卒中后抑郁风险列线图预测模型的构建及验证

Construction and validation of a nomogram prediction model for post-stroke depression

目的建立早期卒中后抑郁风险列线图预测模型,为临床医护人员筛查卒中后抑郁高危患者提供工具。方法对建模组259例卒中患者在卒中后第7~14天收集10项相关危险因素,于卒中后8~10周采用汉密尔顿抑郁量表和蒙哥马利抑郁量表测评卒中患者抑郁状况。利用χ^(2)检验、Lasso回归及logistic回归筛选危险因素建立列线图预测模型,对模型进行内部及外部验证(验证组有82例患者)。结果卒中后抑郁检出率为39.38%;性别、婚姻状况、并存疾病数目、卒中部位、神经功能受损程度、日常生活活动能力是卒中后抑郁的独立危险因素(均P<0.05)。基于上述6个独立危险因素建立的列线图预测模型具有较好的区分度(AUC值:内部验证为0.883,外部验证为0.849)和准确度(Hosmer-Lemeshow检验:内部验证χ^(2)=7.939,P=0.439,外部验证χ^(2)=3.538,P=0.896);决策曲线分析显示预测模型曲线在大于10%的阈值概率区间具有临床实用价值。结论卒中患者发作后8~10周即有较高的卒中后抑郁发生率,构建的列线图模型能够有效预测早期卒中后抑郁风险,利于临床给予针对性干预。

Objective To develop a nomogram prediction model for early screening of post-stroke depression(PSD),and to provide an instrument for screening high risk patients.Methods Ten risk factors were collected from 259 stroke patients(the derivative group)within 7-14 days after stroke onset,and depression was measured using the Hamilton Depression Inventory and the Montgomery Depression Inventory in 8-10 weeks after stroke.The risk factors were determined using Chi-square test,Lasso regression and logistic regression to establish a nomogram prediction model,and the model was validated internally and externally(the validation group,82 patients).Results The detection rate of PSD was 39.38%in the derivative group.Gender,marital status,number of coexisting diseases,lesion location,degree of neurological impairment,and ability to perform activities of daily living were independent risk factors for PSD(all P<0.05).The nomogram prediction model based on the above-mentioned 6 independent risk factors had good discrimination(AUC value:0.883 in internal validation and 0.849 in external validation)and accuracy(Hosmer-Lemeshow test:χ^(2)=7.939,P=0.439 in internal validation,andχ^(2)=3.538,P=0.896 in external validation).Decision curve analysis showed that the prediction model curve had clinical benefit when the threshold probability value was greater than 10%.Conclusion There is a high incidence of PSD in stroke patients 8 to 10 weeks after the onset of stroke.The nomogram prediction model can effectively predict the risk of early PSD and facilitate clinical delivery of targeted interventions.