PO.CL11.02 · 临床研究
以机器学习增强个体化干预：一种更好的方法？
Enhancing individualized interventions with machine learning: A better approach?
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
背景：亚裔美国乳腺癌幸存者面临额外的文化、语言和就医障碍，妨碍了最佳的疼痛自我管理和及时的心理健康护理。这些挑战凸显了对文化响应性、可扩展干预措施的需求。在初步的癌症疼痛管理项目（CAPA）工作基础上，我们开发了癌症疼痛管理：基于技术的干预项目（CAI），它在 CAPA 基础上增加了针对报告抑郁症状幸存者的抑郁聚焦组件，以及用于个体化支持的机器学习功能。
方法：作为一项正在进行的随机对照试验的一部分，106 名有乳腺癌病史的亚裔美国女性被随机分配：58 名分入干预组，48 名分入活性对照组。干预组使用 CAI，活性对照组使用 CAPA。CAI 和 CAPA 均为文化适配的、多组件的、基于网络的干预，结构完全相同，唯 CAI 包含抑郁聚焦内容和机器学习驱动的个性化。主要结局包括疼痛（癌症疼痛管理 [CPM]、简明疼痛量表简表 [BPI-SF]）、症状负荷（纪念症状评估量表简表：MSAS-SF）、抑郁（流行病学研究中心抑郁量表：CES-D）以及生活质量（癌症治疗功能评估量表-乳腺癌：FACT-B）。评估在基线（T0）、1 个月（T1）和 3 个月（T2）时进行。混合效应生长模型检验了组别、时间及交互效应。
结果：在基线时，各组均衡良好；在社会人口学变量、乳腺癌相关特征或主要结局指标方面均未观察到显著差异（所有 p > 0.05）。随时间推移，疼痛（BPI-SF，p < 0.001）、抑郁（CES-D，p < 0.001）和生活质量（FACT-B，p < 0.001）均观察到显著改善。然而，未出现显著的组别或组别×时间交互效应（所有 p > 0.05），表明尽管 CAI 具有机器学习组件，其表现并未优于 CAPA。
结论：两种干预均随时间改善了结局；然而，纳入机器学习驱动个体化的 CAI 显示出比 CAPA 更大的改善，尽管这些差异无统计学意义。这些发现凸显了进一步研究以评估和优化使用机器学习的个性化策略的必要性。
查看英文原文 English abstract
Background: Asian American breast cancer survivors face additional cultural, linguistic, and access barriers that impede optimal pain self-management and timely mental-health care. These challenges underscore the need for culturally responsive, scalable interventions. Building on preliminary Cancer Pain Management Program (CAPA) work, we developed the Cancer Pain Management: A Technology-Based Intervention Program (CAI) that augments CAPA with depression-focused components for survivors reporting depressive symptoms and a machine learning feature for individualized support.
Methods: As part of an ongoing randomized controlled trial, 106 Asian American women with a history of breast cancer were randomized: 58 to the intervention group and 48 to the active control group. The intervention group used the CAI and the active control group used the CAPA. CAI and CAPA were culturally tailored, multi-component, web-based interventions identical in structure, except CAI included depression-focused content and machine learning-driven personalization. Primary outcomes included pain (Cancer Pain Management [CPM], Brief Pain Inventory-short form [BPI-SF]), symptom burden (Memorial Symptom Assessment Scale-Short Form: MSAS-SF), depression (Center for Epidemiologic Studies Depression Scale: CES-D), and quality of life (Functional Assessment of Cancer Therapy Scale-Breast Cancer: FACT-B). Assessments occurred at baseline (T0), 1 month (T1), and 3 months (T2). Mixed-effects growth models tested group, time, and interaction effects.
Results: At baseline, groups were well balanced; no significant differences were observed in sociodemographic variables, breast cancer-related characteristics, or primary outcome measures (all p > 0.05). Significant improvements over time were observed for pain (BPI-SF, p < 0.001), depression (CES-D, p < 0.001), and quality of life (FACT-B, p < 0.001). However, no significant group or group-by-time interaction effects emerged (all p > 0.05), indicating that CAI did not outperform CAPA despite its machine learning component.
Conclusion: Both interventions improved outcomes over time; however, CAI, which incorporated machine learning-driven individualization, showed greater improvements than CAPA, although these differences were not statistically significant. These findings highlight the need for further research to evaluate and optimize personalization strategies using machine learning.
利益披露 Disclosure
W. Chee, None..
J. Baek, None..
D. Kim, None..
S. Ryu, None..
Y. Kim, None..
E. Im, None.