基于机器语言的临床预测模型对高血压脑出血预后的预测

Prognosis Prediction of Hypertensive Intracerebral Hemorrhage with Clinical Prediction Model Based on Machine Language

目的:基于我院高血压脑出血(HICH)患者的数据库,使用机器语言的多项式函数(Polynomial),建立预测预后的数据模型,并进行内部检测。方法:回顾性分析327例HICH患者的临床资料。将出院时改良Rankin评分量表(mRS评分)0~2分归为预后良好组,3~6分归为预后不良组。将327例患者分为两组:80%的患者为训练组、20%的患者为测试组。通过机器语言的多项式函数建立数据模型GHRRSD:入院时的格拉斯哥昏迷评分(AGCS)+入院时的心率(AHR)+入院时的呼吸频率(AR)+入院时的收缩压(ASP)+入院时舒张压(ADP)。基于数据模型,分别得出训练组和测试组的ROC曲线,以曲线下面积(AUC)检测数据模型的有效性。再依据数据模型得出其动态列线图,用以检测其实用性。结果:GHRRSD模型对于患者的预后具有很好的预测性。训练组ROC曲线的最优截断值为0.522,95%可信区间为(0.902,0.740),AUC为0.881;测试组ROC曲线的最优截断值为0.551,95%CI为(0.963,0.806),AUC为0.941。二者进行比较无统计学差异(P=0.087),数据模型能适用于训练组及测试组。GHRRSD模型得出动态列线图,能很好地测算出患者不良预后的概率。结论:使用GHRRSD模型对HICH患者预后进行预测,有助于及时发现患者病情、指导临床诊断和治疗及评估预后。

Objective:To develop a data model for predicting prognosis based on a database of hypertensive cerebral hemorrhage(HICH)patients in our hospital using polynomial functions in machine language and to test it internally.Methods:The clinical data of 327 patients with HICH were retrospectively analyzed.A modified Rankin Rating Scale(mRS)of 0~2 at discharge was categorized as a good prognosis group,and a mRS score of 3~6 was categorized as a poor prognosis group.The 327 patients were divided into two groups:80%of the patients were the training group and 20%of the patients were the test group.The data model GHRRSD was developed by polynomial function in machine language:Glasgow Coma Score at admission(AGCS)+heart rate at admission(AHR)+respiratory rate at admission(AR)+systolic blood pressure at admission(ASP)+diastolic blood pressure at admission(ADP).Based on the data model,the ROC curves of the training and test groups were derived separately,and their area under the curve(AUC)was calculated to test the validity of the data model.Then,the dynamic line graphs were derived based on the data model to test its usefulness.Results:The GHRRSD model has good predictive power for patient prognosis.The Optimal Cutoff Value of the ROC curve in the training group was 0.522,the 95%confidence interval(95%CI)was(0.902,0.740),and the AUC was 0.881;the optimal cutoff value of the ROC curve in the test group was 0.551,the 95%CI was(0.963,0.806).There was no statistically significant difference between the two groups(P=0.087),the data model can be applied to the training group and the test group.The GHRRSD model yielded dynamic nomogram plots,which are a good measure of the probability of poor prognosis in patients.Conclusion:Using the GHRRSD model to predict the prognosis of HICH patients can help to detect patients'conditions,guide clinical diagnosis and treatment and assess prognosis in a timely manner.