Development and Internal Validation of China Mortality Prediction Model in Trauma Based on ICD-10-CM Lexicon: CMPMIT-ICD10
Mortality prediction in trauma is a critical aspect of trauma-related research, as it not only helps in assessing the severity of injuries in individual patients but also serves as a benchmark for evaluating the quality of medical institutions. Accurate prediction models are essential for informing triage decisions, guiding treatment strategies, and improving overall trauma care. In China, hospitals predominantly use the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) to describe injuries. However, there is a lack of suitable prediction models tailored to the Chinese population. This study aims to develop and internally validate a new mortality prediction model based on the ICD-10-CM lexicon and a Chinese trauma database, named the China Mortality Prediction Model in Trauma based on ICD-10-CM lexicon (CMPMIT-ICD10).
The study is a retrospective analysis of data from the Beijing Red Cross Emergency Center, the largest trauma emergency center in Beijing, which handles a significant volume of trauma cases. The data included all trauma patients admitted from January 2012 to July 2018, totaling 40,205 patients. The inclusion criteria were patients with traumatic events listed in Chapter XX of the ICD-10-CM, excluding cases of hanging, suffocation, drowning, poisoning, burning, and electrocution. Patients without baseline information or outcomes were also excluded. The study was non-interventional and based on anonymous registration data.
The development of the CMPMIT-ICD10 model involved several steps. First, the ICD-10 codes for injuries (S00–S99) were classified into 20 new region-severity codes (A1, A2, …, G2, G3) based on the severity and region of injury. This classification was necessary because many ICD-10-CM codes occurred infrequently. Additionally, comorbidity information was extracted from the ICD-10-CM codes, including conditions such as myocardial infarction, congestive heart failure, chronic pulmonary disease, and diabetes. The presence of traumatic shock (T79.401) and coma were also included as binary variables to assess physiological responses to trauma. Coma was defined as a Glasgow Coma Score of 8 or lower, indicating the inability to respond to commands or open eyes in response to stimuli.
The model was developed using logistic regression analysis, with mortality as the outcome variable. The logistic regression model calculated the probability of death based on the coefficients of the predictor variables, including sex, age, region-severity codes, comorbidities, traumatic shock, and coma. Age was categorized into 10-year blocks to account for non-linear relationships with mortality. The model-building process used a forward stepwise procedure with backward elimination of covariates, with significance levels set at 0.05 for entry and 0.10 for elimination. Variables that did not achieve sufficient statistical power were excluded from the final model.
The performance of the CMPMIT-ICD10 model was assessed based on discrimination and calibration. Discrimination, which measures the model’s ability to separate survivors from non-survivors, was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC). Calibration, which reflects the consistency between predicted and observed mortality, was assessed using the Brier score and a calibration plot. Internal validation was performed using the bootstrapping method, which involved drawing 1000 random samples with replacement from the original dataset and testing the model’s performance on these samples.
The study population consisted of 40,205 patients, of whom 39,009 survived and 1,196 died, resulting in a mortality rate of 3%. The characteristics of the patients revealed that older age, male sex, traumatic shock, and coma were associated with a higher risk of mortality. The distribution of the new region-severity codes and comorbidities showed significant correlations with mortality, except for certain codes and conditions such as liver disease and malignancy, which did not reach statistical significance.
The final CMPMIT-ICD10 model included sex, age, specific region-severity codes, comorbidities, traumatic shock, and coma as key predictors of mortality. Each predictor was assigned a score based on its coefficient in the logistic regression model. The total score ranged from 0 to 232, with higher scores indicating a higher risk of mortality. The risk of death was categorized into five levels: extremely low risk (0–47 points, mortality 90 points, mortality >90%). Traumatic shock, coma, and advanced age (>80 years) had the greatest influence on mortality risk. Among the region-severity codes, head injuries (A3 and A4), abdominal injuries (E2 and E3), and spinal cord injuries (F2) were the most significant predictors of mortality. Comorbidities such as congestive heart failure and chronic renal failure also had a substantial impact on mortality risk.
The performance of the CMPMIT-ICD10 model was excellent, with an AUC of 0.964, indicating high discrimination ability. The Brier score was 0.0177, suggesting good calibration. Internal validation using the bootstrapping method yielded similar results, with an AUC of 0.963 and a Brier score of 0.0178. The calibration plot showed minor deviations from the ideal 45-degree line, but the overall calibration was acceptable. The model’s underestimation of mortality before a probability of 0.5 and overestimation after 0.5 were minimal.
An example application of the CMPMIT-ICD10 model demonstrated its practical use. A 72-year-old man with traumatic epidural hemorrhage (S06.4, A4), subarachnoid hemorrhage (S06.7, A3), and coma would have a total score of 73 points, corresponding to a medium risk of mortality. This example illustrates how the model can be used to predict mortality risk based on specific injury codes and patient characteristics.
The CMPMIT-ICD10 model addresses several limitations of existing trauma prediction models. Unlike the Injury Severity Score (ISS), which relies on the Abbreviated Injury Scale (AIS) and requires specialized training for coders, the CMPMIT-ICD10 model is based on ICD-10-CM codes, which are routinely used in Chinese hospitals. This makes the model more accessible and easier to implement in clinical practice. Additionally, the model incorporates comorbidities and acute physiological responses, such as traumatic shock and coma, which are not considered in the Trauma Mortality Prediction Model-ICD10 (TMPM-ICD10).
The study has some limitations. First, the classification of ICD-10 codes into 20 region-severity codes may not capture the full complexity of injury severity. Second, the model was not compared with the TMPM-ICD10 due to differences in ICD-10-CM coding practices between China and other countries. Future research should focus on external validation of the CMPMIT-ICD10 model through multi-center studies to confirm its generalizability and accuracy.
In conclusion, the CMPMIT-ICD10 model represents a significant advancement in trauma mortality prediction for the Chinese population. Its development based on ICD-10-CM codes and a large Chinese trauma database ensures its relevance and applicability in clinical practice. The model’s high discrimination and calibration, along with its simplicity and ease of use, make it a valuable tool for predicting mortality in trauma patients and assessing the quality of trauma care in China.
doi.org/10.1097/CM9.0000000000001371
Was this helpful?
0 / 0