Chinese General Practice ›› 2022, Vol. 25 ›› Issue (20): 2498-2506.DOI: 10.12114/j.issn.1007-9572.2022.0113

• Article • Previous Articles     Next Articles

Development and Improvement of Nomograms Predicting the Prognosis in Patients with Severe Multiple Trauma

  

  1. Department of Emergency, Suzhou Ninth People's Hospital, Suzhou 215200, China
  • Received:2022-01-10 Revised:2022-04-22 Published:2022-07-15 Online:2022-04-28
  • Contact: Xie SHEN
  • About author:
    YIN F, LIU Y, SHEN X. Development and improvement of nomograms predicting the prognosis in patients with severe multiple trauma[J]. Chinese General Practice, 2022, 25 (20) : 2498-2506.

严重多发伤患者预后的影响因素研究及列线图模型的建立和优化

  

  1. 215200 江苏省苏州市第九人民医院急诊科
  • 通讯作者: 沈勰
  • 作者简介:
    殷菲,刘云,沈勰.严重多发伤患者预后的影响因素研究及列线图模型的建立和优化[J].中国全科医学,2022,25(20):2498-2506. [www.chinagp.net] 作者贡献:殷菲提出研究思路和研究目标,设计研究方案,包括LASSO回归、RCS拟合Logistic回归探索各参数、绘制非线性效应列线图等,并负责论文起草、最终版本修订,对论文整体负责。殷菲、刘云负责研究过程的实施,各项数据的采集,包括患者入院时的一般资料,入院24 h内的临床资料,并负责数据的校对、清洗。殷菲、沈勰负责数据的统计学分析、图表绘制,包括运用SPSS软件行单因素、多因素分析,ROC曲线分析,R语言绘制列线图、校准图、ROC曲线图、DCA图等。

Abstract:

Background

Severe multiple trauma prevalence has been increasing recently, which has become the leading cause of labor force loss. Early and rapid assessment of patients' conditions will greatly affect their prognosis, which could be significantly supported by a concise and effective visual scoring system.

Objective

To identify and Screen the prognostic factors of severe multiple trauma, and use them to develop two nomograms, then improve them, and verify their clinical application values.

Methods

Patients with severe multiple trauma were recruited from the general ICU and EICU, Suzhou Ninth People's Hospital, including 321 treated during December 2015 to December 2020 (model group) , and 136 treated during January to August 2021 (validation group) . General data at admission and clinical data within 24 hours of admission were retrospectively collected. Prognosis (successful or unsuccessful treatment result) was assessed at discharge. Prognostic factors of severe multiple trauma were Screened using univariate and LASSO regression, and used to develop models using multivariate Logistic regression with restricted cubic splines, then based on this, two nomograms were developed, and their calibration accuracies were estimated using the bootstrap approach and decision curve analysis (DCA) . The receiver operating characteristic (ROC) analysis with associated AUC values was used to estimate the prognostic value of two nomograms in severe multiple trauma. External verification of the nomograms was carried out in the validation group to evaluate their clinical application values.

Results

(1) In the model group, successful and unsuccessful treatment results occurred in 244 and 77 cases, respectively. LASSO regression with multivariate Logistic regression analyses showed that age (OR=1.028) , Glasgow Coma Score (GCS) (OR=0.616) , arterial lactate (OR=1.202) , platelet count (OR=3.888) and Injury Severity Score (ISS) (OR=1.104) were associated with the prognosis of severe multiple trauma (P<0.05) . Hosmer-Lemeshow test indicated that this model fitted the data well (χ2=2.717, P=0.951) , and was appropriate for developing a static and network-based dynamic nomogram (nomogram 1) . LASSO plus multivariate regression analyses with restricted cubic splines revealed that age and GCS had nonlinear correlation with treatment results (P=0.027, 0.001) , and the fit of this model was satisfactory assessed using Hosmer-Lemeshow test (χ2=2.468, P=0.932) , and was appropriate for developing a static and network-based dynamic nomogram (nomogram 2) . Calibration charts showed that the standard curve fitted well with the probability calibration curves of nomograms 1 and 2 (absolute error=0.010, and 0.019) , indicating that the calibration accuracies of both models were good. The AUC of nomogram 1 in predicting the prognosis of severe multiple trauma was 0.963〔95%CI (0.936, 0.981) 〕with 0.414 was the optimal cut-off value, and that of nomogram 2 was 0.974〔95%CI (0.949, 0.988) 〕 with 0.261 as the optimal cut-off value. Nomogram 2 had a larger AUC value than nomogram 1 (Z=-2.400, P=0.016) . The DCA results showed that under any threshold probability (0-100%) , the net benefit rate of nomogram 2 was higher than that of nomogram 1. (2) In the validation group, successful and unsuccessful treatment results occurred in 104 and 32 cases, respectively. The AUC of nomogram 2 predicting the prognosis of severe multiple trauma was 0.949〔95%CI (0.898, 0.979) 〕. And the model fitted well (χ2=5.813, P=0.668) revealed by Hosmer-Lemeshow test. The AUC of nomogram 2 in predicting the prognosis of severe multiple trauma in model and validation groups had insignificant changes (Z=1.124, P=0.263) .

Conclusion

Age, GCS, arterial lactate, platelet count and ISS were prognostic factors of severe multiple trauma, and the two nomograms in this study based on these five factors had good prognosis predictive value. In particular, the optimized nomogram 2 had higher accuracy (the network-based dynamic version is available at https://yinfxyz.shinyapps.io/dynnomapp2/) , which was rapid, and easy-to-use, and it can help clinicians to identify patients early and improve the prognosis of patients.

Key words: Multiple trauma, Wounds and injuries, Prognosis, Least absolute shrinkage and selection operator regression, Restricted cubic spline, Nomogram, Models, statistical

摘要:

背景

随着社会的发展,创伤导致的严重多发伤患者逐年增多,这对社会劳动力造成的损失已远大于其他疾病。如何早期快速地评估病情将对患者预后产生重要影响,而一个简洁、有效的可视化评估工具可为早期诊断和治疗提供重要依据。

目的

分析和筛选影响严重多发伤患者预后的相关因素,构建列线图模型,进一步调整优化模型,并验证模型的临床应用价值。

方法

选取2015年12月至2020年12月苏州市第九人民医院综合ICU、急诊ICU收治的严重多发伤患者作为建模组人群(n=321),2021年1—8月收治的严重多发伤患者作为预测模型的验证组人群(n=136),两组均根据患者出院时病情分为救治成功和救治失败。回顾性收集患者入院时的一般资料和入院24 h内的临床资料。采用单因素分析、LASSO回归分析筛选影响严重多发伤患者预后的相关变量,运用多因素Logistic回归分析建模、限制性立方样条(RCS)进行优化,绘制列线图,采用Bootstrap方法和临床决策曲线分析(DCA)验证模型的校准度。绘制模型预测严重多发伤患者预后的受试者工作特征(ROC)曲线,利用ROC曲线下面积(AUC)等指标评价新模型。在验证组人群中进行新模型的外部验证,评估其实际临床应用价值。

结果

(1)建模组321例严重多发伤患者中救治成功244例,救治失败77例。LASSO回归结合多因素Logistic回归分析结果显示,年龄(OR=1.028)、格拉斯哥昏迷评分(GCS)(OR=0.616)、动脉血乳酸(LAC)(OR=1.202)、血小板计数(PLT)(OR=3.888)、损伤严重程度评分(ISS)(OR=1.104)是严重多发伤患者预后的影响因素(P<0.05),Hosmer-Lemeshow拟合优度检验结果显示χ2=2.717、P=0.951,绘制静态及网页版动态列线图(即模型1列线图)。LASSO回归结合RCS拟合优化多因素Logistic回归分析结果显示,年龄、GCS与救治结果呈非线性相关,P值(for nonlinear)分别为0.027、0.001,Hosmer-Lemeshow拟合优度检验结果显示χ2=2.468、P=0.932,绘制静态及网页版动态列线图(即模型2列线图)。Bootstrap方法验证模型1、模型2列线图预测效能,校准图显示标准曲线与预测曲线贴合良好,绝对误差分别为0.010、0.019,说明模型的校准度良好。模型1、模型2列线图预测严重多发伤患者预后的AUC分别为0.963〔95%CI(0.936,0.981)〕、0.974〔95%CI(0.949,0.988)〕,最佳截断值分别为0.414、0.261;模型2列线图预测严重多发伤患者预后的AUC大于模型1列线图(Z=-2.400,P=0.016)。DCA结果显示,在任何阈值概率下(0~100%),模型2列线图的净收益率高于模型1列线图。(2)验证组136例严重多发伤患者中救治成功104例,救治失败32例。在验证组人群中,模型2列线图预测严重多发伤患者预后的AUC为0.949〔95%CI(0.898,0.979)〕,Hosmer-Lemeshow拟合优度检验表明该模型拟合良好(χ2=5.813,P=0.668)。模型2列线图在建模组及验证组人群中的AUC比较,差异无统计学意义(Z=1.124,P=0.263)。

结论

年龄、GCS、LAC、PLT、ISS是严重多发伤患者预后的影响因素,本研究基于上述5个因子构建出的列线图具有较好的预测价值,且优化后的模型2列线图准确性更高(网址:https://yinfxyz.shinyapps.io/dynnomapp2/),应用方便快捷,有助于帮助临床医生早期识别患者病情,改善患者预后。

关键词: 多处创伤, 创伤和损伤, 预后, LASSO回归, 限制性立方样条, 列线图, 模型,统计学