Original Contribution

Probability of survival, early critical care process, and resource use in trauma patients^?

Kazuaki Kuwabara PhD, MD ^a,?, Shinya Matsuda PhD, MD ^b, Kiyohide Fushimi PhD, MD ^c, Koichi B. Ishikawa PhD ^d, Hiromasa Horiguchi PhD ^e, Kenji Fujimori PhD, MD ^f

^aDepartment of Health Care Administration and Management, Graduate School of Medical Sciences, Kyushu University,

Fukuoka 812-8582, Japan

^bDepartment of Preventive Medicine and Community Health, University of Occupational and Environmental Health,

Kitakyushu 807-8555, Japan

^cDepartment of Health Policy and Informatics, Tokyo Medical and Dental University Graduate School of Medicine,

Tokyo 113-8510, Japan

^dStatistics and Cancer Control Division, National Cancer Center, Tokyo 104-0045, Japan

^eHealth Management and Policy, University of Tokyo, Graduate School of Medicine, Tokyo 113-8655, Japan

^fDivision of Medical Management, Hokkaido University, Hokkaido 060-8648, Japan

Received 5 December 2008; revised 28 January 2009; accepted 27 February 2009

Abstract

Background: Trauma Injury Severity Score is a frequently used prediction model for mortality. However, few studies have assessed the probability of survival (Ps) and early resource use after trauma. We studied the impact of Ps on early critical care or costs to test its applicability to efficient trauma care. Methods: The relationship between Ps in 8207 trauma patients and patients’ demographics, organ injured, comorbidities, use of critical care, and Total charges during the initial 48 hours was analyzed using multiple regression analyses.

Results: Significant differences were observed among study variables across different Ps. A large variability in total charges was observed and explained by critical care, which Ps was significantly associated with.

Conclusions: Trauma Injury Severity Score offers a tool for estimating resource use and might improve monitoring of early trauma care quality. Measuring the combined effect of Trauma Injury Severity Score and injured organs would refine the methodology for evaluating the trauma care system.

(C) 2010

^? This research project was funded by the Japanese Ministry of Health, Labor and Welfare. Authors have no conflict of interest.

* Corresponding author. Tel.: +81 92 642 6955; fax: +81 92 642 6961.

E-mail address: [email protected] (K. Kuwabara).

Introduction

Most developed countries have struggled to meet the demands of delivering good quality health care, and Intensive care units imperatively have to meet

0735-6757/$ - see front matter (C) 2010 doi:10.1016/j.ajem.2009.02.030

this demand. Although many studies have focused on the mortality of trauma patients, outcomes other than survival, such as functional conditions or reimbursement, require more attention [1,2]. Studies that use risk adjustment with injury-related variables such as Injury Severity Score (ISS) discuss the variability in the process and the costs of trauma care. As such, these studies provide policy implications highlighting high-cost elements in trauma management as well as measuring efficient trauma care interventions [3].

The Trauma Injury Severity Score (TRISS) has been used widely for risk adjustment of mortality in the field of trauma, injury, and critical care to evaluate the quality of trauma care systems. Anatomical, physiologic, and age characteristics are included in the TRISS to quantify the probability of survival (Ps) at admission depending on the severity of injury [4-6]. This score also serves as a screening tool for case identification in quality assurance reviews and as a means to compare outcomes for patients with various traumatic injuries. Calculated Ps for trauma patients could serve as a potential estimate for resource consumption during ICU care or overall hospitalization. In Austria, ICU care has been paid on the basis of ICU staffing or beds and monitored in terms of appropriate intensive care delivery with TRISS, the Therapeutic Intervention Scoring System-28, or the Simplified Acute Physiology Score [7].

However, few studies have documented the association of the ISS or other severity indices with cost, care process, or patient volume [3,8-11]. Christensen et al (2008) [3] reported that a large portion of the cost was explained by length of hospital stay and indicated that shortening of LOS would have the greatest impact on reducing costs for blunt trauma care. As LOS and cost are correlated, these

2 variables should be managed separately to determine which injury-related factors would most influence the resource use. Such an investigation could highlight the most relevant causes for high cost and lead to improvED efficiency and standardized management of trauma care systems. However, additional efforts should be made a priori to reveal acceptable early critical care or resource use for critical periods such as the first 2 days after hospitalization that would be crucial for optimal quality of overall trauma care.

Consequently, it would be possible to determine if patients with a lower Ps or those with a head injury, for example, will need more intensive care input, especially immediately after the trauma. In other words, measuring the impact of Ps on the ICU care process or early resource use would bring quality assurance to trauma care, rationalize cost of ICU care, and have policy implications.

The current study aimed to examine the impact of Ps on resource use and care process on days 1 and 2 after admission to test the applicability of TRISS on estimating costs and use of critical care.

Materials and methods

Study design and setting

A cross-sectional observational study was performed using the Japanese administrative database from a govern- ment project involving the development of a Japanese case- mix classification system. Anonymous claim data with detailed clinical information were collected annually between July 1 and December 31 since 2006. Data from fiscal year 2006 were provided to members of our research team who were engaged in the refinement of case-mix classification in cooperation with the Ministry of Health, Welfare, and Labor.

These data have been used to profile hospital performance and assess hospital payments, and include 82 academic hospitals (80 university hospitals, the National Cancer Center, and the National Cardiovascular Center) and 649 community hospitals. Scattered throughout Japan, these hospitals provide acute care, medical research, and trainee education. This research project was approved by the University of Occupational and Environmental Health Ethics Committee, Fukuoka, Japan.

Variable definition

Study variables included age, sex, use of an ambulance, mechanism of injury (blunt or penetrating), location and number of anatomical injuries, comorbidities, and hospital category (academic or community). We also investigated surgical procedures requiring general anesthesia in the operating room , time (in minutes) in the OR, need for critical care (ie, use of ICU or artificial ventilation), blood transfusion (packed blood cell, in milliliters), and outcome on days 1 and 2 after admission as well as at the time of discharge.

Injury Severity Score was calculated as the sum of the squares of the Abbreviated Injury Scale (AIS, 1998) of the single worst injury in each of the 3 most injured bodily regions. Abbreviated Injury Scale is an ordinal scale ranging from 1 (Minor injury) to 6 (unsalvageable injury). Injury Severity Score ranges between 1 and 75 (maximum). Patients with an AIS severity of 6 in any of the 6 body regions were automatically defined to have an ISS of 75 [5]. Revised Trauma Scores (RTS) were also calculated based on the sum of weighted coded values corresponding to systolic blood pressure, respiratory rate, and Glasgow Coma Scale at admission. The RTS varied from 0 to approximately 8 in noninteger values [6].

We used age (age >=55 or b15 years), mechanism of injury

(blunt or penetrating injury), RTS, and ISS to calculate Ps, and categorized 7 groups based on every .010 interval of Ps. The Ps categories therefore ranged from Ps less than .9400 to Ps of at least .9900 [8]. We also calculated LOS and total charges (TC; US $1 = JPY90) billed during hospitalization. In

Japan, charges for hospital care are determined by a standardized fee-for-service payment system known as the national uniform tariff table, considered to be a good estimate of health care costs [12]. Total charges included fees for physicians and administration as well as for instruments, laboratory, and imaging, and were the sum of consumed service units multiplied by a price per service unit.

We also calculated TC on days 1 and 2 after admission to determine if patients with more critical injuries would spend more resources immediately after admission. Patients were stratified by age into 3 groups: younger than 15 years, 15 to 54 years old, and 55 years or older. This database records 4 comorbidities per patient. To assess the severity of chronic

comorbid conditions, we used the number of preexisting conditions (PEC) from the definition of the Charlson Comorbidity Index (CCI) [13-15]. Risk adjustment is a vital component of health services utilization and outcome studies, where the CCI, which is well validated in many international studies, has been applied [14]. However, the number of PEC was counted instead of CCI itself [15]. Patients were grouped into 3 groups based on the number of PEC: 0, 1, and 2 or more. Patients who died within the first 24 hours, those who were seen only in the outpatient clinic, those with burns or nontraumatic diagnoses, and those who had missing information preventing calculation of Ps were excluded from the analysis.

Table 1 Demographic and clinical characteristics and outcomes by Ps (n, %)

		Ps							P
		.9900 or more	.9800-.9899	.9700-.9799	.9600-.9699	.9500-.9599	.9400-.9499	b.9400
Overall		2600 (31.7)	653 (8.0)	1992 (24.3)	1466 (17.9)	246 (3.0)	262 (3.2)	988 (12.0)
Age	b15 y	533 (20.5)	51 (7.8)	15 (0.8)	3 (0.2)	5 (2.0)	0 (0.0)	79 (8.0)	b.001
	>=55 y	0 (0.0)	385 (59)	1909 (95.8)	1437 (98.0)	222 (90.2)	252 (96.2)	765 (77.4)

Age, mean, SD

	27.3 (14.7)	55.2 (25.2)	72.1 (14.1)	78.1 (12.3)	68.5 (17.9)	73.8 (13.3)	62.7 (24.3)	b.001 a
Male	1831 (70.4)	386 (59.1)	805 (40.4)	407 (27.8)	130 (52.8)	127 (48.5)	594 (60.1)	b.001
Used	1152 (44.3)	442 (67.7)	932 (46.8)	844 (57.6)	184 (74.8)	166 (63.4)	699 (70.7)	b.001
Blunt	2412 (92.8)	580 (88.8)	1822 (91.5)	1447 (98.7)	226 (91.9)	252 (96.2)	869 (88.0)	b.001

Sex Ambulance

Mechanism of injury

Injured organ
	Head and neck	470 (18.1)	230 (35.2)	231 (11.6)	133 (9.1)	109 (44.3)	149 (56.9)	587 (59.4)	b.001
	Face	261 (10.0)	73 (11.2)	86 (4.3)	31 (2.1)	38 (15.4)	7 (2.7)	127 (12.9)
	Chest	223 (8.6)	129 (19.8)	191 (9.6)	100 (6.8)	74 (30.1)	43 (16.4)	241 (24.4)
	Abdomen,	162 (6.2)	66 (10.1)	220 (11.0)	43 (2.9)	34 (13.8)	9 (3.4)	114 (11.5)
	pelvic contents
	Extremity,	1608 (61.8)	174 (26.6)	1402 (70.4)	1248 (85.1)	145 (58.9)	68 (26.0)	451 (45.6)
	pelvic girdle
	External	471 (18.1)	256 (39.2)	195 (9.8)	86 (5.9)	62 (25.2)	7 (2.7)	228 (23.1)
No. of injured organs
1		2165 (83.3)	481 (73.7)	1724 (86.5)	1349 (92)	119 (48.4)	249 (95)	587 (59.4)	b.001
2		322 (12.4)	108 (16.5)	229 (11.5)	72 (4.9)	69 (28.0)	6 (2.3)	205 (20.7)
3		82 (3.2)	38 (5.8)	23 (1.2)	35 (2.4)	39 (15.9)	6 (2.3)	91 (9.2)
4		21 (0.8)	17 (2.6)	9 (0.5)	8 (0.5)	12 (4.9)	1 (0.4)	65 (6.6)
5		4 (0.2)	5 (0.8)	4 (0.2)	1 (0.1)	2 (0.8)	0 (0.0)	22 (2.2)
6		6 (0.2)	4 (0.6)	3 (0.2)	1 (0.1)	5 (2.0)	0 (0.0)	18 (1.8)
No. of PEC
0		2516 (96.8)	572 (87.6)	1583 (79.5)	1033 (70.5)	201 (81.7)	209 (79.8)	819 (82.9)	b.001
1		80 (3.1)	69 (10.6)	334 (16.8)	340 (23.2)	43 (17.5)	47 (17.9)	141 (14.3)
>=2		4 (0.2)	12 (1.8)	75 (3.8)	93 (6.3)	2 (0.8)	6 (2.3)	28 (2.8)
Hospital category
Community		2219 (85.3)	524 (80.2)	1842 (92.5)	1408 (96.0)	206 (83.7)	234 (89.3)	745 (75.4)	b.001
Academic		381 (14.7)	129 (19.8)	150 (7.5)	58 (4.0)	40 (16.3)	28 (10.7)	243 (24.6)
Outcome

Mortality (number on days 1 and 2)

2 (0) 4 (0) 12 (0) 26 (1) 7 (0) 11 (0) 110 (16) .261

^a Compared by analysis of variance. Others by Fisher exact test.

RTS (IQ) 7.84 (0) 7.84 (0)	7.84 (0)	7.84	(0)	7.55 (0.94)	7.84 (0.94)	6.68 (3.1)	b.001 a
ISS (IQ) 4 (1) 1 (15)	4 (0)	9	(0)	12 (4)	16 (7)	16 (16)	b.001 a
Critical care, n (% of overall cases)
Blood transfusion
n (%) 21 (47.7) 31 (66.0)	50 (30.5)	70	(19.8)	24 (57.1)	21 (65.6)	160 (67.8)	b.001
Ventilation
n (%) 10 (76.9) 19 (63.3)	17 (60.7)	11	(36.7)	17 (89.5)	11 (73.3)	195 (85.9)	b.001
ICU care
n (%) 423 (96.8) 192 (95.0)	189 (87.9)	123	(76.9)	87 (94.6)	90 (92.8)	460 (98.3)	b.001
Use of OR
n (%) 666 (68.9) 62 (55.9)	203 (37.5)	101	(20.4)	21 (28.8)	17 (39.5)	210 (66.7)	b.001
a Compared by Kruskal-Wallis test. Others by Fisher exact test. IQ, interquartile range.

Statistical analysis

Table 2 Trauma severity, care volume on days 1 and 2, stratified by Ps

Ps

P

.9900 or more .9800-.9899 .9700-.9799 .9600-.9699 .9500-.9599 .9400-.9499 b.9400

Frequency and proportion of all categorical data including sex, age category, injury mechanism, injured organs and the number of PEC, hospital mortality, hospital type (academic or community hospital), use of ambulance, and study care service were analyzed and compared using Fisher exact test. The proportion of mortality, care such as blood transfusion or

ventilation, ICU admission, and OR use up to day 2 after admission relative to that for the entire hospital stay was calculated. Continuous variables (TC, OR time, and blood transfusion volume on days 1 and 2) were compared across Ps groups using nonparametric tests and displayed in box charts. Age was compared using an analysis of variance. Multiple Linear regression analysis was used to identify the impact of Ps on TC on days 1 and 2. Because the distribution

Fig. 1 Blood transfusion, OR time, and TC (in US dollars) on days 1 and 2.

of TC on days 1 and 2 was right skewed, the data were log₁₀ transformed. multiple logistic regression analysis was used to determine the effect of study variables on use of OR, ventilation, and blood transfusion on days 1 and 2. Age and injury mechanisms were not included in this model to avoid double counting in Ps calculations. Statistics were performed using SPSS (Chicago, IL) 16.0, with a level of significance set a priori at P less than .05.

Results

Demographic and clinical characteristics

Of 1 895 249 patients discharged in 2006, 83 382 trauma- related case-mix patients were identified from 465 hospitals (12 783 cases from 70 academic hospitals and 70 599 from

Table 3 Linear regression analysis of factors associated with log₁₀-transformed TC on days 1 and 2

Estimation		SE	P
Intercept	2.981	0.010	b.001
Male	0.020	0.006	.001
Outcome on days	0.058	0.064	.364
1 and 2 (for alive)
Ambulance	0.043	0.006	b.001
Injured organ (for external)
Head and neck	0.017	0.009	.056
Face	-0.025	0.011	.025
Chest	-0.007	0.009	.444
Abdomen,	-0.012	0.011	.283
pelvic contents
Extremity,	0.059	0.007	b.001
pelvic girdle
Ps (for .9900 or more)
.9800-.9899	0.021	0.012	.074
.9700-.9799	-0.034	0.008	b.001
.9600-.9699	0.011	0.009	.231
.9500-.9599	0.024	0.018	.172
.9400-.9499	0.020	0.017	.245
b.9400	0.047	0.011	b.001
No. of PEC (for zero)
1	-0.030	0.009	.001
>=2	-0.055	0.018	.002
ICU care on days	0.319	0.009	b.001
1 and 2
Use of OR on days	0.482	0.009	b.001
1 and 2
Ventilation on days	-0.008	0.019	.647
1 and 2
Blood transfusion	0.273	0.015	b.001
on days 1 and 2
Hospital (for community)
Academic	0.036	0.010	b.001
F test for the model. P less than .001. Coefficient of determination, 0.502. SE, standard error.

395 community facilities). Among these, 8207 cases with a calculated Ps were eligible in this study (1029 cases from 35 academic hospitals and 7178 from 129 community facilities). The largest group of patients (31.7%) had a Ps of at least

.9900, whereas the smallest group of patients (3.0%) had a Ps of .9500 to .9600. In addition, 12% of patients had Ps less than .9400. The proportion of age, sex, ambulance use, location and number of injured organs, CCI, and surgical technique required was statistically different among Ps categories. Patients with a Ps of .9600 to .9700 had the highest median age and percentage of females. The proportion of injuries in “extremities, pelvic girdle” and “head and neck” regions was greater in the groups with Ps of

.9600 to .9700 and with Ps less than .9400, respectively. Observed mortality at discharge was higher than predicted in groups with Ps of at least .9400, but lower than predicted in the group with Ps less than .9400. The mortality on days 1 and 2 was 16 cases (15%) (Table 1).

Trauma severity, care characteristics, and resource use

Probability of survival was positively correlated with RTS and negatively correlated with ISS, with the exception that the median ISS (interquartile range [IQ]) was 1 (15) in Ps of .9800 to .9899 and that the median RTS (IQ) was 7.55 (0.94) in Ps of

.9500 to .9599. The highest proportion of patients requiring transfusion and ICU admission on days 1 and 2 was in those groups with Ps less than .9400, whereas the highest proportion of patients requiring ventilation was in groups with Ps of

.9500 to .9599. Use of OR on days 1 and 2 was most frequent in groups with Ps of at least .9900. Amount of transfusions, OR time, and TC on days 1 and 2 were statistically different among Ps categories (Table 2 and Fig. 1).

Multivariate analysis

The multiple regression analysis showed that use of OR was the variable with the strongest association with TC on days 1 and 2 and that ICU care and blood transfusion had a greater impact on TC. Among the Ps categories, a Ps less than .9400 or from .9700 to .9799 was a stronger determinant of TC. Among injury location, “extremity, pelvic girdle” had the highest association with TC; and “face” had the lowest association. The coefficient of determination for TC on days 1 and 2 was 0.502 (Table 3).

Mortality cases on days 1 and 2 and Ps less than .9400 were stronger determinants of use of ventilation on days 1 and 2, and Ps from .9400 to .9499 or less than .9400 was a stronger determinant of use of blood transfusion on days 1 and 2. Injury location at the “head and neck” or “chest” was a significant predictor for use of ventilation on days 1 and 2; and injury location at the “abdomen, pelvic contents” or “extremities, pelvic girdles” was a stronger predictor of use of blood transfusion (Table 4).

Table 4 Logistic regression analysis of factors associated with study care process on days 1 and 2

ICU care OR Ventilation Blood transfusion

Odd	(95% CI)		Odd	(95% CI)		Odd	(95% CI)		Odd	(95% CI)
ratio			ratio			ratio			ratio

Sex Female	1.000		1.000		1.000		1.000
Male	1.444	(1.232-1.693)	1.476	(1.278-1.703)	1.383	(0.961-1.992)	0.621	(0.475-0.810)
Outcome Alive	1.000		1.000		1.000		1.000
Mortality	0.310	(0.067-1.433)	0.381	(0.097-1.496)	23.098	(4.621-115.463)	0.985	(0.289-3.361)
Ambulance Not used	1.000		1.000		1.000		1.000
Used	5.643	(4.631-6.875)	0.351	(0.298-0.412)	3.107	(1.645-5.865)	1.971	(1.428-2.721)
Injured organ
External	1.000		1.000		1.000		1.000
Head and neck	5.643	(4.631-6.875)	0.357	(0.28-0.455)	2.559	(1.723-3.799)	0.606	(0.429-0.855)
Face	4.040	(3.371-4.842)	1.413	(1.103-1.811)	1.056	(0.668-1.670)	0.965	(0.626-1.487)
Chest	1.278	(1.012-1.614)	0.402	(0.306-0.528)	1.985	(1.369-2.879)	0.948	(0.674-1.332)
Abdomen,	2.362	(1.931-2.889)	0.860	(0.647-1.143)	0.866	(0.551-1.361)	2.351	(1.651-3.347)
pelvic contents
Extremity,	2.849	(2.233-3.634)	1.593	(1.332-1.906)	0.542	(0.374-0.784)	1.698	(1.265-2.280)
pelvic girdle
Ps
.9900 or more	1.000		1.000		1.000		1.000
.9800-.9899	0.715	(0.602-0.849)	0.410	(0.301-0.559)	3.299	(1.450-7.503)	5.173	(2.784-9.611)
.9700-.9799	1.219	(0.944-1.575)	0.349	(0.290-0.421)	2.605	(1.138-5.965)	4.354	(2.536-7.474)
.9600-.9699	0.654	(0.518-0.826)	0.232	(0.181-0.297)	2.648	(1.047-6.698)	8.905	(5.209-15.223)
.9500-.9599	0.789	(0.603-1.034)	0.273	(0.158-0.470)	7.066	(2.890-17.277)	11.027	(5.586-21.768)
.9400-.9499	1.363	(0.934-1.99)	0.260	(0.148-0.455)	5.412	(2.119-13.823)	15.137	(7.545-30.367)
b.9400	1.824	(1.282-2.594)	0.694	(0.546-0.883)	14.855	(7.439-29.663)	12.103	(7.146-20.499)
No. of PEC
Absent	1.000		1.000		1.000		1.000
1	1.539	(1.213-1.952)	0.768	(0.604-0.975)	1.558	(1.006-2.412)	0.869	(0.611-1.235)
>=2	0.906	(0.715-1.149)	0.331	(0.159-0.689)	0.106	(0.011-1.028)	1.346	(0.677-2.676)
ICU care on days 1 and 2
Absent	1.000		1.000		1.000		1.000
Present		a	1.736	(1.368-2.203)	6.626	(4.181-10.500)	5.009	(3.569-7.030)
OR on days 1 and 2
Absent	1.000		1.000		1.000		1.000
Present	1.376	(1.082-1.75)		a	5.861	(3.997-8.595)	6.027	(4.598-7.900)
Ventilation on days 1 and 2
Absent	1.000		1.000		1.000		1.000
Present	4.510	(2.817-7.221)	7.853	(5.507-11.198)		a	4.510	(3.109-6.542)
Blood transfusion on days 1 and 2
Absent	1.000		1.000		1.000		1.000
Present	4.755	(3.393-6.662)	6.967	(5.309-9.144)	4.447	(2.975-6.646)		a
Hospital
Community	1.000		1.000		1.000		1.000
Academic	9.675	(8.033-11.653)	1.265	(1.019-1.570)	2.311	(1.647-3.242)	0.857	(0.624-1.175)
Hosmer-Lemeshow	b0.001		b0.001		0.791		0.458
goodness of fit
model test
CI indicates confidence interval. a Not included in the model.

Discussion

This study presents descriptive characteristics of trauma patients and assessed the independent effects of Ps on early resource utilization. After controlling for covariates includ-

ing mortality, early indication of critical care except ventilation on days 1 and 2 had a greater impact on TC on days 1 and 2. A Ps less than .9400 or from .9700 to .9799 and injured “extremities, pelvic girdle” or “face” were found to be highly correlated to early TC. Critical care rather than

Ps is a potential predictor of early resource use, although an inverse linear correlation was not observed among Ps groups. However, Ps did significantly influence early indication of ventilation or blood transfusion. When divided according to site of injury, “head and neck” or “chest” impacted early use of ventilation; and “extremities, pelvic girdle” impacted use of blood transfusion.

Traumatic brain injury associated with or without multiple injuries is the frequently assessed parameter in health care research on short- or long-term outcome, the latter of which especially is assumed to be a greater Economic burden in general health [1,2,16,17]. In this study, “head and neck” or lower Ps somewhat corresponding to lower Glasgow Coma Scale could indicate an association with ventilation as a therapeutic intervention, which, however, did not result in higher early resource use. The effect of ventilation may be masked by other aspects of critical care; and injury site or Ps explained the variability of TC up to day 2 to a moderate extent, possibly indicating that trauma-related severity index or anatomically damaged organs are dominated variables in predicting costs in the early phase.

Probability of survival and injured organ were signifi- cantly associated with use of critical care consuming TC on days 1 and 2, which showed a wider range for Ps less than

.9400 (Fig. 1). Total charges on days 1 and 2 showed a narrower range and lower median for Ps from .9600 to

.9699. Probability of survival itself was expected to contain a wide variety of case-mix relevant to trauma, vital signs determining prompt indication or quantity of critical care as well as anatomical damage necessitating some operative repair or enough OR time.

In addition, recent criticisms of ISS or RTS in mortality prediction models at admission have brought into question the validity of Ps and have proposed modifications or new models to enhance accuracy of scoring [18,19]. In the current study, we revealed that injury site significantly differed between the Ps categories, shown in Table 1. For example, “extremity, pelvic girdle” was dominant in Ps from .9600 to

.9699 and associated with females, higher PEC, and transfusion during hospitalization (Table 1). Thus, a discrete AIS or even ISS, if it was the same value, might not predict the precise outcome variation including resource use. Christensen et al (2008) [3] determined costs based on a multivariate regression model using ISS, where the covari- ates explained only 28% of the variation of hospital costs. Coefficient of determination for TC in our study using Ps was 50%, and our model might have superiority of estimating costs. To unravel the miasma of Ps, the interaction of Ps and damaged organs should be evaluated (Annex Table 1). “Head and neck” or “extremities or pelvic girdle” for Ps less than .9400 was a strong predictor of early resuscitative care including critical time-sensitive resource use such as intracranial pressure monitor or embolization. Our model attempted to eliminate both effect of surgical procedures at OR and use of critical care so that conservative management

for injured “chest” or “abdomen, pelvic contents” would not be expected to increase costs during critical periods.

The Therapeutic Intervention Scoring System-28 or the Simplified Acute Physiology Score might be a promising clinical indicator for measuring the appropriate indication for time-sensitive intensive care for critical patients [7]. Assuming that the payments for or monitoring appropriate- ness of ICU care for trauma patients would be performed on the basis of TRISS alone, caution should be paid in cases of Ps less than .9900. Because Ps groups included a wide variety of case-mix, “head and neck” or “pelvic structures” should not be ignored when assessing quality of ICU care. This is due to the fact that careful monitoring of presence of ongoing retroperitoneal bleeding or intracranial pressure may need appropriate or timely use of diagnostic or laboratory tests resulting in greater costs for the efficient care process. A study similar to the current study would overcome to some extent the disadvantage of TRISS caused by the limitation of ISS and concurrently suggest the possibility of sophisticating TRISS to make the most effective use of the trauma care system.

The application of our results must be interpreted in light of several limitations. First, eligible patients represented only 10% of the trauma case-mix population; and the study period was limited to 6 months, which may not reflect a broader population. A comparison of the 10% of cases in the study population and the other 90% of cases might reveal that the study population in general comprised severer cases of older patients, higher rates of ambulance use, and more resources spent overall (Annex Table 2). Because many variables are required to calculate Ps, the study population might comprise patients mainly from hospitals focused on quality of trauma care and acute medicine. Thus, we may have introduced some bias toward institutions that provide injury and critical care medicine. Under the guidance of the Japanese Associa- tion for the Surgery of Trauma, the trauma registry and database used in this study have increased in sample size yearly to represent more hospitals and to more rigorously apply AIS coding, thereby allowing for greater access to information for future studies [20].

Second, we targeted a study population after admission and did not count the resource use spent on emergency department or outpatient facilities. This might under- estimate costs during the early resuscitative period, although practice patterns as the first aid were so limited that they would not influence TC. Our administrative database included the detailed care process before hospi- talization in the same format as in this study, and it is possible to resolve this limitation.

Conclusions

Our analysis demonstrated that use of ICU care, OR, and blood transfusion was associated more strongly with TC on

days 1 and 2 than Ps or injury site. Probability of survival was a strong predictor of use of critical care. Quantity of critical care or TC differed significantly between Ps categories and showed a wide range in Ps less than .9400. “Extremity, pelvic girdles” caused this variation of costs or use of critical care. Probability of survival combined with injury site might contribute to fair payment for ICU care and should be taken into account when measuring quality of ICU

Annex Table 1 (continued)

Estimation SE P Ps .9500-.9599 * abdomen, pelvic contents -0.014 0.051 .779

Ps .9400-.9499 * abdomen, pelvic contents -0.014 0.087 .872

Ps b .9400 * abdomen, pelvic contents 0.002 0.027 .936

Ps N .9900 * extremity, pelvic girdle 0.059 0.010 .000

Ps .9800-.9899 * extremity, pelvic girdle 0.084 0.021 .000

Ps .9700-.9799 * extremity, pelvic girdle 0.019 0.010 .057

Ps .9600-.9699 * extremity, pelvic girdle 0.065 0.011 .000

Ps .9500-.9599 * extremity, 0.041 0.025 .102

pelvic girdle

Ps .9400-.9499 * extremity, 0.039 0.033 .230

pelvic girdle

Ps b .9400 * extremity, pelvic girdle 0.089 0.015 .000

F test for the model. P less than .001. Coefficient of determination, 0.505. Controlling for study variables indicated in Table 3. SE, standard error.

Annex Table 1 (continued)

Estimation SE P

Ps .9600-.9699 ? abdomen, pelvic contents

Ps .9500-.9599 ? abdomen, pelvic contents

Ps .9400-.9499 ? abdomen, pelvic contents

0.010 0.040 .796

-0.014

0.051 .779

-0.014

0.087 .872

Ps .9700-.9799 ? extremity, pelvic girdle

Ps .9600-.9699 ? extremity, pelvic girdle

Ps .9500-.9599 ? extremity, pelvic girdle

Ps .9400-.9499 ? extremity, pelvic girdle

Ps b .9400 ? extremity, pelvic girdle

0.019 0.010 .057

0.065

0.011 .000

0.041

0.025 .102

0.039

0.033 .230

0.089

0.015 .000

F test for the model. P less than .001. Coefficient of determination, 0.505. Controlling for study variables indicated in Table 3. SE, standard error.

Ps .9600-.9699 ? chest	0.045	0.027 .102	Mortality		1062 (66)	173 (17)	.078
Ps .9500-.9599 ? chest	-0.010	0.038 .800	(n on days
Ps .9400-.9499 ? chest	0.023	0.041 .571	1 and 2)
Ps b .9400 ? chest -0.040 0.021 .056 Ps N .9900 ? abdomen, pelvic contents 0.023 0.021 .286 Ps .9800-.9899 ? abdomen, 0.009 0.034 .799 pelvic contents			Critical care, n (% of overall cases) Blood transfusion n (%) 1646 (33.7) 379 (41.2) b.001

Ps .9700-.9799 ? abdomen,	-0.072	0.019 .000
pelvic contents

care. To further refine the cost estimation model in any	Ps b .9400 ? abdomen, pelvic contents	0.002	0.027 .936
period during hospitalization, future studies should confirm	Ps N .9900 ? extremity, pelvic girdle	0.059	0.010 .000
the impact of Ps on resources spent or on indication of critical care by using both TRISS and other variables like	Ps .9800-.9899 ? extremity, pelvic girdle	0.084	0.021 .000

location or number of damaged organs. In turn, such a refined prediction model may offer policy implications on how to optimize facility or staff reallocation or practice patterns in trauma management.

Appendix A

Estimation SE P

Interaction of Ps and injured organ (for Ps N .9900 * external)

Ps .9800-.9899 * external -0.001 0.018 .967

Ps .9700-.9799 * external 0.014 0.020 .470

Ps .9600-.9699 * external -0.044 0.031 .147

Ps .9500-.9599 * external 0.055 0.041 .177

Ps .9400-.9499 * external 0.103 0.108 .343

Ps b .9400 * external -0.040 0.021 .053

Ps N .9900 * head and neck -0.015 0.014 .265

Ps .9800-.9899 * head and neck 0.023 0.019 .234

Ps .9700-.9799 * head and neck 0.040 0.019 .032

Ps .9600-.9699 * head and neck 0.022 0.025 .372

Ps .9500-.9599 * head and neck 0.027 0.030 .364

Ps .9400-.9499 * head and neck 0.046 0.023 .044

Ps b .9400 * head and neck 0.065 0.015 .000

Ps N .9900 * face -0.035 0.017 .039

Ps .9800-.9899 * face 0.003 0.033 .928

Ps .9700-.9799 * face -0.012 0.029 .691

Ps .9600-.9699 * face -0.136 0.050 .007

Ps .9500-.9599 * face -0.002 0.048 .964

Ps .9400-.9499 * face -0.026 0.108 .813

Ps b .9400 * face -0.013 0.026 .628

Ps N .9900 * chest 0.009 0.018 .621

Ps .9800-.9899 * chest 0.002 0.025 .929

Ps .9700-.9799 * chest -0.038 0.020 .055

Ps .9600-.9699 * chest 0.045 0.027 .102

Ps .9500-.9599 * chest -0.010 0.038 .800

Ps .9400-.9499 * chest 0.023 0.041 .571

Ps b .9400 * chest -0.040 0.021 .056

Ps N .9900 * abdomen, pelvic contents 0.023 0.021 .286

Ps .9800-.9899 * abdomen, pelvic contents 0.009 0.034 .799

Ps .9700-.9799 * abdomen, -0.072 0.019 .000

pelvic contents

Ps .9600-.9699 * abdomen, pelvic contents 0.010 0.040 .796

Annex Table 1 Interaction of Ps and injured organ and log10- transformed TC on days 1 and 2

0.020 .470

0.014

-0.001 0.018 .967

Ps .9800-.9899 ? external

Ps .9700-.9799 ? external

Interaction of Ps and injured organ (for Ps N .9900 ? external)

Estimation SE P

Annex Table 2 Comparison of patient characteristics and critical care on days 1 and 2 between the inclusion and exclusion population (n, %)

Ventilation

n (%)

808 (64.2) 281 (77.4) b.001

Ps .9600-.9699 ? external Ps .9500-.9599 ? external	-0.044 0.055	0.031 .147 0.041 .177	Exclusion Inclusion P
Ps .9400-.9499 ? external	0.103	0.108 .343	Overall		75 175 (90.2)	8207 (9.8)
Ps b .9400 ? external	-0.040	0.021 .053	Age	b15 y	7044 (9.4)	686 (8.4)	b.001
Ps N .9900 ? head and neck	-0.015	0.014 .265		>=55 y	41 878 (55.7)	4970 (60.6)
Ps .9800-.9899 ? head and neck	0.023	0.019 .234	Age, mean, SD		53.5 (26.8)	56.4 (26.7)	b.001a
Ps .9700-.9799 ? head and neck	0.040	0.019 .032	Sex
Ps .9600-.9699 ? head and neck	0.022	0.025 .372		Male	38 498 (51.2)	4280 (52.2)	.106
Ps .9500-.9599 ? head and neck	0.027	0.030 .364	Ambulance
Ps .9400-.9499 ? head and neck	0.046	0.023 .044	Used		24 671 (32.8)	4419 (53.8)	b.001
Ps b .9400 ? head and neck	0.065	0.015 .000	Mechanism of injury
Ps N .9900 ? face	-0.035	0.017 .039	Blunt		74 322 (98.9)	7608 (92.7)	b.001
Ps .9800-.9899 ? face	0.003	0.033 .928	No. of PEC
Ps .9700-.9799 ? face -0.012 0.029 .691 Ps .9600-.9699 ? face -0.136 0.050 .007 Ps .9500-.9599 ? face -0.002 0.048 .964 Ps .9400-.9499 ? face -0.026 0.108 .813 Ps b .9400 ? face -0.013 0.026 .628 Ps N .9900 ? chest 0.009 0.018 .621 Ps .9800-.9899 ? chest 0.002 0.025 .929 Ps .9700-.9799 ? chest -0.038 0.020 .055			0 64 514 (85.8) 6933 (84.5) .004 1 8860 (11.8) 1054 (12.8) >=2 1801 (2.4) 220 (2.7) Procedure Present 52 174 (69.4) 5568 (67.8) .004 Hospital Academic 11 558 (15.4) 1029 (12.5) b.001 Outcome

	Annex Table 2	(continued)
	Exclusion Inclusion P
	ICU care
		n (%)	4907 (81.4) 1567 (93.6) b.001
	Use of OR
		n (%)	15 867 (21.8) 1280 (65.8) b.001
	Blood	Median	800 (800) 800 (1200) b.001b
	transfusion,	(IQ)
	mL
	OR time, min	Median	120 (90) 135 (99) b.001b
		(IQ)
	TC on days 1	Median	1173 (1491) 1301 (1862) b.001b
	and 2, US $	(IQ)
	a Compared by t test. b Compared by Mann-Whitney test. Others by Fisher exact test. SD, standard deviation.

References

1. Harris IA, Young JM, Rae H, et al. Predictors of general health after major trauma. J Trauma 2008;64:969-74.
2. Gabbe BJ, Simpson PM, Sutherland AM, et al. Functional measures at discharge. Are they useful predictors of longer term outcomes for trauma registries? Ann Surg 2008;247:854-9.
3. Christensen MC, Ridley S, Lecky FE, et al. Outcomes and costs of blunt trauma in England and Wales. Critical Care 2008;12:R23.
4. Boyd CR, Tolson MA, Copes WS. Evaluating trauma care: the TRISS method. J Trauma 1987;27:370-8.
5. Champion HR, Copes WS, Sacco WJ, et al. A new characterization of injury severity. J Trauma 1990;30:539-46.
6. Champion HR, Sacco WJ, Copes WS, et al. A revision of the trauma score. J Trauma 1989;29:623-9.
7. Organisation und Datenverwaltung fur landesgesundheitsfondsfi- nanzierte Krankenanstalten. Handbuch zur Dokumentation in Landesgesundheitsfonds-finanzierten Krankenanstalten (Anhang1)

   Available at http://www.bmgfj.gv.at/cms/site/attachments/0/7/6/ CH0719/CMS1159518265839/anhang1_-_aufbau_und_inhalt_ der_datenmeldungen.pdf

   Champion HR, Sacco WJ, Copes WS. Injury severity scoring again. J Trauma 1995;38:94-5.

8. Murphy JG, Cayten CG, Ayten MD, et al. Controlling for the severity of injuries in emergency medicine research. Am J Emerg Med 1990;8: 484-91.
9. Ruger JP, Lewis LM, Richter CJ. Identifying high-risk patients for triage and resource allocation in the ED. Am J Emerg Med 2007;25: 794-8.
10. MacLennan P, Jones B, McGwin G, et al. Trauma center resource consumption and mortality. Am Surg 2006;72:1-6.
11. Hirose M, Imanaka Y, Ishizaki T, et al. Profiling hospital performance of laparoscopic cholecystectomy based on the admin- istrative data of four teaching hospitals in Japan. World J Surg 2005; 29:429-35.
12. Sundararajan V, Henderson T, Perry C, et al. New ICD-10 version of the Charlson Comorbidity Index predicted in-hospital mortality. J Clin Epidemiol 2004;57:1288-94.
13. Sundararajan V, Quan H, Halfon P, et al. Cross-national comparative performance of three versions of the ICD-10 Charlson index. Med Care 2007;45:1210-5.
14. Lynne Moore L, Lavoie A, Sage NL, et al. Using information on preexisting conditions to predict mortality from traumatic injury. Ann Emerg Med 2008;52:356-64.
15. Wu J, Sheng L, Ma Y, Gu J, et al. The analysis of risk factors of impacting mortality rate in severe Multiple trauma patients with Posttraumatic acute respiratory distress syndrome. Am J Emerg Med 2008;26:419-24.
16. Colantonio A, Escobar MD, Chipman M, et al. Predictors of postacute mortality following traumatic brain injury in a seriously injured population. J Trauma 2008;64:876-82.
17. Kilgo PD, Osler TM, Meredith W. The worst injury predicts mortality outcome the best: rethinking the role of multiple injuries in trauma outcome scoring. J Trauma 2003;55:599-607.
18. Osler T, Glance L, Buzas JS, et al. A trauma Mortality prediction model based on the anatomic injury scale. Ann Surg 2008;247: 1041-8.
19. AIS: Available at http://www.jast-hp.org/ [access 4 November 2008].