基于Lasso-logistic回归模型构建高龄妊娠期高血压疾病患者不良妊娠结局的列线图模型

Development of a nomogram for adverse pregnancy outcomes in pregnant women of advanced maternal age with hypertensive disorders of pregnancy using Lasso-logistic regression

摘要:

目的 探讨基于Lasso-logistic多因素模型构建预测模型对高龄妊娠期高血压疾病（HDP）患者不良妊娠结局的预测价值。方法 选取2021年1月—2025年6月攀枝花市中心医院收治的342例高龄HDP患者，记录妊娠不良结局发生率。按7∶3的比例将342例患者随机分为训练集（240例）和验证集（102例），将连续变量Z-score标准化，采用Lasso-logistic多因素模型分析训练集数据，分析高龄HDP患者不良妊娠结局的相关因素，构建风险预测模型。采用验证集数据对模型进行验证，以10折交叉验证、λ_min/λ_1se双参数报告、多重插补（m =5）、Box-Tidwell线性检验、方差膨胀因子多重共线性检验、H-L检验10分组、校准曲线、临床决策曲线（DCA）阈值净收益进行分析。结果 训练集和验证集患者中，分别有62例和24例发生不良妊娠结局，发生率分别为25.83%和23.53%。Lasso回归分析筛选产次、糖尿病、前置胎盘、活化部分凝血活酶时间（APTT）、白蛋白（Alb）、胰岛素抵抗指数（HOMA-IR）共6项，将上述因子纳入多因素一般logistic回归分析，结果显示，合并糖尿病［O^R=4.408（95% CI：2.507，7.750）］、前置胎盘［O^R=3.238（95% CI：1.647，6.366）］及HOMA-IR水平高［O^R=2.617，95%CI：1.511，4.533）］均为高龄HDP患者不良妊娠结局的危险因素（P <0.05），Alb水平高［O^R=0.814（95% CI：0.698，0.949）］为高龄HDP患者不良妊娠结局的保护因素（P <0.05）。基于上述因素构建风险预测模型，构建预测方程：1.247+1.483×合并糖尿病+1.175×前置胎盘-0.206×Alb+0.962×HOMA-IR。Hosmer-Lemeshow检验显示，训练集与验证集的预测概率、实际结局比较，差异均无统计学意义（P >0.05）。分别以训练集和验证集数据绘制校准曲线，结果显示，训练集与验证集中实际概率与预测概率的平均绝对误差分别为0.019和0.036，DCA分析结果提示训练集和验证集在0.12～0.84和0.05～0.90中，预测模型的临床净收益高。ROC曲线结果显示，该预测模型预测训练集与验证集高龄HDP患者不良妊娠结局的截断值分别为230.0和225.5，敏感性分别为88.7%（95% CI：0.704，0.912）和87.5%（95% CI：0.743，0.965），特异性分别为74.2%（95% CI：0.718，0.901）和73.1%（95% CI：0.725，0.920）。结论 高龄HDP患者存在不良妊娠结局风险，其发生与患者合并糖尿病、前置胎盘、Alb及HOMA-IR水平有关，临床据此构建的风险预测模型有助于不良妊娠结局高危人群的早期筛查。

Abstract:

Objective To evaluate the predictive value of a model based on multivariable Lasso-logistic regression for adverse pregnancy outcomes in pregnant women of advanced maternal age with hypertensive disorders of pregnancy (HDP).Methods A total of 342 pregnant women of advanced maternal age with HDP treated at Panzhihua Central Hospital between January 2021 and June 2025 were enrolled, and adverse pregnancy outcomes were recorded. The patients were randomly divided 7:3 into a training set (n = 240) and a validation set (n = 102). Continuous variables were standardized using Z scores. A multivariable Lasso-logistic model was used to analyze the training-set data, identify factors associated with adverse pregnancy outcomes, and construct a risk prediction model. The validation set was used to validate the model. Analyses included 10-fold cross-validation, dual reporting of lambda-minimum/lambda-1-standard-error parameters, multiple imputation (m = 5), the Box-Tidwell linearity test, variance inflation factor assessment of multicollinearity, the 10-group Hosmer-Lemeshow test, calibration curves, and decision curve analysis (DCA) of threshold net benefit.Results Adverse pregnancy outcomes occurred in 62 patients in the training set and 24 in the validation set, corresponding to incidences of 25.83% and 23.53%, respectively. Lasso regression selected six variables: parity, diabetes, placenta previa, activated partial thromboplastin time, albumin (Alb), and the homeostatic model assessment of insulin resistance (HOMA-IR). Multivariable logistic regression showed that diabetes [O^R = 4.408 (95% CI: 2.507, 7.750) ], placenta previa [O^R = 3.238 (95% CI: 1.647, 6.366) ], and higher HOMA-IR [O^R = 2.617 (95% CI: 1.511, 4.533) ] were risk factors for adverse pregnancy outcomes (P < 0.05), whereas higher Alb [O^R = 0.814 (95% CI: 0.698, 0.949) ] was a protective factor (P < 0.05). The prediction equation was Logit(P) = 1.247 + 1.483 x diabetes + 1.175 x placenta previa - 0.206 x Alb + 0.962 x HOMA-IR. The Hosmer-Lemeshow test showed no significant differences between predicted probabilities and observed outcomes in either the training or validation set (P > 0.05). Calibration curves showed mean absolute errors of 0.019 and 0.036 in the training and validation sets, respectively. DCA indicated high clinical net benefit at threshold probabilities of 0.12-0.84 in the training set and 0.05-0.90 in the validation set. The model cutoff values were 230.0 and 225.5 in the training and validation sets, respectively; sensitivities were 88.7% (95% CI: 0.704, 0.912) and 87.5% (95% CI: 0.743, 0.965), and specificities were 74.2% (95% CI: 0.718, 0.901) and 73.1% (95% CI: 0.725, 0.920), respectively.Conclusion Pregnant women of advanced maternal age with HDP are at risk of adverse pregnancy outcomes associated with diabetes, placenta previa, Alb, and HOMA-IR. A risk prediction model based on these factors may facilitate early identification of high-risk patients.