Liver transplant candidates whose bodies are aging faster than their birthdays suggest may be dying on the waitlist at lower MELD scores than anyone realized, according to a new study that introduces a machine learning-derived biomarker designed to expose exactly that hidden vulnerability. The research, published in Frontiers in Transplantation, describes the BioWaitIndex, or BWI, a composite measure distilled from twelve routinely collected clinical biomarkers that predicted waitlist death and clinical deterioration in a cohort of 1,502 liver transplant candidates even when their MELD scores did not. The findings arrive at a moment when the gap between organ supply and demand in the United States remains stark: in 2023 alone, nearly 14,700 candidates were added to the liver transplant waiting list, while approximately 10,660 transplants were performed.
The MELD score, the Model for End-Stage Liver Disease, has anchored liver allocation since 2002 and remains the cornerstone of the system. It predicts near-term mortality effectively when assessed dynamically during waitlist follow-up, using objective laboratory measures of liver and kidney dysfunction: bilirubin, INR, creatinine, and sodium. But MELD was designed to capture acute severity from values obtained near the time of organ offer, not a patient’s longer-term trajectory or underlying physiologic reserve. A single MELD value measured at initial transplant evaluation, the new study shows, may miss candidates whose multisystem vulnerability is substantial but whose liver-specific numbers look deceptively reassuring.
The technical architecture behind BWI reflects a deliberate design choice. Rather than training the model in patients with end-stage liver disease, the researchers built it in the general population using the National Health and Nutrition Examination Survey, or NHANES, a nationally representative survey of community-dwelling Americans. A random forest regression model with 500 trees, a minimum node size of five, and four variables considered at each split was trained on a development cohort of 13,830 participants, using chronological age as the training target. This anchoring to age allowed the model to learn expected age-related patterns across multiple biomarker systems without learning liver-disease-specific abnormalities. BWI was then calculated as the difference between the model’s predicted value and the candidate’s chronological age: positive values indicate a biomarker profile predicted to be older than expected, negative values a profile younger than expected.
The twelve biomarkers incorporated into the index span pulmonary, cardiovascular, kidney, metabolic, inflammatory, nutritional, and hematologic domains: albumin, C-reactive protein, white blood cell count, glucose, glycated hemoglobin, alkaline phosphatase, blood urea nitrogen, creatinine, systolic blood pressure, forced expiratory volume in one second, lymphocyte count, and mean corpuscular volume. All are routinely measured in clinical practice and available in the electronic medical record. The trained model was first temporally validated in a separate, non-overlapping sample of 8,151 participants from subsequent NHANES cycles, where BWI was significantly associated with all-cause mortality. It was then applied, without retraining, to an independent, IRB-approved cohort of 1,502 adult liver transplant candidates from Houston Methodist Hospital listed between June 2016 and November 2022, with outcomes followed through December 2024.
The clinical validation results were striking. In multivariable competing-risks analyses with BWI and MELD both measured at initial evaluation, each unit increase in BWI was associated with a greater likelihood of transplantation and a lower risk of waitlist death or clinical deterioration. More tellingly, MELD measured at initial evaluation was not significantly associated with subsequent death or clinical deterioration, whereas BWI remained a robust independent predictor at the same timepoint. The two scores showed only a modest correlation, with low shared variance, indicating that they capture related but largely distinct dimensions of candidate status. Adding BWI to MELD at evaluation improved net reclassification for waitlist mortality by 31.2 percent and for transplantation by 11.1 percent, with consistent associations across sexes.
Permutation-based feature importance analysis revealed that the leading contributors to BWI were systemic rather than liver-specific markers. Forced expiratory volume in one second, systolic blood pressure, and blood urea nitrogen emerged as the dominant drivers, spanning pulmonary, cardiovascular, and renal domains. In the transplant context, FEV1 likely captures respiratory muscle wasting, hepatopulmonary syndrome, and portopulmonary hypertension, all recognized contributors to waitlist mortality independent of MELD. Systolic blood pressure reflects hemodynamic reserve in the setting of cirrhotic cardiomyopathy and splanchnic vasodilation, while blood urea nitrogen integrates renal function, nutritional status, and catabolic state, particularly relevant given the prevalence of hepatorenal syndrome in advanced cirrhosis.
Perhaps the most clinically consequential finding came from the analysis of candidates who died or were removed for clinical deterioration. Those with below-median BWI accounted for a disproportionate share of waitlist deaths, 59.4 percent of the 431 patients who experienced the outcome, despite comprising half the cohort. Their mortality rate was 34.6 percent compared to 23.6 percent among high-BWI patients. Crucially, low-BWI patients who died had significantly lower MELD scores at the time of death, a median of 22 versus 30, and spent more than twice as long on the waitlist, a median of 269 versus 121 days. A dose-response relationship was observed across BWI quartiles: among deaths, median MELD at event increased progressively from the lowest to the highest quartile, while median time on the waitlist decreased from 286 to 88 days.
The researchers emphasize that BWI is not a proxy for chronological age, despite the model being trained to predict age from biomarker patterns. The resulting index captures deviation from expected multisystem function: two candidates of identical chronological age may have markedly different BWI values reflecting divergent systemic reserves. The associations with waitlist outcomes were retained after adjustment for chronological age, addressing the regression-to-the-mean dependence that affects other machine-learning age estimators. The team also evaluated potential algorithmic bias through sex-stratified models, finding identical hazard ratios for males and females for both transplantation and mortality, a result directly relevant to the ongoing policy debate around sex-based disparities in liver allocation.
Previous efforts to quantify systemic reserve beyond organ-specific scores have relied on frailty indices such as the Liver Frailty Index and Fried Frailty Phenotype, which require in-person physical performance assessments that may not be feasible in severely deconditioned patients and are subject to interobserver variability. Molecular assays such as DNA methylation-based measures offer objective quantification of multisystem physiologic status but are constrained by small sample sizes, specialized processing requirements, high cost, and lack of scalability. A scoping review of biomarkers of biological aging in solid organ transplant recipients found research heavily biased toward kidney transplantation, which accounted for 23 of 32 included studies, while only three focused on liver transplant recipients. BWI, by contrast, uses routinely collected clinical measurements and was evaluated in 1,502 waitlisted candidates without requiring specialized molecular assays.
The practical implications are considerable. BWI could serve as an early screening tool at the time of transplant evaluation to identify candidates who may benefit from closer monitoring or expedited reassessment despite moderate MELD scores. Its derivation from routine clinical data makes it potentially suitable for future electronic health record integration after multicenter validation, without requiring machine-learning expertise from end users. The authors caution that the clinical-validation cohort was retrospective and derived from a single transplant center, and that additional prospective and multicenter external validation is required before clinical implementation. Missing data, analyzed with multiple imputation by chained equations, were handled with sensitivity analyses that confirmed the consistency of the findings. Still, the study represents a step toward more holistic risk stratification of liver transplant candidates, one that leverages machine learning and routinely collected data to provide prognostic information that complements the established MELD score, particularly at the time of initial evaluation when the opportunity for clinical intervention is greatest.
Subject of Research: Medicine
Article Title: BioWaitIndex: a novel machine learning-derived biomarker for improved prediction of liver transplant waitlist outcomes
Article References: Patel, K., Connor, A. A., Sanghvi, M., Elaileh, A., Rizk, E., Semaan, S., Dib, Y. E., Todd, J., Moore, L. W., Succar, L., Simon, C. J., Cheah, Y. L., Mobley, C. M., Saharia, A., Patidar, K. R., Kodali, S., Victor, D. W., III, Lindberg, S., Casarin, S., … Ghobrial, R. M. (2026). BioWaitIndex: a novel machine learning-derived biomarker for improved prediction of liver transplant waitlist outcomes. Frontiers in Transplantation, 5, Article 1964045. https://doi.org/10.3389/frtra.2026.1964045
Image Credits: AI Generated
DOI: 10.3389/frtra.2026.1964045
Keywords: liver transplantation, machine learning, BioWaitIndex, MELD, waitlist mortality, biomarker, random forest, NHANES, risk stratification, competing-risk regression, biological aging, transplant allocation
News Source: Teresa Odom. (October 16, 2026). Machine Learning Biomarker BioWaitIndex Predicts Liver Transplant Waitlist Outcomes Beyond MELD. Scienmag.
