Evaluating recipient chimerism levels to identify a threshold above which patients with leukemia are at high risk of relapse after allo-HCT. (A) Highly sensitive chimerism measurements in BMA, PB-derived CD33⁺ myeloid cells, and PB-derived CD3⁺ T cells in the first 540 days after allo-HCT, excluding time points at which relapsed disease was already present and/or time points that self-fulfilled the relapse “prophecy” (see “Methods”). P values (B) and HRs (C) from the Cox proportional hazard regression (unadjusted and adjusted for confounding covariates), with chimerism as a time-dependent covariate, computed over a range of thresholds of chimerism proportions incrementally increasing between 1.0e–5 and 0.5. (D) Probability (cumulative incidence) of relapse with follow-up in a 5-year window (1825 days) after allo-HCT for a representative chimerism proportion threshold. Ticks represent individuals who were censored using the interval censoring approach. A generalized additive model was used to smooth fit the P values (blue line), which were truncated when HRs gave infinite CIs; the HRs gray ribbon represents the 95% CI; Chim, chimerism proportion (recipient); n.max, the maximum number of patients at risk; Not-Det, not detected; adj, adjusted.