Kidney transplantation is the treatment of choice for patients with end-stage renal disease (ESRD).1 One of the main challenges in transplant medicine is prioritizing the best recipients for a kidney transplant according to criteria which would maximize both patient and kidney allograft survival. Previous studies have identified risk factors of patient mortality and graft failure in kidney transplant recipients, including donor kidney status (living vs deceased), age, and race, as well as recipient age, smoking status, race ethnicity, malnutrition inflammation score, comorbidities, acute rejection, delayed graft function, circulating angiopoietin, sleep apnea, and posttransplant proteinuria.2-16 However, a number of these risk factors are measured in the posttransplant period. Studies done by our group have previously identified several pretransplantation risk factors, such as lower muscle mass and serum albumin level, higher body mass index, and alkaline-phosphotase level, hemodialysis versus peritoneal dialysis modality, poor glycemic control, and higher erythropoietin stimulating agent responsiveness index associated with higher risk of adverse outcomes posttransplant such as delayed graft function, allograft loss or death.15,17-25 Physicians often have to urgently select a proper candidate for a kidney transplant using available data. Tools which can inform physicians in the decision-making process by predicting the recipient's chance of overall and allograft survival are needed.
Several prediction scores and calculations have been developed in the last decades to assist physicians.26-41 However, all of these scores are partially based on data obtained after kidney transplantation,30,38-41 or used data from the last century,29,34,38 when the practice and transplant outcomes were different or used incorrect methodology.26 Moreover, most of these studies had defined death-censored allograft failure as the primary outcome of interest and only some of them have focused on the outcome of patient/recipient survival.38-40 To this end, the currently used estimated posttransplant survival (EPTS) score in the United States allocation system was created and is implemented to predict recipients' survival.36,37
To our knowledge, no prediction score has been developed to predict both allograft loss and transplant recipient death based only on data available at the time of transplantation in the 21st century. The purpose of the present study was to develop and robustly validate scores predictive of death-censored allograft failure and recipients' death up to 5 years posttransplantation based on variables which are available at the time of transplantation for kidney transplant recipients across the United States in the 21st century.
MATERIALS AND METHODS
Data Source and Cohort Definition
We linked data of all kidney transplant recipients listed in the Scientific Registry of Transplant Recipients (SRTR) to a list of individuals with ESRD who underwent maintenance hemodialysis treatment from July 2001 to June 2006 in one of the outpatient dialysis facilities of a large dialysis organization (DaVita Inc, before its acquisition of former Gambro dialysis facilities). The study was approved by the Institutional Review Committees of Los Angeles Biomedical Research Institute at Harbor-UCLA, University of California Irvine Medical Center, University of Washington, University of Tennessee Health Science Center, and DaVita Clinical Research. The clinical and research activities being reported are consistent with the Principles of the Declaration of Istanbul as outlined in the 'Declaration of Istanbul on Organ Trafficking and Transplant Tourism.
Clinical and Demographic Measures
The creation of the national DaVita hemodialysis patient cohort has been described previously.42-46 Demographic data and details of medical history were collected, with information on age, sex, race, type of insurance, marital status, presence of diabetes, height, posthemodialysis dry weight (to calculate averaged body mass index), and dialysis vintage. Dialysis vintage was defined as the duration of time between the first day of dialysis treatment and the day of kidney transplantation. Preexisting comorbid conditions, such as coronary artery disease (CAD) and peripheral vascular disease (PAD), were obtained by linking the DaVita database to the Medical Evidence Form 2728 of the United States Renal Data System.47 The transplantation related data, such as donor characteristics, recipients' viral serology, cold ischemic time, and HLA mismatches, were collected from SRTR.
Blood samples were drawn using uniform techniques in all of the DaVita dialysis clinics and were transported to a central laboratory in Deland, FL, typically within 24 hours. All laboratory values were measured by automated and standardized methods. Most laboratory values were measured monthly, including serum urea nitrogen, creatinine, albumin, phosphorus, and alkaline phosphatase. Hemoglobin was measured at least monthly in essentially all patients and weekly to biweekly in most patients. Most blood samples were collected predialysis with the exception of the postdialysis serum urea nitrogen to calculate urea kinetics. The pretransplantation laboratory data from the last quarter before transplantation were used in our calculations.
Characteristics of the study cohort, including all predictors, are summarized as means ± SD or proportions for continuous and categorical variables, respectively. Prediction models were developed for 3 outcomes: (a) mortality, (b) allograft loss (death censored), and (c) a combined outcome of death or allograft loss (a or b), using Cox proportional hazards models. Study follow-up was censored at the end of the study (October 29, 2007). The cohort (N = 15 125) (Figure 1) was divided into a two thirds training/development set (Nd = 10 083) and a one third test/validation set (Nv = 5042). We used multiple imputation (10 imputations) for continuous missing values (27-28%) in the development dataset of recipients' albumin, alkaline phosphatase, hemoglobin, and phosphorus. Missing values (20%) in organ preservation total cold ischemic time were also imputed. Candidate predictors were based on clinical considerations and those used in previous studies.15,19-25 Final models with reduced number of predictors were obtained using backward-selection based on Akaike's information criterion because it has better statistical properties in variable selection compared with P value–based selection,48 and it avoids arbitrary and ineffective selection rules based on P values. To address potential model overfitting (optimism) and also for model calibration, we estimated a linear shrinkage factor (γ) using the bootstrap method applied to the development dataset. Briefly, for each of the 100 bootstrap data sets, the exact development steps described above (Cox regression with Akaike's information criterion backward selection) were fitted. Then, the outcome was regressed on the prognostic score (PS) or linear predictor (Xβ) in a univariate Cox regression. The linear predictor was calculated using the fitted bootstrap coefficients (β) for each patient in the original development data set. The process was repeated to obtain 3 shrinkage factors corresponding to the 3 outcomes. The shrinkage factor γ was used to adjust the final Cox prediction models to correct for model overoptimism as further detailed below.48-51 Furthermore, model calibration was assessed by a group-based goodness-of-fit (GOF) test developed for survival model52 for each prediction model. Briefly, the population was divided into deciles (groups) of the risk score and the group-based GOF test provides an overall assessment of model calibration as well as for each group. Calibration plot for 5-year survival was also examined for each model.
Model prediction was assessed using internal validation on the one-third validation dataset. Model predictive discrimination was assessed using the index of concordance, or C statistic, which accounts for censoring in time-to-event models (for death, allograft loss, or combined event) and is equivalent to the area under the receiver operating characteristic curve for binary outcomes (logistic regression).50,53 Estimate of C and its 95% confidence interval based on the validation data are provided for the 3 outcomes. The final prediction models for each of the 3 outcomes based on the shrunken PS can be used to estimate the predicted probabilities of death, allograft loss or combined event at a given time t (year). That is, the shrunken PS, say PS*, that will be used to predict the outcomes of new/future patients will be PS* = γ Xβ, where β is collection of estimated coefficients in the final prediction model. The predicted survival at time t for new a patient can be obtained as S0(t)exp(PS*), where PS* is the aforementioned calibrated/shrunken PS and S0(t) is the baseline survival estimate from the final model. Analyses were performed in SAS version 9.3 PROC PHREG and R version 2.12 using libraries RMS and SURVIVAL.
Baseline Characteristics of the Cohort and Patients' Outcome
Baseline characteristics of the cohort are shown in Table 1. Briefly, the mean ± SD age was 50 ± 13 years (range, 18-86 years), 61% were men, 36% persons with diabetes, 48%, 28%, and 15% were white, Hispanic, and African American, respectively; and the mean ± SD time on dialysis was 3.6 ± 3.1 years. Median follow-up time was 794 days (interquartile range, 384-1348 days) for combined outcome. There were 1492 deaths (9.9%; mortality rate: 35.9; 95% confidence interval (CI), 34.1-37.7/1000 patient-years), and 1647 graft losses (10.9%; graft loss rate, 41.1; 95% CI, 39.2-43.1/1000 patient-years) during the follow-up period.
Development of the Prediction Score
We developed 2 prediction scores, 1 including donor variables (main score) and the other with only recipients' variables (score for dialysis patients). From the 19 166 transplant events in the SRTR database identified among the study cohort, we excluded transplants which were not the recipients' first transplant and patients with age younger than 18 years or who received the first kidney transplant before July 1, 2001 (Figure 1). The final prediction mortality model coefficients are presented in Table 2. Older recipient age, longer time on dialysis, presence of diabetes, CAD, PAD, and older donor age were associated with increased risk of mortality. The final prediction mortality without donor variables model coefficients are presented in Table S1,http://links.lww.com/TP/B307. The final prediction model coefficients for graft loss are presented in Table 3 and for the combined outcome are presented in Table 4. Younger recipient age, Hispanic ethnicity, hypertension, and glomerulonephritis as cause of ESRD, shorter time on dialysis, recipient's and donor's diabetes, extended criterion donor kidney, and number of HLA mismatch were associated with increased risk of death censored allograft loss (Table 3). Similar risk factors were associated with increased risk of the combined outcome (death or allograft loss), as shown in Table 4. The final prediction mortality without donor variables model coefficients for graft loss are presented in Table S2,http://links.lww.com/TP/B307 and for the combined outcome are presented in Table S3,http://links.lww.com/TP/B307. For comparission, hazard ratios using coefficients from the EPTS score prediction model are presented in Table S4,http://links.lww.com/TP/B307 for all outcomes. The performances of our reduced/simplified models were practically the same as the full models (not shown). In addition, we have performed prediction models after leaving out the variables with missing values. The C statistics from models without these laboratory values are similar, as quantified by the C statistics (not shown). Finally, we have also performed prediction models using multiple imputation for missing values. The C statistics from these models are similar, as quantified by the C statistics (not shown).
Internal Validation and Comparison With Other Prediction Scores
Performance of the prediction score was tested in the validation data set of 5042 patients. Our prediction score for mortality discriminated acceptably, with a C statistic of 0.70 (95% CI, 0.67-0.73) for the main model and 0.70 (95% CI, 0.67-0.72) for the model without donor variables (Table 5). The ability of our new score to discriminate mortality outcomes was better than the EPTS score and the score from Kasiske et al41 and similar to the Cox model based on variables from iChoose Kidney model26 (Table 5). Our main prediction score for allograft loss and for the combined outcome had a C statistic of 0.63 (95% CI, 0.60-0.66) for allograft loss and 0.63 (95% CI, 0.61-0.66) for combined outcome (Table 5). The discrimination ability for these 2 outcomes using our new score was similar or slightly better than the EPTS score and iChoose Kidney model26 and similar to the score from Kasiske et al41 (Table 5). Figure 2 shows the predicted probability of (panel A) mortality, (panel B) graft failure, and (panel C) combined outcome within 5 years of transplant as a function of risk score. The predicted probabilities at 25th, 50th, and 75th percentile of the risk level for mortality are 8.3%, 13.8%, and 22.1%, respectively; for graft failure, 9.6% 13.8%, and 19.4%; for combined outcome, 19.2%, 25.2%, and 33.1%. Model calibration was assessed using the slopes of the prognostic index; slopes of 1.0 represent perfect calibration. Table S5,http://links.lww.com/TP/B307 provides calibration statistics for the group-based GOF tests with the observed number of events and expected/predicted events from each model. There was good overall calibration for the main models for mortality, graft failure, and combined outcome (all P > 0.05). However, for graft failure and combined outcome, the fit was poor for higher deciles of the risk score. Not surprisingly, for models without donor variables, the overall GOF was not as good and similarly poorer fit prediction was observed for several of the higher deciles of the risk score. Calibration plots for 5-year survival (observed vs predicted survival) are provided in Figure S1 (SDC,http://links.lww.com/TP/B307), which shows graphically similar results as the group-based GOF tests.
Using scores from our main model, we present the estimated 1 to 5 years predicted outcome event failure probabilities for several distinct, typical patient characteristics in Table 6. In addition, Table 7 compares the 1- to 5-year event probabilities in 4 typical patients for our current main model score with the scores of the EPTS model and the model from Kasiske et al.41 Results from this table show that, using our score, estimated event probabilities are quite different when patients have many comorbidities (eg, comparing 1A with 2A or 1B with 2B). For example, the predicted 5-year event probabilities for a patient with no comorbidities (1B) compared a patient with all comorbidities present are 21% and 67%, respectively; a greater than 3-fold increase in event failure risk for patients with comorbidities present. The EPTS model, however, does not make this distinction because comorbidities (except for diabetes mellitus) are not included in the model. Similarly, the model from Kasiske et al41 includes a limited number of patient comorbidities in prediction scores and the estimated probabilities therefore do not differ much according to the presence of various comorbid conditions (eg, comparing patient 1A with 2A or 1B with 2B).
The prediction of long-term outcomes in kidney transplantation is a very important issue for a limited resource, not only for managing clinical decisions but also for adequate risk assessment. Predicting which candidate is most eligible and expected to have the greatest longevity for offered allograft kidney can be an extremely helpful tool for physicians making clinical decisions. In this article, we presented a simple clinical score, which includes only variables available at the time of transplantation. All data captured from patients transplanted in the 21st century. This simple clinical score has better or at least the same prediction capability as other currently used prediction scores in the United States despite only pretransplant variables were used.
The main goal for developing this model was to help physicians make decision. The variables included in the model are those that are available to clinicians in everyday clinical practice. These models could help to compare predicted outcomes under various real-life circumstances; for example, when a nephrologist evaluates waitlisted patients with no knowledge of donor-specific information, or when a surgeon needs an urgent determination about which of several potential recipients should receive a kidney once donor information becomes available. We believe the development and assessment of an objective prediction tool, based on systematic data collection and analysis, provides additional help to physicians. It goes without saying that the added value of a prediction tool is not intended to replace clinical judgment/knowledge, but rather to augment it.
Despite previous studies having recognized pretransplantation risk factors for kidney allograft loss or mortality,15,17-25 only few previous prediction score based solely on only these variables,29,32,34,35 and none of them has been developed in the 21st century and focused on both graft and patients' survival. Additional calculations have been performed to calculate life years from transplant,27,28 and scores have been developed for predicting coronary heart disease,30 graft function at 1 year33 or survival after discharge.31 Only few previous scores have been developed based on data from 21st century26,33,35,39-41; however, none of them focused on both graft and patients' survival. Moreover, only a few efforts have been made to describe risk scores for use as prognostic tools to individualize risk of allograft loss or mortality in incident or prevalent transplant recipients.26,38-41 For a prediction score to assume clinical utility, a number of conditions must be met. Of obvious importance, each component of the score should be statistically associated with the assessed clinical outcomes such as allograft loss and mortality. Nonetheless, exact quantification of an individual patient's risk of clinical events requires different statistical approaches from the approaches used to only examine the association between risk factor and event.54,55 For instance, the prediction score should discriminate satisfactorily between the individuals who are experiencing versus those who are not experiencing the clinical endpoints. The C statistic is an adequate method to assess this discrimination. Our new prediction score has acceptable C statistic, especially for the outcome of patient survival. Our score is also able to discriminate outcome risk across different waitlisted transplant candidates at the time of kidney transplantation (main score) or even before transplantation (score without donor variables). Even though our score includes only variables which are actually available at the time of transplantation, the C statistics of our prediction score was better or at least the same for all the studied endpoints (mortality, graft loss and combined of these) than the currently used EPTS score36,37 or from Kasiske et al41 or variables from iChoose Kidney model.26 Although it is important to note that EPTS allows for prediction of mortality in those with prior solid organ transplantation and the Kasiske et al model41 included patients with preemptive kidney transplants and regrafts. In addition, the iChoose Kidney score is based on logistic regression models, which did not take into account the time to event and did not censor for outcome events.26 In our comparison, we used the same variables used in the iChoose Kidney model, but applied a Cox regression model for comparison.26 Another significant advantage of our prediction score is its ability to account for comorbidities, kidney donor related information and different important pretransplant laboratory values, which are associated with posttransplant outcomes.20,21,24Table S6,http://links.lww.com/TP/B307 shows the variables included in several currently avaliable prediction score models in transplant nephrology including our own. Although our score includes the use of more variables than other scores, our score can still be rapidly and efficiently calculated using our website at www.TransplantScore.com. Moreover, as clearly shown in Table 7, taking into account these additional variables results in significant improvement in the ability to predict long-term outcomes in kidney transplant recipients; as the currently used EPTS score36,37 or the Kasiske et al41 score are not able to distinguish between patients with and without comorbidites. Furthermore, while most of the other prediction scores were created to predict allograft loss,38-40 our new prediction score was created to be able to predict not only allograft loss, but graft censored mortality as well.
Our new prediction score was designed to assist physicians in clinical decision making regarding kidney transplants even under urgent circumstances. In addition, we developed a prediction score without donor information, which can be helpful for physicians during the transplant evaluation as well. For prediction of posttransplant graft censored mortality and graft loss 10 predictors were used for each main model. These factors included: recipients' age, cause and length of ESRD, hemoglobin, albumin, selected comorbidities, race, and type of insurance as well as donor characteristics such as donor age, extended criterion donor, diabetes status, number of HLA mismatches (Table S6,http://links.lww.com/TP/B307). Based on the equations used to develop our new prediction score, we created a website at www.TransplantScore.com, where the predicted event probability for a patient can be calculated rapidly and efficiently. This webpage was designed to also be useful on mobile devices both online and off-line, which makes our score applicable even at the patient’s bedside.
Although our score has a marginal increase in the C-statistic over existing score, we note that unfortunately, most of the prediction scores used in transplant nephrology and in general nephrology have similarly low C-statistic. However, it is important to note that we used only pretransplant variables, whereas the rest of the scores used posttransplant variables (which makes the prediction easier). Furthermore, we point out that the C-statistic provides a single-number summary of overall prediction performance which clinical utility should not be solely based on. In addition to pretransplant variables, another important consideration in the fitness for clinical use is that the prediction model incorporates adequate key patient predictive factors able to structurally discriminate among patients' likelihood of death/graft failure event. For instance, a model with only baseline diabetes structurally cannot distinguish varying mortality probabilities for a patient with diabetes mellitus and CAD and PAD or abnormal laboratory results, for instance. A model with a higher C-statistic but which is limited structurally in making adequate individualized predictions may not be appropriate for some clinical applications.
Our study should be noted for several advantages other than those mentioned above. Our prediction score is the only novel score developed from data from the 21st century patients in the United States and focusing on both recipients' and graft survival while previous models such as the EPTS score,36,37 the Kasiske et al prediction score41 and the iChoose Kidney score26 used older data or focused on only one outcome. Moreover, we developed a score without donor data, which can help the dialysis physician to calculate the waitlisted patients expected posttransplant survival. In addition, our cohort size was much larger than the ones used in previous studies to develop prediction scores.38-40 Our score has been developed using data derived from several centers. Center-specific scores, based on data derived from any given center's data, could be more applicable for patients transplanted in the given center. Finally, we created a website www.TransplantScore.com to help physicians easily use our predictive model in everyday practice.
Our study should be qualified for several potential limitations. The prediction models are only as good as the data used in their derivation. In the development dataset in our analyses, data for continuous variables of recipients’ albumin, alkaline phosphatase, hemoglobin and phosphorus were missing in 27-28% of patients, and data on organ preservation total cold ischemic time were missing in 20% of patients. We used multiple imputation (10 imputations) methods to address missing data, although potential bias remains. Additionally, although most demographic variables likely are accurate for recipients and donors, there is always potential nondifferential misclassification bias contributing to type II error in our analyses. Comorbidity data in our study were obtained from the Centers for Medicare and Medicaid Services Medical Evidence Report (form CMS-2728), for which a previous validation study found that comorbid conditions were significantly underreported.47 In addition, we do not have data for important predictors such as midodrine administration.56 Most importantly, our prediction model has yet to be externally validated in other cohorts. To the best of our knowledge, only few previously developed prediction scores were externally validated,35,38,39,57 and only 2 these scores were validated in a different center.39,57 Moreover, neither of these externally validated scores were developed and validated in patients in the United States at the 21st century. Externally validating our score in other cohorts at both the multicenter and individual center level is necessary to ensure the applicability, reliability, and utility of our prediction model for use in potential kidney transplant recipient patients. Moreover, our score was developed using US data from one large dialysis provider; consequently, the applicability of our score for non-US patients and US patients from other dialysis providers might be limited. Further external validation is necessary. Finally, our score can be used only in recipients with first deceased kidney transplantation.
A newly developed prediction tool, which uses 21st century data exclusively available before the time of transplantation to predict patients' and graft survival performs better than currently used tools, such as EPTS. The predicted event risk varies sensibly according to patients' and donors' pretransplant characteristics as well as laboratory measurements and prediction scores accounting for these differences should be implemented.
The authors extend their thanks to the teammates in DaVita clinics who work every day to take care of patients and also to ensure the extensive data collection on which our work is based. The authors also thank DaVita Clinical Research (DCR) for technical assistance in that regard. DCR is committed to advancing the knowledge and practice of kidney care.
1. Wolfe RA, Ashby VB, Milford EL, et al. Comparison of mortality in all patients on dialysis, patients on dialysis awaiting transplantation, and recipients of a first cadaveric transplant. N Engl J Med
2. Ojo AO, Hanson JA, Meier-Kriesche H, et al. Survival in recipients of marginal cadaveric donor kidneys compared with other recipients and wait-listed transplant candidates. J Am Soc Nephrol
3. Troppmann C, Gillingham KJ, Benedetti E, et al. Delayed graft function, acute rejection, and outcome after cadaver renal transplantation. The multivariate analysis. Transplantation
4. Almond PS, Matas A, Gillingham K, et al. Risk factors for chronic rejection in renal allograft recipients. Transplantation
. 1993;55:752–756 discussion 756–757.
5. Tesi RJ, Elkhammas EA, Henry ML, et al. Acute rejection episodes: best predictor of long-term primary cadaveric renal transplant survival. Transplant Proc
. 1993;25(1 Pt 2):901–902.
6. Cosio FG, Hickson LJ, Griffin MD, et al. Patient survival and cardiovascular risk after kidney transplantation: the challenge of diabetes. Am J Transplant
7. Wu C, Evans I, Joseph R, et al. Comorbid conditions in kidney transplantation: association with graft and patient survival. J Am Soc Nephrol
8. Nogueira JM, Haririan A, Jacobs SC, et al. Cigarette smoking, kidney function, and mortality after live donor kidney transplant. Am J Kidney Dis
9. Molnar MZ, Langer RM, Remport A, et al. Roma ethnicity and clinical outcomes in kidney transplant recipients. Int Urol Nephrol
10. Molnar-Varga M, Molnar MZ, Szeifert L, et al. Health-related quality of life and clinical outcomes in kidney transplant recipients. Am J Kidney Dis
11. Molnar MZ, Streja E, Kovesdy CP, et al. Age and the associations of living donor and expanded criteria donor kidneys with kidney transplant outcomes. Am J Kidney Dis
12. Molnar MZ, Lazar AS, Lindner A, et al. Sleep apnea is associated with cardiovascular risk factors among kidney transplant patients. Clin J Am Soc Nephrol
13. Molnar MZ, Kumpers P, Kielstein JT, et al. Circulating Angiopoietin-2 levels predict mortality in kidney transplant recipients: a 4-year prospective case-cohort study. Transpl Int
14. Roodnat JI, Mulder PG, Rischen-Vos J, et al. Proteinuria after renal transplantation affects not only graft survival but also patient survival. Transplantation
15. Streja E, Molnar MZ, Kovesdy CP, et al. Associations of pretransplant weight and muscle mass with mortality in renal transplant recipients. Clin J Am Soc Nephrol
16. Molnar MZ, Czira ME, Rudas A, et al. Association of the malnutrition-inflammation score with clinical outcomes in kidney transplant recipients. Am J Kidney Dis
17. Molnar MZ, Bunnapradist S, Huang E, et al. Association of pre-transplant erythropoiesis-stimulating agent responsiveness with post-transplant outcomes. Nephrol Dial Transplant
18. Molnar MZ, Foster CE 3rd, Sim JJ, et al. Association of pre-transplant blood pressure with post-transplant outcomes. Clin Transplant
19. Molnar MZ, Huang E, Hoshino J, et al. Association of pretransplant glycemic control with posttransplant outcomes in diabetic kidney transplant recipients. Diabetes Care
20. Molnar MZ, Kovesdy CP, Bunnapradist S, et al. Donor race and outcomes in kidney transplant recipients. Clin Transplant
21. Molnar MZ, Kovesdy CP, Bunnapradist S, et al. Associations of pretransplant serum albumin with post-transplant outcomes in kidney transplant recipients. Am J Transplant
22. Molnar MZ, Kovesdy CP, Mucsi I, et al. Association of pre-kidney transplant markers of mineral and bone disorder with post-transplant outcomes. Clin J Am Soc Nephrol
23. Molnar MZ, Mehrotra R, Duong U, et al. Dialysis modality and outcomes in kidney transplant recipients. Clin J Am Soc Nephrol
24. Molnar MZ, Kovesdy CP, Rosivall L, et al. Associations of pre-transplant anemia management with post-transplant delayed graft function in kidney transplant recipients. Clin Transplant
25. Molnar MZ, Kovesdy CP, Mucsi I, et al. Higher recipient body mass index is associated with post-transplant delayed kidney graft function. Kidney Int
26. Patzer RE, Basu M, Larsen CP, et al. iChoose kidney: a clinical decision aid for kidney transplantation versus dialysis treatment. Transplantation
27. Wolfe RA, McCullough KP, Leichtman AB. Predictability of survival models for waiting list and transplant patients: calculating LYFT. Am J Transplant
28. Wolfe RA, McCullough KP, Schaubel DE, et al. Calculating life years from transplant (LYFT): methods for kidney and kidney-pancreas candidates. Am J Transplant
. 2008;8(4 Pt 2):997–1011.
29. Krikov S, Khan A, Baird BC, et al. Predicting kidney transplant survival using tree-based modeling. ASAIO J
30. Israni AK, Snyder JJ, Skeans MA, et al. Predicting coronary heart disease after kidney transplantation: Patient Outcomes in Renal Transplantation (PORT) Study. Am J Transplant
31. Hernández D, Rufino M, Bartolomei S, et al. A novel prognostic index for mortality in renal transplant recipients after hospitalization. Transplantation
32. Lin RS, Horn SD, Hurdle JF, et al. Single and multiple time-point prediction models in kidney transplant outcomes. J Biomed Inform
33. Tiong HY, Goldfarb DA, Kattan MW, et al. Nomograms for predicting graft function and survival in living donor kidney transplantation based on the UNOS Registry. J Urol
34. Goldfarb-Rumyantzev AS, Scandling JD, Pappas L, et al. Prediction of 3-yr cadaveric graft survival based on pre-transplant variables in a large national dataset. Clin Transplant
35. Brown TS, Elster EA, Stevens K, et al. Bayesian modeling of pretransplant variables accurately predicts kidney graft survival. Am J Nephrol
38. Moore J, He X, Shabir S, et al. Development and evaluation of a composite risk score to predict kidney transplant failure. Am J Kidney Dis
39. Shabir S, Halimi JM, Cherukuri A, et al. Predicting 5-year risk of kidney transplant failure: a prediction instrument using data available at 1 year posttransplantation. Am J Kidney Dis
40. Foucher Y, Daguin P, Akl A, et al. A clinical scoring system highly predictive of long-term kidney graft survival. Kidney Int
41. Kasiske BL, Israni AK, Snyder JJ, et al. A simple tool to predict outcomes after kidney transplant. Am J Kidney Dis
42. Molnar MZ, Lukowsky LR, Streja E, et al. Blood pressure and survival in long-term hemodialysis patients with and without polycystic kidney disease. J Hypertens
43. Miller JE, Kovesdy CP, Nissenson AR, et al. Association of hemodialysis treatment time and dose with mortality and the role of race and sex. Am J Kidney Dis
44. Miller JE, Kovesdy CP, Norris KC, et al. Association of cumulatively low or high serum calcium levels with mortality in long-term hemodialysis patients. Am J Nephrol
45. Kalantar-Zadeh K, Streja E, Kovesdy CP, et al. The obesity paradox and mortality associated with surrogates of body size and muscle mass in patients receiving hemodialysis. Mayo Clin Proc
46. Kalantar-Zadeh K, Miller JE, Kovesdy CP, et al. Impact of race on hyperparathyroidism, mineral disarrays, administered vitamin D mimetic, and survival in hemodialysis patients. J Bone Miner Res
47. Longenecker JC, Coresh J, Klag MJ, et al. Validation of comorbid conditions on the end-stage renal disease medical evidence report: the CHOICE study. Choices for Healthy Outcomes in Caring for ESRD. J Am Soc Nephrol
48. Harrell FE. Regression modeling strategies with applications to linear models, logistic regression, and survival analysis
. New York: Springer-Verlag, New York, Inc.; 2001.
49. Steyerberg EW, Eijkemans MJC, Habbema JDF. Application of shrinkage techniques in logistic regression analysis: a case study. Statistica Neerlandica
50. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med
51. Steyerberg EW, Eijkemans MJ, Harrell FE Jr, et al. Prognostic modelling with logistic regression analysis: a comparison of selection and estimation methods in small data sets. Stat Med
52. Crowson CS, Atkinson EJ, Therneau TM. Assessing calibration of prognostic risk scores. Stat Methods Med Res
53. Harrell FE Jr, Califf RM, Pryor DB, et al. Evaluating the yield of medical tests. JAMA
54. Pencina MJ, D'Agostino RB Sr, D'Agostino RB Jr, et al. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med
. 2008;27:157–172 207–112.
55. Cook NR. Statistical evaluation of prognostic versus diagnostic models: beyond the ROC curve. Clin Chem
56. Alhamad T, Brennan DC, Brifkani Z, et al. Pretransplant midodrine use: a newly identified risk marker for complications after kidney transplantation. Transplantation
57. Clayton PA, McDonald SP, Snyder JJ, et al. External validation of the estimated posttransplant survival score for allocation of deceased donor kidneys in the United States. Am J Transplant
Supplemental Digital Content
Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.