Université de Lyon, 69622 Lyon, France.
Cancer Chemother Pharmacol. 2012 Feb;69(2):447-55. doi: 10.1007/s00280-011-1714-9. Epub 2011 Aug 2.
Anticancer drugs often show a narrow therapeutic index and high inter-patient variability, which can lead to the need to adjust doses individually during the treatment. One approach to doing this is to use individual model predictions. Such methods have been proposed to target-specific drug concentrations or blood cell count, both of which are continuous variables. However, many toxic effects are evaluated on a categorical scale. This article presents a novel approach to dose adjustments for reducing a graded toxicity while maintaining efficacy, applied to hand-and-foot syndrome (HFS) induced by capecitabine.
A mixed-effects proportional odds Markov model relating capecitabine doses to HFS grades was individually adjusted at the end of each treatment cycle (3 weeks) by estimating subject-specific parameters by Bayesian MAP technique. It was then used to predict the risk of intolerable (grade ≥ 2) toxicity over the next treatment cycle and determine the next dose accordingly, targeting a predefined tolerable risk. Proof of concept was given by simulating virtual clinical trials, where the standard dose reductions and the prediction-based adaptations were compared, and where the therapeutic effect was simulated using a colorectal tumor inhibition model. A sensitivity analysis was carried out to test various specifications of prediction-based adaptation.
Individualized dose adaptation might reduce the average duration of intolerable HFS by 10 days as compared to the standard reductions (3.8 weeks vs. 5.2 weeks; 27% relative reduction) without compromising antitumor efficacy (both responder rates were 49%). A clinical trial comparing the two methods should include 350 patients per arm to achieve at least 90% power to show a difference in grade ≥2 HFS duration at an alpha level of 0.05.
These results indicate that individual prediction-based dose adaptation based on ordinal data may be feasible and beneficial.
抗癌药物通常具有较窄的治疗指数和较高的个体间变异性，这可能导致在治疗过程中需要单独调整剂量。一种方法是使用个体模型预测。这些方法已经被提出用于靶向特定的药物浓度或血细胞计数，这两种都是连续变量。然而，许多毒性作用是在分类尺度上进行评估的。本文提出了一种新的方法，用于在保持疗效的同时减少分级毒性，应用于卡培他滨引起的手足综合征（HFS）。
通过贝叶斯 MAP 技术估计个体参数，在每个治疗周期（3 周）结束时对卡培他滨剂量与 HFS 分级之间的混合效应比例优势马尔可夫模型进行个体化调整。然后，它被用于预测下一个治疗周期不可耐受（等级≥2）毒性的风险，并相应地确定下一个剂量，以靶向预定的可耐受风险。通过模拟虚拟临床试验来证明概念的可行性，其中比较了标准剂量减少和基于预测的适应性，并且使用结直肠肿瘤抑制模型模拟治疗效果。进行了敏感性分析以测试基于预测的适应性的各种规格。
与标准减少相比，个体化剂量适应性可能使不可耐受的 HFS 的平均持续时间减少 10 天（3.8 周与 5.2 周；相对减少 27%），而不影响抗肿瘤疗效（应答率均为 49%）。比较两种方法的临床试验应包括每臂 350 例患者，以在α水平为 0.05 时至少有 90%的效力显示不可耐受的 HFS 等级≥2 的持续时间的差异。
这些结果表明，基于分类数据的个体基于预测的剂量适应性可能是可行且有益的。
