Various assessment methods based on the ICD-9-CM code, such as the original ICISS, TMPM-ICD9, TRISS, and HARM, among others, have rarely been used [1–3, 5]. The ICISS, TMPM-ICD10, and IMP-ICDX, which are based on the ICD-10-CM code, are assessment techniques that are purely based on anatomical trauma [6–9]. The IMP-ICDX evaluated using three logistic regressions mutual modifications show more accurate prediction results than those of TMPM-ICD10 with a single regression model. Meanwhile, IMP-ICDX has improved its discriminative ability with the incorporation of age, gender, and injury mechanism. But IMP-ICDX does not fully use clinically available information, such as GCS, endotracheal intubation, SCWI, and so on.
The data in this research were derived from NTDB in 2016 [10], which presents a description list of ICD-10-CM codes for trauma. The NTDB is the largest and most reliable publicly available database worldwide with more representative data from different regions and trauma centers in the United States. The TRIMP-ICDX in this study was based on the severity of anatomical injury, incorporating more than 10 auxiliary variables with statistical significance (physiological reserve indicators, such as age, gender, and SCWI; physiological response indicators, such as SBP, RR, GCS, ICU admission, need for mechanical ventilation or emergency surgery, etc.) to create a logistic regression model. The discriminative ability and calibrations of survival and non-survival were superior to those of IMP-ICDX and TRISS. The statistical results were similar for different BRs (Table 2). The specific calculation formula is given in Appendix 2.
The data used in this study were from patients of all ages (from 1 to 89 years old). However, TRISS is only used for patients older than 14 years [16, 17], which may not be appropriate for trauma patients younger than 15 years old. When gender, SCWI, and children under 15 years old were included in the physiological reserve, the prediction results of TRIMP-ICDX were improved. The SCWI can be used as an independent factor to predict the probability of traumatic death [12]. The mechanism of injury and the need for emergency surgery can be understood as indirect indicators of physiological response. Simultaneously, compared with fixed categories, non-parametric regression more accurately explained the relationship between age and traumatic death [12, 18], that is, there is no grouping of age, and the results should be better. Increasing auxiliary variables, such as whether to stay in ICU or whether mechanical ventilation is needed can help to predict the outcome of trauma [9].
Clinically, there are several indications for patients entering the ICU after injury, such as further life support after cardiopulmonary resuscitation, monitoring and treatment after severe trauma, patients requiring mechanical ventilation, etc. Also, there are indications for mechanical ventilation after trauma. For example, patients with post-injury disturbance of consciousness or loss of spontaneous breathing after injury. In general, it is patients with severe injured that require mechanical ventilation and/or admission to the ICU, which can also be regarded as an indicator of indirect physiological response to trauma. This has been confirmed in the existing literature: such as IMP-ICDX [9]. The possibility of death or survival can be assessed in trauma patients who can be diagnosed clearly after admission.
Based on the ISS, TRISS incorporates GCS, SBP, RR, and age in the evaluation of the survival probability of patients [16]. Updated several times, the latest version of TRISS was published in 2011 [17]. Compared with ISS, TRISS predicts outcomes more accurately. This study showed that the lowest survival rate of TRISS exceeded 20%, which is significantly higher than that of TRIMP-ICDX. Probably influenced by ISS, the survival prediction curve of TRISS is above the perfect reference line (Fig. 2).
HARM is based on 80 available variables, such as anatomical indicators and physiological reserves, and is evaluated by independent logistic regression. A study found that compared with TRISS and ICISS, HARM more accurately predicted survival or death from trauma [5]; however, it did not take full advantage of statistically significant physiological response indicators, such as vital signs, mechanical ventilation, and whether surgery was required. This article only used more than 10 available variables, making it easier for clinicians to evaluate the results. The absolute AUC value of TRIMP-ICDX is better than HARM (0.968 versus 0.958).
This study evaluated the overall sample of patients described by the ICD-10-CM code. Cases of blunt injuries or penetrating injuries were not evaluated separately. If the cases were analyzed separately using the proposed method, the discriminative power obtained when using penetrating injuries would be higher than that when using blunt injuries (AUC 0.982 versus 0.965). For different injury mechanisms, the results can be calculated using the same formula without the need to design another method. In this study, the data available in the trauma registry can be used to calculate the probability of death or survival from trauma for each patient by using a computer applet.
Among the methods that apply regression to evaluate the severity of trauma, AIS predot code-based assessment methods, such as TMPM and IMP, yield more accurate prediction results than those of ICD code-based methods, such as TMPM-ICD9, TMPM-ICD10, and IMP-ICDX [8, 9, 19–21]. However, they are all based on the assessment of pure anatomical injuries and do not reflect the contribution of other available clinical information in the results. We speculate that the results of TRIMP-ICDX are similar to the comprehensive evaluation results based on IMP because adding other available information can correct the defects of IMP-ICDX. Therefore, the ICD-10-CM injury diagnostic code should be able to replace the AIS predot code. This approach requires considerable human and material resources as trauma surgery experts need to specifically set AIS predot codes when collecting data, which is difficult even for developed countries and more so for developing ones.