Development and external validation study combining existing models and recent data into an up-to-date prediction model for evaluating kidneys from older deceased donors for transplantation

With a rising demand for kidney transplantation, reliable pre-transplant assessment of organ quality becomes top priority. In clinical practice, physicians are regularly in doubt whether suboptimal offers from older donors should be accepted. Here, we externally validate existing prediction models European population deceased donors, and subsequently developed validated an adverse outcome tool. Recipients grafts 50 years age were included the Netherlands Organ Transplant Registry (NOTR) United States transplant registry 2006-2018. The predicted was composite graft failure, death or chronic disease stage 4 plus within one year after modelled using logistic regression. Discrimination calibration assessed internal, temporal external validation. Seven with same cohorts. NOTR development cohort contained 2510 patients 823 events. validation had 837 used 31987 registry. our full model moderate (C-statistic 0.63), though somewhat better than discrimination seven (average C-statistic 0.57). model’s highly accurate. Thus, since survival performed poorly novel validated, maximum achievable performance donors. These could assist clinicians deciding to accept donor. Kidney transplantation is treatment choice end-stage renal disease, terms life.1Wolfe R.A. Ashby V.B. Milford E.L. et al.Comparison mortality all on dialysis, dialysis awaiting recipients first cadaveric transplant.N Engl J Med. 1999; 341: 1725-1730Crossref PubMed Scopus (3873) Google Scholar,2Liem Y.S. Bosch J.L. Arends L.R. al.Quality life Medical Outcomes Study Short Form 36-Item Health Survey replacement therapy: systematic review meta-analysis.Value Health. 2007; 10: 390-397Abstract Full Text PDF (161) Scholar donor pool lagging behind, acceptance criteria kidneys continue expand.3Port F.K. Bragg-Gresham Metzger al.Donor characteristics associated reduced survival: approach expanding donors.Transplantation. 2002; 74: 1281-1286Crossref (618) Scholar,4Tullius S.G. Rabb H. Improving supply deceased-donor organs transplantation.N 2018; 378: 1920-1929Crossref (72) Grafts recovered who average more comorbidities, come higher rates early dysfunction recipient mortality.5Querard A.H. Foucher Y. Combescure C. outcomes between expanded standard recipients: meta-analysis.Transpl Int. 2016; 29: 403-415Crossref (54) Scholar,6van Ittersum F.J. Hemke A.C. Dekker F.W. al.Increased risk failure Dutch receiving transplant.Transpl 2017; 30: 14-28Crossref (12) decision decline offer largely subjective depends donor-, preservation–, recipient-related characteristics. Discard vary widely individual across geographic areas.7Cooper M. Formica R. Friedewald J. al.Report National Foundation Consensus Conference Decrease Discards.Clin Transplant. 2019; 33: e13419Crossref (40) Scholar, 8Mohan S. Chiles M.C. Patzer R.E. al.Factors leading discard States.Kidney 94: 187-198Abstract (120) 9Mathur A.K. Sands R.L. Wolfe Geographic variation incidence access transplantation.Am 2010; 1069-1080Crossref (89) Reliable pretransplant selection best recipient-to-donor match minimize unjust maximize patient have thus become increasingly important.Various regression-based mathematical been that aim predict transplantation.10Kabore Haller Harambat al.Risk transplantation: review.Nephrol Dial 32: ii68-ii76Crossref (41) As reliably predicting post-transplant prior has proved challenging, several predictors measured during surgery shortly such as iBox score.11Loupy A. Aubert O. Orandi B.J. al.Prediction system allograft loss transplants: international derivation study.BMJ. 366: l4923Crossref (116) Although these might useful monitoring patients, they cannot guide offer. One most (combined mortality) Donor Risk Index (KDRI).12Rao P.S. Schaubel D.E. Guidinger M.K. al.A comprehensive quantification score kidneys: index.Transplantation. 2009; 88: 231-236Crossref (653) Profile (KDPI), derived this KDRI, implemented new US allocation effect 2014.13Stewart Kucheryavaya A.Y. Klassen D.K. al.Changes KAS implementation.Am 16: 1834-1847Crossref (179) Long-term consequences implementation still unknown. Nevertheless, KDPI criticized delayed function increased, dependent age, labeling may cause almost automatic high KDPI.14Massie A.B. Luo X. Lonze B.E. al.Early changes distribution under system.J Am Soc Nephrol. 27: 2495-2501Crossref (70) 15Ruggenenti P. Remuzzi G. Invited letter response to: "Is profile index (KDPI) universal UNOS-specific?.Am 18: 1033-1034Crossref (4) 16Stallone Grandaliano To not discard: art scoring.Clin 12: 564-568Crossref (16) 17Bae Massie rate introduction (KDPI).Am 2202-2207Crossref (104) ScholarIn systems, yet implemented. Similar KDRI developed, but vast majority constructed data States.18Kasiske B.L. Israni Snyder J.J. simple tool transplant.Am Dis. 56: 947-960Abstract (63) 19Vinson A.J. Kiberd B.A. Davis R.B. Tennankore K.K. Nonimmunologic donor-recipient pairing, HLA matching, transplantation.Transplant Direct. 5: e414Crossref (10) 20Molnar M.Z. Nguyen D.V. Chen al.Predictive posttransplantation outcomes.Transplantation. 101: 1353-1364Crossref (30) populations, procedures, policies differ considerably Europe States, there need develop patients. Furthermore, specifically tailored allow improved decision-making regarding which little consensus decline. Therefore, can survival, Northern American received aged ?50 years. Subsequently, improve by developing validating (AO) 1 donors.ResultsExisting modelsFollowing screening, 6 studies, presenting 7 models, considered appropriate (flowchart Supplementary Figure S1). Characteristics shown Table 1.Table 1Characteristics modelsPrediction modelTime horizon, yrOutcomeDevelopment cohortPopulationMean yrReported C-statisticOverall biasModel informationDonor Schold al. (2005)21Schold J.D. Kaplan B. Baliga R.S. Meier-Kriesche H.U. broad spectrum kidneys.Am 2005; 757-765Crossref (152) Scholar—Death/GFUS 1996–2002 (OPTN: SRTR data)First-time, single only, adult Deceased donors——HighRegression coefficients given (unclear what statistical used)Rao (2009),12Rao donor-only model—Death/GFUS 1995–2005 donors—0.62 (IV)HighCox model, HRs givenKasiske (2010)18Kasiske Scholar5Death/GFUS 2000–2006 USRDS data)Single donor380.64 formula givenWatson UKKDRI (2012)22Watson C.J.E. Johnson R.J. Birch simplified transplantation.Transplantation. 2012; 93: 314-318Crossref (84) Scholar9Death/GFEuropean 2000–2007 (UK Registry)Adult recipientsAdult donors49aMedian age.0.62 givenMolnar (2017)20Molnar 2001–2006 (OPTNbOPTN linked facility data.: dialysisDeceased donors390.63 (IV)LowCox givenVinson 3 (2018)19Vinson 2000–2014 recipientsDeceased given—, Unknown/not reported; GF, failure; HR, hazard ratio; IV, internal validation; Index; OPTN, Procurement Transplantation Network; SRTR, Scientific (this set, OPTN supplemented various secondary sources); USRDS, Renal Data System (OPTN Centers Medicare & Medicaid Services).Each use at time allocation. combined each models.a Median age.b data. Open table tab All showed similar C-statistics around 0.63 previous Most bias when Prediction Of Bias ASsessment Tool (PROBAST; S1).23Wolff R.F. Moons K.G.M. Riley R.D. al.PROBAST: assess applicability studies.Ann Intern 170: 51-58Crossref (539) Only 2 provided formula. Included per study, only predictor (Table 2). characteristics.Table 2Final modelsPredictorsAO modelAO data-driven expert modelSchold modelKDRI fullKDRI donor-onlyKasiske modelUKKDRIMolnar modelVinson modelDonor characteristicsAge??????????BMI??Cause death???????Cold ischemic time?????CPR performed?Days hospital?DCD ? CIT???Diabetes mellitus??????Donor cardiac death?????Double Tx?ECD?En bloc Tx?Ethnicity???HCV status???Height??Hypertension history????????Hypotension???Inotrope use???Last serum creatinine?????Left/right kidney?Proteinuria?Sex?Smoking?Warm time???Weight???Recipient characteristicsAge??????Blood hemoglobin?BMI?Cardiovascular disease??Coronary artery disease??Diabetes mellitus????Dialysis duration??????Ethnicity??HCV status?Medical insurance??No. Tx???Peripheral vascular disease?Primary disease????Serum albumin?Sex??Donor-recipientDonor age????Donor-recipient CMV match?Donor-recipient ethnicity difference?Donor-recipient height difference??Donor-recipient weight difference???HLA mismatches????????Peak PRA??AO, outcome; BMI, body mass index; CIT, cold time; CMV, cytomegalovirus; CPR, cardiopulmonary resuscitation; DCD, death; ECD, donor; HCV, hepatitis C virus; HLA, human leukocyte antigen; PRA, panel-reactive antibody; Tx: transplantation. Baseline characteristicsIn total, 3333 (NOTR). split into (2510 patients) (837 patients). From (Organ Network [OPTN]), 31,987 cohort. At baseline, set slightly younger diabetes hypertension substantially fewer donations circulatory 3). More extensive baseline tables, including percentage missing stratified outcome, Tables S2–S4. cohort, total 10.2% (n = 257) experienced 6.9% 172) death, 17.8% 446) (CKD) ?4 year; 9 (<1%) lost follow-up. 8% 67) 3.6% 30) 17.4% 146) CKD ?4; 4% 35) 6.2% 1992) 5.3% 1711) 12.8% 4094) ?4. 200 For AO follow-up assumed outcome.Table 3Baseline cohortCharacteristicsDevelopment (NOTR 2006–2017)(n 2510)Temporal 2017–2018)(n 837)External 2006–2017) 31,987)Donor Age, yr60 (55–65)61 (55–66)56 (53–60) Sex, % male51.456.253.5 Cause %Trauma14.416.921.1Cerebrovascular accident64.456.256.5Anoxia18.124.620.0Other3.12.22.4 DCD donor, %46.158.813.8 Serum creatinine, ?mol/L66 (53–83)64 (52–82)80 (62–106) Proteinuria, %44.449.441.9 kg/m226 (4.7)26 (4.4)29 (6.4) History mellitus, %8.19.312.7 hypertension, %37.538.150.6 Hypotension, %31.521.9— Use inotropic medication, %71.769.951.8 Left kidney, %50.449.849.6 WIT min17 (14–21)15 (13–18)18 (11–27) Cold time, h15.8 (5.8)13.3 (5.7)18.2 (9.2)Recipient (49–67)62 (51–69)60 (51–66) male60.663.662.2 (4.7)27 (4.4)27 (4.8) Primary %Diabetes mellitus14.018.332.6Hypertension20.622.425.9Glomerular nephritis16.617.811.4Cystic disease14.79.97.7Other34.231.522.4 Diabetes %21.526.843.0 Time mo39 (25–57)25 (15–42)40 (13–66) ?1 Previous transplant, %12.915.28.9Donor-recipient Total No. mismatches3 (2–4)3 (2–4)5 (4–5) Peak PRA0 (0–0)0 (0–13)BMI, donation NOTR, Registry; WIT, warm ischemia time.Data mean (SD) median (interquartile range). Laboratory values SI units converted conventional follows: creatinine mg/dL, multiply 0.011. Validation results modelsIn validated. (graft combined) therefore outcome. study (NOTR), predictive ranged poor mediocre. 0.538 (UKKDRI) 0.611 (Vinson model), 0.565 4). models’ C-statistic, 0.587), unsurprising considering Overall, seen recent Vinson al.19Vinson Models conservatively updated calibration, generally reasonable (Supplementary S5 Figures S2–S5). model.Table 4External results: Harrel 1-year end point: deathModelNOTR 2006–2017NOTR 2017–2018OPTN 2006–2017Schold0.562 (0.532–0.591)0.555 (0.495–0.615)0.577 (0.567–0.586)KDRIfull (Rao)0.572 (0.542–0.601)0.560 (0.495–0.625)0.592 (0.582–0.601)KDRIdonor-only (Rao)0.571 (0.541–0.600)0.559 (0.495–0.623)0.590 (0.581–0.600)Kasiske0.584 (0.556–0.612)0.547 (0.484–0.610)0.609 (0.599–0.618)UKKDRI (Watson)0.544 (0.515–0.574)0.538 (0.473–0.603)0.552 (0.542–0.562)Molnar0.566 (0.537–0.596)0.575 (0.515–0.636)0.578 (0.569–0.588)Vinson0.598 (0.569–0.626)0.573 (0.510–0.636)0.611 (0.601–0.620)KDRI, Network.Data (95% confidence interval). modelsFor newly point, least following candidate predictors, predefined research team, well additionally suggested nephrologist panel included. This resulted 28 predictors. regression backward inclusion 14 top-ranked 2. ranking 10 nephrologists S6. general, lot nephrologists, although agreed important predictors.Discrimination moderate, nevertheless models. 0.635 0.630 validation, respectively. 0.628 0.624, respectively, lower 0.609 0.619, respectively 5). Calibration good, tended overpredict higher-risk 5 1). Without recalibration, overpredicted risks S7 S6). sensitivity analyses, also built predicts poorer (data shown).Table 5AO models: resultsModel measureDevelopment 2006–2017)Internal validationTemporal 2017–2018)External 2006–2017)aCorrection factor added recalibrate incidence. without recalibration Material.AO CI)0.680 (0.657–0.703)0.646bNo CIs computed it concerns bootstrap shrinkage corrected C-statistic.0.635 (0.593–0.678)0.630 (0.622–0.637) slope10.8090.8850.739 intercept0–0.125–0.319–0.366 large, %cCalibration large vs. observed.32.8 32.832.9 32.832.2 27.521.9 21.1AO C-statistic0.667 (0.644–0.690)0.637bNo C-statistic.0.628 (0.586–0.669)0.624 (0.617–0.631) slope10.8130.9090.796 intercept0–0.122–0.284–0.286 32.833.0 32.831.8 27.521.7 C-statistic0.658 (0.634–0.682)0.638bNo C-statistic.0.609 (0.566–0.653)0.619 (0.612–0.627) slope10.8690.7760.761 intercept0–0.087–0.391–0.327 indicates CI, interval; Network.a Correction Material.b No C-statistic.c observed. Clinical modelsAn individual’s probability having years, calculated formulas Material R-script. Both North provided. predictions hypothetical S8. example defined independent exemplify realistic scenarios ranging ideal poor. aid reject diagnostic properties thresholds 6. specificity high; correctly generates low do experience AO. However, low, meaning many cases missed model. recipient-donor pairs risk, less half will get (the positive value <50%). usually (high negative value). solely examples. fully determine enhance physicians’ process.Table 6Sensitivity, specificity, PPV, NPV based modelThresholdsSensitivity, %Specificity, %PPV, %NPV, %No. < thresholdNo. false negativesNo. ? positivesP 70%0.499.850.072.683522921P 65%0.999.333.372.683122864P 60%2.299.045.572.8826225116P 55%4.398.045.573.08152202212P 50%8.794.939.273.37862105131P 45%19.189.140.074.572818611066P 40%35.781.241.876.9641148196114AO, NPV, value; P, risk; value.Calculated “No. threshold” would number accepted kidneys. negatives transplanted "No. threshold" rejected positives did year. DiscussionIn current ?4) addition, studies employing advanced methods choosing broader definition shorter horizon. selected clinically relevant population, mak