Nephrol Dial Transplant. 2011 Sep;26(9):3013-8. doi: 10.1093/ndt/gfq788. Epub 2011 Feb 3.
Renal allograft half-life time (t½) is the most straightforward representation of long-term graft survival. Since some statistical models overestimate this parameter, we compare different approaches to evaluate t½.
Patients with a 1-year functioning graft transplanted in Spain during 1990, 1994, 1998 and 2002 were included. Exponential, Weibull, gamma, lognormal and log-logistic models censoring the last year of follow-up were evaluated. The goodness of fit of these models was evaluated according to the Cox-Snell residuals and the Akaike's information criterion (AIC) was employed to compare these models.
We included 4842 patients. Real t½ in 1990 was 14.2 years. Median t½ (95% confidence interval) in 1990 and 2002 was 15.8 (14.2-17.5) versus 52.6 (35.6-69.5) according to the exponential model (P < 0.001). No differences between 1990 and 2002 were observed when t½ was estimated with the other models. In 1990 and 2002, t½ was 14.0 (13.1-15.0) versus 18.0 (13.7-22.4) according to Weibull, 15.5 (13.9-17.1) versus 19.1 (15.6-22.6) according to gamma, 14.4 (13.3-15.6) versus 18.3 (14.2-22.3) according to the log-logistic and 15.2 (13.8-16.6) versus 18.8 (15.3-22.3) according to the lognormal models. The AIC confirmed that the exponential model had the lowest goodness of fit, while the other models yielded a similar result.
The exponential model overestimates t½, especially in cohorts of patients with a short follow-up, while any of the other studied models allow a better estimation even in cohorts with short follow-up.
肾移植半寿期（t½）是长期移植物存活的最直接表现。由于某些统计模型高估了该参数，因此我们比较了评估 t½ 的不同方法。
纳入了 1990 年、1994 年、1998 年和 2002 年在西班牙接受 1 年功能正常的移植物移植的患者。评估了截止最后 1 年随访的指数、Weibull、伽马、对数正态和对数逻辑模型。根据 Cox-Snell 残差评估这些模型的拟合优度，并采用赤池信息量准则（AIC）比较这些模型。
我们纳入了 4842 名患者。1990 年的真实 t½为 14.2 年。1990 年和 2002 年指数模型的中位 t½（95%置信区间）分别为 15.8（14.2-17.5）和 52.6（35.6-69.5）（P<0.001）。1990 年和 2002 年，Weibull 模型分别为 14.0（13.1-15.0）和 18.0（13.7-22.4），伽马模型分别为 15.5（13.9-17.1）和 19.1（15.6-22.6），对数逻辑模型分别为 14.4（13.3-15.6）和 18.3（14.2-22.3），对数正态模型分别为 15.2（13.8-16.6）和 18.8（15.3-22.3），但其他模型没有观察到差异。AIC 证实指数模型的拟合优度最低，而其他模型的拟合结果相似。
指数模型高估了 t½，特别是在随访时间较短的患者队列中，而任何其他研究模型即使在随访时间较短的队列中也能得到更好的估计。
