Abstract

Lead is a common toxin which has detrimental effects on human health. Since lead poisoning is not associated with specific symptoms, diagnosing elevated blood lead concentration (EBLC) should be taken seriously. The purpose of this study was to propose a prediction model for EBLC based on demographic and clinical variables through a decision-tree model.

In this cross-sectional study, 630 subjects (above 40 years old) living in South Khorasan Province, Iran in 2017 were selected via cluster random sampling method. From among the 630 participants who met the inclusion criteria, 70% (N = 456) were chosen randomly to achieve a set for developing the decision tree and multiple logistic regression (MLR). The other 30% (N = 174) were placed in a holdout sample to examine the function of the decision tree and MLR models. The predictive performance for various models was studied using the Receiver Operating Characteristic (ROC) curve.

In the decision tree model, the parameters of hematocrit (HCT), White Blood Cell (WBC), Red Blood Cell (RBC), Mean corpuscular volume (MCV), creatinine concentration, abdominal pain, gender, route of administration, and history of cigarette smoking were the most critical factors in identifying people at risk of EBLC. The HCT concentration was the most critical variable, which was chosen as the root node of the tree. Based on the ROC curve, the decision tree model had better predictive accuracy than the logistic regression model.

Our results indicated that the decision tree model offers far greater predictive precision than the logistic regression model. Doctors should pay more attention to some factors including the hematological parameters such as MCV, RBC, HCT, leukocytosis, creatinine levels, male sex, history of cigarette, and opium consumption for the screening of EBLCs.

摘要

铅是一种常见的毒素，对人体健康有不利影响。由于铅中毒与特定症状无关，因此应认真对待血铅浓度升高（EBLC）的诊断。本研究的目的是通过决策树模型提出基于人口统计学和临床​​变量的 EBLC 预测模型。

在这项横断面研究中，通过整群随机抽样方法选择了 2017 年生活在伊朗南呼罗珊省的 630 名受试者（40 岁以上）。从符合纳入标准的 630 名参与者中，随机选择 70% ( N  = 456) 来获得一组用于开发决策树和多元逻辑回归 (MLR)。其他 30% ( N  = 174) 被放置在一个保留样本中，以检查决策树和 MLR 模型的功能。使用接收器操作特征 (ROC) 曲线研究了各种模型的预测性能。

在决策树模型中，血细胞比容 (HCT)、白细胞 (WBC)、红细胞 (RBC)、平均红细胞体积 (MCV)、肌酐浓度、腹痛、性别、给药途径和历史吸烟是识别有 EBLC 风险的人的最关键因素。HCT浓度是最关键的变量，被选为树的根节点。基于ROC曲线，决策树模型比逻辑回归模型具有更好的预测精度。

我们的结果表明，决策树模型比逻辑回归模型提供了更高的预测精度。医生在筛查 EBLC 时应多加注意一些因素，包括 MCV、RBC、HCT、白细胞增多、肌酐水平、男性、吸烟史和鸦片消费等血液学参数。