各种辅助治疗对低级别胶质瘤（LGG）的有效性仍存在不确定性。预测个体治疗效果 (ITE) 并提供治疗建议的机器学习 (ML) 模型可以帮助根据每位患者的需求定制治疗方案。我们试图使用 ML 模型来辨别 LGG 患者放疗 (RT) 或放化疗 (CRT) 的个体适合性对 10 个 ML 模型进行了评估，这些模型经过训练可以推断 4,042 名 LGG 患者的 ITE。我们将遵循模型提供的治疗建议的患者与未遵循模型提供的治疗建议的患者进行了比较。为了降低治疗选择偏差的风险，我们采用了逆概率治疗加权（IPTW）。平衡生存套索网络（BSL）模型在我们测试的所有模型中显示出最显着的保护效果（风险比（HR）：0.52， 95% CI，0.41-0.64；IPTW 调整后 HR：0.58，95% CI，0.45-0.74；限制平均生存时间 (DRMST) 差异：9.11，95% CI，6.19-12.03；IPTW 调整后 DRMST：9.17 ，95% CI，6.30-11.83）。 CRT 在“推荐 CRT”组中表现出保护作用（IPTW 调整后 HR：0.60，95% CI，0.39-0.93），但在“推荐 RT”组中表现出不良作用（IPTW 调整后 HR：1.64， 95% CI，1.19-2.25）。此外，模型预测年轻患者以及病变重叠或肿瘤跨越中线的患者更适合 CRT（HR：0.62，95% CI，0.42-0.91；IPTW 调整后的 HR：0.59，95% CI，0.36-0.97 ）。我们的研究结果强调了 BSL 模型在指导 LGG 患者辅助治疗选择方面的潜力，有可能延长生存时间。这项研究强调了机器学习在定制患者护理、了解治疗选择的细微差别以及推进个性化医疗方面的重要性。版权所有：© 2024 Zhu 等人。这是一篇根据知识共享署名许可条款分发的开放获取文章，允许在任何媒体上不受限制地使用、分发和复制，前提是注明原始作者和来源。

There is ongoing uncertainty about the effectiveness of various adjuvant treatments for low-grade gliomas (LGGs). Machine learning (ML) models that predict individual treatment effects (ITE) and provide treatment recommendations could help tailor treatments to each patient's needs.We sought to discern the individual suitability of radiotherapy (RT) or chemoradiotherapy (CRT) in LGG patients using ML models.Ten ML models, trained to infer ITE in 4,042 LGG patients, were assessed. We compared patients who followed treatment recommendations provided by the models with those who did not. To mitigate the risk of treatment selection bias, we employed inverse probability treatment weighting (IPTW).The Balanced Survival Lasso-Network (BSL) model showed the most significant protective effect among all the models we tested (hazard ratio (HR): 0.52, 95% CI, 0.41-0.64; IPTW-adjusted HR: 0.58, 95% CI, 0.45-0.74; the difference in restricted mean survival time (DRMST): 9.11, 95% CI, 6.19-12.03; IPTW-adjusted DRMST: 9.17, 95% CI, 6.30-11.83). CRT presented a protective effect in the 'recommend for CRT' group (IPTW-adjusted HR: 0.60, 95% CI, 0.39-0.93) yet presented an adverse effect in the 'recommend for RT' group (IPTW-adjusted HR: 1.64, 95% CI, 1.19-2.25). Moreover, the models predict that younger patients and patients with overlapping lesions or tumors crossing the midline are better suited for CRT (HR: 0.62, 95% CI, 0.42-0.91; IPTW-adjusted HR: 0.59, 95% CI, 0.36-0.97).Our findings underscore the potential of the BSL model in guiding the choice of adjuvant treatment for LGGs patients, potentially improving survival time. This study emphasizes the importance of ML in customizing patient care, understanding the nuances of treatment selection, and advancing personalized medicine.Copyright: © 2024 Zhu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.