脑卒中患者短期预后预测模型的构建及验证

Construction and validation of a short-term prognostic prediction model for mild to moderate stroke patients

  • 摘要:
    背景 早期识别脑卒中患者短期不良事件的高危因素和高风险人群,对于提高患者的预后和康复进程至关重要。
    目的 分析脑卒中患者短期预后的影响因素,构建预测模型及列线图,为制定针对性的康复干预措施提供参考。
    方法 采用回顾性病例对照研究方法,纳入2018年1月— 2020年12月在解放军总医院第一医学中心就诊并符合入选标准的脑卒中患者。以改良Barthel指数评价短期预后结果,使用Logistic回归从年龄、性别、个人病史、合并症等指标筛选出最佳预测变量,建立预测模型,构建列线图。使用ROC曲线下面积、Hosmer-Lemeshow检验、校准曲线,比较模型与临床量表的预测能力,验证其区分度和校准度。选取2021年1月— 2023年1月符合入选标准的脑卒中患者作为验证集,对预测模型进行验证。
    结果 本研究的训练集纳入120例患者,验证集纳入65例患者。多因素Logistic回归结果显示,年龄(OR=1.033,95% CI:0.990 ~ 1.062,P=0.041)、肺部感染(OR=1.724,95% CI:1.652 ~ 1.880,P=0.007)、改良RANKIN量表(modified rankin scale,mRS)评分(OR=1.970,95% CI:1.353 ~ 2.872,P<0.001)、医学研究理事会量表总分(健侧肢体)(MRC sum scores,MRC-SS)(OR=0.854,95% CI:0.744 ~ 0.982,P=0.012)与脑卒中短期预后不良独立关联。据此构建预测模型,ROC曲线下面积为0.809(95% CI:73.7% ~ 89.2%),敏感度为0.803,特异度为0.763。Hosmer-Lemeshow检验值为P=0.385;校准曲线显示预测值与实际值之间存在显著一致性。与美国国立卫生研究院卒中量表评分相比,模型显示出明显更高的预测能力(AUC:0.809 vs 0.613,P=0.004)。模型经验证集验证,表现出与训练集结果有相似的预测价值(AUC=0.784,95% CI:0.665 ~ 0.902)。
    结论 本研究通过年龄、肺部感染、改良RANKIN量表、健侧医学研究理事会量表评分4种指标建立预测模型,对脑卒中患者入院后两周的短期预后进行预测,模型预测能力良好,有助于临床早期康复干预,改善患者长期预后。

     

    Abstract:
    Background Early identification of high-risk factors and populations for short-term adverse events in stroke patients is crucial for improving prognosis and recovery outcomes.
    Objective To analyze the factors influencing short-term prognosis in stroke patients and develop a predictive model and nomogram, so as to provide references for formulating targeted rehabilitation interventions.
    Methods A retrospective case-control study was conducted, including stroke patients who admitted to our hospital from January 2018 to December 2020 and met the inclusion criteria. Modified Barthel Index was applied to evaluate short-term prognosis, and key predictors were identified from clinically relevant variables (age, sex, past medical history, comorbidities) using multivariate logistic regression. A prognostic prediction model was developed based on the identified variables, followed by the construction of a clinical nomogram. The predictive performance of the model versus clinical scales was assessed using the area under the ROC curve (AUC), Hosmer-Lemeshow goodness-of-fit test, and calibration curves to evaluate discrimination and calibration accuracy. Stroke patients who met the inclusion criteria from January 2021 to January 2023 were selected as a validation cohort to verify the predictive model.
    Results The training set of this study included 120 patients and the validation set included 65 patients. The results of multifactorial logistic regression showed that age (OR=1.033, 95% CI: 0.990 - 1.062, P=0.041), lung infection (OR=1.724, 95% CI: 1.652 - 1.880, P=0.007), modified Rankin Scale (mRS) score (OR=1.970, 95% CI: 1.353 - 2.872, P < 0.001) and Medical Research Council scale total score (healthy limb) (MRC sum scores, MRC-SS) (OR=0.854, 95% CI: 0.744 - 0.982, P=0.012) were independently correlated with poor short-term prognosis of stroke. A prediction model was constructed accordingly, and the area under the ROC curve (AUC) was 0.809 (95% CI: 73.7% - 89.2%), with a sensitivity of 0.803 and a specificity of 0.763. The Hosmer-Lemeshow test value was P=0.385; and the calibration curve showed a significant concordance between the predicted and actual values. The model showed significantly higher predictive power compared with the National Institutes of Health Stroke Scale score (AUC: 0.809 vs 0.613, P=0.004). The model was validated against the validation set and showed similar predictive value to the training set results (AUC=0.784, 95% CI: 0.665 - 0.902), the short-term prognostic clinical prediction model for stroke had good stability.
    Conclusion This study develops a prediction model utilizing four variables (age, pulmonary infection status, modified Rankin Scale score, and contralateral Medical Research Council Scale score) to forecast short-term prognosis at two weeks post-admission in stroke patients. The model demonstrates favorable predictive performance, potentially facilitating early clinical rehabilitation interventions and improving long-term patient outcomes.​

     

/

返回文章
返回