Claims-Based Phenotyping of Acute Kidney Transplant Rejection in a National Medicare Cohort Authors: Cameron Ekanayake, Miko Yu, Karthik Natarajan, Syed A Husain, Chetan Sree Muppavarapu, Jesse Schold, Sumit Mohan Published: 2026 ABSTRACT AbstractIntroductionNational transplant registries record acute rejection (AR) without the granularity needed to study rejection heterogeneity and prognosis at scale. Whether AR can be stratified by type and severity in national claims data using a claims-based algorithm, and whether these phenotypes correlate with graft outcomes, remains unclear.MethodsWe conducted a cohort study using the US Renal Data System (USRDS) data including adult kidney transplant recipients from 2012–2021. AR within 12 months posttransplant was identified using International Classification of Diseases codes for transplant complications with evidence of antirejection treatment within 2 weeks. Episodes were categorized by type and severity based on treatment regimen. Biopsy-associated AR was defined by renal biopsy within 14 days before the treatment. Death-censored graft survival was assessed using Kaplan-Meier methods and multivariable Cox proportional hazards models.ResultsAmong 92,845 recipients, 5712 AR episodes were identified (69.4% mild acute-cellular rejection [ACR], 8.8% severe ACR, 19.7% antibody-mediated rejection [AMR], and 2.1% mixed [ACR + AMR]). Among 3052 biopsy-confirmed episodes, 3-year death-censored graft survival differed meaningfully across phenotypes (87.9% mild ACR, 80.7% severe ACR, 73.5% AMR, and 68.4% mixed [ACR + AMR] [P < 0.05]), with trends persisting after adjustment for recipient and donor characteristics.ConclusionClaims-based phenotyping can identify clinically meaningful AR subtypes and prognostic gradients at national scale, enabling more granular national rejection surveillance.Keywords: acute rejection, administrative claims data, allograft survival, antibody mediated rejection, cellular rejection, kidney transplantation, phenotype, United States renal data system Kidney transplantation is the optimal therapy for patients with end-stage kidney disease, offering superior survival and quality of life compared with dialysis.1,2 Despite improvements in immunosuppressive agents, AR remains a major adverse event for transplant recipients during the first posttransplant year.3,4 Various forms of AR have been well-characterized, ranging from ACR to AMR, with additional forms exhibiting overlapping features. Additional forms with varying degrees of severity and mechanistic complexity continue to emerge as clinically-distinct phenotypes in the literature.5,6 Although each rejection subtype varies in its immunopathologic mechanisms and response to therapy, there is broad consensus on initial management strategies: high-dose corticosteroids are used for mild ACRs and lymphocyte-depleting agents, i.v. Ig, and plasmapheresis are used for more severe ACRs or AMRs.7,8 Despite the clinical importance of these entities, our understanding of long-term outcomes after rejection remains incomplete. Although current evidence suggests that, broadly, AMRs indicates worse outcomes than ACRs,9, 10, 11, 12 most evidence is constrained to single-center studies with small sample size and generalizability. National registry data, such as those captured by the Organ Procurement and Transplant Network, lack the granularity needed to study rejection-related outcomes in a way that is representative of real-world rejection heterogeneity.13,14 For example, rejection events are recorded as a binary “yes or no” variable without distinguishing type, grade, or treatment in the United Network for Organ Sharing registry. Moreover, previous studies have called into question the completeness of this data, with a recent comparison of the multiple national registry data finding a significant lack of concordance of documented rejection events between registries.15 Administrative claims data offer a potential solution to allow researchers to identify AR events among patients who have Medicare as their primary payer16 with more granularity offered by classic registry data. The USRDS, which integrates Medicare claims for nearly all the US kidney transplant recipients, captures detailed information on hospitalizations, diagnostic procedures, and therapeutic interventions. We have previously shown it is possible to identify biopsy-proven rejection episodes using diagnostic and therapeutic codes with a claims-based algorithm.15 Assessing the ability of this approach to further classify rejection events by type and severity could be used to evaluate the extent to which claims data may be used to study rejection-related outcomes in a granular manner at a national scale. In this study, we aimed to describe the incidence and outcomes of AR classified by type and severity by a claims-based algorithm in a large, representative national cohort. Methods Study Design We performed a retrospective cohort study using the standard analysis files from the USRDS. We identified adult (age > 18 years) first-time, kidney-only, transplant recipients from 2012 to 2021 (n = 163,971). Patients who had Medicare as their primary insurance were identified through claims data using the payer history file. We excluded patients who did not have Medicare as their primary insurance payer at the time of transplant (n = 71,126) which yielded a final analytical cohort of 92,845 patients. Claims-Based AR Subtype Phenotyping Records of AR were independently identified using a claims-based algorithm that uses claims data that is contained in separate files within the USRDS standard analysis files. We used the institutional and physician or supplier data files to filter claims using their International Classification of Diseases codes, Current Procedural Terminal codes, and National Drug Codes. We used the Part D Medicare billing data files to identify additional claims by their National Drug Codes. Our claims-based algorithm defined AR as evidence of ACR or AMR with concurrent (i.e., within 14 days of treatment) International Classification of Diseases-9 or -10 billing codes for kidney transplant complications (996.81, T86.1, T86.10-.11, and T86.19) within 12 months of transplant date. Given that there are standard regimens for different types and grades of rejections,7,8 AR was further subclassified into 4 types (mild ACR, severe ACR, AMR, and mixed (ACR + AMR)) based on the treatment criteria found in Table 1. Mild ACR was inferred by evidence of i.v. steroids (dexamethasone, hydrocortisone, prednisolone, and methylprednisolone) alone, and severe ACR was inferred by evidence of i.v. thymoglobulin within 14 days of evidence of receiving i.v. steroids. Evidence of a thymoglobulin claim without a concurrent steroid claim was also considered severe ACR. AMR was inferred by evidence of a procedure for plasmapheresis within 14 days of receiving i.v. Ig. Evidence of a plasmapheresis claim without a concurrent i.v. Ig claim was also considered AMR. For individuals with multiple rejection episodes within the first posttransplant year, classification was based on the most severe qualifying phenotype, and patients were assigned a single rejection category for survival analyses. Table 1Treatment criteria for each AR typeType of ARTreatment criteriaACR (mild)Steroids onlyACR (severe)Thymoglobulin ± SteroidsAMRPlasmapheresis ± IVIGMixed ACR + AMRSteroids ± Thymoglobulin AND Plasmapheresis ± IVIGACR, acute cellular rejection; AMR, antibody-mediated rejection; AR, acute rejection; IVIG, i.v. Ig. Treatment criteria for each AR type ACR, acute cellular rejection; AMR, antibody-mediated rejection; AR, acute rejection; IVIG, i.v. Ig. Analyses Primary outcomes were death-censored graft survival and overall graft survival (a composite of graft failure and death) assessed with Kaplan-Meier models for each rejection type. Time 0 was defined as the date of antirejection treatment, and follow-up was censored at 3 years after treatment. Patients without valid (> 0-day) follow-up time were excluded. Survival probabilities were estimated at 1, 2, and 3 years, and differences between the rejection phenotypes were evaluated using log-rank tests. Given that some of the therapies we identified as rejection treatments in claims data may have other posttransplant indications (e.g. treatment of delayed graft function or management of autoimmune cytopenias unrelated to rejection), we conducted the same analyses for AR events identified by our algorithm, specifically for those events that had evidence of biopsy and those that did not have evidence of biopsy to determine whether a biopsy near the time of treatment conveyed any prognostic value when attempting to classify and grade rejections in claims data. AR-related biopsies across each rejection subtype were identified using Current Procedural Terminal code for renal biopsy (50200) but limited to those that were 14 days before the treatment to avoid the inclusion of biopsies that were not associated with a rejection. We used multivariable Cox proportional hazards models to assess the association between phenotype and death-censored graft failure within both subcohorts. Each AR phenotype was modeled as a time-varying exposure, with follow-up beginning at the time of transplant; recipients contributed person-time to the no-rejection group until the date of antirejection treatment and to the corresponding rejection phenotype group thereafter. Models were adjusted for key recipient and donor demographic and clinical covariates. Covariates were selected a priori based on established clinical relevance and previous literature evaluating risk factors for graft outcomes. The proportional hazards assumption was evaluated using Schoenfeld residual-based tests. This study was approved by the institutional review board at Columbia University Medical Center. Informed consent was not required. Statistical significance was defined by a 2-sided α of < 0.05. Analyses were performed using STATA/MP version 19.0 (StataCorp, College Station, TX). The data reported here have been supplied by the USRDS.17 The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as an official policy or interpretation of the US government. Results Comparison of Claims-Based AR Subtypes A total of 92,845 adults who were first-time kidney transplant recipients between 2012 and 2021 were included in the analysis (Figure 1). We identified 5712 rejection episodes using the presence of various rejection therapies as outlined in the methods. Recipient characteristics are described in Table 2. Recipients identified as having an AR episode within 12 months of transplant and those that were identified as not having an AR episode within 12 months of transplant were very similar. These recipients had a median age of 56 (interquartile range, 42–64 vs. 45–65) years and were similar with respect to male sex as defined by sex assigned at birth (59.0% vs. 62.1%), White race (58.4% vs. 59.2%), having received a deceased donor kidney (81.7% vs. 79.7%), and the number of human leukocyte antigen mismatches (5 vs. 4). Diagram of study cohort. ACR, acute cellular rejection; AMR, antibody-mediated rejection; AR, acute rejection. Table 2Recipient characteristics of patients were independently identified in the US Renal Data System claims-based algorithm for AR episodes occurring within 12 months of transplantRecipient characteristicsAR within 12 mos (n = 5712)No AR within 12 mos (n = 87,133)Recipient age (yr), median (IQR)54 (42–64)56 (45–65)Sex, n (%)Female2342 (41.0%)33,064 (38.0%)Male3370 (59.0%)54,069 (62.1%)Race, n (%)Asian261 (4.6%)5281 (6.1%)Black1993 (34.9%)27,922 (32.0%)White3336 (58.4%)51,568 (59.2%)Other/multi122 (2.1%)2362 (2.7%)Donor type, n (%) Deceased4667 (81.7%)69,398 (79.7%) Living1045 (18.3%)17,735 (20.4%)No. of HLA mismatches5 (4–5)4 (3–5)Creatinine levels (mg/dl) At 6-mo follow-up1.5 (1.2–1.9)1.3 (1.1–1.6) At 12-mo follow-up1.5 (1.2–1.9)1.3 (1.0–1.6)AR, acute rejection; HLA, human leukocyte antigen; IQR, interquartile range. Recipient characteristics of patients were independently identified in the US Renal Data System claims-based algorithm for AR episodes occurring within 12 months of transplant AR, acute rejection; HLA, human leukocyte antigen; IQR, interquartile range. Among the 5712 rejection events, 3966 (69.4%) were mild ACRs, 502 (8.8%) were severe ACRs, 1123 (19.7%) were AMRs, and 121 (2.1%) were concomitant mixed ACRs and AMRs. Baseline recipient and donor characteristics stratified by rejection phenotype (no AR, mild ACR, severe ACR, AMR, and mixed [ACR + AMR]) are presented in Supplementary Table S1. At 3 years, death-censored graft survival remained highest among recipients with mild acute ACR(89.9%) and severe ACR(88.5%). Death-censored graft survival was lower for those with AMR (74.4%) and lowest for those with mixed ACR and AMR (71.1%) (P < 0.001) (Supplementary Table S2 and Figure 2). A total of 3052 rejection events had evidence of a biopsy within 14 days before beforethe treatment and 2660 received the treatment without prior biopsy. A breakdown by rejection type for each subgroup can be found in Figure 1. Among the outcomes for patients who received rejection treatment after a biopsy, at 3 years after rejection, those with evidence of a mixed ACR + AMR rejection had the lowest death-censored graft survival (68.4%), whereas those with suspected AMR alone had higher death-censored graft survival (73.5%). Patients with treatments suggestive of ACR did the best, whereas those with mild ACR had better death-censored graft survival than those with severe ACR (87.9% vs. 80.7%, respectively) (P < 0.001) (Table 3 and Figure 3a). Graft survival estimates at 1, 2, and 3 years for each rejection subtype are displayed in Table 3. Comparison of death-censored graft survival between acute rejection types identified using the US Renal Data System claims-based algorithm. Kaplan-Meier curves are unadjusted. ACR, acute cellular rejection; AMR, antibody-mediated rejection. Table 3Death-censored graft survival and overall graft survival at each after rejection timepoint for rejection episodes with evidence of biopsy within 14-days before rejection treatmentOutcome typeYr since ARMild ACRSevere ACRAMRMixed (ACR + AMR)Death-censored graft survival probability (yr)196.2%91.5%88.7%85.0%292.5%85.8%81.0%71.8%387.9%80.7%73.5%68.4%Overall graft survival probability (yr)193.4%87.5%85.6%80.4%287.4%79.9%75.7%65.3%381.0%73.8%66.7%60.6%ACR, acute cellular rejection; AMR, antibody-mediated rejection. Death-censored graft survival and overall graft survival at each after rejection timepoint for rejection episodes with evidence of biopsy within 14-days before rejection treatment ACR, acute cellular rejection; AMR, antibody-mediated rejection. Comparison of death-censored graft survival between acute rejection types identified using the US Renal Data System claims-based algorithm (a) with evidence of biopsy before treatment and (b) without evidence of biopsy before treatment. Kaplan-Meier curves are unadjusted. ACR, acute cellular rejection; AMR, antibody-mediated rejection. Outcomes were similar in patients receiving rejection treatment without prior biopsy, except for ACRs. When biopsy was not done before rejection treatment, death-censored graft survival at 3 years was highest for severe ACR (93.9%), followed by mild ACR (92.4%) (P < 0.001) (Supplementary Table S3 and Figure 3b). In adjusted Cox proportional hazards models treating rejection phenotypes as time-varying exposures, biopsy-confirmed mild ACR, severe ACR, AMR, and mixed (ACR + AMR) were associated with significantly increased risk of death-censored graft failure compared with nonrejectors (hazard ratio, 1.84, 2.50, 4.18, and 4.59, respectively; Table 4). Among recipients treated for rejection without evidence of a prior biopsy, mild ACR, AMR, and mixed rejection similarly had higher risk of death-censored graft failure (hazard ratio, 1.43, 4.62, and 4.00, respectively) whereas severe ACR was not associated with worse outcomes after adjustment (Supplementary Table S4). For sensitivity analysis, we additionally fit a pooled multivariable model including both biopsy-confirmed and nonbiopsy-confirmed rejection episodes while adjusting for biopsy status as a covariate (Supplementary Table S5). Results were unchanged. Biopsy-confirmed rejection was independently associated with a modestly increased risk of graft failure (hazard ratio, 1.19 [95% confidence interval, 1.04–1.38] [P = 0.014]). Across models, higher 6- and 12-month serum creatinine, greater human leukocyte antigen mismatch, deceased-donor type, older donor age, and Black donor race were independently associated with increased risk. Schoenfeld residual testing did not show meaningful violations for the rejection phenotype indicators, the primary exposures of interest, although statistical deviations were observed for select adjustment covariates. In a sensitivity analysis restricted to recipients aged 18–65 years, findings were largely unchanged (Supplementary Tables S6 and S7). Table 4Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients with evidence of biopsy within 14 days before the rejection treatmentPredictorHR95% CIP-valueAR phenotype (ref = No AR) Mild ACR1.841.56–2.22< 0.001 Severe ACR2.501.76–3.54< 0.001 AMR4.183.50–4.99< 0.001 ACR + AMR4.592.98–7.060.001Recipient age (per 5 yr)0.950.94–0.96< 0.001Recipient Race (ref = White) Black1.241.16–1.32< 0.001 Asian0.800.69–0.920.002 Other0.850.71–1.030.099Recipient ethnicity (ref = non-Hispanic) Hispanic0.990.91–1.080.873Recipient gender (ref = Female) Male0.860.81–0.91< 0.001Recipient blood type (ref = Type O) A1.070.89–1.300.450 B1.110.92–1.340.261 AB0.910.76–1.110.386CPRA (ref = 0–19) 20–491.150.86–1.540.351 50–791.040.68–1.600.844 80–971.060.69–1.640.781 98–1001.410.80–2.500.228Donor age (per 5 yr)1.041.03–1.05< 0.001Donor Race (ref = White) Black1.261.17–1.35< 0.001 Asian1.120.94–1.330.191 Other0.940.73–1.210.637Donor ethnicity (ref = non-Hispanic) Hispanic1.070.99–1.160.101Donor gender (ref = Female) Male0.920.87–0.970.004Donor blood type (ref = Type O) A0.940.70–1.270.699 B0.880.65–1.200.419 AB1.050.78–1.430.738Donor type (ref = deceased donor) Living donor0.580.53–0.63< 0.001HLA mismatch (per mismatch unit)1.051.03–1.07< 0.0016-mo Creatinine (mg/dl)1.591.56–1.62< 0.00112-mo Creatinine (mg/dl)1.251.24–1.26< 0.001ACR, acute cellular rejection; AMR, antibody-mediated rejection; AR, acute rejection; CI, confidence interval; CPRA, calculated panel reactive antibody; HLA, human leukocyte antigen; HR, hazard ratio. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients with evidence of biopsy within 14 days before the rejection treatment ACR, acute cellular rejection; AMR, antibody-mediated rejection; AR, acute rejection; CI, confidence interval; CPRA, calculated panel reactive antibody; HLA, human leukocyte antigen; HR, hazard ratio. In another sensitivity check, we repeated these analyses for a composite outcome of death and graft failure (overall graft survival) with similar results. See Supplementary Tables S2, S3, S8, S9 and Supplementary Figures S1 and S2 for results. Discussion In this cohort study using the USRDS claims data, we developed an algorithm to identify and classify AR episodes based on type and severity. Our results confirm worsening outcomes over time with increasing severity of rejection in the first-year after posttransplant and that outcomes after AMR are inferior to outcomes after ACR. Moreover, patients who received treatment for both ACR and AMR rejections had worse outcomes. This is the first analysis using Medicare claims data to identify types of rejection episodes in such a granular manner, something that has been possible using national registry data because of reporting/collecting/quality limitations inherent in these data sources. We further show the prognostic value of various types and grades of AR based on the timing and treatment regimen. Together, these findings show the feasibility of a claims-based algorithm to identify phenotype ARs in a granular manner at scale, offering a practical solution to the current limitations inherent to registry analyses which do not capture rejection episodes comprehensively or with any additional information on type or severity, when attempting to study long-term outcomes following AR robustly. Among patients treated for rejection, we found evidence of a biopsy within 2 weeks before rejection treatment in 53.4% of patients. Among this subcohort, allograft outcomes differed meaningfully across rejection phenotypes in a pattern consistent with biological plausibility and current evidence from single-center studies.9, 10, 11, 12 Episodes classified by our algorithm as AMR were associated with significantly worse 3-year graft survival than all forms of ACR, whereas mixed rejection carried the poorest prognosis overall. Among ACR episodes, those categorized as severe had inferior outcomes compared with mild episodes. To our knowledge, this is the first study to show these relationships at a national level within the US, the largest and most diverse transplant system in the world.18,19 Although smaller studies from other national registries with more granular rejection data, such as the ANZDATA registry, have reported similar patterns,3 our findings extend this evidence by showing that claims-based phenotyping can distinguish clinically meaningful rejection subtypes and prognostic differences across a large, real-world population. For rejection episodes without biopsy confirmation within 2 weeks of treatment, we observed that severe ACRs paradoxically had slightly better outcomes than mild ACRs after 3 years. It is possible that differences between mild and severe ACR in this subcohort were attenuated because of possible misclassification in the absence of histology or differential treatment with respect to the level of immunosuppression that was continued after the rejection episode. The excellent outcomes in patients who received thymoglobulin in the absence of a biopsy suggests that either these individuals were highly responsive to therapy or that some milder cases of ACR received thymoglobulin as well because of clinician concerns about the extent of the injury. The favorable death-censored graft survival observed among nonbiopsy severe ACR should be interpreted cautiously, and the corresponding overall graft survival analyses may better capture combined risk of graft loss and death. Our classification of AMR was based on the treatment patterns, specifically the use of plasmapheresis and i.v. Ig, rather than histologic or immunologic criteria. Although this approach likely captures clinically significant AMR that was treated, it may preferentially identify more severe or overt cases and may fail to detect less intensively treated or atypical AMR presentations. As such, our phenotypes reflect real-world treatment decisions and clinical recognition rather than biopsy-defined diagnoses alone. These findings further emphasize that severity and subtype classifications using claims data should be done with temporal association of a allograft biopsy to the initiation of treatment where phenotypic discrimination was strongest and prognostic gradients were most consistent. Despite advancements in immunosuppressive drugs and surgical-techniques leading to improvements in short-term outcomes,4,20,21 improvements in long-term outcomes have stagnated.4,21,22 Improving long-term outcomes will require improvements in when rejection episodes occur, how they are treated and the subsequent management of the allograft. However, this will require large granular datasets that provide insight into the type of rejection and the choice of therapy. Prior investigations of rejection treatment outcomes have largely been limited to small, single-center cohorts because national registry data do not capture the therapeutic context of rejection episodes, only their occurrence and subsequent graft outcomes.23, 24, 25 Leveraging claims data provides a unique opportunity to link each episode of acute rejection to the specific antirejection therapies administered on a national scale, thereby anchoring prognostic analyses to real-world treatment patterns at scale. This granularity may enable systematic evaluation of how different therapeutic approaches influence graft survival across rejection phenotypes, offering an evidence base to guide treatment selection and to enhance patient counseling regarding expected outcomes following treated rejection. Lastly, our findings underscore the need for more granular information on rejection episodes and treatment response within the national transplant registry. At present, the Organ Procurement and Transplant Network registries capture rejection as a binary yes or no field without details on rejection subtype, timing, or therapeutic interventions, and reporting quality remains highly variable across centers suggesting that the limited granularity is further compounded by the inconsistent reporting. This lack of specificity precludes meaningful differentiation between patients who respond to therapy and those who do not, limiting the registry’s utility for both clinical benchmarking and research. Although our study points to the potential of administrative datasets such as that used in this study, improving the completeness and granularity of rejection reporting should be a priority for future iterations of the Organ Procurement and Transplant Network data collection framework. Such enhancements would enable more accurate national surveillance of allograft outcomes and foster data-driven strategies to optimize posttransplant care. This study is not without its limitations. First, although claims data enable national-scale analyses, they lack direct clinical detail and are inherently vulnerable to misclassification as previously noted. Our reliance on procedure and diagnosis codes may capture empiric treatments administered for nonrejection events, and biopsy timing was inferred from billing records rather than pathology reports. Although requiring biopsy before treatment improves confidence that the episode reflected clinically significant allograft pathology, it is possible that recurrent glomerular disease, like recurrent FSGS, may have been inadvertently included in the subset of patients thought to have an AMR. However, we believe this to be a relatively infrequent event. Similarly, although requiring biopsy within a 14-day window improved specificity and prognostic validity, this approach may exclude true rejections treated empirically or at centers with limited biopsy capacity, thereby underestimating overall incidence. Additionally, a key limitation of this study is the absence of a patient-level gold standard against which to directly validate our claims-based phenotypes (e.g., comparison with histopathology data from electronic health records). Such validation would ideally allow us to assess the performance of our algorithm including its sensitivity, specificity, and positive predictive value. However, our data use agreements preclude linkage of individual patients to identifiable institutional records, making traditional chart-level validation infeasible at a national scale. In this context, we instead adopted a construct validity framework, evaluating whether our phenotypes derived from treatment patterns show clinically coherent and prognostically meaningful gradients. Therefore, our approach represents a pragmatic alternative for studying rejection heterogeneity at scale, while highlighting the need for future data infrastructure that enables linkage between registry, claims, and clinical data sources for more rigorous validation. Second, our cohort was restricted to Medicare beneficiaries. Although most of the ESRD patients in the US are Medicare beneficiaries,26 our study does not consider Medicare advantage or patients using private insurance because of the absence of administrative claims from these groups. Finally, as with all observational analyses, unmeasured confounding such as differences in adherence may influence the observed associations. Broadly, our findings align with a growing recognition that real-world data sources can provide valuable evidence in settings where traditional clinical data are incomplete or infeasible to obtain at scale. Regulatory agencies, including the US Food and Drug Administration, have increasingly supported the use of real-world evidence to inform clinical and policy decision-making, particularly when randomized or fully adjudicated data are not available.27, 28, 29 In transplantation, where national registries lack granularity and linkage to detailed clinical data remains limited, administrative claims represent a pragmatic and scalable data source. Although such data require careful interpretation, our ability to recover clinically coherent and prognostically meaningful rejection phenotypes underscores their potential to complement existing data systems and to inform both clinical understanding and policy development. In summary, our study shows that, by using claims data, we can reproducibly assess type and severity of acute rejection events. Using this approach, our study clearly shows that type and severity of acute rejection influence short and medium-term outcomes. By distinguishing between type and severity in a national cohort, our approach may help bridge the gap between limitations present in registry surveillance and real-world clinical heterogeneity, highlighting the potential value of claims data to supplement the registry. Ultimately, these findings highlight the importance of strengthening the Organ Procurement and Transplant Network registry through sustained investment and more active oversight to ensure it remains a robust resource for guiding clinical practice and shaping transplant policy. Disclosure SM receives personal fees from Sanofi, Kidney International Reports (Deputy Editor), Visterra, DEKA research and Specialist Direct. The other authors of this manuscript have no conflicts of interest to report. Acknowledgments 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 USRDS or the US Government. The authors would also like to acknowledge the Mark and Jane Nielsen fund. Funding SM is supported by DK116066, EB032910, DK126739, DK130058. SAH was supported by NIDDK grant K23DK133729. Data Availability Statement The data that support the findings of this study are available on request from the USRDS. Restrictions apply to the sharing of these data because of privacy or ethical restrictions. Footnotes Supplementary File (PDF) Figure S1. Comparison of overall graft survival between acute rejection types identified using the US Renal Data System claims-based algorithm. Kaplan-Meier curves are unadjusted. Figure S2. Comparison of overall graft survival between acute rejection types identified using the US Renal Data System claims-based algorithm (A) with evidence of biopsy before treatment and (B) without evidence of biopsy before treatment. Kaplan-Meier curves are unadjusted. Table S1. Characteristics of recipient patients were independently identified in the US Renal Data System claims-based algorithm for AR episodes occurring within 12 months of transplant, by rejection phenotype. Table S2. Death-censored graft survival and overall graft survival at each after rejection timepoint for rejection episodes identified by claims-based algorithm. Table S3. Death-censored graft survival and overall graft survival at each postrejection timepoint for rejection episodes without evidence of biopsy within 14 days before the rejection treatment. Table S4. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients without evidence of biopsy within 14 days before the treatment. Table S5. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients with and without evidence of biopsy within 14 days before the treatment, controlling for biopsy-confirmation status. Table S6. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients with evidence of biopsy within 14 days before the treatment restricted to patients between the ages of 18 and 65. Table S7. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients without evidence of biopsy within 14 days before the treatment restricted to patients between the ages of 18 and 65. Table S8. Multivariable Cox proportional hazards model for 3-year overall graft survival among patients with evidence of biopsy within 14 days before the rejection treatment. Table S9. Multivariable cox proportional hazards model for 3-year overall graft survival among patients without evidence of biopsy within 14 days before the rejection treatment. Contributor Information Cameron Ekanayake, Email: cde2121@cumc.columbia.edu. Sumit Mohan, Email: sm2206@cumc.columbia.edu. Supplementary Material Supplementary File (PDF) Figure S1. Comparison of overall graft survival between acute rejection types identified using the US Renal Data System claims-based algorithm. Kaplan-Meier curves are unadjusted. Figure S2. Comparison of overall graft survival between acute rejection types identified using the US Renal Data System claims-based algorithm (A) with evidence of biopsy before treatment and (B) without evidence of biopsy before treatment. Kaplan-Meier curves are unadjusted. Table S1. Recipient characteristics of patients were independently identified in the US Renal Data System claims-based algorithm for AR episodes occurring within 12 months of transplant, by rejection phenotype. Table S2. Death-censored graft survival and overall graft survival at each after rejection timepoint for rejection episodes identified by claims-based algorithm. Table S3. Death-censored graft survival and overall graft survival at each after rejection timepoint for rejection episodes without evidence of biopsy within 14 days before the rejection treatment. Table S4. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients without evidence of biopsy within 14 days before the treatment. Table S5. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients without and without evidence of biopsy within 14 days before the treatment, controlling for biopsy-confirmation status. Table S6. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients with evidence of biopsy within 14 days before the treatment, restricted to patients between the ages of 18 and 65. Table S7. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients without evidence of biopsy within 14 days before the treatment restricted to patients between the ages of 18 and 65. Table S8. Multivariable Cox proportional hazards model for 3-year overall graft survival among patients with evidence of biopsy within 14 days before the rejection treatment. Table S9. Multivariable Cox proportional hazards model for 3-year overall graft survival among patients without evidence of biopsy within 14 days before the rejection treatment. References Associated Data Supplementary Materials Supplementary File (PDF) Figure S1. Comparison of overall graft survival between acute rejection types identified using the US Renal Data System claims-based algorithm. Kaplan-Meier curves are unadjusted. Figure S2. Comparison of overall graft survival between acute rejection types identified using the US Renal Data System claims-based algorithm (A) with evidence of biopsy before treatment and (B) without evidence of biopsy before treatment. Kaplan-Meier curves are unadjusted. Table S1. Recipient characteristics of patients were independently identified in the US Renal Data System claims-based algorithm for AR episodes occurring within 12 months of transplant, by rejection phenotype. Table S2. Death-censored graft survival and overall graft survival at each after rejection timepoint for rejection episodes identified by claims-based algorithm. Table S3. Death-censored graft survival and overall graft survival at each after rejection timepoint for rejection episodes without evidence of biopsy within 14 days before the rejection treatment. Table S4. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients without evidence of biopsy within 14 days before the treatment. Table S5. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients without and without evidence of biopsy within 14 days before the treatment, controlling for biopsy-confirmation status. Table S6. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients with evidence of biopsy within 14 days before the treatment, restricted to patients between the ages of 18 and 65. Table S7. Multivariable Cox proportional hazards model for 3-year death-censored graft failure among patients without evidence of biopsy within 14 days before the treatment restricted to patients between the ages of 18 and 65. Table S8. Multivariable Cox proportional hazards model for 3-year overall graft survival among patients with evidence of biopsy within 14 days before the rejection treatment. Table S9. Multivariable Cox proportional hazards model for 3-year overall graft survival among patients without evidence of biopsy within 14 days before the rejection treatment. Data Availability Statement The data that support the findings of this study are available on request from the USRDS. Restrictions apply to the sharing of these data because of privacy or ethical restrictions.