A Multivariable Mixed-Effects Framework for Predicting Disease Progression in Acute Myeloid Leukemia Using Peripheral Hematology, FLT3 Expression, and Patient Compliance
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
The clinical course of acute leukemias and related hematologic neoplasms is defined by aggressive clonal proliferation, severe hematopoietic disruption, and recurrent molecular alterations. Although serial bone marrow biopsies, multiparameter flow cytometry, and next-generation sequencing remain standard surveillance modalities, procedural invasiveness, high costs, and operational turnaround times limit their frequency. Minimally invasive, scalable biomarkers—integrating routine hematologic indices, targeted gene expression, circulating oncoprotein levels, and patient compliance metrics—may provide an accessible framework for dynamic monitoring.
Methods
In this longitudinal cohort study of 66 patients across 272 serial encounters (discrete follow-up intervals: days 0, 14, 41, 71, and 78), we assessed the prognostic value of peripheral complete blood counts (white blood cell count [WBC], hemoglobin [Hb]), molecular indicators ( FLT3, NPM1 , and DNMT3A transcript levels; targeted soluble FLT3 quantification), and medication adherence. The primary clinical endpoint was longitudinal progression to critical deterioration or disease-related mortality. Generalized linear mixed-effects models (GLMM) with patient-specific random intercepts were constructed to account for intra-subject repeated measures. Predictors were standardized per 1-SD increment. Model performance was assessed via the Akaike Information Criterion (AIC) and the Concordance index (C-index).
Results
Leukocyte expansion correlated inversely with hemoglobin concentration ( r = -0.877, p < 0.001), reflecting bone marrow infiltration and secondary erythroid suppression. Multivariable mixed-effects modeling confirmed that elevated WBC count (adjusted odds ratio [aOR] = 1.52 per 1-SD increase; 95% CI: 1.24–1.86; p < 0.001) and reduced medication adherence (aOR = 0.58; 95% CI: 0.44–0.76; p < 0.001) independently predicted adverse clinical transition. Integrating targeted molecular markers ( FLT3 expression and soluble protein levels) with routine hematologic indices and adherence demonstrated superior calibration and discrimination over single-domain models (ΔAIC ≥ 116; Harrell’s C-index = 0.81).
Conclusions
Readily accessible peripheral hematologic indices, targeted molecular quantification, and medication adherence provide complementary prognostic information regarding disease activity. Integrating these low-burden parameters into clinical workflows offers an efficient, scalable paradigm for early risk stratification in hematologic malignancies.
