应用数据挖掘技术构建无机钙磷制剂相容性效应面预测模型  

Compatibility effect surface prediction model of inorganic calcium and phosphorus preparations in paren⁃teral nutrition solution

作　　者：解根育 杨丝露 田绍武 徐帆 XIE Genyu;YANG Silu;TIAN Shaowu;XU Fan(Graduate School,Kunming Medical Unisivity,Yunnan Kunming 671500,China;Department of Pharmacy,Fuwai Cardiovascular Hospital of Yunnan Province,Yunnan Kunming 671500,China;Department of Clinical Pharmacy,No.920 Hospital of Joint Logistics Support Force,Yunnan Kunming 671500,China)

机构地区：[1]昆明医科大学研究生院,云南昆明671500 [2]云南省阜外心血管病医院药剂科,云南昆明671500 [3]解放军联勤保障部队第九二〇医院临床药学科,云南昆明671500

出　　处：《中国医院药学杂志》2024年第6期714-719,共6页Chinese Journal of Hospital Pharmacy

基　　金：基础研究计划(编号:202101AY070001-293)。

摘　　要：目的:为安全合理使用无机钙、磷制剂组方肠外营养液,有效评判肠外营养处方中是否存在产生钙磷沉淀的风险,提供实验室数据及预测计算模型支持。方法:应用自研动态法微粒监测装置对无机钙、磷浓度、温度、pH值、复方氨基酸种类及浓度、葡萄糖浓度7个影响因素全部水平相互组合实验,并运用R软件建立逻辑回归、随机森林算法和支持向量机3种肠外营养液中无机钙磷制剂相容性效应面预测模型,对比评价3种算法,优选模型并验证拟合度。结果:(1)实验共获取14 400组有效数据集。(2)通过对3种模型的评价对比,随机森林算法构建的肠外营养液中无机钙磷制剂相容性效应面模型准确率最高、预测效果最好。(3)基于随机森林算法的模型分析显示钙浓度、磷浓度、pH值、温度、葡萄糖浓度与无机钙磷沉淀的产生呈正相关,其中钙浓度对沉淀的产生影响最大,2类复方氨基酸的浓度与无机钙磷沉淀的产生呈负相关。(4)基于随机森林算法构建的效应面模型进行实验分析验证,预测结果与实验结果拟合度高。结论:该研究应用现代分析技术研发了可动态监测无机钙磷制剂配伍后沉淀产生的装置,并构建了肠外营养液中无机钙磷制剂相容性效应面模型。应用该模型,输入相应参数,便可预测出该配方组合是否会有无机钙磷沉淀产生,为临床医务人员安全合理开具肠外营养处方提供实验室依据及计算模型支持。OBJECTIVE To provide laboratory data and predictive model supports for safe and rational uses of inorganic cal⁃cium and phosphorus parenteral nutrition solution and effectively evaluate the risk of calcium and phosphorus precipitation in paren⁃teral nutrition prescription.METHODS The self-developed dynamic particle monitoring device was utilized for combining seven influencing factors of inorganic calcium,phosphorus ion concentration,temperature,pH value,type/concentration of compound amino acid and glucose concentration at all levels.The compatibility effect surface prediction models of inorganic calcium and phos⁃phorus preparations in parenteral nutrition solution of logistic regression,random forest algorithm and support vector machine were established by R software and three algorithms were compared.Optimized model was employed for verifying the degree of fit.RESULTS A total of 14400 effective data sets were obtained in the experiment.Through the evaluations of three models,the compatibility effect surface model of inorganic calcium and phosphorus in a parenteral nutrition solution based upon a random forest algorithm exhibited the highest accuracy and the optimal prediction effect.Model analysis based upon a random forest algorithm revealed that calcium concentration,phosphorus concentration,pH value,temperature and glucose concentration were correlated positively with inorganic calcium and phosphorus precipitation.And calcium concentration had the greatest effect on inorganic cal⁃cium and phosphorus precipitation.The concentrations of two kinds of compound amino acids were correlated negatively with the precipitation of inorganic calcium and phosphorus.The effect surface model generated by random forest algorithm was utilized.It indicated that a significant level of agreement existed between projected results and actual observations.CONCLUSION A device may dynamically monitor the precipitation of inorganic calcium and phosphorus preparations.And compatibility effect sur⁃face model of inorganic cal

关 键 词：肠外营养液 无机钙磷沉淀 逻辑回归 随机森林算法 支持向量机 效应面 效应面模型 

分 类 号：R917[医药卫生—药物分析学]