Monitoring for relapse with an artificial intelligence (AI)-powered peripheral blood-based tool called AlloHeme demonstrated greater sensitivity in predicting relapse after hematopoietic cell transplantation (HCT) in patients with acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS) than standard donor engraftment (chimerism) or measurable residual disease testing. Findings from a 2-year analysis of the ACROBAT study were presented during the 2026 Transplantation & Cellular Therapy Tandem Meetings of ASTCT and CIBMTR.¹

“The ACROBAT results suggest that blood-based AlloHeme monitoring offers a simple, effective strategy for identifying relapse risk and enabling early intervention [in patients with acute myeloid leukemia and myelodysplastic syndrome],” said presenting author Ran Reshef, MD, MSc, Professor of Medicine at Columbia University and Director of Cell Therapy, Blood and Marrow Transplantation Program at Herbert Irving Comprehensive Cancer Center.

Background and Study Methods

Dr. Reshef explained that existing tools for monitoring relapse and measurable residual disease have limitations. For example, short tandem repeat–polymerase chain reaction–based chimerism testing may be useful for assessing engraftment, but it is not leukemia-specific and has limited sensitivity. Multiparametric flow cytometry is rapid and widely available but cannot be standardized because it is highly operator-dependent, has limited sensitivity, and requires bone marrow sampling.²

“We clearly need a tool that is highly sensitive, universal, and ideally noninvasive,” Dr. Reshef said.

Researchers conducted a prospective, observational cohort study to evaluate AlloHeme, a blood-based, next-generation sequencing–driven multi-analyte test, for detecting relapse after transplantation in patients with acute lymphoblastic leukemia, acute myeloid leukemia, and myelodysplastic syndrome.

In the ACROBAT study, 285 patients were enrolled and monitored for relapse for up to 2 years after HCT, with testing performed biweekly for the first 3 months, monthly for the next 3 months, and quarterly from months 9 through 24. The assay’s performance was assessed at 24 months using leave-one-out cross-validation to evaluate its robustness.

Testing involved peripheral blood sampling with next-generation sequencing of more than 400 single-nucleotide polymorphisms, supported by a proprietary AI algorithm. Dr. Reshef explained that the algorithm incorporates whole-blood chimerism and sorted myeloid cell chimerism (CD33) to detect early relapse. At least two consecutive tests were required to generate a definitive positive or negative result.

The study focused on patients with acute myeloid leukemia or myelodysplastic syndrome, yielding an analytical cohort of 198 participants. (Too few patients with acute lymphoblastic leukemia experienced relapse to be evaluated for the study.) Of these, 40 experienced disease relapse, 14 withdrew or were lost to follow-up, 26 had nonrelapse mortality, and 118 completed 2 years of follow-up. Dr. Reshef noted that this cohort was representative of a standard transplant population.

Results

Patients who tested positive with AlloHeme at 2, 3, or 6 months posttransplant were significantly more likely to relapse than those who tested negative, “indicating that the positive test is an excellent marker of impending relapse,” Dr. Reshef said.

At 2 months posttransplant, the hazard ratio (HR) for relapse was 5.9 (95% confidence interval [CI] = 2.1–16.8; P < .001); at 3 months, it was 6.3 (95% CI = 2.6–15.4; P < .001); and at 6 months, 11.9 (95% CI = 5.0–28.2; P < .001).

For relapse or death, the HR was 5.2 at 2 months (95% CI = 2.5–11.2; P = .010), 6.2 at 3 months (95% CI = 3.1–12.4; P < .001), and 4.1 at 6 months (95% CI = 1.7–9.6; P < .001).

In terms of assay performance, 43 patients had a positive AlloHeme result; of these, 9 (21%) did not experience relapse. Among the 115 patients with a negative result, 6 (5%) relapsed.

The test achieved 85% sensitivity and 92% specificity, with a positive predictive value of 79% and a negative predictive value of 95%. The area under the receiver operating characteristic curve was 0.89 and was consistent across subgroups. Dr. Reshef noted that these findings were based on a median of 11 tests per patient, rather than the protocol-defined 14, as compliance with the recommended testing schedule was challenging. He commented that assay performance might have improved with better compliance, more frequent testing, and longer follow-up.

AlloHeme detected relapse a median of 41 days before clinical relapse (range, 19–85 days). The mean time was 101 days (standard deviation, 153 days). No significant difference in lead time was observed between the groups of patients with acute myeloid leukemia and myelodysplastic syndrome.

Comparative Context

Compared with real-world approaches for detecting relapse, the ACROBAT study showed that AlloHeme performed better. Real-world chimerism testing showed a sensitivity of 60% (P < .05) and a median lead time of 0 days, whereas next-generation sequencing-based chimerism testing had a sensitivity of 53% and a median lead time of 16 days, according to the study authors.

Compared with real-world measurable residual disease testing, AlloHeme demonstrated superior performance. Bone marrow–based multiparametric flow cytometry had a sensitivity of 54% (P < .05) and a negative predictive value of 85%, whereas molecular measurable residual disease testing had a sensitivity of 48% (P < .05) and a negative predictive value of 82%. Overall, measurable residual disease testing had a sensitivity of 57% (P < .05) and a negative predictive value of 85%.

Source – ASCO Post