伏立康唑浓度超出参考范围上限的风险预测模型的建立与评估

Establishment and evaluation of a risk prediction model for voriconazole plasma concentration exceeding the reference therapeutic range

关键词:  伏立康唑  血药浓度  影响因素  风险预测模型

Key words:  voriconazole  blood concentration  influencing factors  risk prediction model

摘要:

目的 研究伏立康唑浓度超出参考治疗范围上限的危险因素并建立风险预测模型。方法 选择常州市第一人民医院2019年1月-2023年6月使用伏立康唑治疗侵袭性真菌感染并进行治疗药物监测的住院患者，将浓度在0.5~5.0mg·L-1内的患者列为达标组，浓度高于5.0mg·L-1的患者列为超限组，收集年龄、性别、BMI、基础疾病、吸烟史、饮酒史、伏立康唑用药情况、伏立康唑血药浓度、合并用药、血常规及肝肾功能等。采用单因素分析及二元logistic回归分析筛选伏立康唑血药浓度超出参考治疗范围上限的危险因素，构建列线图模型并进行评价。结果 二元logistic回归结果显示年龄（OR=3.458，P=0.001）、性别（OR=0.398，P=0.032）、剂量（OR=1.560，P=0.002）、白蛋白（OR=0.352，P=0.007）、C反应蛋白（OR=2.299，P=0.023和OR=4.898，P=0.002）是伏立康唑血药浓度是否超限的独立影响因素。基于此构建的列线图模型的ROC曲线下面积为0.802（95%CI为0.744~0.859）。采用Bootstrap自抽样法重复抽样1000次对模型进行内部验证，结果显示平均绝对误差0.021，预测值与观察值之间一致性较好。当风险阈值约在10%-92%范围内时，使用列线图模型预测伏立康唑浓度超出参考治疗范围上限的发生概率可产生临床净收益。结论 本研究基于二元logistic回归筛选出可能导致伏立康唑浓度超限的五个因素，即年龄、性别、剂量、白蛋白及C反应蛋白，并建立列线图预测模型，可助力医师快速识别伏立康唑浓度超标的风险人群。

Abstract:

OBJECTIVE To study the risk factors of voriconazole concentration exceeding the upper limit of the reference therapeutic range, and establish a risk prediction model. Methods Hospitalized patients who used voriconazole for invasive fungal infection and monitored the drug concentration were selected in The First People""s Hospital of Changzhou from January 2019 to June 2023. Patients whose concentration is within 0.5~5.0 mg·L-1 were classified as the compliance group, and whose voriconazole concentration higher than 5.0 mg· L-1 were classified as the over-standard group, clinical data were collected, including age, gender, body mass index, underlying diseases, smoking and drinking history, voriconazole medication, voriconazole concentration, combined medication, blood routine, liver and kidney function, etc. Univariate analysis and binary logistic regression analysis were used to study the risk factors of voriconazole blood concentration exceeding the upper limit of the reference treatment range, and constructed a nomogram prediction model, evaluated the discrimination, accuracy and the clinical benefit of the model. Results The results of binary logistic regression analysis showed that age (OR=3.458,P=0.001), gender(OR=0.398, P=0.032), dose(OR=1.560, P=0.002), albumin (OR= 0.352, P=0.007) and C-reactive protein(OR=2.299, P=0.023 and OR=4.898,P=0.002) were independent influencing factors of whether the plasma concentration of voriconazole exceeded the limit. Based on this, the area under the ROC curve of the nomogram model is 0.802（95%CI为0.744~0.859）, which indicated a high degree of discrimination of the nomogram prediction model, and the bootstrap self-sampling method is used to repeat sampling for 1000 times to verify the internal verification of the model, and the average absolute error is 0.021, the results showed the good consistency between the predicted value and the observed value. When the risk threshold is in the range of about 10% to 92%, using a nomogram model to predict the probability of VRZ concentration exceeding the upper limit of the reference treatment range can yield a net clinical benefit. Conclusion Based on binary logistic regression, five factors of excessive voriconazole concentration were screened out, namely age, gender, dose, low albumin and C-reactive protein, and established a nomogram prediction model, it can help physicians identify the risk groups with excessive voriconazole concentration quickly.