Chen, Dong

ABSTRACT We examined the prognostic value of routine NGS data in newly diagnosed acute myeloid leukemia (AML) treated with intensive (7 + 3 backbone) induction, specified by AML subtype. A contemporary (2015–2025) series of 545 Mayo Clinic patients (median age 56 years, females 44%) was considered. Median follow‐up was 49 months with 341 (63%) allogeneic hematopoietic stem cell transplantations (AHSCT) recorded. AML subtypes included core‐binding factor (CBF; N = 72; 13%), primary non‐CBF ( N = 403; 74%), post‐myelodysplastic syndromes (MDS) or post‐myelodysplastic/myeloproliferative neoplasms (post‐MDS‐MDS/MPN; N = 28; 5%), post‐MPN ( N = 15; 3%), and therapy‐related (t‐AML; N = 27; 5%). Corresponding complete remission rates, with/without count recovery (CR/CRi), were 89%, 76%, 64%, 27%, and 78% ( p < 0.01) and 5‐year transplant‐censored survival rates 68%, 57%, 21%, 0%, and 55% ( p < 0.01). In multivariable analyses, the prognostic value of specific mutations was mostly limited to primary non‐CBF AML where adverse karyotype (OR 1.9; p = 0.04) predicted inferior and FLT3 ‐ITD (OR 0.4; p < 0.01) or NPM1 MUT / FLT3 WT (OR 0.1; p < 0.01) superior CR/CRi while KRAS MUT (HR 8.8; p < 0.01), TP53 MUT (HR 5.4; p < 0.01), and TET2 MUT (HR 2.3; p = 0.01) predicted inferior and NPM1 MUT / FLT3 WT (HR 0.3; p = 0.01) superior survival. Prognostication in intensively‐treated AML should start with subtype specification and recognition of the limited value of NGS in non‐primary AML. Post‐MPN AML is particularly associated with dismal outcomes and should be prognostically distinguished from post‐MDS‐MDS/MPN AML. In primary non‐CBF AML, in addition to previously established risk factors, the favorable impact of FLT3 ‐ITD on achieving CR/CRi and the unfavorable impact of KRAS MUT and TET2 MUT on transplant‐censored survival were noted and require confirmation from additional studies.

ABSTRACT Background Platelet factor 4‐polyanion enzyme‐linked immunosorbent assays (ELISAs) are considered highly sensitive for diagnosing heparin‐induced thrombocytopenia (HIT), such that current practice guidelines recommend use of ELISA‐negative results to exclude HIT. Once HIT is ruled out, alternative, non‐heparin‐based anticoagulant treatments are ceased, and heparin reintroduction frequently occurs. Methods Antigen‐based and PF4‐dependent functional testing were used to study PF4/polyvinylsulfonate ELISA‐negative platelet‐activating antibodies in HIT‐suspected patients. Results Three patients with clinical presentations consistent with HIT tested negative in an ELISA using PF4‐polyvinylsulfonate (PF4/PVS), an antigenic target very commonly used for HIT antibody screening. All three patients demonstrated PF4‐dependent platelet activation in functional testing that was sensitive to blockade of platelet FcγRIIa receptors and inhibited by high concentrations of heparin, consistent with pathogenic HIT antibodies. Functional testing‐based screening of 500 ELISA‐negative patients identified three additional patients whose sera activated platelets in a PF4‐ and FcγRIIa‐dependent manner and had clinical histories consistent with HIT. Five of the six ELISA‐negative HIT patients were re‐exposed to heparin, which precipitated a decrease in platelet counts in all re‐exposed patients, and one patient developed a new thrombus. Conclusions Recognition of ELISA‐negative HIT is critical to avoid harm due to the cessation of alternative anticoagulation therapy and re‐exposure of these patients to heparin.

Abstract Background Interpretation of multi-analyte laboratory profiles, such as lupus anticoagulant (LAC) testing, requires integration of numerous test results and clinical information. However, this process is time-consuming, requires domain knowledge, and has limited scalability. Given that rule-based systems are rigid and difficult to maintain and large language models pose risks of “hallucinations,” we developed and evaluated a novel retrieval-based approach to automate LAC profile interpretation to improve efficiency while maintaining fidelity and reliability for clinical use. Methods Our method processes laboratory values through ensemble random forest (RF) models to retrieve the most similar expert interpretation from a curated database of 347 previously interpreted LAC profiles. The system's performance was assessed through technical accuracy metrics, a work flow efficiency timing study, and expert review of interpretation accuracy for held-out cases. Results Technical validation demonstrated high accuracy in retrieving correct interpretative elements (median: 93% accuracy, interquartile range [IQR]: 67% to 100%). Retrieval accuracy was highly correlated with the prevalence of a given element (ρ = 0.8307). In a timing study of 93 consecutive LAC profiles, ensemble-assisted interpretation achieved a median time of 22 s per report (IQR: 16 s to 31 s), representing a 78.6% total reduction compared to manual interpretation (median: 118 s, IQR: 96 s to 172 s; t = 18.61, P &lt; 0.001). Furthermore, 92.5% of AI-assisted interpretations were completed within 60 s, exceeding the target of 85%. Expert review confirmed 100% concordance for LAC presence/absence. Conclusions The proposed approach significantly improves efficiency of complex laboratory interpretation for LAC profiles, while maintaining consistency and high fidelity. Future studies will explore the method’s inherent extensibility to the interpretation of other multi-analyte laboratory profiles.

Abstract Objective Gastrointestinal (GI) symptoms are common in systemic mastocytosis (SM). We hereby report the clinicopathologic and molecular features of GI mastocytosis. Methods The study includes 104 patients with suspected SM. Of 258 GI biopsy ­specimens, 33 (20 patients, 19%) were mastocytosis-positive, with 19 (16 patients) initially missed. Results Tryptase (n = 29) was weakly positive/negative in two-thirds of the mastocytosis-positive GI biopsy specimens. Additional immunohistochemistry (n = 23) showed positive mast cell expression for CD30 (n = 2), CD123 (n = 7), and PD-L1 (n = 12), as well as increased PD-1–positive cells (n = 4). Sixty patients had positive bone marrow (BM) for SM (SM-BM+), including 19 with positive GI biopsy specimens (SM-GI+). Almost all (43/44) SM-BM– patients were SM-GI–. The SM-GI vs SM-BM status showed high positive predictive value (95%) and specificity (98%). Patients with SM with mastocytosis-positive vs mastocytosis-negative GI biopsy specimens had higher serum tryptase levels (P = .04). KIT D816V mutation was detected in 7 of 13 and 12 of 13 mastocytosis-positive GI biopsy specimens using allele-specific polymerase chain reaction and droplet digital polymerase chain reaction (ddPCR), respectively. Conclusions GI mastocytosis is highly predictive of SM in BM; however, most SM-BM+ patients are SM-GI–. The SM-BM– patients are very unlikely to be SM-GI+. Because GI neoplastic mast cells are often tryptase negative, diagnostic evaluation should include CD117 and CD25 immunohistochemistry, as well as KIT D816V mutation analysis by ddPCR as needed.

ABSTRACTThe current study includes 910 patients with systemic mastocytosis (SM) seen at the Mayo Clinic from 1968 to 2024. The primary objective was to examine the prognostic contribution of the International Consensus Classification (ICC), in the context of the Mayo Alliance Prognostic System (MAPS) for SM. World Health Organization classification (WHO‐HAEM5) subcategories included (i) indolent/smoldering SM (ISM/SSM; N = 518), (ii) SM associated with another hematological neoplasm (SM‐AHN; N = 273), with the latter including both myeloid and lymphoid neoplasms, (iii) aggressive SM (ASM; N = 106), and (iv) WHO‐defined mast cell leukemia (MCL; N = 13), which included mast cells with both “mature” and “immature” morphology. The ICC‐defined subcategories were mostly similar with the exception that SM‐AHN was replaced by SM associated with another myeloid neoplasm (SM‐AMN; N = 235) and WHO‐defined MCL was replaced by ICC‐defined MCL (N = 8), which included only those with immature MC morphology. Overall survival (OS) was similar between WHO‐defined MCL (median 1.8 years) vs. SM‐AHN (median 2.3 years; p = 0.3) but significantly different between ICC‐defined MCL (median 0.08 years) vs. SM‐AMN (median 2.0 years; p < 0.01). Significant difference in OS was also apparent between ICC‐defined SM‐AMN and SM associated with lymphoid neoplasm (SM‐ALN; median 8.1 years; HR 3.4; p < 0.01). Multivariable analysis confirmed the inter‐independent prognostic contributions of both ICC‐defined morphologic subcategories and MAPS risk variables, including age > 60 years, anemia, alkaline phosphatase > ULN, and platelet count < 150 × 109/L (p < 0.03 in all instances); the same was not true for WHO‐HAEM5‐defined SM subcategories. The addition of mutational information into the multivariable model resulted in ousting anemia and inclusion of ASXL1 (p < 0.01), SRSF2 (p = 0.01), and NRAS (p = 0.01) mutations as additional risk factors. Classification of SM by ICC is prognostically more accurate, compared to WHO‐HAEM5, and strengthens the prognostic contribution of morphology to current clinical and molecular risk models.

Abstract Background The lupus anticoagulant (LAC) is an important laboratory criterion in the diagnosis of antiphospholipid antibody syndrome. LAC testing at the Mayo Clinic Special Coagulation Laboratory includes up to 13 tests, which are interpreted by trained physicians to identify the presence of LAC and rule out anticoagulant interferences. This feasibility study explored the use of two deep neural network (DNN) architectures for multilabel classification of LAC profiles as a first step toward automating interpretation. Methods Seven thousand two hundred and two retrospective cases were randomly split (64:16:20) for training, validation, and test, respectively. LAC positivity by dilute Russell's viper venom time (LAC-DRVVT) and activated partial thromboplastin time (LAC-APTT) and the presence of warfarin (WAR) and heparin (HEP) were adjudicated by one expert. DNN architectures included: single-output DNNs using domain-knowledge input feature selection and a single-column multioutput DNN using all 13 inputs. Results Domain-knowledge-naïve multioutput DNN achieved similar or improved performance for all 4 label prediction tasks compared with domain knowledge optimized DNNs: F1 scores of 0.977 vs 0.968 for LAC-DRVVT, 0.954 vs 0.945 for LAC-APTT, 0.961 vs 0.957 for HEP, and 0.995 vs 0.977 for WAR, respectively. Conclusions The comparable performance of the 4 domain knowledge optimized DNNs and the multioutput DNN using all 13 input features suggests that the DNN may learn feature importance or mapping to a task without explicit input selection. Given its relative simplicity and versatility, the multioutput DNN is the preferred choice for implementation in a clinical laboratory to standardize LAC diagnosis.

ABSTRACTIntroductionVon Willebrand factor (VWF) multimer analysis is essential for diagnosing and classifying von Willebrand disease (VWD) but requires expert interpretation and is subject to inter‐rater variability. We developed an automated image analysis pipeline using deep learning to improve the reproducibility and efficiency of VWF multimer pattern classification.MethodsWe trained a YOLOv8 deep learning model on 514 gel images (6168 labeled instances) to classify VWF multimer patterns into 12 classes. The model was validated on 192 images (2304 instances) and tested on an independent set of 94 images (1128 instances). Images underwent preprocessing, including histogram equalization, contrast enhancement, and gamma correction. Two expert raters provided ground truth classifications.ResultsThe model achieved 91% accuracy compared to Expert 1 (macro‐averaged precision = 0.851, recall = 0.757, F1‐score = 0.786) and 87% accuracy compared to Expert 2 (macro‐averaged precision = 0.653, recall = 0.653, F1‐score = 0.641). Inter‐rater agreement was very high between experts (κ = 0.883), with strong agreement between the model and Expert 1 (κ = 0.845) and good agreement with Expert 2 (κ = 0.773). The model performed exceptionally well on common patterns (F1 > 0.93) but showed lower performance on rare subtypes.ConclusionAutomated VWF multimer analysis using deep learning demonstrates high accuracy in pattern classification and could standardize the interpretation of VWF multimer patterns. While not replacing expert analysis, this approach could improve the efficiency of expert human review, potentially streamlining laboratory workflow and expanding access to VWF multimer testing.