Data_Sheet_1_A machine-learning approach to estimating public intentions to become a living kidney donor in England: Evidence from repeated cross-sectional survey data.docx

BackgroundLiving kidney organ donors offer a cost-effective alternative to deceased organ donation. They enable patients with life-threatening conditions to receive grafts that would otherwise not be available, thereby creating space for other patients waiting for organs and contributing to reducing overall waiting times for organs. There is an emerging consensus that an increase in living donation could contribute even more than deceased donation to reducing inequalities in organ donation between different population sub-groups in England. Increasing living donation is thus a priority for National Health Service Blood and Transplant (NHSBT) in the United Kingdom.MethodsUsing the random forest model, a machine learning (ML) approach, this study analyzed eight waves of repeated cross-sectional survey data collected from 2017 to 2021 (n = 14,278) as part of the organ donation attitudinal tracker survey commissioned by NHSBT in England to identify and help predict key factors that inform public intentions to become living donors.ResultsOverall, around 58.8% of the population would consider donating their kidney to a family member (50.5%), a friend (28%) or an unknown person (13.2%). The ML algorithm identified important factors that influence intentions to become a living kidney donor. They include, in reducing order of importance, support for organ donation, awareness of organ donation publicity campaigns, gender, age, occupation, religion, number of children in the household, and ethnic origin. Support for organ donation, awareness of public campaigns, and being younger were all positively associated with predicted propensity for living donation. The variable importance scores show that ethnic origin and religion were less important than the other variables in predicting living donor intention.ConclusionFactors influencing intentions to become a living donor are complex and highly individual in nature. Machine learning methods that allow for complex interactions between characteristics can be helpful in explaining these decisions. This work has identified important factors and subgroups that have higher propensity for living donation. Interventions should target both potential live donors and recipients. Research is needed to explore the extent to which these preferences are malleable to better understand what works and in which contexts to increase live organ donation.

背景：活体肾脏器官捐献者提供了相较于遗体器官捐献的经济高效替代方案。他们使面临生命威胁的患者得以获得原本无法获取的移植器官，从而为等待器官的其他患者腾出空间，并有助于缩短整体器官等待时间。一种共识正在逐渐形成，即活体捐献的增加可能比遗体捐献更有效地减少英格兰不同人口子群体之间在器官捐献方面的不平等。因此，提高活体捐献率成为英国国家医疗服务体系血液与移植（NHSBT）的首要任务。
方法：本研究采用随机森林模型这一机器学习（ML）方法，分析了从2017年至2021年收集的八轮重复横断面调查数据（n = 14,278），这些数据是作为NHSBT在英国委托进行的器官捐献态度追踪调查的一部分，旨在识别并预测影响公众成为活体捐献者意愿的关键因素。
结果：总体而言，大约58.8%的公众会考虑将肾脏捐给家庭成员（50.5%）、朋友（28%）或陌生人（13.2%）。机器学习算法确定了影响成为活体肾脏捐献者意愿的重要因素，按重要性递减的顺序排列，包括对器官捐献的支持、对器官捐献宣传活动的了解、性别、年龄、职业、宗教信仰、家庭中儿童的数量以及种族起源。对器官捐献的支持、对公众活动的了解以及年龄较轻均与预测的活体捐献倾向呈正相关。变量重要性得分表明，种族起源和宗教信仰在预测活体捐献意愿方面的重要性低于其他变量。
结论：影响成为活体捐献者意愿的因素复杂且极具个性化特征。允许特征之间进行复杂交互的机器学习方法有助于解释这些决策。本研究已确定了重要因素和倾向性较高的亚组。干预措施应针对潜在捐献者和接受者。需要进一步研究以探究这些偏好可调节的程度，从而更好地理解何种策略在何种情境下能有效地增加活体器官捐献。