AI-Powered Diagnostic Advances Organ Transplant Surveillance

By Deborah Borfitz 

September 29, 2026 | Precision medicine company CareDx has released a string of AI-driven diagnostic platforms since July of 2025 with the launch of AlloSure Plus. The patient-specific kidney transplant rejection risk prediction tool delivers actionable insights to reduce unnecessary biopsies, enable earlier treatment, and personalize care, thereby helping transplant centers improve outcomes, for which CareDx was awarded an Innovative Practices Award earlier this year at the Bio-IT World Conference & Expo.  

Since then, the company has enlarged the AI-powered ecosystem to include AlloHeme, a surveillance solution designed for bone marrow transplant relapse tracking in blood cancers, and VANTx, a platform for solid organ transplant-related cohort analysis, clinical trial design, and research, as was reported at the event by Jing Huang, chief data and AI officer. “We move fast because we know those patients need us,” she says. 

Being a transplant patient is a complex journey that begins with getting listed, followed by the matching and surgery phase and subsequent lifelong immunosuppressant management. One wrong move along the trajectory can have costly or fatal consequences, says Huang. 

The award-winning AlloSure Plus integrates clinical variables commonly used to predict organ stress— eGFR (estimated glomerular filtration rate), history of prior rejection, proteinuria, and kidney graft instability—with a cutting-edge biomarker called donor-derived, cell-free DNA. AI is used to generate a synergistic prediction score, Huang explains, which gets pulled directly into the Epic electronic health record for easy access by treating physicians.  

AlloSure Plus has been clinically validated in over 2,700 kidney transplant biopsies, covering a heterogenous population that includes pediatric and adult patients at U.S. and European transplant centers, first-time and sequential transplant patients, and biopsies performed for cause or as surveillance (Nature Medicine, DOI: 10.1038/s41591-024-03087-3). This means its predictions are “relevant, applicable, and generalizable,” says Huang. The entire spectrum of transplant rejection types is represented, not just the “easy cases.”  

The accuracy of predictions made by AlloSure Plus was measured visually using the ROC (receiver operating characteristic) curve where 0.5 represents a random guess. The cell-free DNA shed from the donor organ, by itself, is already “way better than any clinical variable by itself [just over 0.7],” but when combined with AI reached almost 0.82—a first in the world of transplant rejection prediction, she notes. 

More recent data presented at the 2026 American Transplant Congress showed that AlloSure Plus differentiated biopsy-confirmed rejection from non-rejection, with an AUROC of 0.79, further supporting its role in enhancing rejection risk assessment beyond donor-derived cell-free DNA alone.  

GPS-style Relapse Prediction 

Landmark results with the AlloHeme platform were presented in February 2026 at the Tandem Meetings, one of the largest scientific conferences dedicated to bone marrow transplantation, cellular therapy, and gene therapy. As with AlloSure Plus, the biomarker platform uses next-generation sequencing (NGS) and a proprietary AI algorithm, but for the prediction of relapse after bone marrow transplant, says Huang.  

The algorithm functions similarly to a GPS in that relapse prediction scores are dynamically tied to how far out the patients are from receiving a bone marrow transplant and their NGS level at testing times measuring the balance between the recipient's cancer-prone DNA and the donor’s healthy transplanted DNA in the blood. “It’s personalized and follows them through their whole [post-transplant] journey,” she says.  

Blood-based NGS monitoring has a risk-adapted frequency for the first 24 months after the transplant, says Huang. While relapses typically happen within the first year, a couple percent occur between the first and second year and, if patients are relapse-free after that point, they are usually out of the woods. 

Based on an evaluation via the ACROBAT prospective observational study, AlloHeme achieved a 0.89 ROC curve value, based on its sensitivity (85%) and specificity (92%), Huang reports. Among the nine false-positives, one patient had donor lymphocyte infusion, two patients started maintenance therapy, and one relapsed literally one week after the two-year follow-up period. Four of the six false-negatives missed a couple of their scheduled tests, she adds, meaning lack of testing adherence impeded AlloHeme’s relapse detection capabilities.  

In terms of lead time, AlloHeme was also found to perform better than standard-of-care measurable residual disease testing in predicting relapse risk, says Huang. “Basically, the test becomes very useful even just after two months of the treatment.” AlloHeme detected a positive signal a median of 41 days before clinical relapse, providing a “golden window” of opportunity for physicians to intervene and prevent that from occurring.   

VANTx is the latest AI-powered platform released by CareDx, which uses data and AI to provide a vantage point for better data-driven insights, she says. It is already being used by more than 100 kidney transplant centers and over 30 heart transplant centers, and lung features are slated for development later this year. It resides at the top of the ecosystem, using AI to analyze aggregate data.  

Actionable Insights 

CareDx innovates in partnership with transplant centers, says Huang, using dynamic risk prediction to guide patient management using traffic light colors to signal if relapse risk is high (red), intermediate (yellow), or low (green). At one of those centers, patients transplanted between 2023 and early 2024 were compared to those transplanted in the latter half of 2024 and all of 2025 over the course of one year post-treatment.  

VANTx reveals important insights that would otherwise be difficult to detect. At this center, patients in the earlier management period were initially distributed as 6% high risk, 37% intermediate risk, and 57% low risk. In the later period, the high-risk population was nearly twice as large at baseline (11%), while 23% were intermediate risk and 67% were low risk. The longitudinal trends, particularly when contrasted across the two periods, reveal an unexpected yet compelling story, Huang says.  

In the earlier period, the proportion of high-risk patients more than doubled, rising from 6% to over 14% by one year, she continues. In contrast, during the later period, the high-risk proportion declined sharply—from 11% to 2% by month three and to nearly zero thereafter. Together, these contrasting trajectories highlight a substantial, previously unrecognized improvement in patient outcome.  

It took some investigation to find the explanation, says Huang. Physicians were initially only actively treating the high-risk patients but more recently started doing the same for intermediate-risk individuals “and it’s working.” The high- and intermediate-risk patients have all since moved to the low-risk category, and the story of how those data-driven, center-level insights were optimized will be the subject of a forthcoming paper.  

At the patient level, Huang points to findings from the KOAR (Kidney Allograft Outcomes AlloSure Registry) study evaluating the clinical utility of AlloSure, the donor-derived cell-free DNA blood test upon which AlloSure Plus is built, in kidney transplant recipients. The focus was on how the biomarker helps clinicians assess and tailor immunosuppression medication doses based on real-time risk.    

The biomarker is enabling a shift in dosing toward a personalized medicine model based on actual biological response rather than rigid demographic formulas such as weight and age, she says. The reporting tools for AlloSure utilize different sized circles to indicate if the dose should be lower, the same, or higher than the standardized dose, while the traffic light colors are used to denote rejection risk. 

AlloSure almost always detects elevated risk before its segue into a serious medical situation, but “the testing is only as good as how good you adhere to the test,” Huang emphasizes. In the KOAR study, rejection risk was signaled more than three weeks before a full clinical episode of acute rejection.