Department of Pediatrics, 2nd Faculty of Medicine, Charles University in Prague and Motol University Hospital, Prague.
Laboratory of Growth Regulators, Palacky University Olomouc and Institute of Experimental Botany AS CR, Olomouc.
J Pediatr Gastroenterol Nutr. 2019 Oct;69(4):e105-e110. doi: 10.1097/MPG.0000000000002436.
Therapeutic drug monitoring of thiopurine erythrocyte levels is not available in all centers and it usually requires quite a long time to obtain the results. The aims of this study were to build a model predicting low levels of 6-thioguanine and 6-methylmercaptopurine in pediatric inflammatory bowel disease (IBD) patients and to build a model to predict nonadherence in patients treated with azathioprine (AZA).
The study consisted of 332 observations in 88 pediatric IBD patients. Low AZA dosing was defined as 6-thioguanine levels <125 pmol/8 × 10 erythrocytes and 6-methylmercaptopurine levels <5700 pmol/8 × 10 erythrocytes. Nonadherence was defined as undetectable levels of 6-thioguanine and 6-methylmercaptopurine <240 pmol/8 × 10 erythrocytes. Data were divided into training and testing part. To construct the model predicting low 6-thioguanine levels, nonadherence, and the level of 6-thioguanine, the modification of random forest method with cross-validation and resampling was used.
The final models predicting low 6-thioguanine levels and nonadherence had area under the curve, 0.87 and 0.94; sensitivity, 0.81 and 0.82; specificity, 0.80 and 86; and distance, 0.31 and 0.21, respectively, when applied on the testing part of the dataset. When the final model for prediction of 6-thioguanine values was applied on testing dataset, a root-mean-square error of 110 was obtained.
Using easily obtained laboratory parameters, we constructed a model with sufficient accuracy to predict patients with low 6-thioguanine levels and a model for prediction of AZA treatment nonadherence (web applications: https://hradskyo.shinyapps.io/6TG_prediction/ and https://hradskyo.shinyapps.io/Non_adherence/).
并非所有中心都能进行硫嘌呤红细胞水平的治疗药物监测，而且通常需要相当长的时间才能获得结果。本研究的目的是建立一个预测儿科炎症性肠病（IBD）患者 6-硫鸟嘌呤和 6-甲基巯基嘌呤水平较低的模型，并建立一个预测巯嘌呤（AZA）治疗患者不依从的模型。
该研究包括 88 例儿科 IBD 患者的 332 次观察。低 AZA 剂量定义为 6-硫鸟嘌呤水平<125 pmol/8×10 个红细胞和 6-甲基巯基嘌呤水平<5700 pmol/8×10 个红细胞。不依从定义为 6-硫鸟嘌呤和 6-甲基巯基嘌呤<240 pmol/8×10 个红细胞的未检测水平。数据分为训练和测试部分。为了构建预测低 6-硫鸟嘌呤水平、不依从和 6-硫鸟嘌呤水平的模型，使用了带有交叉验证和重采样的随机森林方法的修改。
当应用于数据集的测试部分时，预测低 6-硫鸟嘌呤水平和不依从的最终模型的曲线下面积分别为 0.87 和 0.94；灵敏度分别为 0.81 和 0.82；特异性分别为 0.80 和 0.86；距离分别为 0.31 和 0.21。当将最终的 6-硫鸟嘌呤值预测模型应用于测试数据集时，得到了 110 的均方根误差。
使用易于获得的实验室参数，我们构建了一个具有足够准确性的模型，可以预测低 6-硫鸟嘌呤水平的患者和一个预测 AZA 治疗不依从的模型（网络应用程序：https://hradskyo.shinyapps.io/6TG_prediction/ 和 https://hradskyo.shinyapps.io/Non_adherence/）。
