Over 90000 patients are currently on the waitlist to receive kidney transplant,1 but only about 20000 patients were transplanted in 2016 with deceased donation rate of kidneys remaining stable in the past 10 years,2 so the demand for kidneys is critical. Unfortunately, in the same year, the discard rate of kidneys recovered from deceased adult donors was about 20%, whereas >8 000 patients were removed from the waitlist due to deteriorating health or death.2 Therefore, current listing and transplantation strategies at transplant centers need to be reevaluated to decrease the number of unwanted removals and discard rate of deceased-donor kidneys.
These strategies could be improved by considering transplant candidates’ functional statuses. This information is readily available in the Scientific Registry of Transplant Recipients (SRTR) and Organ Procurement and Transplantation Network (OPTN) data and it is measured by the Karnofsky performance score (KPS), using an 11-point rating scale ranging from 0 (dead) to 100 (normal). Table 1 summarizes the scores. Developed as an assessment tool in oncology,3 KPS gained traction as an independent risk factor for mortality in the renal settings, such as for acute renal failure,4 end-stage renal disease (ESRD),5 and dialysis.6-8 Including KPS in the current risk-adjusted Cox regression models would improve their accuracies, motivating transplant centers to modify their current transplantation strategies.
Estimated from the risk-adjusted Cox models, the expected 1- and 3-year graft and patient survivals9,10 set the performance standards that transplant centers have to satisfy and thus influence their strategies.11 The expected survival statistics are provided to the Membership and Professional Standards Committee of the OPTN and then used to evaluate transplant center performances.9,12 If a transplant center failed to meet performance standards, it is flagged, thereby undergoing a review process by United Network of Organ Sharing and/or Centers for Medicare and Medicaid Services (CMS). Consequently, the center may face unintended consequences, such as decertification by CMS, decrease in transplant patient volume and candidate referrals, and loss of insurance contracts.11 Lack of improvement might shut it down. Therefore, an underperforming center needs to change its listing and transplantation strategies. Schold et al surveyed that low-performing centers were significantly more likely to increase recipient and donor selection criteria,13 thereby restricting access to kidney transplantation and significantly linking them with reduced transplant volumes.14 To mitigate these unintended consequences, risk-adjusted Cox models must be as accurate as possible to prevent transplant centers from being falsely flagged.
Because functional status is not included in any of the current kidney-related risk-adjustment models,15 the aim of this study is to analyze its predictive power measured in KPS on 1- and 3-year posttransplant survival for deceased-donor kidney transplant (DDKT) recipients. In addition, posttransplant survival outcomes are evaluated for patients transplanted with low-quality kidneys across different KPS strata. The improved Cox models could serve as insightful tools that drive efficacious kidney transplantation decisions.
MATERIALS AND METHODS
This study used data from SRTR. The SRTR data system includes data on all donors, waitlisted candidates, and transplant recipients in the US, submitted by the members of the OPTN. The Health Resources and Services Administration, U.S. Department of Health and Human Services provides oversight to the activities of the OPTN and SRTR contractors.
The data set was used to perform statistical analyses on DDKT recipients who were at least 18 years old at the time of transplant and were not subsequently retransplanted. The latest follow-up status date posttransplant in the given cohort was September 2, 2015. From this cohort, 2 data sets were created for 1- and 3-year survival analyses, respectively.
One data set consisted of 72 839 DDKT recipients transplanted from January 1, 2007, to September 2, 2014. In this data set, each patient had at least 1-year follow-up. Patients were right-censored if they were lost to follow-up within 1 year posttransplant or survived past it. This data set was used for the 1-year survival model.
Another data set consisted of 53 242 DDKT recipients transplanted from January 1, 2007, to September 2, 2012. Each patient had at least 3-year follow-up. Patients were right-censored if they were lost to follow-up within 3 years posttransplant or survived past it. This data set was used for the 3-year survival model.
Karnofsky performance score at transplant and most of the covariates used in SRTR risk-adjustment models15 for DDKT recipients were already included in the SRTR data set. Most of the recipients’ information was gathered from the Transplant Candidate Registration (TCR) and Transplant Recipient Registration (TRR) forms.16 The TCR records information at the time of listing such as ethnicity and education; TRR records information at the time of initial transplant admission such as age, kidney diagnoses, and KPS at transplant. Donor’s information was recorded at the time of organ donation in the Deceased Donor Registration form.16 Computed using the available covariates in the SRTR data set, the only derived covariates were estimated glomerular filtration rate (eGFR),17 kidney donor risk index (KDRI),18 total ESRD time, and donor’s body mass index (BMI). KPS was stratified into 6 strata: 10 to 30, 40 to 50, 60 to 70, 80 to 90, 100, and “unknown.” These strata were based on similar definitions and Ma et al’s work,19 which mapped KPS to the Eastern Cooperative Oncology Group scores. The stratification is summarized in Table 1. The covariates are summarized in Tables 2 (1 y) and 3 (3 y).
A considerable number of the covariates used in survival model building had missing data, but no observation was omitted for having “unknown” or missing values in any of the considered covariates. Missing data were addressed for categorical covariates and numeric covariates separately. If a categorical covariate had at least 1% missing or unknown, these values were treated together as 1 separate category. If it had <1% missing or unknown and the other values were “P/N” (positive/negative) or “Y/N” (yes/no), then the missing and unknown values were imputed using multiple imputation by chained equations.20 Multiple imputation by chained equation was also applied to numeric covariates with missing data. The imputation method used logistic regression for categorical covariates with “P/N” or “Y/N” values and predictive mean matching for numeric covariates. Imputation onto a data set produced 10 multiply imputed data sets, each with distinct model estimates of the missing values. In total, 20 multiply imputed data sets were obtained, 10 each from the 1- and 3-year survival data sets. These data sets were used for the statistical analyses.
The statistical analyses were conducted using the R programming language.21 Imputation was performed using the mice package20 and survival modeling was performed using the survival package.22 For both the 1- and 3-year survival data sets, log-rank tests were applied to examine pairwise comparison of survival times between KPS strata. For each imputed data set, 2 multivariate Cox proportional hazard models were fitted: the standard and augmented models. The standard model was the SRTR risk-adjustment model,15 either first-year outcomes (47 total covariates) or 3-year outcomes (61 total covariates), while the augmented model was the same model with 5 additional indicator covariates, each corresponding to a KPS stratum. “Unknown” or missing values that appear at least 1% within a categorical covariate are set as the reference level before fitting the model. Four sets of 10 models were fitted: standard 1-year survival, augmented 1-year survival, standard 3-year survival, and augmented 3-year survival. The coefficients of the models were averaged to obtain the pooled coefficients, which were used to construct the pooled Cox proportional hazard model. The pooled coefficients were recorded in Tables S1-S4, SDC (http://links.lww.com/TP/B619).
The log-likelihood, Akaike information criterion (AIC),23 Bayesian information criterion (BIC),24 and Harrell’s c statistic25 were computed using the pooled Cox model. To evaluate model calibration improvement, Meng and Rubin’s method26,27 was used to perform the likelihood ratio test. To compare the differences in AICs, BICs, and c statistics between the standard and augmented models, paired t tests with 9-degrees of freedom were performed.
To estimate the expected survival curves conditional on a covariate of interest (eg, KPS), the unique combinations of categorical covariates, excluding the conditioned covariate if categorical, were listed along with their number of frequencies within the corresponding data set of the pooled Cox model. The list was converted to a data set, where each observation had its categorical covariates as one of the unique combinations and its numerical covariates as the averages of their respective, nonmissing values within the data set after splitting the original SRTR data set and before imputation. Each observation in the newly constructed data set has the same value for the covariate of interest being conditioned on. For each observation of the new data set, the pooled Cox model was used to estimate its survival curve. The adjusted expected survival curve was computed as the weighted average of these survival curves using the normalized frequencies as the weights.
This study was obtained under IRB approval (STU00204041) from Northwestern University Institutional Review Board. All identifiers were removed upon data receipt for the purposes of this study.
Study Cohort Characteristics
The KPS distributions between the 1- and 3-year survival cohorts were similar. The most prevalent KPS value was 80 to 90 (approximately 54%), followed by 60 to 70 (approximately 23%–25%) and 100 (approximately 14%–15%). Recipients with KPS of 50 or less comprised approximately 3% of the cohort. In addition, 3% of the cohort reported their KPS values as unknown. Therefore, approximately 94% of the recipients were capable of self-care. The specific percentages of each KPS stratum for each cohort are listed in Tables 2 and 3. Among the 1-year survival cohort, 2766/72 839 recipients died (Table 2); among the 3-year survival cohort, 4879/53 242 recipients died (Table 3). Table 4 counts the number of deaths and provides the prevalence rates for each KPS stratum for both the 1- and 3-year survival data sets. Although the prevalence rates for KPS strata 10 to 30 and 40 to 50 are extremely low that they may compromise model accuracy, the numbers of events (or deaths) per independent variables (EPV) for both 1- and 3-year models exceed 10, ensuring accurate estimation of the regression coefficients.28 Specifically, the EPV for the augmented 1-year Cox model is 2766/52 = 55.32, while the EPV for the augmented 3-year Cox model is 4879/66 = 73.92. Even if they exceed 40–50, the threshold ensures minimal bias in estimating regression coefficients of binary covariates with very low prevalence.29
One-year Patient Survival
Table 5 summarizes the statistical results for the SRTR and augmented models for the 1-year survival data set. After including KPS into the SRTR model, the c statistics of the augmented model improved by 0.01. Although incremental, the increase was statistically significant (P < 0.001) according to the paired t test. Furthermore, KPS improved model calibration since the average log-likelihoods of the augmented model using the original coefficients and pooled coefficients were higher than the average log-likelihoods of the SRTR model. By the likelihood ratio test, the improvement in model calibration was statistically significant (P < 0.001), indicating that KPS was a significant predictor of 1-year posttransplant survival. KPS also improved the predictive accuracy of the model, as indicated by the decrease in AIC and BIC. The differences in AICs and BICs between the SRTR and augmented models were significant (P< 0.001 for both). More interestingly, because BIC penalized against models with large number of variables, the significant improvement in BIC indicated that including KPS still strengthened the predictive accuracy of the model. Table 6 confirmed specifically that KPS 10 to 30, KPS 80 to 90, and KPS 100 were independent predictors of 1-year posttransplant survival. More importantly, it showed that KPS 10 to 30 has an extremely high hazard ratio (HR) of 11.25, whereas KPS 40 to 50 has an HR of 1.28 and the other strata have an HR <1.00.
Table 7 shows the P values for the pairwise log-rank tests between the KPS strata. The survival differences between any pair of KPS strata are statistically significant (P < 0.05 for all pairs). Validating these differences, Figure 1 shows the adjusted expected 1-year survival curves conditional on each KPS stratum at transplant and their 95% confidence intervals estimated using the augmented model with pooled coefficients. According to the figure, DDKT recipients with KPS scores 10 to 30 have the worst survivals, with their 1-year expected survival probabilities being below 65%. However, KPS 40 to 50 recipients have significantly better expected survival, with their expected 1-year survival probabilities being approximately 95%. As KPS score increases, the survival probabilities improve.
Figure 2 shows the families of adjusted expected survival curves conditioned on KDRI and on each KPS stratum at transplant. In the figure, KDRI was converted to kidney donor profile index (KDPI),30 where each integer value from 0 to 100 corresponds to a range of KDRI values. The figure shows that KPS 10 to 30 recipients have the largest range in expected 1-year survival probability from approximately 76% with the best quality kidneys (KDPI = 0) to approximately 26% with the worst quality kidneys (KDPI = 100). Even if KPS 10 to 30 recipients were transplanted with the best quality kidneys, their expected survival probabilities are still worse than the probabilities of KPS 40 to 50 recipients transplanted with the worst quality kidneys. For DDKT recipients with KPS 40–50, their expected 1-year survival probabilities range from above 95% to approximately 85%. Transplanted with KDPI of 99 or less kidneys, DDKT recipients are expected to live with above 90% survival probability in 1 year. As KPS improves, the minimum and maximum expected 1-year survival probabilities increase and the ranges become narrower.
Three-year Patient Survival
Table 8 summarizes the statistical results for the SRTR and augmented models with the 5 additional KPS indicator covariates for the 3-year survival data set. Including KPS in the 3-year survival model showed similar improvements as it did for the 1-year survival model. The c statistic of the augmented model improved by 0.005, but paired t test verified its statistical significance (P < 0.001). Despite the incremental improvement in model discrimination, KPS was validated to be a significant predictor of 3-year posttransplant survival. By likelihood ratio test, model calibration improved statistically significantly (P < 0.001). Furthermore, predictive accuracy improved as conveyed by the lower AIC and BIC values after including KPS. Improvements in AIC and BIC values were again statistically significant (P < 0.001 for both). Table 9 confirmed specifically that KPS 10 to 30, KPS 40 to 50, and KPS 100 are independent predictors of 3-year posttransplant survival. KPS 10 to 30 has an extremely high HR being 10.38, whereas the other KPS strata have an HR <1.5.
Table 10 shows the P values for the pairwise log-rank tests between the KPS strata. The survival differences between any pair of KPS strata are statistically significant (P < 0.05 for all pairs). Validating these differences, Figure 3 shows the adjusted expected 3-year survival curves conditional on each KPS stratum at transplant and their 95% confidence intervals estimated using the augmented model with pooled coefficients. KPS 10 to 30 recipients have the worst expected survival probabilities in 3 years, significantly decreasing to below 50% from after 1 year. In the next KPS stratum 40 to 50, although the expected survival probabilities decrease over time, the rate of decrease in expected survival probabilities is not as severe as for the KPS stratum 10 to 30. The expected 3-year survival probabilities remain above 85%. As for the other strata, the expected 3-year survival probabilities are above 90%.
Figure 4 shows the families of adjusted expected 3-year survival curves conditional on KDRI and each KPS stratum at transplant. Like in Figure 2, KDRI was converted to KDPI. With the best kidney quality (KDPI = 0), DDKT recipients with KPS 10 to 30 have at most 60% expected 3-year survival probabilities. In the next stratum KPS 40 to 50, DDKT recipients’ expected survival probabilities are much better because with the worst quality kidney (KDPI = 100), their expected survival probabilities are as low as 75%. However, if they were transplanted with KDPI of 99 or less kidneys, their expected survival probabilities are >80%. As for the other strata, expected 3-year survival probabilities are above 80%.
Previous studies have attempted to improve posttransplant graft/patient survival models for DDKT recipients by including additional covariates such as comorbidities not considered by the SRTR. Jassal et al31 and Grosso et al32 included comorbidities indices into their risk adjustment models and reported that Charlson Comorbidity Index was a significant independent predictor of kidney posttransplant mortality. Wu et al33 confirmed that the Charlson Comorbidity Index was correlated with graft loss and patient death after renal transplantation. Weinhandl et al34 gathered a list of 31 comorbid conditions, which includes the Elixhauser conditions35 and cardiac arrhythmias, from CMS and incorporated them in the risk-adjustment Cox models for graft survivals trained on 1992 to 2005 data. Pelletier et al36 focused on incorporating cardiovascular comorbidities and demonstrated improvement in the 1-year posttransplant model for kidney graft survival. Unfortunately, the comorbidity indices and most of the comorbid conditions are unavailable in both the SRTR and OPTN data set. An alternative covariate that summarizes the overall impact of comorbidities onto a recipient’s health is functional status, which measures the patient’s ability to perform normal daily activities and to maintain health and well-being.37
Kutner et al38 and Reese et al39,40 studied the impact of functional status on posttransplant survival for kidney transplant recipients. In their statistical analyses, functional status was recorded using the Physical Functioning (PF) domain of the Medical Outcomes Study 36-Item Short Form Health Survey,41 which was self-reported by the patient instead of physician-reported. Furthermore, the PF scores used in the statistical analyses may not be up-to-date as they were recorded a few months before transplant. Despite these limitations, the authors demonstrated that PF was an independent predictor of posttransplant survival for kidney transplant recipients. However, PF scores are unavailable in the SRTR and OPTN data.
Unlike the PF scores, KPS is physician-reported and thus more reliable. It was validated by Mor et al42 as an accurate measure for functional status. KPS has been examined as an independent predictor for posttransplant survival for recipients of other solid organs. For example, Dolgin et al stratified KPS into 3 strata and demonstrated that it was a reliable predictor of posttransplant mortality for liver transplant recipients.43 Grimm et al44 examined the impact of KPS onto posttransplant survival outcomes for lung transplant recipients and found it to be a significant independent predictor for 1-year mortality. In addition, Kilic et al45 assessed that KPS was a significant predictor of posttransplant survival outcomes for lung retransplant recipients. Contributing to the KPS literature in transplantation, this study examined the impact of KPS on posttransplant survival outcomes for DDKT recipients and determined whether KPS could help make better transplantation decisions.
According to the cohort analysis, only about 3% of transplant recipients had KPS 40 to 50 at the time of transplant. This proportion is comparable to the proportion of KPS at listing from April 1, 2005 (the earliest date when KPS was recorded in SRTR46) to September 2, 2014, which is summarized in Table 11. The small proportion suggested that the transplant centers may be risk averse in listing and even transplanting KPS 40 to 50 patients, but the survival analyses demonstrated that they have a high probability of surviving at least 3 years, even if they were transplanted with low-quality kidneys. Based on the survival curves in Figures 2 and 4, the differences in survival probabilities between recipients with KPS 10–30 and KPS 40 to 50 could be as large as 50% for 1 year and 60% for 3 years if they were transplanted with the same quality kidneys. On the other hand, the differences in survival probabilities between recipients with KPS 40 to 50 and other higher KPS stratum were <10% for both 1 and 3 years if transplanted with the same quality kidneys. Although transplanting a KPS 40 to 50 candidate is riskier than a higher KPS candidate, it is not as severe as transplanting a candidate with KPS 10 to 30. If more KPS 40 to 50 patients were listed and transplanted, KPS 10 to 30 patients could be encouraged to undergo prehabilitation to improve their functional statuses, potentially improving their accessibility to DDKT and bettering their posttransplant survival outcomes.
For KPS 40 to 100 recipients, being transplanted with a low-quality kidney can still yield a high survival probability. According to Figures 2 and 4, recipients transplanted with KDPI ≤99 kidneys have survival probabilities of 90% or greater for 1 year and 80% or greater for 3 years. These results could dispel concerns that being transplanted with low-quality kidneys result in low survival probabilities for KPS 40 to 100 candidates. Among the waitlisted candidates, the willingness to accept a kidney with KDPI >85 was at 49.9% in 2014, but it decreased to 45.7% in 2016.2 This behavior may have attributed to the high discard rate of low-quality kidneys. Nearly 60% of kidneys with KDPI >85 were discarded in 2016,2 yet these discarded kidneys could have been transplanted to viable candidates, such as those with KPS 40 to 50.
This study has some limitations. For example, KPS is susceptible to observer bias and its evaluation can vary within and across transplant centers. The second limitation is the missing values in the SRTR data. Finally, KPS was originally developed for oncology patients and it may not be entirely applicable to ESRD patients because disease symptoms and progression differ between them.
Despite these limitations, this study reveals the potential utility of functional status in making decisions of transplanting patients given their physical health and offered kidney qualities. It shows that functional status provides a more objective measure than the “eyeball test” that may have discriminated against older patients from being listed and transplanted.47,48 Furthermore, it contributes to the growing literature that justify functional status as a potential predictor of postsurgical survival and recovery outcomes. Robinson et al demonstrated that functional independence measured using the Katz Index of independence was a significant predictor of 6-month postoperative mortality in elderly patients.49 Lawrence et al50 showed that better functional status nearly always predicted better recovery in shorter time after major abdominal surgery for elderly patients. In a related study, Brouquet et al51 showed that mobility and physical function for daily living activities were independent predictors of postoperative delirium for elderly patients who underwent major abdominal surgery. In liver transplantation within the United Kingdom and Ireland patient population, Jacob et al found that functional status measured by Eastern Cooperative Oncology Group had an impact on posttransplant outcome.52
To refine the analyses of this study, KPS’s recording requires standardization within and across transplant centers and validation of interrater reliability in the ESRD setting, as it has been done for palliative care53,54 and oncology.55,56 However, if a more reliable functional status measure, especially for the renal setting, were developed, then similar analyses using this alternative measure can still be performed and used to assist in making transplantation decisions. Overall, this study encourages further investigation of the use of functional status to help improve current transplant center strategies in transplanting more ESRD patients, maximizing listed patients’ survival outcomes, and reducing the kidney discard rate.
This research was supported in part through the computational resources and staff contributions provided for the Quest high performance computing facility at Northwestern University, which is jointly supported by the Office of the Provost, Office for Research, and Northwestern University Information Technology.
The data reported here have been supplied by the Minneapolis Medical Research Foundation (MMRF) as the contractor for the SRTR. The interpretation and reporting of these data are the responsibility of the author(s) and in no way should be seen as an official policy of or interpretation by the SRTR or U.S. Government.
1. HRSA. Organ Procurement and Transplantation Network data reports. 2018. Available at https://optn.transplant.hrsa.gov/data/
2. Hart A, Smith JM, Skeans MA, et al.OPTN/SRTR 2016 annual data report: kidney. Am J Transplant. 2018;18S118–113.
3. Karnofsky DA, Burchenal JHMacLeod CMThe clinical evaluation of chemotherapeutic agents in cancer. In: Evaluation of chemotherapeutic agents. 1949:New York: Columbia University Press191–205.
4. Perez Valdivieso JR, Bes-Rastrollo M, Monedero P, et al.Karnofsky performance score in acute renal failure as a predictor of short-term survival. Nephrology (Carlton). 2007;12:533–538.
5. McClellan WM, Anson C, Birkeli K, et al.Functional status and quality of life: predictors of early mortality among patients entering treatment for end stage renal disease. J Clin Epidemiol. 1991;44:83–89.
6. Ifudu O, Paul HR, Homel P, et al.Predictive value of functional status for mortality in patients on maintenance hemodialysis. Am J Nephrol. 1998;18:109–116.
7. Okada T, Nakao TPhysical functional status and factors contributing to disability in Japanese chronic dialysis patients. Nephrol Ther. 1998;4:195–203.
8. Bossola M, Marino C, Di Napoli A, et al.Functional impairment and risk of mortality in patients on chronic hemodialysis: results of the Lazio dialysis registry. J Nephrol. 2018;31:593–602.
9. Dickinson DM, Shearon TH, O’Keefe J, et al.SRTR center-specific reporting tools: posttransplant outcomes. Am J Transplant. 2006;6:1198–1211.
10. Snyder JJ, Salkowski N, Kim SJ, et al.Developing statistical models to assess transplant outcomes using national registries: the process in the United States. Transplantation. 2016;100:288–294.
11. U.S. Centers for Medicare & Medicaid Services. CMS conditions of participation. Available at https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/CertificationandComplianc/Transplant.html
. Accessed April 13, 2018.
12. Pelletier RP, Henry MLProgram specific reports: friend or foe?—the intended and unintended consequences of scientific registry of transplant recipient program specific reports. Curr Transplant Rep. 2014;1:86–90.
13. Schold JD, Arrington CJ, Levine GSignificant alterations in reported clinical practice associated with increased oversight of organ transplant center performance. Prog Transplant. 2010;20:279–287.
14. Schold JD, Buccini LD, Srinivas TR, et al.The association of center performance evaluations and kidney transplant volume in the United States. Am J Transplant. 2013;13:67–75.
15. SRTR. SRTR risk adjustment model documentation: waitlist and Posttransplant outcomes. Available at https://www.srtr.org/reports-tools/risk-adjustment-models-posttransplant-outcomes/
16. Massie AB, Kucirka LM, Segev DLBig data in organ transplantation: registries and administrative claims. Am J Transplant. 2014;14:1723–1730.
17. Levey AS, Stevens LA, Schmid CH, et al.A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150:604–612.
18. Rao PS, Schaubel DE, Guidinger MK, et al.A comprehensive risk quantification score for deceased donor kidneys: the kidney donor risk index. Transplantation. 2009;88:231–236.
19. Ma C, Bandukwala S, Burman D, et al.Interconversion of three measures of performance status: an empirical analysis. Eur J Cancer. 2010;46:3175–3183.
20. van Buuren S, Groothuis-Oudshoorn KMice: multivariate imputation by chained equations in R. J Stat Softw. 2011;45:1–68.
21. Team RC. R: a language and environment for statistical computing. 2018. Available at https://www.R-project.org
22. Therneau TM, Lumley TPackage ‘survival’. R Package Version. 2017;2:41–43.
23. Collett DModelling Survival Data in Medical Research. 2015.CRC Press;
24. Volinsky CT, Raftery AEBayesian information criterion for censored survival models. Biometrics. 2000;56:256–262.
25. Harrell FE, Lee KL, Mark DBMultivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387.
26. Meng X-L, Rubin DBPerforming likelihood ratio tests with multiply-imputed data sets. Biometrika. 1992;79:103–111.
27. Van Buuren SFlexible Imputation of Missing Data. 2012.New York: CRC Press;
28. Peduzzi P, Concato J, Feinstein AR, et al.Importance of events per independent variable in proportional hazards regression analysis II. Accuracy and precision of regression estimates. J Clin Epidemiol. 1995;48:1503–1510.
29. Austin PC, Allignol A, Fine JPThe number of primary events per variable affects estimation of the subdistribution hazard competing risks model. J Clin Epidemiol. 2017;83:75–84.
31. Jassal SV, Schaubel DE, Fenton SSBaseline comorbidity in kidney transplant recipients: a comparison of comorbidity indices. Am J Kidney Dis. 2005;46:136–142.
32. Grosso G, Corona D, Mistretta A, et al.Predictive value of the Charlson comorbidity index in kidney transplantation. 2012.Paper presented at: Transplantation proceedings;
33. Wu C, Evans I, Joseph R, et al.Comorbid conditions in kidney transplantation: association with graft and patient survival. J Am Soc Nephrol. 2005;16:3437–3444.
34. Weinhandl ED, Snyder JJ, Israni AK, et al.Effect of comorbidity adjustment on CMS criteria for kidney transplant center performance. Am J Transplant. 2009;9:506–516.
35. Elixhauser A, Steiner C, Harris DR, et al.Comorbidity measures for use with administrative data. Med Care. 1998;36:8–27.
36. Pelletier RP, Phillips GS, Rajab A, et al.Effects of cardiovascular comorbidity adjustment on SRTR risk-adjusted Cox proportional hazard models of graft survival. Transplantation. 2014;97:686–693.
37. Leidy NKFunctional status and the forward progress of merry-go-rounds: toward a coherent analytical framework. Nurs Res. 1994;43:196–202.
38. Kutner NG, Zhang R, Bowles T, et al.Pretransplant physical functioning and kidney patients’ risk for posttransplantation hospitalization/death: evidence from a national cohort. Clin J Am Soc Nephrol. 2006;1:837–843.
39. Reese PP, Bloom RD, Shults J, et al.Functional status and survival after kidney transplantation. Transplantation. 2014;97:189.
40. Reese PP, Shults J, Bloom RD, et al.Functional status, time to transplantation, and survival benefit of kidney transplantation among wait-listed candidates. Am J Kid Dis. 2015;66:837–845.
41. Ware JE Jr, Sherbourne CDThe MOS 36-item short-form health survey (SF-36): I. Conceptual framework and item selection. Med Care. 1992;30:473–483.
42. Mor V, Laliberte L, Morris JN, et al.The Karnofsky performance status scale: an examination of its reliability and validity in a research setting. Cancer. 1984;53:2002–2007.
43. Dolgin NH, Martins PN, Movahedi B, et al.Functional status predicts postoperative mortality after liver transplantation. Clin Transplant. 2016;30:1403–1410.
44. Grimm JC, Valero V 3rd, Kilic A, et al.Preoperative performance status impacts perioperative morbidity and mortality after lung transplantation. Ann Thorac Surg. 2015;99:482–489.
45. Kilic A, Beaty CA, Merlo CA, et al.Functional status is highly predictive of outcomes after redo lung transplantation: an analysis of 390 cases in the modern era. Ann Thorac Surg. 2013;96:1804–1811.
46. Orman ES, Ghabril M, Chalasani NPoor performance status is associated with increased mortality in patients with cirrhosis. Clin Gastroenterol Hepatol. 2016;14:1189–1195.
47. Segev DL, Kucirka LM, Oberai PC, et al.Age and comorbidities are effect modifiers of gender disparities in renal transplantation. J Am Soc Nephrol. 2009;20:621–628.
48. Abecassis M, Bridges ND, Clancy CJ, et al.Solid-organ transplantation in older adults: current status and future research. Am J Transplant. 2012;12:2608–2622.
49. Robinson TN, Eiseman B, Wallace JI, et al.Redefining geriatric preoperative assessment using frailty, disability and co-morbidity. Ann Surg. 2009;250:449–455.
50. Lawrence VA, Hazuda HP, Cornell JE, et al.Functional independence after major abdominal surgery in the elderly. J Am Coll Surg. 2004;199:762–772.
51. Brouquet A, Cudennec T, Benoist S, et al.Impaired mobility, ASA status and administration of tramadol are risk factors for postoperative delirium in patients aged 75 years or more after major abdominal surgery. Ann Surg. 2010;251:759–765.
52. Jacob M, Copley LP, Lewsey JD, et al.Functional status of patients before liver transplantation as a predictor of posttransplant mortality. Transplantation. 2005;80:52–57.
53. Zimmermann C, Burman D, Bandukwala S, et al.Nurse and physician inter-rater agreement of three performance status measures in palliative care outpatients. Support Care Cancer. 2010;18:609–616.
54. Myers J, Gardiner K, Harris K, et al.Evaluating correlation and interrater reliability for four performance scales in the palliative care setting. J Pain Symptom Manage. 2010;39:250–258.
55. Schag CC, Heinrich RL, Ganz PAKarnofsky performance status revisited: reliability, validity, and guidelines. J Clin Oncol. 1984;2:187–193.
56. Liem BJ, Holland JM, Kang MY, et al.Karnofsky performance status assessment: resident versus attending. J Cancer Educ. 2002;17:138–141.