Injury, Int. J. Care Injured 46 (2015) 1784–1789
Contents lists available at ScienceDirect
Injury journal homepage: www.elsevier.com/locate/injury
Current state of trauma care in China, tools to predict death and ICU admission after arrival to hospital Guilan Kong a, Xiaofeng Yin b, Tianbing Wang b, Richard Body c, Yu-Wang Chen d, Jing Wang a, Liying Cao e, Shouling Wu e, Jingli Gao e, Guosheng Wang e, Yonghua Hu a, Baoguo Jiang b,* a
Medical Informatics Center, Peking University, Beijing 100191, China Department of Trauma and Orthopaedics, Peking University People’s Hospital, Beijing 100044, China c Emergency Department, Manchester Royal Infirmary, Oxford Road, Manchester M13 9WL, UK d Decision and Cognitive Sciences Research Centre, The University of Manchester, Manchester M15 6PB, UK e Kailuan Hospital, Tangshan City, Hebei Province 063000, China b
A R T I C L E I N F O
A B S T R A C T
Article history: Accepted 1 June 2015
Background: In China, a nationwide emergency system takes charge of pre-hospital emergency services, and it adopts a proximity principle to send trauma patients to the nearest hospitals. However, many severely injured patients have been sent to low level hospitals with no capability to treat severe trauma. Thus those patients with high probability of in-hospital death or intensive care unit (ICU) admission need to be identified in the emergency department (ED) for optimal utilisation of hospital resources and better patient outcomes. The purpose of the study was to develop a computerised tool to aid ED physicians’ prediction of in-hospital death and ICU admission for trauma patients after arrival to hospital. Methods: We reviewed a sample of 1,299 trauma patients who had been directly sent to the ED at Kailuan Hospital, North China. After excluding those cases with incomplete data entry, information of 1,195 patients was employed for analysis. The primary outcome was severe trauma that either resulted in death in hospital or in ICU admission. We proposed to use a complementary approach to combine the Pre-Hospital Index (PHI), the Trauma Index (TI), and the Glasgow Coma Score (GCS) in a decision support system (DSS) to assess trauma and predict in-hospital death and ICU admission. The sensitivity, specificity, over-triage rate, and under-triage rate were used as measurements to compare system performances of the DSS with the three scoring tools. Results: Among the 1,195 patients, 30 (2.5%) had severe trauma. The proposed DSS showed the best sensitivity (66.7%; 95% CI: 49.8–83.6%) among all the four studied tools. The TI (sensitivity 50.0%, 95% CI: 32.2–67.8%) performed slightly better than the GCS (sensitivity 46.7%, 95% CI: 28.9–64.5%), while both the TI and GCS performed better than the PHI (sensitivity 30.0%, 95% CI: 13.5–46.5%). The performance differences between the DSS and the three extant scoring tools were statistically significant. Conclusions: The proposed DSS outperformed the extant trauma scoring systems. It has a strong potential to help ED physicians identify severe trauma, optimally utilise hospital resources, and recommend appropriate triage and treatment strategies for trauma patients that have strong possibilities for in-hospital death and ICU admission. ß 2015 Published by Elsevier Ltd.
Keywords: Trauma Decision support system In-hospital death ICU admission Sensitivity Specificity
Introduction Trauma is an increasingly significant public health issue around the globe. It has become one of the leading causes of mortality and
* Corresponding author. Tel.: +86 10 88324570; fax: +86 10 88324570. E-mail address: [email protected] (B. Jiang). http://dx.doi.org/10.1016/j.injury.2015.06.002 0020–1383/ß 2015 Published by Elsevier Ltd.
disability worldwide [1,2]. Trauma accounts for 16% of the global burden of disease and 16,000 people die from injury every day. In addition millions more are temporarily or permanently disabled due to trauma. Without appropriate actions road traffic injuries are predicted to become the third leading cause of global morbidity and mortality [3]. Death and disability from trauma is especially high in low- and middle-income countries where approximately 90% of the total burden of trauma in the world occurs. In developed
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
countries, different levels of trauma centres were established to treat trauma with varying degrees of severity. Initial assessment and outcome prediction for trauma patients after arrival to hospital can help identify patients who may need intensive care unit (ICU) admission or may die in hospital and ensure that severely injured patients can get appropriate treatment, which may lead to optimal utilisation of hospital resources and a reduced level of mortality and disability. Unfortunately, there are no nationwide initial trauma assessment guidelines or tools to aid physicians in pre-hospital environments or the emergency departments (EDs) of many less developed countries, e.g., Ghana [4] and China. This study examines China as a case study. A pre-hospital ‘120’ (the ambulance call number used in China) emergency system currently operates throughout the country. A typical, local ‘120’ emergency system usually consists of an emergency command centre, one or more emergency command stations supervised by the command centre, and local hospitals. The emergency command centre takes charge of ‘120’ calls and dispatches ambulances and pre-hospital rescuers to accident sites to transport trauma patients to the nearest hospitals. There are no trauma centres in China, and ranked hospitals are analogous to ranked trauma centres in developed countries. Hospitals are ranked according to medical services and management, quality and safety of clinical care, and technical level and efficiency [5]. As a result, higher level hospitals receive more government funding and can provide better diagnostic care and therapeutic options for trauma patients than lower level hospitals. On the contrary, low level hospitals have little or no capability to treat complicated trauma cases. Some severe trauma patients sent to low level hospitals by the ‘120’ emergency system will have to be retransferred to higher level hospitals. Trauma patient outcomes in China are relatively poor compared with those in developed countries, partially due to the time wasted in transportation. More than 400,000 people die in China each year from injury, and trauma is the fifth leading cause of death after malignant tumours and cardiac, cerebral, and respiratory diseases. It is the most common cause of death of young people aged 18–40 years in China [6]. Being confronted with the increasing challenges in trauma care, National Health and Family Planning Commission (NHFPC) of China and the Trauma Society of Chinese Medical Association have jointly launched a national research project to develop a computerised trauma assessment tool to aid ED physicians to assess trauma and predict in-hospital death and ICU admission. This study is part of that project. The focus of this study was on assessment of trauma and prediction of in-hospital death and ICU admission for patients after their arrival to hospital. Various severity scoring tools [7] have been developed from physiological, anatomical, or composite perspectives to assess trauma and predict in-hospital death and ICU admission. Since detailed diagnoses of trauma cannot be made upon patients’ arrival to hospital, anatomical scoring systems are not applicable in this study. Frequently used physiological trauma scoring systems in the literature include the Pre-Hospital Index (PHI) [8], Trauma Index (TI) [9,10], Glasgow Coma Score (GCS) [11,12], and the Revised Trauma Score (RTS) [13]. Studies [9,10,14,15] show that there is a good correlation between the severity score assigned by any of the above three systems PHI, TI, and GCS and the actual severity of trauma. Patients who are assessed with severe trauma by any of the PHI, TI and GCS systems have significantly high probabilities of in-hospital death or ICU admission. Thus PHI, TI, and GCS can be used to predict which trauma patients are most likely to die in hospital or will be admitted to the ICU. Based on previous experiences of using knowledge-based, clinical decision support systems (CDSSs) [16,17] to aid physicians to make clinical decisions, we proposed to integrate the PHI, TI and GCS scoring systems in a decision support system (DSS) to help ED
1785
physicians carry out rapid initial trauma assessment and predict in-hospital death and ICU admission upon patients’ arrival to hospital. The RTS was not taken into consideration in this study because it combines the GCS with respiratory rate and systolic blood pressure (SBP); the latter two variables are measured by the PHI and both the GCS and PHI are included in the DSS. Our hypothesis was that the proposed DSS should perform better than the PHI, TI, and GCS in identifying patients with strong possibilities for in-hospital death and ICU admission. It can improve the accuracy of ED physicians’ predictions of in-hospital death and ICU admission for trauma patients and provide them with appropriate trauma management strategies. Consequently, it can help to optimise the utilisation of hospital resources and reduce mortality and disability associated with trauma. This hypothesis has been supported by the empirical study. Materials and methods Data sources and study subjects This study was carried out at Kailuan Hospital, Tangshan City, a typical city located in Hebei province, North China, where all ambulance services are supplied by the ‘120’ pre-hospital emergency system. The size of Tangshan is about 13,472 km2, and it hosts a population of 7.2 million. The city was fully rebuilt after a 7.8 magnitude earthquake in 1976. Kailun Hospital is one of six top-ranked hospitals in Tangshan. The hospital has around 1,000 beds, and it has the capability to treat patients with the most severe trauma. Patients were included for analysis if they met the following criteria: (a) directly sent to the ED from the accident site; (b) clinical data required by the PHI, TI, and GCS were well-recorded upon the arrival to ED; and (c) corresponding in-hospital data could be retrieved. No further restrictions were made about the severity or characteristics of the cases. There were 1,299 trauma patients directly sent to the ED at Kailuan Hospital in the sample period. Out of 1,299 trauma patients included in the study, 1,195 (92.0%) patients had both the ED data and in-hospital data. The remaining 104 patients had either missing ED data or missing inhospital data, and thus had to be excluded from the data analysis. The ED data recorded at Kailuan Hospital contain information on basic demographics, dates of arrival to the ED, the location, cause, and type of body injury, body temperature, respiratory rate, pulse rate, blood pressure, eye-opening status, verbal response, and motor response. The retrieved in-hospital data include length of stay in the ICU, length of stay in hospital, and the patient’s condition at discharge. The primary outcome of this study was a composite of both inhospital death and ICU admission, which was defined as severe trauma. Ethical review of this study was approved by the biomedical ethics committee of the first author’s institution and the participating hospital. Identifying severe trauma using a DSS by integrating the PHI, TI, and GCS The PHI and TI scores increase with trauma severity, while the GCS score decreases with severity. The three scoring systems stratify trauma patients into three groups, namely severe, moderate, and minor, according to severity scores assigned by respective systems based on patients’ clinical signs and symptoms. The score range for each severity group of the three scoring systems is shown in Table 1 [8–12]. Studies show that patients with severe trauma, identified by any of the three scoring systems, have strong possibilities for
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
1786
Table 1 Score range for Severe, Moderate, and Minor of the PHI, TI, and GCS systems.
PHI TI GCS
Table 2 ‘IF-THEN’ rules for combing the PHI, TI, and GCS systems.
Severe
Moderate
Minor
IF
THEN
8–20 17–30 3–8
4–7 10–16 9–12
0–3 5–9 13–15
‘Severe’ by any system of PHI, TI, and GCS ‘Moderate’ by all systems of PHI, TI, and GCS ‘Moderate’ by GCS, and ‘Moderate’ by either PHI or TI ‘Moderate’ by GCS, and ‘Minor’ by both systems of PHI and TI ‘Minor’ by GCS, and ‘Moderate’ by either PHI or TI ‘Minor’ by all systems of PHI, TI, and GCS
Severe Severe Severe Moderate Moderate Minor
in-hospital death or ICU admission, while the probabilities of adverse events for patients with moderate or minor trauma are low [9,10,14,15,18]. Since each scoring system has its merits, specifically, the PHI focuses on measurements of vital signs, the TI takes into account measurements of trauma-specific signs, and the GCS focuses on measurements of consciousness, we proposed to integrate the PHI, TI, and GCS in a complementary way that will identify patients with strong possibilities for in-hospital death and ICU admission. The rationale behind the proposed DSS is that severe trauma patients have significantly high probability of inhospital death and ICU admission and that mortality and disability of trauma can be reduced if timely and proper trauma triage and treatment strategies are provided for those patients in the ED. To this end, we invited five trauma experts from Peking University People’s Hospital and five experts in emergency medicine from Beijing Emergency Medical Center to form an advisory panel and organised small meetings to discuss how to combine the PHI, TI, and GCS to provide a rapid initial assessment of patient trauma and predict in-hospital death and ICU admission. We adopted a modified Delphi approach [19] to seek consensus. First, we requested each expert to come up with a proposal for combinational use of the three scores. Second, we requested each expert to defend his or her proposal using his or her best knowledge. Third, we asked the expert panel to vote for the best proposal. After several meetings, the expert panel achieved consensus and proposed to use a decision tree to combine the PHI, TI, and GCS scores to determine the severity of injury in the ED, thus predicting in-hospital death and ICU admission based on the
severity level. The decision tree for combining the PHI, TI, and GCS is shown in Fig. 1. There are six IF-THEN rules that can be derived from the proposed decision tree, and the rules are described in Table 2. It is clear from Fig. 1 and Table 2 that although the same set of trauma severity categories (severe, moderate, and minor) are utilised by the proposed DSS and its component systems, the coverage of resultant severe trauma in the DSS is broader than that of severe trauma in the PHI, TI, and GCS. Particularly, the proposed DSS is more cautious in identifying patients with severe trauma than the three scoring systems. That is to say, in addition to severe trauma identified by any one of the three scoring tools, moderate trauma identified by the GCS and either of the PHI and TI simultaneously will be considered as severe trauma in the DSS. Consequently, the coverage of resultant moderate and minor trauma in the DSS is narrower than that of moderate and minor trauma in the three scoring systems. Because patients with moderate or minor trauma identified by the PHI, TI, and GCS have low probability of in-hospital death and ICU admission, and a proportion of patients with moderate trauma identified by the three extant scoring tools was assessed with severe trauma in the DSS, the probability of adverse events linked with moderate or minor trauma in the DSS will be even lower than any of the three extant scoring tools. This shows that the three severity categories of the four tools are necessary for assessing trauma severity in the DSS.
Fig. 1. The decision tree for combining the PHI, TI, and GCS systems.
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
The knowledge base of the DSS contains those combination rules as shown in Table 2. The DSS was designed to have clientserver structure. The client-side can be installed as software on a handheld device to equip ED physicians, and the server-side software can be installed on the hospital’s hard server. The clientside aids the capture of patient demographics, vital signs, and other available clinical data required for severity assessment; it also communicates with the server-side and helps the hospital trauma team evaluate the patient’s degree of trauma severity and know whether the patient has a strong possibility for in-hospital death or ICU admission.
Table 3 Characteristics of studied trauma data. Variable
Subgroup
Non-severe trauma patients (N:1165)
Severe trauma patients (N:30)
Gender
Male Female
936(80.3%) 229(19.7%)
26(86.7%) 4(13.3%)
Age (years)
Mean age (SD)
43.1(16.0)
40.0(17.5)
Trauma cause
Traffic accident Blunt force injury Sharp object injury Fall Crush/burial Explosion/fire Burn/electrical injury Frozen injury Sports injury Battle injury/gunshot injury Other
298(25.6%) 290(24.9%) 115(9.9%) 291(25.0%) 85(7.3%) 8(0.7%) 14(1.2%) 0 3(0.3%) 0 61(5.2%)
22(73.3%) 0 0 5(16.7%) 2(6.7%) 0 0 0 0 0 1(3.3%)
Analysis We extracted the PHI, TI, and GCS scores for each trauma patient on the basis of his or her recorded ED data. The purpose of the proposed DSS was to identify patients who have strong possibilities for adverse events, so as to suggest proper trauma management strategies for them. Based on the proposed decision tree as shown in Fig. 1, patients with severe trauma assessed by the DSS were quickly identified from the collected dataset. Further, to measure the prediction performance of the DSS, we identified which patients had actually either died in hospital or were admitted to the ICU. The next step was to calculate performance measures, such as sensitivity, specificity, and over- and undertriage rates of the DSS. We realised that calculation of sensitivity, specificity, and overand under- triage rates required the classification tools to provide binary categories; however, there are three severity categories in the DSS and the three scoring systems. As discussed before, patients with severe trauma have significantly high probability of in-hospital death or ICU admission, while patients with moderate or minor trauma have low probability of adverse events; therefore, we bundled moderate and minor trauma as non-severe trauma in the four studied tools for the following data analysis. Accordingly, the DSS together with the three scoring systems were changed to be binary classification tools for calculating performance measures including sensitivity, specificity, over-triage rate (1 specificity), and under-triage rate (1 sensitivity). We used SPSS to analyse the collected trauma data. Firstly, we used definitive cutoff values in the PHI, TI, and GCS systems to classify patients into severe, moderate, or minor trauma groups respectively. Secondly, based on the severity results generated by the three extant scoring systems, we used the proposed DSS to reclassify the patients into severe, moderate, or minor groups based on the decision tree as shown in Fig. 1. Thirdly, we bundled moderate and minor trauma as non-severe trauma for each trauma assessment tool and used the sensitivity, specificity, over-triage rate, and under-triage rate in identifying severe trauma to measure
1787
and compare the PHI, TI, GCS, and the proposed DSS. Finally, the McNemar test was employed to compare assessment performances of all four prediction tools.
Results Among the studied cases, 1,195 patients, 14 (1.2%) died in hospital, 21 (1.8%) were admitted to the ICU, 1,177 (98.5%) fully recovered at discharge, and 2 had improved at discharge. Based on the recorded death and ICU admission, we identified 30 (2.5%) patients with actual severe trauma and 1,165 (97.5%) with actual non-severe trauma. The characteristics of the studied data are described in Table 3. In the excluded 104 cases, 76 (73.08%) were male, 28 (26.92%) were female: the mean age of the male patients is 39.3 (SD = 17.3), and the mean age of the female patients is 34.5 (SD = 14.8). Also, none of them had been admitted to ICU and none died. As the excluded patients were a very small portion of the collected cases and contained randomly missing data, the exclusion would not affect the results of the study. Table 4 presents the performance measures including sensitivity, specificity, over-triage rate, under-triage rate and corresponding 95% confidence intervals (CI) of the PHI, TI, GCS, and the proposed DSS in identifying severe trauma, which may result in either in-hospital death or ICU admission. For convenience, the numbers of identified cases of severe trauma, non-severe trauma, over-assessed non-severe trauma, and under-assessed severe trauma are shown in Table 4, too.
Table 4 Sensitivity, specificity, over-triage rate, and under-triage rate of the PHI, TI, GCS, and the proposed DSS for predicting ICU admission or death in hospital. PHI
TI
GCS
The proposed DSS
Sensitivity
Value 95% CI Identified severe trauma (N)
30.0% 13.5–46.5% 9
50.0% 32.2–67.8% 15
46.7% 28.9–64.5% 14
66.7% 49.8–83.6% 20
Specificity
Value 95% CI Identified non-severe trauma (N)
99.8% 99.6–100% 1163
94.4% 93.0–95.8% 1100
99.2% 98.6–99.8% 1156
93.7% 92.3–95.1% 1091
Over-triage rate
Value 95% CI Over-assessed non-severe trauma (N)
0.2% 0–0.4% 2
5.6% 4.2–7.0% 65
0.8% 0.2–1.4% 9
6.3% 4.9–7.7% 74
Under-triage rate
Value 95% CI Under-assessed severe trauma (N)
70.0% 53.5–86.5% 21
50.0% 32.2–67.8% 15
53.3% 35.5–71.1% 16
33.3% 16.4–50.2% 10
1788
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
Based on the performance comparison results as shown in Table 4, the proposed DSS had the best values for sensitivity (66.7%; 95% CI: 49.8–83.6%) and under-triage rate (33.3%, 95% CI: 16.4–50.2%) among the studied physiological scoring tools. The TI (sensitivity: 50.0%, 95% CI: 32.2–67.8%; under-triage rate: 50.0%, 95% CI: 32.2–67.8%) performed slightly better than the GCS (sensitivity: 46.7%, 95% CI: 28.9–64.5%; under-triage rate 53.3%, 95% CI: 35.5–71.1%), while both the TI and GCS performed better than the PHI (sensitivity 30.0%, 95% CI: 13.5–46.5%; under-triage rate 70.0%, 95% CI: 53.5–86.5%). Inevitably, the specificity decreased when the sensitivity increased. Results from the McNemar test showed that the performance differences between the proposed DSS and the three scoring tools, namely the PHI, TI, and GCS were statistically significant (DSS vs TI: p = 0.004; DSS vs GCS: p < 0.001; DSS vs PHI: p < 0.001). Discussion and conclusions The purpose of the study was to develop a computerised tool to help ED physicians identify patients with strong possibilities for inhospital death or ICU admission. In this study, we proposed to employ a decision tree to integrate the three extant physiological scoring systems, namely the PHI, TI, and GCS into a composite DSS. The performance of the proposed DSS was tested using a retrospective set of trauma data collected at the ED of Kailuan Hospital in Tangshan, a city located in North China. Based on the results (see Table 4), the proposed DSS performed best among the four studied tools in predicting in-hospital death and ICU admission. The superiority of the DSS may be explained by the following. First, the three extant scoring systems have respective merits and deficiencies. The PHI [8] was designed to assess trauma from a physiological perspective, yet its assessment factors consist only of vital signs. The TI [9,10] was designed for assessing trauma not only from a physiological perspective but also from a trauma specific perspective, thus it contains some trauma-specific factors, such as body region of injury and type of injury, besides vital signs. For example, a penetrating injury would be assigned a score of 3 for the injury type in the TI scoring system. However, the PHI and TI may miss some severe head injuries without obvious abnormal physiologic signs. The GCS [11,12], specifically designed to assess the severity of head injuries, focuses on consciousness, while other factors, such as the type of injury and vital signs are not taken into consideration; thus, the GCS may miss some patients who suffer severe trauma but do not exhibit disorders of consciousness. Second, the DSS adopts a cautious approach in combining the PHI, TI, and GCS. The DSS considers a patient to have severe trauma under the following two conditions: (a) labelled severe by any component system; or (b) simultaneously labelled moderate by the GCS and either the PHI or TI. Thus the GCS can complement the TI and PHI in identifying severe head trauma without obvious abnormal physiologic signs, and the TI and PHI can complement the GCS in identifying severe trauma without head injury. As a result, the proposed DSS makes the three physiological scoring systems complement each other to achieve the best performance. In the study, the sensitivity of the proposed DSS (66.7%; 95% CI: 49.8–83.6%) is lower than some trauma assessment models proposed by other researchers such as Cox et al. [20]. The relatively low sensitivity in the study may be due to the following reasons. Cox et al. [20] developed their triage tool in a pre-hospital environment and used much broader criteria to define severe trauma, which included death, injury severity score (ISS) > 15, ICU admission, and urgent surgeries. In this study, we developed the prediction tool in the ED and used stricter criteria, including only in-hospital death and ICU admission to define actual severe trauma. Thus, the number of patients that were identified to have died in hospital or were admitted to the ICU was small (N = 30). As
can be found from Table 3, the incidence rate of death or ICU admission was only 2.5% among trauma patients directly sent to Kailuan Hospital. This may affect sensitivity. Regarding over-triage (non-severe patients assessed as severe) and under-triage (severe patients assessed as non-severe), the values of over-triage (1 specificity) and under-triage (1 sensitivity) rates of the DSS were 6.3% and 33.3%, respectively. As the purpose of inhospital death and ICU admission prediction was to optimise the management of trauma patients and utilisation of hospital resources, a relatively low over-triage rate is good for interhospital transfer and ICU resource allocation. Regarding the calculation efficiency of the proposed DSS, a clarification is necessary. It seems that the decision tree employed in the DSS is a bit complicated for physicians. But on the client-side of the DSS, whether or not a patient has a strong possibility for inhospital death or ICU admission can be predicted and displayed as soon as necessary patient data have been inputted because the rules for combining the PHI, TI, and GCS systems are implemented on the client-side. Since calculation of the PHI, TI, GCS, and the decision tree on the DSS client-side has no need for data transmitted from the server-side, it works just like a mobile phone counter and takes no time. ED physicians have no need to know the details of the inner logic of the DSS. As the client-side and the server-side of the DSS communicate in real time, the hospital IT professional in charge of the trauma system can know the degree of severity of any trauma patient in the ED; thus, they can inform hospital trauma experts in a timely manner to provide appropriate treatment for the patient. Alternatively, they can guide transfer of one patient to a higher level hospital if the patient has severe trauma and the current hospital has no capability to treat the patient. This will ensure that severely injured patients can be treated quickly and well, and the limited hospital resources can be utilised optimally. In addition, a clarification about bundling moderate and minor trauma as non-severe trauma for calculating performance measures is needed. In the proposed decision tree, the four trauma severity assessment tools stratify trauma patients into three groups: severe, moderate, and minor. Since all the three severity categories are necessary antecedent status for those combinational rules as shown in Table 2, the three severity categories are needed for the four tools in assessing trauma severity and predicting in-hospital death and ICU admission. However, to use sensitivity, specificity, over-triage rate, and undertriage rate to measure the stratification or prediction performance of the four tools, we need to change the four tools to binary classification systems with two severity categories. As severe trauma is linked with strong possibilities for in-hospital death or ICU admission, and moderate and minor trauma is linked with low probability of adverse events, we bundled moderate and minor trauma in the four studied tools as non-severe trauma in data analysis for system performance measurement and comparison. This study develops a novel trauma assessment and adverse events prediction tool by incorporating the strengths of three extant severity scoring systems, namely the PHI, TI, and GCS. The findings show that the proposed DSS outperformed the PHI, TI, and GCS in identifying trauma patients with strong possibilities for inhospital death and ICU admission with statistical significance. It is the first study in China to help design and develop a computerised tool that predicts in-hospital death and ICU admission for trauma patients in the ED. The limitation of the study is that it was a retrospective study. To test the applicability of current research, future research needs to prospectively test the DSS. To conclude, the proposed DSS outperformed the extant severity scoring systems in predicting in-hospital death and ICU admission. It has a strong potential to help ED physicians identify severe trauma and send severely injured patients to ICU or higher
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
level hospitals to receive proper trauma care. In our future research agenda, we plan to prospectively test the proposed DSS before it is implemented in a real environment. Conflict of interest
[6]
[7] [8]
We declare that there is no conflict of interest with any financial organization regarding the material discussed in the manuscript entitled ‘‘Current state of trauma care in China, tools to predict death and ICU admission after arrival to hospital’’.
[9] [10] [11]
Acknowledgments [12]
This study was supported by a grant from the National Natural Science Foundation of China (grant no. 81301296). This study was also supported by a grant from the National Health and Family Planning Commission of China (grant no. 201002014). This study was also supported by a project funded by the Ministry of Education of China (grant no. 13YJC630066).
[14]
References
[16]
[1] Mock C, Lormand JD, Goosen J, Joshipura M, Peden M. Guidelines for essential trauma care. Geneva: World Health Organization; 2004. [2] Sasser S, Varghese M, Kellermann A, Lormand JD. Prehospital trauma care systems. Geneva: World Health Organizations; 2005. [3] Peden M, Scurfield R, Sleet D, Mohan D, Hyden AA, Jarawan E. World report on road traffic injury prevention: summary. Geneva: World Health Organization; 2004. [4] Mock CN, Tiska M, Adu-Ampofo M, Boakye G. Improvements in prehospital trauma care in an African country with no formal emergency medical services. The Journal of Trauma: Injury Infection and Critical Care 2002;53(1):90–7. [5] Xu B, Liu H, Su N, Kong G, Bao X, Li J, et al. Association between winter season and risk of death from cardiovascular diseases: a study in more than half a
[13]
[15]
[17]
[18]
[19] [20]
1789
million inpatients in Beijing, China. BMC Cardiovascular Disorders 2013;13:93. Zong Z-W, Li N, Cheng T-M, Ran X-Z, Shen Y, Zhao Y-F, et al. Current state and future perspectives of trauma care system in mainland China. Injury 2011;42(9):874–8. Camp LV, Yates DW. Trauma scoring. In: Soreide E, Grande CM, editors. Prehospital trauma care. New York: Taylor & Francis; 2005. p. 153–67. Koehler JJ, Baer LJ, Malafa SA, Meindertsma MS, Navitskas NR, Huizenga JE. Prehospital Index A scoring system for field triage of trauma victims. Annals of Emergency Medicine 1986;15(2):178–82. Kirkpatrick JR, Youmans RL. Trauma Index, an aide in the evaluation of injury victims. The Journal of Trauma 1971;11(8):711–4. Ogawa M, Sugimoto T. Rating severity of the injured by ambulance attendants: field research of trauma index. The Journal of Trauma 1974;14(11):934–7. Teasdale G, Jennett B. Assessment of coma and impaired consciousness. The Lancet 1974;304(7872):81–4. Wang MC, Gennarelli TA. Classification systems. In: Jallo J, Loftus CM, editors. Neurotrauma and critical care of the brain. New York: Thieme Medical Publishers; 2009. Champion HR, Sacco WJ, Copes WS, Gann DS, Gennarelli TA, Flanagan ME. A revision of trauma score. The Journal of Trauma 1989;29(5):623–9. Mahdian M, Fazel MR, Fakharian E, Akbari H, Mahdian S. Cerebral state index versus Glasgow coma scale as a predictor for in-hospital mortality in braininjured patients. Chinese Journal of Traumatology 2014;17(4):220–4. Tamim H, Joseph L, Mulder D, Battista RN, Lavoie A, Sampalis JS. Field triage of trauma patients: improving on the Prehospital Index. The American Journal of Emergency Medicine 2002;20(3):170–6. Kong GL, Xu DL, Liu XB, Yang JB. Applying a belief rule-base inference methodology to a guideline-based clinical decision support system. Expert Systems 2009;26(5):391–408. Kong GL, Xu DL, Body R, Yang JB, Mackway-Jones KRH, Carley S. A belief rulebased decision support system for clincial risk assessment of cardiac chest pain. European Journal of Operational Research 2012;219(3):564–73. Sampalis JS, Tamim H, Nikolis A, Lavoie A, Williams JI. Predictive validity and internal consistency of the pre-hospital index measured on-site by physicians. Accident Analysis & Prevention 1996;28(6):675–84. Linstone HA, Turoff M. The Delphi method: techniques and applications. London: Addison-Wesley; 1975. Cox S, Currell A, Harriss L, Barger B, Cameron P, Smith K. Evaluation of the Victorian state adult pre-hospital trauma triage criteria. Injury 2012;43(5): 573–81.
Contents lists available at ScienceDirect
Injury journal homepage: www.elsevier.com/locate/injury
Current state of trauma care in China, tools to predict death and ICU admission after arrival to hospital Guilan Kong a, Xiaofeng Yin b, Tianbing Wang b, Richard Body c, Yu-Wang Chen d, Jing Wang a, Liying Cao e, Shouling Wu e, Jingli Gao e, Guosheng Wang e, Yonghua Hu a, Baoguo Jiang b,* a
Medical Informatics Center, Peking University, Beijing 100191, China Department of Trauma and Orthopaedics, Peking University People’s Hospital, Beijing 100044, China c Emergency Department, Manchester Royal Infirmary, Oxford Road, Manchester M13 9WL, UK d Decision and Cognitive Sciences Research Centre, The University of Manchester, Manchester M15 6PB, UK e Kailuan Hospital, Tangshan City, Hebei Province 063000, China b
A R T I C L E I N F O
A B S T R A C T
Article history: Accepted 1 June 2015
Background: In China, a nationwide emergency system takes charge of pre-hospital emergency services, and it adopts a proximity principle to send trauma patients to the nearest hospitals. However, many severely injured patients have been sent to low level hospitals with no capability to treat severe trauma. Thus those patients with high probability of in-hospital death or intensive care unit (ICU) admission need to be identified in the emergency department (ED) for optimal utilisation of hospital resources and better patient outcomes. The purpose of the study was to develop a computerised tool to aid ED physicians’ prediction of in-hospital death and ICU admission for trauma patients after arrival to hospital. Methods: We reviewed a sample of 1,299 trauma patients who had been directly sent to the ED at Kailuan Hospital, North China. After excluding those cases with incomplete data entry, information of 1,195 patients was employed for analysis. The primary outcome was severe trauma that either resulted in death in hospital or in ICU admission. We proposed to use a complementary approach to combine the Pre-Hospital Index (PHI), the Trauma Index (TI), and the Glasgow Coma Score (GCS) in a decision support system (DSS) to assess trauma and predict in-hospital death and ICU admission. The sensitivity, specificity, over-triage rate, and under-triage rate were used as measurements to compare system performances of the DSS with the three scoring tools. Results: Among the 1,195 patients, 30 (2.5%) had severe trauma. The proposed DSS showed the best sensitivity (66.7%; 95% CI: 49.8–83.6%) among all the four studied tools. The TI (sensitivity 50.0%, 95% CI: 32.2–67.8%) performed slightly better than the GCS (sensitivity 46.7%, 95% CI: 28.9–64.5%), while both the TI and GCS performed better than the PHI (sensitivity 30.0%, 95% CI: 13.5–46.5%). The performance differences between the DSS and the three extant scoring tools were statistically significant. Conclusions: The proposed DSS outperformed the extant trauma scoring systems. It has a strong potential to help ED physicians identify severe trauma, optimally utilise hospital resources, and recommend appropriate triage and treatment strategies for trauma patients that have strong possibilities for in-hospital death and ICU admission. ß 2015 Published by Elsevier Ltd.
Keywords: Trauma Decision support system In-hospital death ICU admission Sensitivity Specificity
Introduction Trauma is an increasingly significant public health issue around the globe. It has become one of the leading causes of mortality and
* Corresponding author. Tel.: +86 10 88324570; fax: +86 10 88324570. E-mail address: [email protected] (B. Jiang). http://dx.doi.org/10.1016/j.injury.2015.06.002 0020–1383/ß 2015 Published by Elsevier Ltd.
disability worldwide [1,2]. Trauma accounts for 16% of the global burden of disease and 16,000 people die from injury every day. In addition millions more are temporarily or permanently disabled due to trauma. Without appropriate actions road traffic injuries are predicted to become the third leading cause of global morbidity and mortality [3]. Death and disability from trauma is especially high in low- and middle-income countries where approximately 90% of the total burden of trauma in the world occurs. In developed
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
countries, different levels of trauma centres were established to treat trauma with varying degrees of severity. Initial assessment and outcome prediction for trauma patients after arrival to hospital can help identify patients who may need intensive care unit (ICU) admission or may die in hospital and ensure that severely injured patients can get appropriate treatment, which may lead to optimal utilisation of hospital resources and a reduced level of mortality and disability. Unfortunately, there are no nationwide initial trauma assessment guidelines or tools to aid physicians in pre-hospital environments or the emergency departments (EDs) of many less developed countries, e.g., Ghana [4] and China. This study examines China as a case study. A pre-hospital ‘120’ (the ambulance call number used in China) emergency system currently operates throughout the country. A typical, local ‘120’ emergency system usually consists of an emergency command centre, one or more emergency command stations supervised by the command centre, and local hospitals. The emergency command centre takes charge of ‘120’ calls and dispatches ambulances and pre-hospital rescuers to accident sites to transport trauma patients to the nearest hospitals. There are no trauma centres in China, and ranked hospitals are analogous to ranked trauma centres in developed countries. Hospitals are ranked according to medical services and management, quality and safety of clinical care, and technical level and efficiency [5]. As a result, higher level hospitals receive more government funding and can provide better diagnostic care and therapeutic options for trauma patients than lower level hospitals. On the contrary, low level hospitals have little or no capability to treat complicated trauma cases. Some severe trauma patients sent to low level hospitals by the ‘120’ emergency system will have to be retransferred to higher level hospitals. Trauma patient outcomes in China are relatively poor compared with those in developed countries, partially due to the time wasted in transportation. More than 400,000 people die in China each year from injury, and trauma is the fifth leading cause of death after malignant tumours and cardiac, cerebral, and respiratory diseases. It is the most common cause of death of young people aged 18–40 years in China [6]. Being confronted with the increasing challenges in trauma care, National Health and Family Planning Commission (NHFPC) of China and the Trauma Society of Chinese Medical Association have jointly launched a national research project to develop a computerised trauma assessment tool to aid ED physicians to assess trauma and predict in-hospital death and ICU admission. This study is part of that project. The focus of this study was on assessment of trauma and prediction of in-hospital death and ICU admission for patients after their arrival to hospital. Various severity scoring tools [7] have been developed from physiological, anatomical, or composite perspectives to assess trauma and predict in-hospital death and ICU admission. Since detailed diagnoses of trauma cannot be made upon patients’ arrival to hospital, anatomical scoring systems are not applicable in this study. Frequently used physiological trauma scoring systems in the literature include the Pre-Hospital Index (PHI) [8], Trauma Index (TI) [9,10], Glasgow Coma Score (GCS) [11,12], and the Revised Trauma Score (RTS) [13]. Studies [9,10,14,15] show that there is a good correlation between the severity score assigned by any of the above three systems PHI, TI, and GCS and the actual severity of trauma. Patients who are assessed with severe trauma by any of the PHI, TI and GCS systems have significantly high probabilities of in-hospital death or ICU admission. Thus PHI, TI, and GCS can be used to predict which trauma patients are most likely to die in hospital or will be admitted to the ICU. Based on previous experiences of using knowledge-based, clinical decision support systems (CDSSs) [16,17] to aid physicians to make clinical decisions, we proposed to integrate the PHI, TI and GCS scoring systems in a decision support system (DSS) to help ED
1785
physicians carry out rapid initial trauma assessment and predict in-hospital death and ICU admission upon patients’ arrival to hospital. The RTS was not taken into consideration in this study because it combines the GCS with respiratory rate and systolic blood pressure (SBP); the latter two variables are measured by the PHI and both the GCS and PHI are included in the DSS. Our hypothesis was that the proposed DSS should perform better than the PHI, TI, and GCS in identifying patients with strong possibilities for in-hospital death and ICU admission. It can improve the accuracy of ED physicians’ predictions of in-hospital death and ICU admission for trauma patients and provide them with appropriate trauma management strategies. Consequently, it can help to optimise the utilisation of hospital resources and reduce mortality and disability associated with trauma. This hypothesis has been supported by the empirical study. Materials and methods Data sources and study subjects This study was carried out at Kailuan Hospital, Tangshan City, a typical city located in Hebei province, North China, where all ambulance services are supplied by the ‘120’ pre-hospital emergency system. The size of Tangshan is about 13,472 km2, and it hosts a population of 7.2 million. The city was fully rebuilt after a 7.8 magnitude earthquake in 1976. Kailun Hospital is one of six top-ranked hospitals in Tangshan. The hospital has around 1,000 beds, and it has the capability to treat patients with the most severe trauma. Patients were included for analysis if they met the following criteria: (a) directly sent to the ED from the accident site; (b) clinical data required by the PHI, TI, and GCS were well-recorded upon the arrival to ED; and (c) corresponding in-hospital data could be retrieved. No further restrictions were made about the severity or characteristics of the cases. There were 1,299 trauma patients directly sent to the ED at Kailuan Hospital in the sample period. Out of 1,299 trauma patients included in the study, 1,195 (92.0%) patients had both the ED data and in-hospital data. The remaining 104 patients had either missing ED data or missing inhospital data, and thus had to be excluded from the data analysis. The ED data recorded at Kailuan Hospital contain information on basic demographics, dates of arrival to the ED, the location, cause, and type of body injury, body temperature, respiratory rate, pulse rate, blood pressure, eye-opening status, verbal response, and motor response. The retrieved in-hospital data include length of stay in the ICU, length of stay in hospital, and the patient’s condition at discharge. The primary outcome of this study was a composite of both inhospital death and ICU admission, which was defined as severe trauma. Ethical review of this study was approved by the biomedical ethics committee of the first author’s institution and the participating hospital. Identifying severe trauma using a DSS by integrating the PHI, TI, and GCS The PHI and TI scores increase with trauma severity, while the GCS score decreases with severity. The three scoring systems stratify trauma patients into three groups, namely severe, moderate, and minor, according to severity scores assigned by respective systems based on patients’ clinical signs and symptoms. The score range for each severity group of the three scoring systems is shown in Table 1 [8–12]. Studies show that patients with severe trauma, identified by any of the three scoring systems, have strong possibilities for
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
1786
Table 1 Score range for Severe, Moderate, and Minor of the PHI, TI, and GCS systems.
PHI TI GCS
Table 2 ‘IF-THEN’ rules for combing the PHI, TI, and GCS systems.
Severe
Moderate
Minor
IF
THEN
8–20 17–30 3–8
4–7 10–16 9–12
0–3 5–9 13–15
‘Severe’ by any system of PHI, TI, and GCS ‘Moderate’ by all systems of PHI, TI, and GCS ‘Moderate’ by GCS, and ‘Moderate’ by either PHI or TI ‘Moderate’ by GCS, and ‘Minor’ by both systems of PHI and TI ‘Minor’ by GCS, and ‘Moderate’ by either PHI or TI ‘Minor’ by all systems of PHI, TI, and GCS
Severe Severe Severe Moderate Moderate Minor
in-hospital death or ICU admission, while the probabilities of adverse events for patients with moderate or minor trauma are low [9,10,14,15,18]. Since each scoring system has its merits, specifically, the PHI focuses on measurements of vital signs, the TI takes into account measurements of trauma-specific signs, and the GCS focuses on measurements of consciousness, we proposed to integrate the PHI, TI, and GCS in a complementary way that will identify patients with strong possibilities for in-hospital death and ICU admission. The rationale behind the proposed DSS is that severe trauma patients have significantly high probability of inhospital death and ICU admission and that mortality and disability of trauma can be reduced if timely and proper trauma triage and treatment strategies are provided for those patients in the ED. To this end, we invited five trauma experts from Peking University People’s Hospital and five experts in emergency medicine from Beijing Emergency Medical Center to form an advisory panel and organised small meetings to discuss how to combine the PHI, TI, and GCS to provide a rapid initial assessment of patient trauma and predict in-hospital death and ICU admission. We adopted a modified Delphi approach [19] to seek consensus. First, we requested each expert to come up with a proposal for combinational use of the three scores. Second, we requested each expert to defend his or her proposal using his or her best knowledge. Third, we asked the expert panel to vote for the best proposal. After several meetings, the expert panel achieved consensus and proposed to use a decision tree to combine the PHI, TI, and GCS scores to determine the severity of injury in the ED, thus predicting in-hospital death and ICU admission based on the
severity level. The decision tree for combining the PHI, TI, and GCS is shown in Fig. 1. There are six IF-THEN rules that can be derived from the proposed decision tree, and the rules are described in Table 2. It is clear from Fig. 1 and Table 2 that although the same set of trauma severity categories (severe, moderate, and minor) are utilised by the proposed DSS and its component systems, the coverage of resultant severe trauma in the DSS is broader than that of severe trauma in the PHI, TI, and GCS. Particularly, the proposed DSS is more cautious in identifying patients with severe trauma than the three scoring systems. That is to say, in addition to severe trauma identified by any one of the three scoring tools, moderate trauma identified by the GCS and either of the PHI and TI simultaneously will be considered as severe trauma in the DSS. Consequently, the coverage of resultant moderate and minor trauma in the DSS is narrower than that of moderate and minor trauma in the three scoring systems. Because patients with moderate or minor trauma identified by the PHI, TI, and GCS have low probability of in-hospital death and ICU admission, and a proportion of patients with moderate trauma identified by the three extant scoring tools was assessed with severe trauma in the DSS, the probability of adverse events linked with moderate or minor trauma in the DSS will be even lower than any of the three extant scoring tools. This shows that the three severity categories of the four tools are necessary for assessing trauma severity in the DSS.
Fig. 1. The decision tree for combining the PHI, TI, and GCS systems.
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
The knowledge base of the DSS contains those combination rules as shown in Table 2. The DSS was designed to have clientserver structure. The client-side can be installed as software on a handheld device to equip ED physicians, and the server-side software can be installed on the hospital’s hard server. The clientside aids the capture of patient demographics, vital signs, and other available clinical data required for severity assessment; it also communicates with the server-side and helps the hospital trauma team evaluate the patient’s degree of trauma severity and know whether the patient has a strong possibility for in-hospital death or ICU admission.
Table 3 Characteristics of studied trauma data. Variable
Subgroup
Non-severe trauma patients (N:1165)
Severe trauma patients (N:30)
Gender
Male Female
936(80.3%) 229(19.7%)
26(86.7%) 4(13.3%)
Age (years)
Mean age (SD)
43.1(16.0)
40.0(17.5)
Trauma cause
Traffic accident Blunt force injury Sharp object injury Fall Crush/burial Explosion/fire Burn/electrical injury Frozen injury Sports injury Battle injury/gunshot injury Other
298(25.6%) 290(24.9%) 115(9.9%) 291(25.0%) 85(7.3%) 8(0.7%) 14(1.2%) 0 3(0.3%) 0 61(5.2%)
22(73.3%) 0 0 5(16.7%) 2(6.7%) 0 0 0 0 0 1(3.3%)
Analysis We extracted the PHI, TI, and GCS scores for each trauma patient on the basis of his or her recorded ED data. The purpose of the proposed DSS was to identify patients who have strong possibilities for adverse events, so as to suggest proper trauma management strategies for them. Based on the proposed decision tree as shown in Fig. 1, patients with severe trauma assessed by the DSS were quickly identified from the collected dataset. Further, to measure the prediction performance of the DSS, we identified which patients had actually either died in hospital or were admitted to the ICU. The next step was to calculate performance measures, such as sensitivity, specificity, and over- and undertriage rates of the DSS. We realised that calculation of sensitivity, specificity, and overand under- triage rates required the classification tools to provide binary categories; however, there are three severity categories in the DSS and the three scoring systems. As discussed before, patients with severe trauma have significantly high probability of in-hospital death or ICU admission, while patients with moderate or minor trauma have low probability of adverse events; therefore, we bundled moderate and minor trauma as non-severe trauma in the four studied tools for the following data analysis. Accordingly, the DSS together with the three scoring systems were changed to be binary classification tools for calculating performance measures including sensitivity, specificity, over-triage rate (1 specificity), and under-triage rate (1 sensitivity). We used SPSS to analyse the collected trauma data. Firstly, we used definitive cutoff values in the PHI, TI, and GCS systems to classify patients into severe, moderate, or minor trauma groups respectively. Secondly, based on the severity results generated by the three extant scoring systems, we used the proposed DSS to reclassify the patients into severe, moderate, or minor groups based on the decision tree as shown in Fig. 1. Thirdly, we bundled moderate and minor trauma as non-severe trauma for each trauma assessment tool and used the sensitivity, specificity, over-triage rate, and under-triage rate in identifying severe trauma to measure
1787
and compare the PHI, TI, GCS, and the proposed DSS. Finally, the McNemar test was employed to compare assessment performances of all four prediction tools.
Results Among the studied cases, 1,195 patients, 14 (1.2%) died in hospital, 21 (1.8%) were admitted to the ICU, 1,177 (98.5%) fully recovered at discharge, and 2 had improved at discharge. Based on the recorded death and ICU admission, we identified 30 (2.5%) patients with actual severe trauma and 1,165 (97.5%) with actual non-severe trauma. The characteristics of the studied data are described in Table 3. In the excluded 104 cases, 76 (73.08%) were male, 28 (26.92%) were female: the mean age of the male patients is 39.3 (SD = 17.3), and the mean age of the female patients is 34.5 (SD = 14.8). Also, none of them had been admitted to ICU and none died. As the excluded patients were a very small portion of the collected cases and contained randomly missing data, the exclusion would not affect the results of the study. Table 4 presents the performance measures including sensitivity, specificity, over-triage rate, under-triage rate and corresponding 95% confidence intervals (CI) of the PHI, TI, GCS, and the proposed DSS in identifying severe trauma, which may result in either in-hospital death or ICU admission. For convenience, the numbers of identified cases of severe trauma, non-severe trauma, over-assessed non-severe trauma, and under-assessed severe trauma are shown in Table 4, too.
Table 4 Sensitivity, specificity, over-triage rate, and under-triage rate of the PHI, TI, GCS, and the proposed DSS for predicting ICU admission or death in hospital. PHI
TI
GCS
The proposed DSS
Sensitivity
Value 95% CI Identified severe trauma (N)
30.0% 13.5–46.5% 9
50.0% 32.2–67.8% 15
46.7% 28.9–64.5% 14
66.7% 49.8–83.6% 20
Specificity
Value 95% CI Identified non-severe trauma (N)
99.8% 99.6–100% 1163
94.4% 93.0–95.8% 1100
99.2% 98.6–99.8% 1156
93.7% 92.3–95.1% 1091
Over-triage rate
Value 95% CI Over-assessed non-severe trauma (N)
0.2% 0–0.4% 2
5.6% 4.2–7.0% 65
0.8% 0.2–1.4% 9
6.3% 4.9–7.7% 74
Under-triage rate
Value 95% CI Under-assessed severe trauma (N)
70.0% 53.5–86.5% 21
50.0% 32.2–67.8% 15
53.3% 35.5–71.1% 16
33.3% 16.4–50.2% 10
1788
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
Based on the performance comparison results as shown in Table 4, the proposed DSS had the best values for sensitivity (66.7%; 95% CI: 49.8–83.6%) and under-triage rate (33.3%, 95% CI: 16.4–50.2%) among the studied physiological scoring tools. The TI (sensitivity: 50.0%, 95% CI: 32.2–67.8%; under-triage rate: 50.0%, 95% CI: 32.2–67.8%) performed slightly better than the GCS (sensitivity: 46.7%, 95% CI: 28.9–64.5%; under-triage rate 53.3%, 95% CI: 35.5–71.1%), while both the TI and GCS performed better than the PHI (sensitivity 30.0%, 95% CI: 13.5–46.5%; under-triage rate 70.0%, 95% CI: 53.5–86.5%). Inevitably, the specificity decreased when the sensitivity increased. Results from the McNemar test showed that the performance differences between the proposed DSS and the three scoring tools, namely the PHI, TI, and GCS were statistically significant (DSS vs TI: p = 0.004; DSS vs GCS: p < 0.001; DSS vs PHI: p < 0.001). Discussion and conclusions The purpose of the study was to develop a computerised tool to help ED physicians identify patients with strong possibilities for inhospital death or ICU admission. In this study, we proposed to employ a decision tree to integrate the three extant physiological scoring systems, namely the PHI, TI, and GCS into a composite DSS. The performance of the proposed DSS was tested using a retrospective set of trauma data collected at the ED of Kailuan Hospital in Tangshan, a city located in North China. Based on the results (see Table 4), the proposed DSS performed best among the four studied tools in predicting in-hospital death and ICU admission. The superiority of the DSS may be explained by the following. First, the three extant scoring systems have respective merits and deficiencies. The PHI [8] was designed to assess trauma from a physiological perspective, yet its assessment factors consist only of vital signs. The TI [9,10] was designed for assessing trauma not only from a physiological perspective but also from a trauma specific perspective, thus it contains some trauma-specific factors, such as body region of injury and type of injury, besides vital signs. For example, a penetrating injury would be assigned a score of 3 for the injury type in the TI scoring system. However, the PHI and TI may miss some severe head injuries without obvious abnormal physiologic signs. The GCS [11,12], specifically designed to assess the severity of head injuries, focuses on consciousness, while other factors, such as the type of injury and vital signs are not taken into consideration; thus, the GCS may miss some patients who suffer severe trauma but do not exhibit disorders of consciousness. Second, the DSS adopts a cautious approach in combining the PHI, TI, and GCS. The DSS considers a patient to have severe trauma under the following two conditions: (a) labelled severe by any component system; or (b) simultaneously labelled moderate by the GCS and either the PHI or TI. Thus the GCS can complement the TI and PHI in identifying severe head trauma without obvious abnormal physiologic signs, and the TI and PHI can complement the GCS in identifying severe trauma without head injury. As a result, the proposed DSS makes the three physiological scoring systems complement each other to achieve the best performance. In the study, the sensitivity of the proposed DSS (66.7%; 95% CI: 49.8–83.6%) is lower than some trauma assessment models proposed by other researchers such as Cox et al. [20]. The relatively low sensitivity in the study may be due to the following reasons. Cox et al. [20] developed their triage tool in a pre-hospital environment and used much broader criteria to define severe trauma, which included death, injury severity score (ISS) > 15, ICU admission, and urgent surgeries. In this study, we developed the prediction tool in the ED and used stricter criteria, including only in-hospital death and ICU admission to define actual severe trauma. Thus, the number of patients that were identified to have died in hospital or were admitted to the ICU was small (N = 30). As
can be found from Table 3, the incidence rate of death or ICU admission was only 2.5% among trauma patients directly sent to Kailuan Hospital. This may affect sensitivity. Regarding over-triage (non-severe patients assessed as severe) and under-triage (severe patients assessed as non-severe), the values of over-triage (1 specificity) and under-triage (1 sensitivity) rates of the DSS were 6.3% and 33.3%, respectively. As the purpose of inhospital death and ICU admission prediction was to optimise the management of trauma patients and utilisation of hospital resources, a relatively low over-triage rate is good for interhospital transfer and ICU resource allocation. Regarding the calculation efficiency of the proposed DSS, a clarification is necessary. It seems that the decision tree employed in the DSS is a bit complicated for physicians. But on the client-side of the DSS, whether or not a patient has a strong possibility for inhospital death or ICU admission can be predicted and displayed as soon as necessary patient data have been inputted because the rules for combining the PHI, TI, and GCS systems are implemented on the client-side. Since calculation of the PHI, TI, GCS, and the decision tree on the DSS client-side has no need for data transmitted from the server-side, it works just like a mobile phone counter and takes no time. ED physicians have no need to know the details of the inner logic of the DSS. As the client-side and the server-side of the DSS communicate in real time, the hospital IT professional in charge of the trauma system can know the degree of severity of any trauma patient in the ED; thus, they can inform hospital trauma experts in a timely manner to provide appropriate treatment for the patient. Alternatively, they can guide transfer of one patient to a higher level hospital if the patient has severe trauma and the current hospital has no capability to treat the patient. This will ensure that severely injured patients can be treated quickly and well, and the limited hospital resources can be utilised optimally. In addition, a clarification about bundling moderate and minor trauma as non-severe trauma for calculating performance measures is needed. In the proposed decision tree, the four trauma severity assessment tools stratify trauma patients into three groups: severe, moderate, and minor. Since all the three severity categories are necessary antecedent status for those combinational rules as shown in Table 2, the three severity categories are needed for the four tools in assessing trauma severity and predicting in-hospital death and ICU admission. However, to use sensitivity, specificity, over-triage rate, and undertriage rate to measure the stratification or prediction performance of the four tools, we need to change the four tools to binary classification systems with two severity categories. As severe trauma is linked with strong possibilities for in-hospital death or ICU admission, and moderate and minor trauma is linked with low probability of adverse events, we bundled moderate and minor trauma in the four studied tools as non-severe trauma in data analysis for system performance measurement and comparison. This study develops a novel trauma assessment and adverse events prediction tool by incorporating the strengths of three extant severity scoring systems, namely the PHI, TI, and GCS. The findings show that the proposed DSS outperformed the PHI, TI, and GCS in identifying trauma patients with strong possibilities for inhospital death and ICU admission with statistical significance. It is the first study in China to help design and develop a computerised tool that predicts in-hospital death and ICU admission for trauma patients in the ED. The limitation of the study is that it was a retrospective study. To test the applicability of current research, future research needs to prospectively test the DSS. To conclude, the proposed DSS outperformed the extant severity scoring systems in predicting in-hospital death and ICU admission. It has a strong potential to help ED physicians identify severe trauma and send severely injured patients to ICU or higher
G. Kong et al. / Injury, Int. J. Care Injured 46 (2015) 1784–1789
level hospitals to receive proper trauma care. In our future research agenda, we plan to prospectively test the proposed DSS before it is implemented in a real environment. Conflict of interest
[6]
[7] [8]
We declare that there is no conflict of interest with any financial organization regarding the material discussed in the manuscript entitled ‘‘Current state of trauma care in China, tools to predict death and ICU admission after arrival to hospital’’.
[9] [10] [11]
Acknowledgments [12]
This study was supported by a grant from the National Natural Science Foundation of China (grant no. 81301296). This study was also supported by a grant from the National Health and Family Planning Commission of China (grant no. 201002014). This study was also supported by a project funded by the Ministry of Education of China (grant no. 13YJC630066).
[14]
References
[16]
[1] Mock C, Lormand JD, Goosen J, Joshipura M, Peden M. Guidelines for essential trauma care. Geneva: World Health Organization; 2004. [2] Sasser S, Varghese M, Kellermann A, Lormand JD. Prehospital trauma care systems. Geneva: World Health Organizations; 2005. [3] Peden M, Scurfield R, Sleet D, Mohan D, Hyden AA, Jarawan E. World report on road traffic injury prevention: summary. Geneva: World Health Organization; 2004. [4] Mock CN, Tiska M, Adu-Ampofo M, Boakye G. Improvements in prehospital trauma care in an African country with no formal emergency medical services. The Journal of Trauma: Injury Infection and Critical Care 2002;53(1):90–7. [5] Xu B, Liu H, Su N, Kong G, Bao X, Li J, et al. Association between winter season and risk of death from cardiovascular diseases: a study in more than half a
[13]
[15]
[17]
[18]
[19] [20]
1789
million inpatients in Beijing, China. BMC Cardiovascular Disorders 2013;13:93. Zong Z-W, Li N, Cheng T-M, Ran X-Z, Shen Y, Zhao Y-F, et al. Current state and future perspectives of trauma care system in mainland China. Injury 2011;42(9):874–8. Camp LV, Yates DW. Trauma scoring. In: Soreide E, Grande CM, editors. Prehospital trauma care. New York: Taylor & Francis; 2005. p. 153–67. Koehler JJ, Baer LJ, Malafa SA, Meindertsma MS, Navitskas NR, Huizenga JE. Prehospital Index A scoring system for field triage of trauma victims. Annals of Emergency Medicine 1986;15(2):178–82. Kirkpatrick JR, Youmans RL. Trauma Index, an aide in the evaluation of injury victims. The Journal of Trauma 1971;11(8):711–4. Ogawa M, Sugimoto T. Rating severity of the injured by ambulance attendants: field research of trauma index. The Journal of Trauma 1974;14(11):934–7. Teasdale G, Jennett B. Assessment of coma and impaired consciousness. The Lancet 1974;304(7872):81–4. Wang MC, Gennarelli TA. Classification systems. In: Jallo J, Loftus CM, editors. Neurotrauma and critical care of the brain. New York: Thieme Medical Publishers; 2009. Champion HR, Sacco WJ, Copes WS, Gann DS, Gennarelli TA, Flanagan ME. A revision of trauma score. The Journal of Trauma 1989;29(5):623–9. Mahdian M, Fazel MR, Fakharian E, Akbari H, Mahdian S. Cerebral state index versus Glasgow coma scale as a predictor for in-hospital mortality in braininjured patients. Chinese Journal of Traumatology 2014;17(4):220–4. Tamim H, Joseph L, Mulder D, Battista RN, Lavoie A, Sampalis JS. Field triage of trauma patients: improving on the Prehospital Index. The American Journal of Emergency Medicine 2002;20(3):170–6. Kong GL, Xu DL, Liu XB, Yang JB. Applying a belief rule-base inference methodology to a guideline-based clinical decision support system. Expert Systems 2009;26(5):391–408. Kong GL, Xu DL, Body R, Yang JB, Mackway-Jones KRH, Carley S. A belief rulebased decision support system for clincial risk assessment of cardiac chest pain. European Journal of Operational Research 2012;219(3):564–73. Sampalis JS, Tamim H, Nikolis A, Lavoie A, Williams JI. Predictive validity and internal consistency of the pre-hospital index measured on-site by physicians. Accident Analysis & Prevention 1996;28(6):675–84. Linstone HA, Turoff M. The Delphi method: techniques and applications. London: Addison-Wesley; 1975. Cox S, Currell A, Harriss L, Barger B, Cameron P, Smith K. Evaluation of the Victorian state adult pre-hospital trauma triage criteria. Injury 2012;43(5): 573–81.
