SOAR和GWTG-Stroke死亡预测模型的验证与扩展应用

发布时间：2018-05-03 18:57

本文选题：GWTG-Stroke预测模型 + SOAR预测模型　；参考：《首都医科大学》2014年硕士论文

【摘要】：目的应用中国国家卒中登记数据库（China National Stroke Register, CNSR）对卒中类型-牛津郡社区卒中项目-年龄-卒中前mRS评分模型（Stroke subtype,Oxfordshire Community Stroke Project Classification, Age, and prestroke Rankinstroke，SOAR）评分和跟着指南走（Get With the Guidelines Stroke，GWTG-Stroke）死亡预测评分进行外部验证，以明确上述两个预测模型能否应用于中国人群，同时探讨上述两个模型预测远期死亡的能力。资料与方法在CNSR中筛选符合原文准入及排除标准的子数据库。主要终点事件为在院死亡，次要终点为30天、3个月、6个月、1年死亡。对CNSR中子数据库的人群特征及终点事件与原文人群进行单因素分析和比较，其中连续变量的比较应t检验，分类变量应用卡方检验，P0.01提示两组差异有统计学意义。应用多因素Logistic回归的方法对模型的辨别能力和校正能力进行评估，其中辨别能力通过受试者工作曲线下面积（C值）及95%可信区间进行评定，C值越接近1，提示预测能力越好；校正能力应用Pearson相关系数比较预测的及观察到的死亡事件的拟合程度，相关系数0.9提示拟合优度好。结果 1.对于GWTG-Stroke模型的验证 CNSR中的人群与GWTG的人群相比存在显著差异。中国人群较为年轻，男性比例较高，通过私人交通工具到院的比例较高（P0.001）；患房颤、心脏瓣膜置换、既往卒中/TIA、冠状动脉病、糖尿病、周围血管病、高血压和脂代谢紊乱的比例明显较低（P0.001）。CNSR的在院死亡率为6.3%，明显低于GWTG的8.5%，其差异具有统计学意义。无美国国立卫生研究院卒中量表(NationalInstitute of Health stroke scale, NIHSS)评分的模型预测在院死亡的C值为0.76(0.75-0.78)，有NIHSS评分模型的为0.86(0.84-0.88)。有NIHSS评分的模型预测30天、3个月、6个月和1年死亡率的C值分别为0.86（0.84-0.88），0.84（0.83-0.86），0.83（0.81-0.84），0.82（0.80-0.83），无NIHSS评分的模型预测中远期死亡的C值在0.71-0.76之间。这两个模型预测在院死亡时的拟合优度欠佳（Pearson相关系数分别为0.213、0.689）。 2.对于SOAR模型的验证 CNSR中的人群与SOAR的人群相比，中国人群较为年轻，男性比例较高，发病前mRS评分较低（P0.001）；OCSP分型之间存在明显差异。CNSR的在院死亡率为4.7%，明显低于SOAR的19.8%，其差异具有统计学意义。SOAR模型预测在院、30天、3个月、6个月、1年死亡的C值及95%可信区间分别为0.73（0.71-0.75），0.71（0.69-0.73），，0.71（0.69-0.72），0.71（0.69-0.72），0.71（0.69-0.72）。并且具有较好的拟合优度（Pearson相关系数分别为0.9）。结论 1. GWTG-Stroke涉及的两个模型，即无NIHSS模型和有NIHSS均能够预测中国卒中患者在院死亡。 2. GWTG-Stroke涉及的两个模型，能够预测中国卒中患者30天、3个月、6个月及1年的死亡，其中有NIHSS模型能够预测能力更好。 3. SOAR预测模型能预测中国人群卒中在院死亡及30天、3个月、6个月及1年的死亡。
[Abstract]:Purpose Use of China National Stroke Register, CNSR) to evaluate Stroke subtype Oxfordshire Community Stroke Project Classification, Age, and prestroke Rankinstroke-SOARAR-Stroke Type-Oxage in Oxfordshire Community Stroke projects and follow the guidelines to estimate the death of Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-Stroke-death The prediction score was externally validated, In order to determine whether the two prediction models can be applied to Chinese population, and to explore the ability of these two models to predict long-term mortality. Data and methods Screen subdatabases in CNSR that meet the original access and exclusion criteria. The primary end point event was hospital death, and the secondary end point was 30 days, 3 months, 6 months and 1 year. Univariate analysis and comparison of population characteristics and endpoint events between CNSR neutron database and the original population showed that the comparison of continuous variables should be t test, and the difference between the two groups was statistically significant by chi-square test (P0.01). The discriminative ability and correction ability of the model were evaluated by using multivariate Logistic regression method. The better the predictive ability was, the closer the C value of discrimination was assessed by the area under the operating curve (C value) and 95% confidence interval. Pearson correlation coefficient was used to compare the fitting degree of predicted and observed death events, and the correlation coefficient was 0. 9 indicating good fit. Result 1. Verification of GWTG-Stroke Model There was significant difference between the population of CNSR and that of GWTG. The Chinese population is relatively young, the proportion of men is higher, the proportion of people coming to hospital by private means of transportation is higher (P0.001); patients with atrial fibrillation, heart valve replacement, previous stroke / TIA, coronary artery disease, diabetes mellitus, peripheral vascular disease, The hospital mortality of hypertension and lipid metabolism disorder was significantly lower than that of GWTG (P 0.001N. CNSR), which was significantly lower than that of GWTG (8.5%). The model without the National Institute of Health stroke scale, NIHSS) score of the National Institutes of Health (NIH) was used to predict the death in hospital with a C value of 0.76 ~ 0.75 ~ 0.78, and a NIHSS score of 0.86 ~ 0.84-0.88. The C values of the model with NIHSS score were 0.86 ~ 0.84-0.88 ~ 0.83-0.86 ~ 0.86 ~ 0.83-0.86 ~ 0.84 ~ 0.81-0.84 ~ 0.82 ~ 0.80-0.83 respectively. The C value of the model without NIHSS score was between 0.71-0.76 for medium and long term mortality, which was predicted by the model of 30 days, 3 months, 6 months and 1 year, respectively. The Pearson correlation coefficients of the two models were 0.213 ~ 0.689g respectively. 2. Verification of SOAR Model The population in CNSR is younger than that in SOAR, and the proportion of men is higher. There was a significant difference between the mRS scores before onset and P0.001OCSP classification. The mortality rate in hospital was 4.70.It was significantly lower than that of SOAR 19.8.The difference was statistically significant. The Soar model predicted the C value and 95% confidence interval of death in hospital for 30 days, 3 months, 6 months, and 1 year were 0.730.71-0.750.750.730.730.71-0.730.71-0.71 and 0.69-0.72ng 0.779-0.72P, respectively. And the Pearson correlation coefficient with good fitting degree was 0.9. Conclusion 1. The two models involved in GWTG-Stroke, no NIHSS model and NIHSS, were able to predict nosocomial death of stroke patients in China. 2. The two models involved in GWTG-Stroke were able to predict 30 days, 3 months, 6 months and 1 year of death in Chinese stroke patients, among which NIHSS models were better predictors. 3. The SOAR predictive model can predict hospital death and 30 days, 3 months, 6 months and 1 year death of stroke in Chinese population.
【学位授予单位】：首都医科大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R743.3

【相似文献】