RESUMO
Objective:To construct a classification and regression tree which can be used to guide the tracheostomy for traumatic cervical spinal cord injury (TCSCI) based on the identification of the risk factors for TCSCI.Methods:The 498 patients with TCSCI were retrospectively analyzed who had been treated at Department of Orthopedics, The Second Hospital Affiliated to Army Medical University from January 2009 to December 2018. There were 403 males and 86 females, with an age of (50.2±13.6) years. Of the patients, 69 received tracheostomy and 420 did not. The gender, age, smoking history, injury cause, neurological level of injury (NLI), American Spinal Cord Injury Association (ASIA) grade, injury severity score (ISS), thoracic injuries, prior pulmonary diseases, prior basic diseases, and operative approaches of the patients were statistically analyzed by single factor analysis. After the independent risk factors for tracheostomy were analyzed by binary logistic regression, the classification and regression tree was developed which could be used to guide the tracheostomy.Results:The logistic regression analysis showed age>50 years ( OR=4.744, 95% CI: 1.802 to 12.493, P=0.002), NLI at C 4 and above ( OR=23.662, 95% CI: 8.449 to 66.268, P<0.001), ASIA grade A ( OR=40.007, 95% CI: 12.992 to 123.193, P<0.001), and ISS score>16 ( OR=10.502, 95% CI: 3.909 to 28.211, P<0.001) were the independent risk factors for the tracheotomy. The classification and regression tree revealed that ASIA grade A and NLI at C 4 and above were the first and second decision nodes, which had a strong predictive effect on tracheostomy. 86.84% of the patients with ASIA grade A and NLI at C 4 and above underwent tracheostomy. Conclusion:Our classification and regression tree shows that NLI at C 4 and above and ASIA grade A have a strong guiding effect on tracheotomy for TCSCI.
RESUMO
Objective:To explore grouping methods of subdividing adjacent diagnosis related groups(DRGs) by introducing patient clinical complexity level(PCCL) principle and provide references for exploring DRG grouping method in line with context in China.Methods:Clinical complexity level of each complication was assigned by clinicians.PCCL model was selected to calculate the scores of clinical complexity cases.Each adjacent DRG was subdivided into DRG groups by classification and regression trees(CART) model.The rank-sum test was applied to test the statistical significances of the grouping results.Results:9 surgical adjacent DRGs were subdivided into 18 DRG groups.There were statistical significances in the differences of hospitalization expenses and length of stay among different DRG groups in each adjacent group.Conclusion:PCCL model showed high performance in DRG subdivision.The unification of the quality of medical records and coding were the key factors to ensure the reasonable grouping results.