Centre de Recherche en Transplantation et immunologue, UMR 1064, INSERM, Université de Nantes, Nantes, France.
Institut de Transplantation Urologie Nephrologie, CHU Nantes, Nantes, France.
Eur J Epidemiol. 2018 Mar;33(3):275-286. doi: 10.1007/s10654-017-0322-3. Epub 2017 Oct 30.
When a patient is registered on renal transplant waiting list, she/he expects a clear information on the likelihood of being transplanted. Nevertheless, this event is in competition with death and usual models for competing events are difficult to interpret for non-specialists. We used a horizontal mixture model. Data were extracted from two French dialysis and transplantation registries. The "Ile-de-France" region was used for external validation. The other patients were randomly divided for training and internal validation. Seven variables were associated with decreased long-term probability of transplantation: age over 40 years, comorbidities (diabetes, cardiovascular disease, malignancy), dialysis longer than 1 year before registration and blood groups O or B. We additionally demonstrated longer mean time-to-transplantation for recipients under the age of 50, overweight recipients, recipients with blood group O or B and with pre-transplantation anti-HLA class I or II immunization. Our model can be used to predict the long-term probability of transplantation and the time in dialysis among transplanted patients, two easily interpretable parts. Discriminative capacities were validated on both the internal and external (AUC at 5 years = 0.72, 95% CI from 0.68 to 0.76) validation samples. However, calibration issues were highlighted and illustrated the importance of complete re-estimation of the model for other countries. We illustrated the ease of interpretation of horizontal modelling, which constitutes an alternative to sub-hazard or cause-specific approaches. Nevertheless, it would be useful to test this in practice, for instance by questioning both the physicians and the patients. We believe that this model should also be used in other chronic diseases, for both etiologic and prognostic studies.
当患者在肾移植等待名单上登记时，他/她期望获得有关移植可能性的明确信息。然而，这一事件与死亡竞争，通常用于竞争事件的模型对于非专业人士来说难以解释。我们使用了水平混合模型。数据从两个法国透析和移植登记处提取。“法兰西岛”地区用于外部验证。其他患者被随机分为训练和内部验证。七个变量与长期移植概率降低相关：年龄超过 40 岁、合并症（糖尿病、心血管疾病、恶性肿瘤）、登记前透析时间超过 1 年以及血型为 O 或 B。我们还证明了年龄在 50 岁以下的受者、超重受者、血型为 O 或 B 的受者以及移植前存在 HLA Ⅰ类或Ⅱ类免疫的受者的平均移植时间更长。我们的模型可用于预测移植的长期概率和移植患者的透析时间，这两个部分易于解释。在内部和外部验证样本（5 年 AUC = 0.72，95%CI 为 0.68 至 0.76）中均验证了区分能力。然而，校准问题突出表明，为其他国家重新估计模型的重要性。我们说明了水平建模易于解释，这是替代亚风险或特定原因方法的一种选择。然而，在实践中测试这一点将是有用的，例如通过询问医生和患者。我们认为该模型也应在其他慢性疾病中用于病因学和预后研究。
