	No	766,499 (99.9)	732,296 (99.9)	34,203 (99.7)	<.001
	Yes	616 (.1)	499 (.1)	117 (.3)
Coagulopathy	No	689,645 (89.9)	662,095 (90.4)	27,550 (80.3)	<.001
	Yes	77,470 (10.1)	70,700 (9.6)	6,770 (19.7)
On chemotherapy	No	763,727 (99.6)	729,736 (99.6)	33,991 (99)	<.001
	Yes	3,388 (.4)	3,059 (.4)	329 (1)
CHF	No	730,765 (95.3)	699,785 (95.5)	30,980 (90.3)	<.001
	Yes	36,350 (4.7)	33,010 (4.5)	3,340 (9.7)
CRF	No	754,304 (98.3)	721,290 (98.4)	33,014 (96.2)	<.001
	Yes	12,811 (1.7)	11,505 (1.6)	1,306 (3.8)
Dialysis	No	766,268 (99.9)	732,054 (99.9)	34,214 (99.7)	<.001
	Yes	847 (.1)	741 (.1)	106 (.3)
CVA	No	739,854 (96.4)	707,465 (96.5)	32,389 (94.4)	<.001
,	Yes	27,261 (3.6)	25,330 (3.5)	1,931 (5.6)
DM	No	628,661 (82)	602,498 (82.2)	26,163 (76.2)	<.001
	Yes	138,454 (18)	130,297 (17.8)	8,157 (23.8)
Disseminated cancer	No	759,590 (99)	726,004 (99.1)	33,586 (97.9)	<.001
	Yes	7,525 (1)	6,791 (.9)	734 (2.1)
Functionally dependent	No	731,723 (95.4)	699,348 (95.4)	32,375 (94.3)	<.001
	Yes	35,392 (4.6)	33,447 (4.6)	1,945 (5.7)
MI	No	753,223 (98.2)	719,985 (98.3)	33,238 (96.8)	<.001
	Yes	13,892 (1.8)	12,810 (1.7)	1,082 (3.2)
PVD	No	761,137 (99.2)	727,275 (99.2)	33,862 (98.7)	<.001
	Yes	5,978 (.8)	5,520 (.8)	458 (1.3)
HTN	No	419,061 (54.6)	401,993 (54.9)	17,068 (49.7)	<.001
	Yes	348,054 (45.4)	330,802 (45.1)	17,252 (50.3)
Obesity	No	721,273 (94)	689,069 (94)	32,204 (93.8)	.132
	Yes	45,842 (6)	43,726 (6)	2,116 (6.2)
COPD	No	735,017 (95.8)	702,530 (95.9)	32,487 (94.7)	<.001
	Yes	32,098 (4.2)	30,265 (4.1)	1,833 (5.3)
Chronic steroid use	No	759,962 (99.1)	726,024 (99.1)	33,938 (98.9)	<.001
	Yes	7,153 (.9)	6,771 (.9)	382 (1.1)
Cirrhosis	No	757,981 (98.8)	724,991 (98.9)	32,990 (96.1)	<.001
	Yes	9,134 (1.2)	7,804 (1.1)	1,330 (3.9)
Dementia	No	730,907 (95.3)	698,668 (95.3)	32,239 (93.9)	<.001
	Yes	36,208 (4.7)	34,127 (4.7)	2,081 (6.1)
Abbreviated injury scale of body regions
Abdomen	0	671,131 (87.5)	642,790 (87.7)	28,341 (82.6)	<.001
	1	18,059 (2.4)	17,270 (2.4)	789 (2.3)
	2	30,777 (4)	28,895 (3.9)	1,882 (5.5)
	3	30,034 (3.9)	28,445 (3.9)	1,589 (4.6)
	4	13,031 (1.7)	11,918 (1.6)	1,113 (3.2)
	5	4,083 (.5)	3,477 (.5)	606 (1.8)
Brain	0	480,126 (62.6)	467,762 (63.8)	12,364 (36)	<.001
	1	7,425 (1)	7,326 (1)	99 (.3)
	2	51,482 (6.7)	50,485 (6.9)	997 (2.9)
	3	104,195 (13.6)	100,530 (13.7)	3,665 (10.7)
	4	87,018 (11.3)	81,427 (11.1)	5,591 (16.3)
	5	36,869 (4.8)	25,265 (3.4)	11,604 (33.8)
Face	0	623,654 (81.3)	598,055 (81.6)	25,599 (74.6)	<.001
	1	82,291 (10.7)	78,103 (10.7)	4,188 (12.2)
	2	55,658 (7.3)	51,654 (7)	4,004 (11.7)
	3	5,265 (.7)	4,795 (.7)	470 (1.4)
	4	247 (0)	188 (0)	59 (.2)
Neck	0	752,954 (98.2)	719,759 (98.2)	33,195 (96.7)	<.001
	1	5,229 (.7)	4,941 (.7)	288 (.8)
	2	3,411 (.4)	3,131 (.4)	280 (.8)
	3	4,625 (.6)	4,215 (.6)	410 (1.2)
	4	763 (.1)	647 (.1)	116 (.3)
	5	133 (0)	102 (0)	31 (.1)
Pelvis	0	730,740 (95.3)	698,491 (95.3)	32,249 (94)	<.001
	2	30,136 (3.9)	28,362 (3.9)	1,774 (5.2)
	3	6,239 (.8)	5,942 (.8)	297 (.9)
Spine	0	743,880 (97)	710,797 (97)	33,083 (96.4)	<.001
	1	6,207 (.8)	6,014 (.8)	193 (.6)
	2	972 (.1)	964 (.1)	8 (0)
	3	7,534 (1)	7,159 (1)	375 (1.1)
	4	5,099 (.7)	4,871 (.7)	228 (.7)
	5	3,423 (.4)	2,990 (.4)	433 (1.3)
Thorax	0	520,347 (67.8)	498,065 (68)	22,282 (64.9)	<.001
	1	19,238 (2.5)	18,253 (2.5)	985 (2.9)
	2	33,729 (4.4)	32,229 (4.4)	1,500 (4.4)
	3	165,810 (21.6)	159,171 (21.7)	6,639 (19.3)
	4	23,239 (3)	21,103 (2.9)	2,136 (6.2)
	5	4,752 (.6)	3,974 (.5)	778 (2.3)
Extremities	0	436,277 (56.9)	412,591 (56.3)	23,686 (69)	<.001
	1	57,056 (7.4)	54,179 (7.4)	2,877 (8.4)
	2	85,146 (11.1)	81,797 (11.2)	3,349 (9.8)
	3	187,343 (24.4)	183,032 (25)	4,311 (12.6)
	4	1,274 (.2)	1,182 (.2)	92 (.3)
	5	19 (0)	14 (0)	5 (0)

Note: Q1 = first quartile; Q3 = third quartile; ISS = injury severity score; GCS = Glasgow coma scale; HTN = hypertension on medication; CHF = congestive heart failure; CRF = chronic renal failure; CVA = cerebrovascular accident; MI = myocardial infarction; PVD = peripheral vascular disease; COPD = chronic obstructive pulmonary disease; MVT = motor vehicle trauma; GSW = gunshot wound; SW = stab wound; hypotension = initial systolic blood pressure <90 mmHg; AIS = abbreviated injury scale goes from 1 to 5; n = number; (%) = percentage.

Multiple Logistic Regression Analysis

Initially all the variables mentioned above in the univariate analysis were included into the study. With backward elimination methods, the final model was selected. The final model included patient’s characteristics, mechanism of injury, injury severity, GCS, presence of hypotension at initial presentation, and comorbidities. The estimated β coefficient provided the change of log odds of mortality for every unit or category change in the predictor variables (Table 2).

Table 2.

Multivariable Analysis of the Risk of Mortality.

Variable	β Coefficient	95% CI for β		OR	95% CI for OR		P-Value
(Intercept)	−6.921	−7.080	−6.762				<.001
Age in years
45-54	.612	.556	.668	1.844	1.744	1.951	<.001
55-64	1.069	1.015	1.123	2.912	2.758	3.074	<.001
65-74	1.576	1.519	1.633	4.837	4.569	5.121	<.001
75-89	2.160	2.103	2.217	8.672	8.195	9.177	<.001
Sex (male)	.254	.223	.286	1.289	1.249	1.331	<.001
Race
Asian	.349	.175	.523	1.417	1.191	1.686	<.001
Black or African American	.155	.005	.304	1.167	1.005	1.356	.043
Other race	.222	.069	.376	1.249	1.071	1.456	.005
White	.282	.139	.426	1.326	1.149	1.531	<.001
Injury severity score (ISS)
15-24	.812	.771	.852	2.252	2.162	2.345	<.001
25-34	2.061	2.022	2.099	7.851	7.555	8.159	<.001
35-45	2.558	2.497	2.619	12.912	12.151	13.722	<.001
46-75	3.048	2.951	3.145	21.075	19.129	23.220	<.001
Mechanism of injury
fall	.282	.233	.331	1.326	1.262	1.393	<.001
GSW	1.332	1.262	1.401	3.788	3.534	4.060	<.001
Motor vehicle trauma	.148	.096	.200	1.160	1.101	1.222	<.001
Pedestrian hit by	.398	.324	.472	1.489	1.383	1.602	<.001
Glasgow coma scale (GCS)
03-08	2.646	2.611	2.681	14.096	13.609	14.601	<.001
09-12	1.369	1.311	1.426	3.930	3.711	4.161	<.001
Hypotension (SBP <90 mmHg)	.780	.727	.832	2.181	2.069	2.299	<.001
Comorbidities
Ascites	1.050	.773	1.328	2.859	2.165	3.774	<.001
Coagulopathy	.385	.346	.425	1.470	1.413	1.530	<.001
Receiving chemotherapy	.677	.519	.835	1.969	1.681	2.306	<.001
Congestive heart failure (CHF)	.586	.534	.638	1.797	1.705	1.893	<.001
Chronic renal failure (CRF)	.854	.775	.932	2.349	2.171	2.540	<.001
Receiving dialysis for (CRF)	1.192	.916	1.467	3.293	2.500	4.338	<.001
Stroke (CVA)	.130	.066	.194	1.139	1.068	1.214	<.001
Diabetes mellitus (DM)	.204	.168	.240	1.226	1.183	1.271	<.001
Disseminated cancer (disc_canc)	.667	.561	.773	1.948	1.753	2.166	<.001
Myocardial infarction (MI)	.241	.156	.326	1.272	1.169	1.385	<.001
Peripheral vascular disease (PVD)	.245	.118	.372	1.278	1.125	1.451	<.001
No hypertension (HTN)	.082	.049	.115	1.085	1.050	1.122	<.001
Obesity	.143	.083	.204	1.154	1.086	1.226	<.001
Chronic pulmonary obstructive disease (COPD)	.284	.220	.348	1.328	1.246	1.416	<.001
Cirrhosis	1.436	1.353	1.520	4.205	3.868	4.572	<.001
Dementia	.075	.014	.135	1.077	1.014	1.145	.016

Log-odds of mortality = −6.92079 +.61219*(age: 45-54) +1.06887*(age: 55-64) +1.57638*(age: 65-74) +2.16011*(age: 75-89) +.25418*(sex: male) +.34884*(race: Asian) +.15471*(race: black or African American) +.22235*(race: other race) +.28247*(race: white) +.81182*(iss: 15-24) +2.06063*(iss: 25-34) +2.55819*(iss: 35-45) +3.04811*(iss: 46-75) +.2821*(mech: fall) +1.33185*(mech: GSW) +.14834*(mech: MVT) +.39796*(mech: pedestrian hit by) +2.64591*(gcs: 03-08) +1.36853*(gcs: 09-12) +.77978*(sbp<90) +1.05032*(ascites) +.38529*(coagulopathy) +.67741*(chemo) +.58593*(chf) +.85382*(crf) +1.19174*(dialysis) +.13015*(cva) +.20365*(dm) +.66703*(dis_canc) +.24081*(mi) +.24502*(pvd) +.08187*(no.HTN) +.14316*(obesity) +.28364*(copd) +1.43628*(cirrhosis) +.07457*(dementia)

Probability of mortality $= \frac{EXP (LOG - Odds of mortality)}{1 + EXP (LOG - Odds of mortality)}$

The final model can predict the mortality from .59% to 99.99%. Our model showed patients’ age more than 75 years old has an approximately 9-fold increase chance of mortality compared to age less than 45 years old. Gunshot wound, Asian race, male gender, higher severity of injuries, and some comorbidities including chronic renal failure on dialysis and cirrhosis have the highest chances of mortality. Our model was further tested for the appropriate fit to predict mortality by creating a receiver operating characteristics (ROC) curve, and an area under the curve (AUC) was generated. The AUC of TMS (0.892, 95% CI: 0.888-0.896) was better than TRISS (0.864, 95% CI: 0.859-0.869, P<0.0001) and ASCOT (0.841, 95% CI: 0.835-0.846, P<0.0001), respectively, in blunt injury (Figure 1).

Figure 1.

Receiver operating characteristics (ROC) curve of TMS, TRISS, and ASCOT for prognostic model performance in predicting the mortality in blunt injury from the testing data set.

When ROC and AUC of TMS was compared with TRISS and ASCOT score in penetrating injury, TMS prediction (AUC: 0.949, 95% CI: 0.940-0.958) was better in penetrating injury when compared with TRISS (0.942, 95% CI: 0.934-0.951, P=0.030) and ASCOT (0.924, 95% CI: 0.912-0.936, P<0.0001), respectively (Figure 2).

Figure 2.

Receiver operating characteristics (ROC) curve of TMS, TRISS, and ASCOT for prognostic model performance in predicting the mortality in penetrating injury from the testing data set.

Furthermore, the training data set (80%) and testing data (20%) set were compared on the baseline characteristics and mortality. There was no significant difference found between the two data sets regarding characteristics and mortality (Table 3). Approximately 154,000 (20%) of patients’ data that were used to validate the risk model showed similar predictability as the training data set showed.

Table 3.

Comparison between Training Data and Testing Data.

Variable	Value	Training Data	Testing Data	P-Value
		(n = 613,439)	(n = 153,676)
Age category (years), n (%)
	16-44	182 940 (29.8)	45 671 (29.7)	.359
	45-54	83 022 (13.5)	20 683 (13.5)
	55-64	98 401 (16)	24 490 (15.9)
	65-74	92 564 (15.1)	23 403 (15.2)
	75-89	156 512 (25.5)	39 429 (25.7)
Sex, n (%)				.938
	Female	233 978 (38.1)	58 598 (38.1)
	Male	379 461 (61.9)	95 078 (61.9)
Race, n (%)				.289
	NH/AI	6865 (1.1)	1683 (1.1)
	Asian	10 208 (1.7)	2604 (1.7)
	Black or African American	76 782 (12.5)	19 509 (12.7)
	Other race	40 113 (6.5)	10 055 (6.5)
	White	479 471 (78.2)	119 825 (78)
ISS, n (%)				.338
	1-14	399 825 (65.2)	100 360 (65.3)
	15-24	138 232 (22.5)	34 512 (22.5)
	25-34	60 868 (9.9)	15 298 (10)
	35-45	11 608 (1.9)	2799 (1.8)
	46-75	2906 (.5)	707 (.5)
Mechanism, n (%)				.27
	Others	125 608 (20.5)	31 242 (20.3)
	Fall	307 210 (50.1)	77 294 (50.3)
	GSW	28 297 (4.6)	7036 (4.6)
	MVT	129 299 (21.1)	32 224 (21)
	Pedestrian hit by	23 025 (3.8)	5880 (3.8)
GCS, n (%)				.753
	3-8	48 860 (8)	12 223 (8)
	9-12	17 760 (2.9)	4504 (2.9)
	13-15	546 819 (89.1)	136 949 (89.1)
Hypotension, n (%)				.401
	No	595 927 (97.1)	149 227 (97.1)
	Yes	17 512 (2.9)	4449 (2.9)
Comorbidities, n (%)
Ascites	No	612 937 (99.9)	153 562 (99.9)	.37
	Yes	502 (.1)	114 (.1)
Coagulopathy	No	551 638 (89.9)	138 007 (89.8)	.158
	Yes	61 801 (10.1)	15 669 (10.2)
On chemotherapy	No	610 707 (99.6)	153 020 (99.6)	.339
	Yes	2732 (.4)	656 (.4)
CHF	No	584 355 (95.3)	146 410 (95.3)	.835
	Yes	29 084 (4.7)	7266 (4.7)
CRF	No	603 179 (98.3)	151 125 (98.3)	.74
	Yes	10 260 (1.7)	2551 (1.7)
Dialysis	No	612 770 (99.9)	153 498 (99.9)	.502
	Yes	669 (.1)	178 (.1)
CVA	No	591 754 (96.5)	148 100 (96.4)	.078
	Yes	21 685 (3.5)	5576 (3.6)
DM	No	502 699 (81.9)	125 962 (82)	.871
	Yes	110 740 (18.1)	27 714 (18)
Disseminated cancer	No	607 395 (99)	152 195 (99)	.452
	Yes	6044 (1)	1481 (1)
MI	No	602 351 (98.2)	150 872 (98.2)	.661
	Yes	11 088 (1.8)	2804 (1.8)
PVD	No	608 653 (99.2)	152 484 (99.2)	.869
	Yes	4786 (.8)	1192 (.8)
HTN	No	335 153 (54.6)	83 908 (54.6)	.81
	yes	278 286 (45.4)	69 768 (45.4)
Obesity	No	576 776 (94)	144 497 (94)	.961
	yes	36 663 (6)	9179 (6)
COPD	No	587 772 (95.8)	147 245 (95.8)	.996
	yes	25 667 (4.2)	6431 (4.2)
Cirrhosis	No	606 074 (98.8)	151 907 (98.8)	.113
	yes	7365 (1.2)	1769 (1.2)
Dementia	No	584 583 (95.3)	146 324 (95.2)	.188
	yes	28 856 (4.7)	7352 (4.8)
Mortality	No	586 068 (95.5)	146 727 (95.5)	.313
	yes	27 371 (4.5)	6949 (4.5)

Note: NH/AI = Native Hawaiian or other Pacific Islander/American Indian; ISS = injury severity score; GCS = Glasgow coma scale; HTN = hypertension on medication; CHF = congestive heart failure; CRF = chronic renal failure; CVA = cerebrovascular accident; MI = myocardial infarction; PVD = peripheral vascular disease; COPD = chronic obstructive pulmonary disease; MVT = motor vehicle trauma; GSW = gunshot wound; SW = stab wound; hypotension = initial systolic blood pressure <90 mmHg; n = number; (%) = percentage.

The TMS was obtained first multiplying the β coefficient of the variable by factor 10 to get the nearest whole number and sum up all the numbers to get the total score of the patient. The range of score goes from 0 to 188 with the predicted mortality from .59% to 99.99% (Table 4). A user-friendly calculator has been developed for the point of care physician to calculate the risk of mortality at the bedside (Supplement Figure 1).

Table 4.

Score Assignment.

Variable	Coefficient from Multiple Logistic Regression Model	Assigned Score
Age category (years)
45-54	.612	6
55-64	1.069	11
65-74	1.576	16
75-89	2.160	22
Sex (male)	.254	3
Race
Asian	.349	3
Black or African American	.155	2
Other race	.222	2
White	.282	3
ISS
15-24	.812	8
25-34	2.061	21
35-45	2.558	26
46-75	3.048	30
Mechanism
fall	.282	3
GSW	1.332	13
MVT	.148	1
Pedestrian hit by	.398	4
GCS
03-08	2.646	26
09-12	1.369	14
Hypotension	.780	8
Ascites	1.050	11
Coagulopathy	.385	4
On chemotherapy	.677	7
CHF	.586	6
CRF	.854	9
Dialysis	1.192	12
CVA	.130	1
DM	.204	2
Disseminated cancer	.667	7
MI	.241	2
PVD	.245	2
No HTN	.082	1
Obesity	.143	1
COPD	.284	3
Cirrhosis	1.436	14
Dementia	.075	1

Note: ISS = injury severity score; GCS = Glasgow coma scale; HTN = hypertension on medication; CHF = congestive heart failure; CRF = chronic renal failure; CVA = cerebrovascular accident; MI = myocardial infarction; PVD COPD = chronic obstructive pulmonary disease; MVT = motor vehicle trauma; GSW = gunshot wound; SW = stab wound; hypotension = initial systolic blood pressure <90 mmHg; n = number; (%) = percentage.

Figure 3 shows the observed and predicted percentages of mortality based on the training data set. The observed percentages were calculated based on the number of patients within each value of score. The predicted percentages were based on the estimated logistic regression model. It shows that the scores predictability of mortality with actual mortality was similar.

Figure 3.

Relationship of score with predicted vs observed mortality in the training data set.

Figure 4 shows the predicted percentages of mortality based on the training data set (labeled with “Training Data Set” in blue line, 80% of the data). The predicted percentages were based on the estimated logistic regression model. The observed percentages were calculated based on the number of patients within each value of score in the testing data set (labeled with “Testing Data Set” with black line, 20% of the data). The figure shows that the training data set mortality falls within the confidence interval of the testing data set in relation to scores.

Figure 4.

Observed probability of mortality in testing data set and predicted mortality in training data set by the risk score.

Discussion

The study showed 4.5% of trauma victims died and 95.5% survived. Advanced age, male sex, Asian race, lower GCS, higher ISS, and certain comorbidities have highly correlated with mortality.

The development of risk score from the training data set and further validating with 153,676 (20%) randomly selected data set showed that the model predicts mortality from .59% to 99.99%. When the comparison of variables was performed between the training data set and testing data set, we found no difference between the two data sets on patients’ characteristics, injury type, severity, comorbidities, and overall mortality (Table 3). We tested the TMS in the training and testing data set and found no obvious difference in predicting mortality.

Our risk model takes into account all the previously known factors of mortality^11,20-24 and added new variables (race and comorbidities) that increase the predictability of mortality. Some of the previously reported factors that have been directly related to mortality, for example, ISS score alone may not necessarily predict the mortality accurately.^25,26

The previously developed RTS takes into account only GCS, systolic blood pressure, and respiratory rate.⁸ RTS was developed more than 30 years ago mainly for triaging injured patients to transport to either specialized trauma center vs non-trauma center. Later, another score was developed around the same time adding age into the other factors making it more robust.⁹ About 10 years later after developing RTS, the same group came up with another score, A Severity Characterization of Trauma (ASCOT) score. The authors claimed a better scoring system in blunt mechanism.¹⁰ However, ASCOT was not good for predicting mortality in penetrating injuries. Our TMS system showed improvement in predictability of mortality in blunt and in penetrating injuries as well when compared with TRISS and ASCOT scoring. The AUCs of TMS, TRISS and ASCOT were 0.892 (95% CI: 0.888-0.896), 0.864 (95% CI: 0.859-0.869) and 0.841 (95% CI: 0.835-0.846), respectively, in blunt injury. Similarly, TMS prediction was better in penetrating injury when compared with TRISS and ASCOT. The AUCs were 0.949 (95% CI: 0.940-0.958), 0.942 (95% CI: 0.934-0.951) and 0.924 (95% CI: 0.912-0.936), respectively. Although the TMS has better predictability, the differences in AUC are not huge. The counterargument is that adding comorbidities which is one of the contributing factors in patient mortality helps identify those patients who are likely to die or survive based on the patient’s prior history. For example, if a 50-year-old Caucasian patient is presented with GCS score of 15 and ISS of 14 would have a 3.43% mortality risk without any comorbidities, whereas the same patient with an only history of cirrhosis would increase the probability of death to a 12.98%. Therefore, it is important to realize that certain comorbid conditions can lead to an increase the mortality many folds as evident by the above example.

All above studies included some variables of the anatomical distribution of injury, physiological characteristics, age of the patient, and mechanism of injury. However, none of them considers the pre-existing condition of the patient that can impact the outcome of the patients.^27,28 Our novel scoring system, TMS, is one of the most comprehensive scoring systems to date that includes patients’ demography, mechanism of injury, severity of injury, some physiological parameters, and comorbidities to identify the risk of in-hospital mortality. Our study added age into a different category instead of dichotomized the age at 60 as other had used it.¹¹ We analyzed the TMS to find the predictability of mortality in observed vs expected mortality and the scores were most part falling in the midline. When TMS was compared with actual mortality in testing and training data set, the predictability of the testing data set was close to training data set findings. Availability of TMS calculator to the point of care physician helps the physician to determine the prognosis of the trauma patient early and in certain circumstances, early involvement of palliative care consultation (Supplement Figure 1).

The first trauma scoring system was introduced almost 40 years ago. To this date, our community is still searching for a better scoring system to determine the mortality of trauma victims which is a testament that a comprehensive scoring system is needed that can not only accurately predict mortality early but is also easy to use by point of care providers. We believe, our scoring system could offer the most up to date comprehensive scoring system to our providers.

Limitations

Our study was performed from the TQIP database but suffered from some limitations. We performed internal validation by randomly selecting the 20% of the data set that showed the same predictability as the training data set. Our data set tried to include every region in the United States and included all races and gender for the generalizability of the predictor model. Our study, however, is lacking external validation.

Conclusion

Our novel TMS accurately predicts mortality in the majority of the trauma victims. TMS predictability is better than TRISS and ASCOT. TMS should be used in place of TRISS in blunt and in penetrating injury as well. Using this scoring system, one can calculate the high-risk patient in all types of trauma patients. This tool will help point of care physicians to allocate appropriate resources for the care of the patient. This scoring system can also help guide the shared decision-making process in certain circumstances.

Application. A TMS calculator has been developed and will be available to all providers to ensure easiness of calculating the prognosis of injured patients.

Supplemental Material

sj-tif-1-asu-10.1177_00031348221078980 – Supplement Material for Prediction of Trauma Mortality Incorporating Pre-injury Comorbidities into Existing Mortality Scoring Indices

Supplemental Material, sj-tif-1-asu-10.1177_00031348221078980 for Prediction of Trauma Mortality Incorporating Pre-injury Comorbidities into Existing Mortality Scoring Indices by Nasim Ahmed and Yen-Hong Kuo in The American Surgeon

Footnotes

Acknowledgments

Elli Gourna Paleoudis, MS, PhD has proofread the article and performed the final editing.

Author’s Contribution

Nasim Ahmed (NA) conceived and designed the study.

NA was responsible for retrieving the study data, while Yen-Hong Kuo (YHK) performed the data analysis.

NA and YHK both contributed to manuscript writing and both were responsible for all follow-up manuscript revisions.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Ethical Approval

All procedures followed were accordance with the ethical standards of the Institutional Review Board of Hackensack Meridian Health and with the Helsinki Declaration of 1975, as revised in 2008. Since the data of NTDB/TQIP are de-identified patient’s information available to the researchers, the study was exempted from Meridian Health IRB review.

Informed Consent

Given this retrospective study was performed from the de-identified National database from the American College of Surgeons that is available to all researchers, this study was exempt from the IRB review as per policy and no informed consent was required.

Supplementary Material

Supplementary Material for this article is available online.

References

. National Center for Injury Prevention and Control. “Deaths: Final Data for 2007,” National Vital Statistics Reports 58, no. 19; 2010. (accessed date April 30, 2020) http://webappa.cdc.gov/sasweb/ncipc/mortrate10_sy.html.

Violence and Injury Prevention . 2020. https://www.who.int/violence_injury_prevention/key_facts/en/Access. (access date April 30, 2020).

Cost of Injuries and Violence in the United States . 2020. https://www.cdc.gov/injury/wisqars/overview/cost_of_injury.html. (access date April 30, 2020)

Murad

Larsen

Husum

. Prehospital trauma care reduces mortality. Ten-year results from a time-cohort and trauma audit study in Iraq. Scand J Trauma Resuscitation Emerg Med. 2012;20:13. doi:10.1186/1757-7241-20-13.

MacKenzie

Rivara

Jurkovich

Salkever

Scharfstein

. A national evaluation of the effect of trauma-center care on mortality. N Engl J Med. 2006;354:366-378. doi:10.1056/NEJMsa052049.

Durham

Pracht

Orban

Lottenburg

Tepas

Flint

. Evaluation of a mature trauma system. Ann Surg. 2006;243(6):775-5. doi:10.1097/01.sla.0000219644.52926.f1.

Kelley

Morrison

. Palliative care for the seriously ill. N Engl J Med. 2015;373(8):747-755.

Champion

Sacco

Copes

Gann

Gennarelli

Flanagan

. A revision of the Trauma Score. J Trauma. 1989;29(5):623-629. doi:10.1097/00005373-198905000-00017. PMID:2657085

Boyd

Tolson

Copes

. Evaluating trauma care: The TRISS method. Trauma score and the injury severity score. J Trauma. 1987;27(4):370-378. PMID: 3106646.

10.

Champion

HRFACS

Copes

Sacco

, et al. Improved predictions from a severity characterization of trauma (ASCOT) over trauma and injury severity score (TRISS). J Trauma Inj Infect Crit Care. 1996;40(1):42-49.

11.

Sartorius

Le Manach

David

, et al. Mechanism. glasgow coma scale, age, and arterial pressure (MGAP): A new simple prehospital triage score to predict mortality in trauma patients. Crit Care Med. 2010;38:831-837.

12.

Kondo

Abe

Kohshi

Tokuda

Cook

Kukita

. Revised trauma scoring system to predict in-hospital mortality in the emergency department: Glasgow coma scale, age, and systolic blood pressure score. Crit CarePMCID. 2011;15(4):R191PMC3387633. doi:10.1186/cc10348. PMID: 21831280

13.

Kunitake

Kornblith

Cohen

Callcut

. Trauma early mortality prediction tool (TEMPT) for assessing 28-day mortality. Trauma Surg Acute Care Open. 2018;3(1):e000131. doi:10.1136/tsaco-2017-000131.

14.

Wang

Chen

Chien

, et al. Impact of comorbidities on the prognoses of trauma patients: Analysis of a hospital-based trauma registry database. PLoS One. 2018;13(3):e0194749. doi:10.1371/journal.pone.0194749.

15.

Boyd

Tolson

Copes

. Evaluating trauma care: The TRISS method. Trauma score and the injury severity score. J Trauma. 1987;27(4):370-378. PMID: 3106646.

16.

Champion

Copes

Sacco

Lawnick

Bain

Gann

Gennarelli

Mackenzie

Schwaitzberg

. A new characterization of injury severity. J Trauma. 1990 May;30(5):539-45; discussion 545-6. PMID: 2342136.

17.

DeLong

Clarke-Pearson

. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837-845. PMID: 3203132.

18.

Robin

Turck

Hainard

, et al. pROC: An open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinf. 2011;12:77. doi:10.1186/1471-2105-12-77.

19.

R Core Team . R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2018. URL https://www.R-project.org/.

20.

Haider

Efron

Haut

Chang

Paidas

Cornwell

. Mortality in adolescent girls vs boys following traumatic shock: An analysis of the national pediatric trauma registry. Arch Surg. 2007;142(9):875-880.

21.

Baker

O'Neill

. The injury severity score: an update. J Trauma. 1976;16(11):882-885.

22.

MacLeod

Lynn

McKenney

Jeroukhimov

Cohn

. Predictors of mortality in trauma patients. Am Surg. 2004;70(9):805-810. PMID: 15481299.

23.

Haider

Chang

Efron

Haut

Crandall

Cornwell

. Race and insurance status as risk factors for trauma mortality. Arch Surg. 2008;143(10):945-949. doi:10.1001/archsurg.143.10.945.

24.

Morris

MacKenzie

Edelstein

. The Effect of Preexisting conditions on mortality in trauma patients. JAMA. 1990;263(14):1942-1946. doi:10.1001/jama.1990.03440140068033.

25.

Baker

O'Neill

Haddon

Jr Long

. The injury severity score: A method for describing patients with multiple injuries and evaluating emergency care. J Trauma. 1974;14:187-196.

26.

Deng

Tang

Xue

, et al. Comparison of the ability to predict mortality between the injury severity score and the new injury severity score: A meta-analysis. Int J Environ Res Publ Health. 2016;13(8):825. doi:10.3390/ijerph13080825.

27.

Kirshenbom

Ben-Zaken

Albilya

Niyibizi

Bala

. Older age, comorbid illnesses, and injury severity affect immediate outcome in elderly trauma patients. J Emergencies, Trauma, Shock. 2017;10(3):146-150. doi:10.4103/JETS.JETS_62_16.

28.

Niven

Kirkpatrick

Ball

Laupland

. Effect of comorbid illness on the long-term outcome of adults suffering major traumatic injury: A population-based cohort study. Am J Surg. 2012;204(2):151-156. doi:10.1016/j.amjsurg.2012.02.014. Epub 2012 May 18. PMID: 22608670.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.16 MB