Evaluation of CITOR NTCP models in HNC patients treated at MDACC

<b>Background and purpose:</b> Normal tissue complication probability (NTCP) models can be used to guide radiation therapy (RT) treatment planning and select the optimal technique to minimize side-effects. The comprehensive individual toxicity risk (CITOR) profile encompasses NTCP models for 22 common side-effects experienced by head and neck cancer (HNC) patients. To investigate the generalizability of these models, even with different side-effect assessments, and to promote integration into clinical practice, this study aimed to evaluate the performance of several CITOR models in HNC patients treated with RT at the MD Anderson Cancer Center (MDACC).<b>Material/methods:</b> In total, 407 patients from a prospective registry study with oropharyngeal cancer or an unknown primary HNC tumor treated with definitive RT with or without systemic therapy between 2015 and 2022 were included. NTCP models predicting dysphagia, aspiration, xerostomia, sticky saliva, taste loss, speech problems, oral pain, and fatigue at 6 and 12 months after RT were evaluated. All side-effects were patient-rated using the MDASI-HN, except dysphagia which was reported by clinicians using PSS-HN diet normalcy score. Model performance was appraised by discrimination (area under the curve (AUC)), calibration, and explained variance (R²).<b>Results</b>: CITOR models showed moderate-to-high performance in the MDACC cohort (mean AUC=0.67, often slight overestimation of the predicted risk, range R²=0.02-0.28). NTCP models for dysphagia, xerostomia, sticky saliva, and fatigue were the top performing models. Models for aspiration, taste loss and speech problems performed moderately well, which was partly explained by lower incidences.<b>Conclusion: </b>Despite differences between the CITOR development and MDACC validation cohorts, including use of different side-effect assessment tools, most models exhibited moderate-to-high performance, demonstrating that the dose-effect relations as described in the models were generalizable. Therefore, this study supports further integration of these NTCP models in clinical practice, especially of the NTCP models predicting dysphagia, xerostomia, sticky saliva, and fatigue.

背景与目的：正常组织并发症概率（Normal Tissue Complication Probability, NTCP）模型可用于指导放射治疗（Radiation Therapy, RT）计划制定，并筛选最优技术以最小化不良反应风险。综合个体毒性风险（Comprehensive Individual Toxicity Risk, CITOR）概况涵盖了针对头颈部肿瘤（Head and Neck Cancer, HNC）患者常见的22种不良反应的NTCP模型。为探究此类模型在不同不良反应评估方式下的泛化性，并推动其在临床实践中的整合应用，本研究旨在评估多款CITOR模型在安德森癌症中心（MD Anderson Cancer Center, MDACC）接受放射治疗的头颈部肿瘤患者中的应用性能。
材料与方法：本研究纳入2015年至2022年间，来自一项前瞻性登记研究的407例口咽癌或原发灶不明头颈部肿瘤患者，所有患者均接受根治性放射治疗，联合或不联合全身治疗。本研究针对放疗后6、12个月的吞咽困难、误吸、口干症、唾液黏稠、味觉减退、言语障碍、口腔疼痛及疲劳共8种不良反应对应的NTCP模型进行了性能评估。所有不良反应均由患者通过MD安德森症状量表-头颈部版（MDASI-HN）进行自评，仅吞咽困难由临床医师通过头颈部患者进食状态量表-饮食正常评分（PSS-HN）进行报告。模型性能通过区分度（受试者工作特征曲线下面积（Area Under the Curve, AUC））、校准度及解释方差（决定系数R²）进行评价。
结果：CITOR模型在MDACC队列中展现出中至高的预测性能（平均AUC=0.67，对预测风险存在轻微高估，R²范围为0.02~0.28）。其中，针对吞咽困难、口干症、唾液黏稠及疲劳的NTCP模型表现最优。针对误吸、味觉减退与言语障碍的模型表现中等，这一结果部分可由较低的不良反应发生率加以解释。
结论：尽管CITOR模型的开发队列与MDACC验证队列存在差异，包括使用了不同的不良反应评估工具，但大多数模型仍展现出中至高的性能，证明模型所描述的剂量-效应关系具有泛化性。因此，本研究支持将此类NTCP模型进一步整合至临床实践中，尤其针对预测吞咽困难、口干症、唾液黏稠及疲劳的NTCP模型。