We read the article by Udomkarnjananun et al and appreciate the author’s effort to establish the first Asian kidney transplant (KT) prediction model for long-term outcomes.1 As the authors mentioned in the article, most of the previous prediction models were derived from Caucasian cohorts. Considering the differences in ethnicity, insurance, and pharmacokinetics, this prediction model seems to be useful for predicting outcomes in Asian KT recipients. However, this model might not be fully applicable in current Asian recipients and for long-term outcomes because of two reasons. First, a prediction model is evaluated in terms of the abilities of discrimination and calibration. Regarding calibration, this model derived in Thailand would work well in other Asian countries. However, it might be difficult to discriminate the probability of outcome in each individual since most Asian countries, except China and Thailand, perform predominantly living-donor KTs (LDKTs) (Figure 1).2 Although this model includes donor type as a predictor, characteristics of LDKT in Asian countries might show a high level of collinearity except the recipient’s age, cause of end-stage kidney disease, and comorbidity. For instance, the proportion of positive hepatitis C virus antibodies in donors is <2%, and the use of tacrolimus as the maintenance immunosuppressive medication is over 85% in Japan. Furthermore, donor’s serum creatinine levels have very narrow SD (0.8 ± 0.1 mg/dL), and most patients have extremely low levels of panel-reactive antibody (Annual Progress Report from the Japanese Renal Transplant Registry. In Japanese). Thus, it would be difficult to discriminate the probability of outcomes in some Asian countries where LDKTs predominate DDKTs.FIGURE 1.: Proportion of donor types among Asian countries in 2019. This figure is referred to as Ref. 2.Second, prediction models in KT mainly have three purposes (decision making before transplantation, and short-term and long-term prognosis). If the decision making for the patients and their families whether KT is better than dialysis therapy in terms of prognosis is the main concern, only available information at the time of KT can be incorporated into the prediction model as in “iChoose Kidney.”3 The other US prediction score by Molnar et al4 (http://transplantscore.com/), which was referred as a comparison in the article by Udomkarnjananun et al, was created not only for shared decision-making purposes but also for short-term prediction.4 Therefore, it selected only pre-KT information. On the other hand, if the prediction of long-term outcomes (5 y and above) is the priority as in the model by Udomkarnjananun et al, the addition of relevant post-KT information as well as pre-KT variables such as allograft biopsy findings and rejection events would much improve the predictive accuracy. In fact, a recent prediction model for kidney allograft loss (long-term outcome) by Loupy et al,5 which consists of only post-KT information (such as kidney function, proteinuria, histologic findings, and human leukocyte antigen antibody profiling), has 0.8 of C-statistics for a 7-year prediction, which is regarded as a great discrimination ability. Albeit utility of using only post-KT information for predicting long-term outcomes, short-term prognosis (early death or graft loss) could not be taken into account, which corresponds to survivor bias. Thus, to develop the most appropriate and applicable prediction model in Asia, we suggest the model would take the type of donor and post-KT information into account, which would make the prediction ability much better.
Read full abstract
Aug 11, 2021
Karsten Suhre + 6
Open Access
