Seheult, Jansen N.

Background: Flow cytometry (FC) is widely utilized for the identification of minimal residual disease (MRD) in B acute lymphoblastic leukemia (BALL), providing an essential laboratory indicator for prognostic assessment and therapeutic management. Unfortunately, the high level of expertise required, long manual analysis time, and complexity of FC data files limit the offering of this testing to selected reference laboratories. Methods: We developed an artificial intelligence (AI) pipeline to automatically enhance unaltered FC files, minimizing the expertise and manual analysis time needed to detect BALL-MRD on a single-tube 10-color panel (CD10, CD19, CD20, CD22, CD24, CD34, CD38, CD45, CD58, CD66c). Raw FC files corresponding to peripheral blood or bone marrow specimens from 171 BALL MRD positive cases (MRD < 5%, median = 0.11%, 107 patients), and 89 MRD-negative cases (63 patients) were processed through our CCADDAS (Clustering and Classification of All events, Dimensionality reduction, Downsampling and Aberrancy Scaling) pipeline on a cloud environment (Google Vertex AI). Automated processing steps included elimination of acquisition errors (FlowCut), state-of-the-art clustering (PARC), dimensionality reduction (UMAP), cluster-based anomaly detection compared to negative controls (24 bone marrow aspirates and 7 peripheral bloods), and cluster-informed downsampling with preservation of low-event subsets. In addition, a deep neural network (DNN) trained on expert-defined subpopulations from negative controls was included for automatic gating of normal subsets (Tensorflow v2). AI-enhanced data files were analyzed using a general purpose flow cytometry analysis software (Kaluza v 3.5, Beckman Coulter), and %MRD estimates compared to reported results based on expert analysis of original FC files. Results: Cluster-informed downsampling reduced the number of cells per case to be manually analyzed from 1.2 million to 155,884 cells on average (87% cellularity reduction); resulting in a smaller FC data file (from 64.5 MB to 13.7 MB; 79% data reduction). Importantly, low-level MRD events were adequately preserved after downsampling (median 100% retention for both MRD<0.01% and MRD 0.01-1%). True number of events were accurately estimated on gated subsets using an “upsampling factor” parameter [observed # events x mean (upsampling factor)]. Gating of normal subsets was completely automated using a DNN classifier parameter. Identification of MRD was aided by an AI-generated “aberrancy scale” parameter that discriminated BALL (median: 991) from both benign mononuclear cells (median: 486, p < 0.0001) and background B-lymphoid cells (median: 379, p < 0.0001), with a performance (AUC: 0.98 and 0.94 for mononuclear and B-lymphoid cells, respectively) superior to CD20 (0.54 and 0.69), CD38 (0.60 and 0.81), CD45 (0.75 and 0.74), CD58 (0.52 and 0.89), CD66c (0.56 and 0.72) and all other antigens studied. The use of AI-enhanced files reduced manual analysis time from 14.16 minutes (SD = 3.7) to 1.05 minutes (SD = 0.6) per case, on average (p < 0.0001) (93% reduction). BALL MRD was detected in all positive cases above clinically relevant threshold (≥0.01%, as recommended by NCCN guidelines), and in 28 of 32 (88%) cases below this threshold, with excellent quantitative correlation with conventional analysis on original FC files (linear regression slope = 1.18, intercept = 0.04%, R squared = 0.92, P < 0.0001). Conclusion: We introduce a largely unsupervised AI pipeline that transforms raw BALL-MRD FC data into a markedly smaller, AI-annotated and software-agnostic FC file, including comparison to normal controls. This CCADDAS pipeline simplifies and accelerates detection of BALL-MRD in clinical diagnostics, reducing number of cells analyzed by 87% and manual analysis time by 93%, without impacting test performance. Moreover, CCADDAS can be utilized for any labroatory-developed BALL-MRD assay, and its small-sized export is compatible with any clinical FC software, computing platform and analysis strategy. Adoption of CCADDAS is likely to facilitate the implementation of BALL-MRD FC analysis by more clinical laboratories.

Introduction: Ticagrelor is a widely used reversible platelet adenosine 5'-diphosphate (ADP) P2Y12 receptor inhibitor used in patients with acute coronary syndrome and coronary interventions. Ticagrelor treated patients may be concomitantly exposed to heparin treatment for prophylaxis/treatment of thrombosis, and heparin induced thrombocytopenia (HIT), a thrombotic thrombocytopenic adverse event may be suspected in such patients. Objective: Here we describe the serological profile of a patient on ticagrelor treatment and strong clinical suspicion of HIT who presented with negative serotonin release assay (SRA) results. A follow up systematic study on the impact of ticagrelor on this gold standard HIT assay is presented. Methods: Remnants from 23 patient samples submitted for clinical SRA testing (5- SRA negative; 6- “low positive” [20-50%]; 5- “moderate positive” [>50 to <90%]; 7- “strong positive” [>90%]) were spiked with ticagrelor in concentrations of 100, 250, 500, and 1000 ng/mL before SRA re-testing. Testing was performed using heat inactivated serum, with low (0.1 units/mL) or high (100 units/mL) heparin and washed platelets (Leger, THSNA 2022). A positive SRA result was defined as ≥20% serotonin release with the low heparin condition and >50% inhibition of serotonin release with high concentration of heparin. A subset of SRA positive patient samples that tested falsely negative upon ticagrelor addition were subsequently incubated with activated charcoal and retested in the SRA to determine if ticagrelor interference was reversible using this methodology. Results: The index patient on ticagrelor had a high 4Ts score, ELISA optical density >1.0 but tested negative in the SRA. Re-testing of this patient with an activated charcoal treated sample demonstrated a striking change in SRA to a strong positive result (91% serotonin release). Upon testing of ticagrelor-spiked samples in the SRA negative patient cohort, as expected, the drug had no effect on results, with all 5 samples persistently staying SRA negative. Of the 6 patients that had low positive SRA results, all converted to SRA negative with ticagrelor concentrations >500 ng/mL. One of these patients became SRA negative at 100 ng/mL and 3 at 250 ng/mL. Of the 5 patients with moderately positive SRA results, 2 converted to SRA negative with ticagrelor concentrations >100 ng/mL, and 4 of 5 patients became negative with spiking at 500 ng/mL. Finally, of the 7 patients with strong positive SRA results, 1 became falsely negative in the SRA at a ticagrelor concentration of 1000 ng/mL. Activated charcoal was incubated with 6 samples that demonstrated negative SRA results with drug concentration of 500 ng/mL. Two samples reverted to SRA positive, while 4 samples remained persistently negative. Conclusion: In preliminary studies, ticagrelor has been shown to interfere with the heparin induced platelet activation (HIPA) assay but its impact on the SRA has been relatively unexplored. Here, we demonstrate that patients on ticagrelor may present with spuriously false HIT functional assay testing by the SRA, particularly if the HIT antibody is not strongly platelet activating. Activated charcoal treatment was only able to reverse drug interference in a subset of samples, even at drug concentrations below peak levels recorded in patients (approx. 800ng/mL). Given these findings, we recommend de-emphasizing the utility of the SRA in ticagrelor treated patients with reliance more on criteria such as pre-test probability of HIT (e.g. 4Ts score), and ELISA results in order to facilitate an accurate diagnosis of HIT.

Introduction: Laboratory identification of a lupus anticoagulant (LAC) requires evidence of: a) prolongation of a phospholipid-dependent screening test, usually a LAC-responsive dilute Russell's viper venom time (DRVVT) or activated partial thromboplastin time (APTT); b) inhibition on mixing with normal pooled plasma (NPP); c) confirmation of phospholipid-dependent inhibition by repeating the prolonged test with excess phospholipid. Despite decades of assay development and re-design and numerous attempts at standardization, diagnosis of a LAC remains challenging due to the heterogeneity of antiphospholipid antibodies, the marked variations in reagents, inconsistencies in post-analytic processes, and not least, the multitude of interferences, such as anticoagulant drugs, which influence test results. Guidelines recommend interruption of anticoagulation, if possible, prior to LAC testing to prevent obtaining false test results. While this is attempted in some instances for direct-acting anticoagulants, it is typically not considered for warfarin, and it is also not always possible. Our laboratory has observed a concerning pattern of potentially false-negative DRVVT screen/ confirm ratios in samples collected from patients on warfarin anticoagulation or with a factor deficiency (e.g. hypoprothrombinemia). The goal of this study is to replicate this observation using in vitro dilution experiments, mixing LAC-positive samples with either NPP or warfarin plasma. Methods: DRVVT Screen, Mix and Confirm assays (Precision Biologic, Dartmouth, NS) were performed using archived LAC-positive plasmas from 5 unique patients not on anticoagulation, and titered with normal pooled plasma (NPP) (Precision Biologic, Dartmouth, NS) to a screen/confirm ratio of 1.20-1.35, near the assay cutoff for DRVVTC (<1.20) and analyzed on the ACL TOP 750 (Werfen, Bedford, MA). Warfarin donor plasma was then mixed in the same dilution scheme with each of the LAC-positive plasmas so that the final dilution factors for LAC-NPP and LAC-Warfarin contrived samples were the same. The DRVVT screen/ confirm ratios of the LAC-NPP and LAC-Warfarin plasma contrived samples were compared. Additionally, a DRVVT Mix Screen and Mix Confirm were performed after mixing in a 1:1 ratio with NPP. Results: The average prothrombin time (PT) in the LAC-Warfarin samples was significantly longer than the LAC-NPP samples (20.7±8.5s vs 10.6±0.2s, p=0.03). All 5 LAC-NPP contrived samples had a positive DRVVT screen/ confirm ratio >1.20, while all of the LAC-Warfarin contrived samples had a negative DRVVT screen/ confirm ratio (1.26±0.04 vs 1.02±0.11, p<0.01). When the DRVVT screening time was divided by the DRVVT Mix Confirm time, 4 of the 5 LAC-Warfarin contrived samples reverted to positive using a cutoff of 1.20 (values of 1.30, 1.21, 1.21, 1.02 and 1.23 for LAC-Warfarin samples) and a similar pattern was observed with the LAC-NPP samples (values of 1.32, 1.29, 1.27, 1.14, 1.27, respectively, for LAC-NPP samples). Conclusion: This study confirms our anecdotal observation that warfarin and/ or factor deficiencies in the common pathway can result in a false-negative DRVVT screen/ confirm ratio using the Precision Biologic LA Sure and LA Check reagents, similar to findings described for apixaban. The DRVVT assay is widely considered to be the most sensitive or responsive to LAC and this in vitro observation raises questions about whether negative DRVVT screen/ confirm ratios can reliably be used to exclude the presence of a LAC in patients on warfarin using this reagent system. Larger studies are required using pre-treatment and on-treatment paired specimens to confirm these findings, and to further explore the utility of the DRVVT Screen/Mix Confirm ratio in this patient population.

Background: Flow cytometry (FC) is widely utilized for the identification of minimal residual disease (MRD) in chronic lymphocytic leukemia (CLL), providing an essential laboratory indicator for prognostic assessment and therapeutic management. Unfortunately, the high level of expertise required, long manual analysis time, and complexity of FC data files limit the offering of this testing to selected reference laboratories. Methods: We developed an artificial intelligence (AI) pipeline to automatically enhance unaltered FC files, minimizing the expertise and manual analysis time needed to detect CLL MRD on a single-tube 10-color panel (CD5, CD19, CD20, CD22, CD38, CD43, CD45, CD200, kappa and lambda). Raw FC files corresponding to peripheral blood or bone marrow specimens from 166 CLL MRD positive cases (MRD < 5%, median = 0.17%, 122 patients), and 61 MRD-negative cases (45 patients) were processed through our CCADDAS (Clustering and Classification of All events, Dimensionality reduction, Downsampling and Aberrancy Scaling) pipeline on a cloud environment (Google Vertex AI). Automated processing steps included elimination of acquisition errors (FlowCut), state-of-the-art clustering (PARC), dimensionality reduction (UMAP), cluster-based anomaly detection compared to negative controls (15 bone marrow aspirates and 14 peripheral bloods), and cluster-informed downsampling with preservation of low-event subsets. In addition, a deep neural network (DNN) trained on expert-defined subpopulations from negative controls was included for automatic gating of normal subsets (Tensorflow v2). AI-enhanced data files were analyzed using a general purpose flow cytometry analysis software (Kaluza v 3.5, Beckman Coulter), and %MRD estimates compared to reported results based on expert analysis of original FC files. Results: Cluster-informed downsampling reduced the number of cells per case to be manually analyzed from 1.1 million to 165,010 cells on average (85% cellularity reduction); resulting in a smaller FC data file (from 62.1 MB to 15.2 MB; 75% data reduction). Importantly, low-level MRD events were adequately preserved after downsampling (median 100% retention for both MRD<0.01% and MRD 0.01-1%). True number of events were accurately estimated on gated subsets using an “upsampling factor” parameter [observed # events x mean (upsampling factor)]. Gating of normal subsets was completely automated using a DNN classifier parameter. Rapid identification of suspected MRD was aided by an AI-generated “aberrancy scale” parameter that discriminated CLL MRD (mean ± 2SD: 969 ± 144) from background B-lymphoid cells (mean ± 2SD: 395 ± 401) (p<0.0001) (AUC = 0.98), with a performance superior to CD5 (AUC = 0.93), CD20 (AUC = 0.83), CD43 (AUC = 0.8) and CD200 (AUC = 0.7). Immunoglobulin light chain-positive CLL MRD was visually identified on a pre-calculated UMAP plot as a single-colored cluster, after dual coloring for kappa and lambda. The use of AI-enhanced files reduced manual analysis time from 11.6 minutes (SD = 4.5) to 0.8 minutes (SD = 0.4) per case, on average (p < 0.01) (93% reduction). CLL MRD was detected in all positive cases above clinically relevant threshold (≥0.01%, as recommended by NCCN guidelines), and in 17 of 21 (81%) cases below this threshold, with excellent quantitative correlation with conventional analysis on original FC files (linear regression slope = 1.02, intercept = 0.001%, R squared = 0.98, P < 0.0001). Conclusion: We introduce a largely unsupervised AI pipeline that transforms raw CLL-MRD FC data into a markedly smaller, AI-annotated and software-agnostic FC file, including comparison to normal controls. This CCADDAS pipeline simplifies and accelerates detection of CLL-MRD in clinical diagnostics, reducing number of cells analyzed by 85% and manual analysis time by 93%, without impacting test performance. Moreover, CCADDAS can be utilized for any laboratory-developed CLL-MRD assay, and its small-sized export is compatible with any clinical FC software, computing platform and analysis strategy. Adoption of CCADDAS is likely to facilitate the implementation of CLL MRD FC analysis by more clinical laboratories.

Context.— Generative artificial intelligence (GAI) is a promising new technology with the potential to transform communication and workflows in health care and pathology. Although new technologies offer advantages, they also come with risks that users, particularly early adopters, must recognize. Given the fast pace of GAI developments, pathologists may find it challenging to stay current with the terminology, technical underpinnings, and latest advancements. Building this knowledge base will enable pathologists to grasp the potential risks and impacts that GAI may have on the future practice of pathology. Objective.— To present key elements of GAI development, evaluation, and implementation in a way that is accessible to pathologists and relevant to laboratory applications. Data Sources.— Information was gathered from recent studies and reviews from PubMed and arXiv. Conclusions.— GAI offers many potential benefits for practicing pathologists. However, the use of GAI in clinical practice requires rigorous oversight and continuous refinement to fully realize its potential and mitigate inherent risks. The performance of GAI is highly dependent on the quality and diversity of the training and fine-tuning data, which can also propagate biases if not carefully managed. Ethical concerns, particularly regarding patient privacy and autonomy, must be addressed to ensure responsible use. By harnessing these emergent technologies, pathologists will be well placed to continue forward as leaders in diagnostic medicine.

Context.— Generative artificial intelligence (GAI) technologies are likely to dramatically impact health care workflows in clinical pathology (CP). Applications in CP include education, data mining, decision support, result summaries, and patient trend assessments. Objective.— To review use cases of GAI in CP, with a particular focus on large language models. Specific examples are provided for the applications of GAI in the subspecialties of clinical chemistry, microbiology, hematopathology, and molecular diagnostics. Additionally, the review addresses potential pitfalls of GAI paradigms. Data Sources.— Current literature on GAI in health care was reviewed broadly. The use case scenarios for each CP subspecialty review common data sources generated in each subspecialty. The potential for utilization of CP data in the GAI context was subsequently assessed, focusing on issues such as future reporting paradigms, impact on quality metrics, and potential for translational research activities. Conclusions.— GAI is a powerful tool with the potential to revolutionize health care for patients and practitioners alike. However, GAI must be implemented with much caution considering various shortcomings of the technology such as biases, hallucinations, practical challenges of implementing GAI in existing CP workflows, and end-user acceptance. Human-in-the-loop models of GAI implementation have the potential to revolutionize CP by delivering deeper, meaningful insights into patient outcomes both at an individual and a population level.

Abstract Background Fibrinogen plays a critical role in hemostasis and is essential to clot formation. Changes in the quantity or activity of fibrinogen can lead to disorders affecting coagulation. Thus, it is essential to precisely quantify fibrinogen levels to differentiate between hyperfibrinogenemia in hypercoagulable disorders and dysfibrinogenemia, hypo/afibrinogenemia, and irregular bleeding. Current diagnostic tests that use antibodies to detect fibrinogen or activity-based measures are susceptible to interference from heparin contamination, hemolysis, and hyperbilirubinemia. Here, we evaluated the use of targeted proteomics for developing a quantitative assay for fibrinogen in plasma samples. Methods Four peptides each for alpha and gamma chains of fibrinogen were synthesized using FMOC chemistry on a Liberty Blue (CEM Corp) peptide synthesizer. A targeted multiple reaction monitoring (MRM) assay was developed and optimized using Dionex Ultimate 3000 HPLC connected to a TSQ Altis triple quadrupole mass spectrometer (Thermo Scientific). Evaluations were conducted using a range of standards, such as pure protein, WHO plasma and commercially available plasma with known fibrinogen amounts for the accurate quantitation of fibrinogen. Calibration curves were generated using different dilutions in various matrices. The repeatability and precision of the assay were evaluated across five different days. The plasma samples were processed and digested in a 96-well plate and subjected to C18 clean-up using a Bravo automated liquid handler (Agilent). All samples were spiked-in with corresponding heavy-labeled peptides prior to digestion with trypsin. Multiple reaction monitoring assays were performed on tryptic digests of plasma samples. Results Calibration curves using commercial plasma for all peptides showed linearity from 20 mg/dl to 1,308 mg/dl. This covers the entire clinical range including normal (200-350 mg/dl), deficient states (<100 mg/dl) and excess production (>350 mg/dl) that can be used for evaluation of bleeding disorders including consumption coagulopathy and hyperfibrinogenemia. Based on the overall performance in terms of linearity, limit of detection/quantitation, carryover and coefficient of variation (CV) across five different days, two peptides per chain were finally selected for developing the quantitative assay. The intra-assay variability was <5% and intra-day CV was ∼20%. This assay is currently being tested on a variety of clinical samples representing a range of fibrinogen levels. Conclusions Our study focused on assessing whether multiplexed MRM assay could be implemented to measure fibrinogen levels in plasma. Our data indicate the feasibility of MRM-based assays for determining fibrinogen levels as an alternative to existing non-mass spectrometry-based activity or antigen assays.

Abstract Background Laboratory identification of lupus anticoagulant (LAC) is critical for diagnosing antiphospholipid antibody syndrome, a common cause of unprovoked thromboembolic events and pregnancy morbidity. Commonly used clot-based assays for LAC include the dilute Russell’s viper venom time (DRVVT) and activated partial thromboplastin time (APTT), but these are susceptible to anticoagulant interferences. Despite standardization efforts, LAC interpretation remains challenging due to pre- and post-analytical complexity and limited expertise. This study explored the use of deep neural networks (DNN) to facilitate LAC diagnosis. Methods Retrospective data for LAC profiling with 13 features performed at Mayo Clinic between 2021 and 2022 was collected. The profiles of 7,202 patients were randomly divided into training (n=4608), validation (n=1153), and test cohorts (n=1441). Diagnostic outputs including LAC positivity (determined by DRVVT or APTT) and the presence of warfarin (WAR) and heparin (HEP), were confirmed by expert assessment. Three DNN architectures were developed and evaluated (Figure 1): (1) four single output DNN models with a domain knowledge driven selection of input features, (2) four single output DNN models using the same 13 inputs across all models, and (3) a multi-output DNN model. Model performance was evaluated using the F1 score and the index of balanced accuracy (IBA) for each output label. Results Architecture 1 achieved high diagnostic accuracy with IBA values of 0.989-1.000 for all four outputs and F1 scores of 0.945-0.991. Architecture 2 exhibited F1 scores of 0.902-0.973 and IBA values of 0.935-0.992. Architecture 3 achieved F1 scores of 0.933-0.967 and IBA values of 0.941-0.983. Conclusions All three architectures accurately classify LAC and common anticoagulation effects. The comparable performance of architectures 1 and 2 suggests that manual feature extraction is unnecessary. Given its relative simplicity and versatility, architecture 3 is the preferred choice for implementation in a clinical laboratory to standardize LAC diagnosis.

AbstractBackgroundUsing low titer group O whole blood (LTOWB) is increasingly popular for resuscitating trauma patients. LTOWB is often RhD‐positive, which might cause D‐alloimmunization and hemolytic disease of the fetus and newborn (HDFN) if transfused to RhD‐negative females of childbearing potential (FCP). This simulation determined the number of life years gained by the FCP and her future children if she was resuscitated with LTOWB compared with conventional component therapy (CCT).MethodsThe model simulated 500,000 injured FCPs of each age between 0 and 49 years with LTOWB mortality relative reductions (MRRs) compared with components between 0.1% and 25%. For each surviving FCP, number of life years gained was calculated using her age at injury and average life expectancy for American women. The number of expected future pregnancies for FCPs that did not survive was also based on her age at injury; each future child was assigned the maximum lifespan unless they suffered perinatal mortality or serious neurological events from HDFN.ResultsThe LTOWB group with an MRR 25% compared with CCT had the largest total life years gained. The point of equivalence for RhD‐positive LTOWB compared to CCT, where life years lost due to severe HDFN was equivalent to life years gained due to FCP survival/future childbearing, occurred at an MRR of approximately 0.1%.ConclusionIn this model, RhD‐positive LTOWB resulted in substantial gains in maternal and child life years compared with CCT. A >0.1% relative mortality reduction from LTOWB offset the life years lost to HDFN mortality and severe neurological events.