Citations

Abstract Background Minimal Residual Disease (MRD) measurement in multiple myeloma (MM) is becoming the standard of care. Currently, bone marrow (BM) is the only source used for determining MRD. Blood-based MRD detection is desirable since blood is less invasive and less expensive to obtain than BM, enabling more frequent MRD monitoring. A serum liquid chromatography quadrupole time of flight (LC-QTOF) method has shown a significantly lower limit of detection than our current MALDI-TOF (Mass-Fix) method. While Mass-Fix is more sensitive and specific than immunofixation electrophoresis, the LC-QTOF (HR-Mass-Fix) has demonstrated the potential to detect M-proteins at even lower concentrations. Methods 203 Mayo Clinic Rochester patients who had a clinically ordered BM MRD assessment by next generation flow (NGF) cytometry (2 x 10-6), a blood sample available within 7 days of the BM biopsy, and a pre-treatment M-protein mass measured by Mass-Fix were included. Blood samples were analyzed by both Mass-Fix (MALDI-TOF) and HR-Mass-Fix (LC-QTOF). For Mass-Fix, immunoglobulin-enriched extracts were analyzed on a benchtop MALDI-TOF mass spectrometer (Microflex, Bruker) in linear positive ion mode (4000-32000 m/z). For HR-Mass-Fix, kappa and lambda light chain-enriched eluates were analyzed using an Agilent 1290 Infinity II HPLC connected to a Sciex ZenoTOF 7600 mass spectrometer with positive ESI (1000-2500 m/z). Residual M-proteins were identified by mass match within ± 1 Da of the pre-treatment measurement. Agreement between blood and BM MRD status was assessed using overall percent agreement and Cohen's kappa. The effect of M-protein half-life on blood/BM discordance was evaluated by comparing isotype distributions across concordance groups using Fisher's exact test. Results The cohort consisted of 175 myeloma, 18 AL amyloidosis, 4 LPL/lymphoma, and 6 other plasma cell disorders. All blood Mass-Fix positive samples were also positive by HR-Mass-Fix (100% concordance). HR-Mass-Fix detected residual M-proteins in 44 additional samples (61%) that were negative by Mass-Fix, increasing the overall blood MRD positivity rate from 58% to 84%. The BM NGF MRD positivity rate was 45%. Blood-BM agreement by Mass-Fix was 73% (kappa 0.46) which decreased to 61% (kappa 0.21) with HR-Mass-Fix, driven primarily by HR-Mass-Fix blood(+)/BM NGF(-) discordant cases. M-protein half-life appeared to influence discordance patterns. Immunoglobulin half-lives are: IgG 23 days, IgA 6 days, IgM 5 days, and free light chain 0.25 days. In HR-Mass-Fix blood(+)/BM(-) discordant patients, IgG was significantly overrepresented compared to the rest of the cohort (70% vs 46%, p=0.003). In concordant positive patients, the isotype distribution did not significantly differ from the overall cohort. These findings suggest that longer half-life M-proteins (IgG) persist in blood after BM clearance due to immunoglobulin recycling. Conclusion HR-Mass-Fix significantly increases M-protein detection compared to standard Mass-Fix, with the tested cohort demonstrating a rescue rate of 61% among Mass-Fix negative samples. Immunoglobulin recycling is a likely explanation for blood(+)/BM(-) discordance. Serial measurements will be needed to distinguish residual M-protein from ongoing M-protein production. Blood-based M-protein measurement offers a less invasive, lower-cost alternative to bone marrow collection.

Abstract Background Health disparities affecting Black/African American communities remain a persistent public health challenge, particularly for chronic conditions such as diabetes, cardiovascular disease, and kidney disease. The Love Wins + Health initiative represents a collaborative, community-engaged strategy to reduce these disparities through point-of-care testing (POCT), health education, and culturally responsive outreach. This program was a partnership between Mayo Clinic’s Department of Laboratory Medicine and Pathology (DLMP), Mayo Clinic Community Internal Medicine, and the Rochester, MN Branch of NAACP. It was implemented during the 2025 Juneteenth celebration, an event serving more than 1,000 community members. Phlebotomists played a central role in delivering accessible diagnostic services in a trusted community setting. Volunteer phlebotomists from Mayo Clinic Laboratory Services provided specimen collection and testing support as the primary point of patient interaction. Beyond technical expertise, they functioned as ambassadors of patient-centered care, reinforcing respect, dignity, and cultural sensitivity during testing encounters. This role was essential to establishing trust and ensuring specimen integrity and accurate results. Methods Free testing was promoted through community posters, and an informational health-focused townhall was coordinated with the Rochester, MN Branch of NAACP. The testing event was held June 14, 2025, at Martin Luther King Jr. Park in Rochester, MN. Participants registered on-site and completed an intake form capturing demographics and health history. Three trained phlebotomists performed venipuncture whole blood collections in Mayo Clinic’s Mobile Health vehicle. CLIA-waived POCT was conducted by trained laboratory technologists using Afinion hemoglobin A1c (HbA1c), Cholestech LDX lipid profile, and i-STAT creatinine platforms (Abbott Laboratories). On-site physicians reviewed results and discussed follow-up recommendations. A post-collection survey assessed participant impact. Percentages of abnormal results were calculated, and survey responses were tabulated. The Mayo Clinic Institutional Review Board determined the initiative met criteria for quality improvement and did not require formal review. Results Fifty-one individuals participated (median age 42 years; range 18–81). Participants were 62% female, 36% male, and 2% nonbinary; 27.5% identified as Black/African American. Testing revealed a high prevalence of cardiometabolic abnormalities. Elevated low-density lipoprotein cholesterol was observed in 57.1% of participants, elevated non–high-density lipoprotein cholesterol in 52.0%, and elevated triglycerides in 31.4%. Markers of renal dysfunction, including elevated creatinine and reduced estimated glomerular filtration rate (eGFR), were present in 17.6%. Elevated HbA1c was identified in 13.7% and elevated glucose in 7.8%. Black participants demonstrated higher rates of abnormal HbA1c (20.0% vs 12.1%) and reduced eGFR (20.0% vs 18.2%) compared with non-Black participants. Post-event surveys indicated strong engagement: 84.3% reported improved understanding of their health, 92.2% would recommend screening to others, and 56.9% planned follow-up care. Conclusion The Love Wins + Health initiative demonstrates the value of community-based POCT in identifying undiagnosed cardiometabolic and renal risk factors. The high prevalence of abnormal findings underscores persistent diagnostic care gaps and the need for accessible screening. Immediate result review and on-site clinical consultation supported timely education and referral. This collaborative, community-centered model offers a scalable approach to expanding preventive care access and advancing health equity among populations disproportionately affected by chronic disease.

Abstract Background Glucose-6-phosphate dehydrogenase (G6PD) enzyme activity is ordered in patients with elevated white blood cell (WBC) counts to assess for enzyme deficiency prior to administration of Rasburicase, a medication protective against tumor lysis syndrome that is contraindicated in G6PD deficient persons. It is well known that G6PD deficiency can be masked by reticulocytes, but effect of leukocytosis is not well-understood. Herein we evaluated increased WBC counts in various conditions and compared G6PD values, with longitudinal follow-up when available. Methods G6PD enzyme activity was performed on 78 unique patients with elevated WBC. Anticoagulated whole blood was washed using a cellulose-packed column. G6PD activity was measured using a Siemens Advia XPT chemistry analyzer. G6PD catalyzes the oxidation of glucose-6-phosphate to 6-phosphogluconate. Simultaneously, nicotinamide adenine dinucleotide phosphate (NADP) is reduced to NADPH causing an increase in absorbance which is measured at 340 nm. G6PD levels are adjusted to the hemoglobin present in the hemolysate. Complete blood counts (CBC) were performed using a Sysmex XN-9100 analyzer. Results Our cohort contained 56 myeloid predominant (MP) [19 AML, 14 reactive leukocytosis, 12 CML, 7 chronic myeloid neoplasms, 2 CMML, and 2 MPAL (B/myeloid)]; and 22 lymphoid predominant (LP) [14 CLL, 4 CLPD-NOS, 2 mantle cell lymphoma, 1 B-ALL and 1 T-ALL] disorders. G6PD activity ranged from <0.5-40.2 U/gHb and 8.0-14.9 U/gHb in the MP and LP groups, respectively. Only one G6PD deficient patient was identified (AML). Very high G6PD levels (>20 U/gHb) occurred in 7 MP disorders, 2 of which G6PD activity normalized after WBC reduction. No cases of masked G6PD deficiency were identified. Twenty-eight cases with two timepoints show similar G6PD activity before and after leukocyte normalization. Three show a mild increase, which is assumed from reticulocytosis or RBC transfusion, two of which had borderline low G6PD activity prior to therapy. Conclusion Masking of G6PD deficiency by elevated WBC count was not observed in our cohort. Enzyme levels were within normal limits/mildly elevated with markedly elevated WBC counts in the LP group regardless of cell maturity. The MP group affected enzyme activity much more than the LP group. When multiple timepoints were available after therapy when WBC count decreased, a similar G6PD activity level was measured in both groups.Rarely G6PD levels were shown to mildly increase with decreased WBC count which is presumed from increased reticulocytes after treatment.Markedly high G6PD levels (>20 U/gHb) were restricted to myeloid blood disorders or reactive leukocytosis with markedly elevated granulocytes and/or monocytes.A single G6PD deficient patient showed very low G6PD activity even when the WBC was elevated.

Abstract Background Liquid chromatography mass spectrometry (LC-MS) has become a common platform for the implementation of lab-developed tests (LDTs). MS can offer a fast pathway to assay development with low cost-per-assay, multiple assays per platform, and no costly analyte specific reagents. Therapeutic monoclonal antibodies (tmAb) used in the treatment of autoimmune disorders and cancers are routinely measured by MS for therapeutic drug monitoring. Recently high-resolution accurate mass (HRAM) MS has been proven useful for the detection of tmAbs and also endogenous monoclonal antibodies in multiple myeloma minimal residual disease (MRD). While our tmAb and MRD development was performed using a microflow single-channel LC integrated with a Sciex TripleTOF 5600, we looked to instrumentation with higher capacity for implementation; triple quads or orbitraps that could be integrated with multiplexing, ultra-high-performance liquid chromatography (UHPLC) allowing for throughput and savings in turnaround time (TAT). Our goal was to investigate the possibility of integrating a multiplexing 2-channel UHPLC with a HRAM MS for added future efficiency in TAT and capacity in testing volumes. Methods We integrated a Thermo Scientific Transcend LX Multichannel (2-channel) UHPLC with our current Sciex ZenoTOF 7600 HRAM MS. Thermo Scientific provided the details for the contact closure cable, TMO link, and software needed. Thermo Scientific Aria MX was used for managing and controlling pump operations, XCalibur for batch building and Sciex OS for data acquisition and analysis. Residual clinical extracted runs previously analyzed and released on our single channel Agilent Infinity II UHPLC integrated with the Sciex ZenoTOF 7600 MS were reinjected onto a Sciex ZenoTOF 7600 MS using the 2-channel Thermo Scientific LC. Results The LC method for our quantitative risankizumab serum assay was translated from the Agilent to the Thermo LC without any changes to the gradient. We were successfully able to run across 2 channels using the 10-minute LC method with a 3-minute MS window to allow for multiplexing. Analytical sensitivity (measured as peak area counts) was comparable between systems. A N=22 patient comparison covering the measuring interval from 1 to 100 mcg/mL gave linear regression with a slope 0.971 and intercept of 0.7991; R2=0.9979. While the 2-channel LC could decrease the TAT in half, there is still the potential to implement a 3-channel LC with the 3-minute MS window from a 10-minute LC method. MRD is a qualitative assay with current LC run time of 13 minutes. Of the N=14 samples run for this comparison, there was 100% agreement for positivity and equivalent sensitivity compared to the single channel LC. By adding the 2-channel LC, we would be able to complete a 96-well plate for MRD in 12 hours instead of the current 24 hours. Conclusion We have successfully shown proof-of-concept for multiplexing a 2-channel UHPLC integrated with a HRAM TOFMS instrument, which has not been accomplished before. This will provide improved efficiency in turnaround time and open instrument capacity for future increases in testing volume without compromise in test cost, analytical sensitivity, or need for additional MS instrumentation.

Abstract Background Comprehensive lipoprotein analysis and interpretation is a cornerstone of cardiovascular disease (CVD) prevention. Often, multiple platforms and assays are used to quantify lipoprotein parameters and apolipoproteins. Growing economic pressure in healthcare systems is driving the need for efficient technologies capable of measuring multiple analytes in a single analytical step. Nuclear magnetic resonance (NMR) enables simultaneous quantification of lipoprotein subclasses and related biomarkers in one integrated measurement. Although apolipoprotein B (ApoB) has previously been described as an NMR-derived parameter, its implementation has been limited to research-use-only (RUO) applications, restricting routine clinical adoption. Herein, we present the IVD-grade analytical validation of an AXINON® NMR-based ApoB, benchmarked independently against two routine clinical chemistry analyzers, aiming to enable?future?integration into clinical routine diagnostics. Methods NMR analytical experiments for precision, linearity, and method comparison were performed using the AXINON® ApoB research use only modules (numares AG, Regensburg, Germany) at two different sites. Method comparison was performed against two different commercially available clinical analyzers (Thermofisher Optilite and Roche Cobas c501). Samples were derived from clinical residual serum collected in plain serum clot tubes (RED: n=241) or serum separator tubes (SST: n=66). Within-laboratory coefficients of variation (CVWL) and repeatability (CVr) were determined by replicate measures (n=14) on different runs. Linearity was assessed by mixing samples (n=14) with a high and low ApoB concentration at various concentrations. Results Imprecision was determined at various concentrations of ApoB (range 35.5 – 145.5 mg/dL); CVWL was <3.8% and CVr was <2.9%, for all sites and concentrations. The linear slope was between 0.97 and 1.01, bias was <4.5mg/dL and correlation was >0.99 for both sites. The mean absolute error (MAE) was 7 mg/dL compared to Roche and 5.5 mg/dL for Optilite measured ApoB. The percentage of measurements within 15% of Roche ApoB was > 83% Red and >86% for SST: all samples (n=34) were within 15% of Optilite ApoB. Conclusion Our data support that AXINON® NMR-based ApoB shows comparable precision, and concordance with commercially available ApoB immunoassays. ApoB NMR profiling in addition to lipoprotein particle number and lipoprotein size may provide expanded utility for lipid management.

Abstract Background Hypermobile Ehlers–Danlos syndrome (hEDS) and hypermobility spectrum disorders (HSD) lack objective laboratory biomarkers, contributing to prolonged diagnostic delays and reliance on subjective clinical criteria. Ritelli et al. recently reported disease-associated ∼45–52 kDa fibronectin (FN) and type I collagen (COL1) plasma fragments detected by immunoprecipitation (IP) and Western blotting, suggesting a potential protein-based diagnostic marker. However, Western blotting is a laborious, low throughput technique not ideally suited for routine clinical analysis. Alternatively, liquid chromatography-mass spectrometry has the potential to enable translation of this observation into a scalable, quantitative routine clinical assay. Methods Serum samples from individuals with hEDS/HSD and controls were analyzed using IP with the same commercial anti-FN and anti-COL1 antibodies reported by Ritelli et al., covalently coupled to magnetic beads. Two complementary LC–MS/MS workflows were evaluated: (1) direct in-solution trypsin digestion of immunoprecipitated proteins followed by discovery proteomics, and (2) SDS-PAGE separation of IP-enriched material, excision of the ∼45–55 kDa gel region, in-gel trypsin digestion, and LC–MS/MS analysis. Peptide identification was performed using high-resolution LC–MS/MS with database searching at 1% false discovery rate. Candidate disease-associated peptides identified by discovery software were subjected to stringent manual validation, including extracted ion chromatogram inspection, signal-to-noise assessment, isotopic pattern evaluation, and reproducibility across biological replicates. Results Across both analytical workflows, fibronectin- and collagen I–derived peptides were reproducibly detected following immunoenrichment, with many peptides exhibiting well-defined chromatographic peaks, appropriate m/z, isotopic distributions, and consistent retention times. Database searching resulted in identification of multiple candidate peptides with robust signal intensity and apparent statistically significant differences between hEDS/HSD and control samples, suggesting potential disease association at the initial screening level. However, subsequent manual review of extracted ion chromatograms in TraceFinder with re-integration based on peak area demonstrated that the majority of these discovery-stage candidates exhibited comparable signal intensity in both cases and controls, indicating detection of shared protein regions rather than disease-specific fragments. These apparent group differences were largely attributable to limitations in database searching or low-level noise influencing discovery statistics rather than true biological separation. Remaining candidates failed analytical validation due to low signal strength, poorly defined chromatographic peaks, inconsistent detection across samples, or mismatch with expected m/z and isotopic profiles. Analysis of peptides derived from the excised ∼45–55 kDa gel region yielded similar findings, with either uniformly detected peptides across groups or signals insufficient for reliable quantification. No peptides demonstrated case-restricted presence, reproducible enrichment, or fragment-specific mapping consistent with a disease-associated fibronectin or collagen I fragmentation signature. Consequently, no candidates met predefined analytical and biological criteria required for advancement to targeted assay development. Conclusion Using antibody reagents and enrichment strategies aligned with prior Western blot–based reports, LC–MS/MS analysis did not identify reproducible fibronectin or collagen I peptide signatures specific to hEDS/HSD. These findings highlight challenges in translating gel-based fragment observations into peptide-level mass spectrometry assays and underscore the importance of rigorous analytical validation. Further work is needed to clarify whether the reported fragments represent low-abundance, heterogeneous proteoforms, pre-analytical artifacts, or disease features not readily captured by bottom-up proteomics.

Abstract Background Clinical laboratories receive frequent inbound phone calls from healthcare providers related to test ordering, specimen status, and result availability. In high-volume clinical laboratory environments, limited phone triage specificity may result in high call volumes reaching laboratory staff that could be addressed through alternative routing or standardized messaging. This project evaluated provider phone call volume and call type distribution within a high-volume centralized clinical laboratory call center and assessed the impact of targeted redesign of phone triage options. Methods Inbound provider phone calls answered by staff in a centralized clinical laboratory call center were prospectively tracked using standardized tally sheets. Calls were categorized into four predefined call types: requests for transfer to another area or operator misdirects, questions regarding tests not performed in the central clinical laboratory, requests for patient results or confirmation of specimen receipt, and questions regarding test add-on eligibility or specimen availability. Pre-intervention phone triage consisted of a limited 2 option call routing menu routing calls bidirectionally for patient result inquiries versus all other laboratory questions. Based on pre-intervention call category data, a revised phone menu was implemented that included a standardized message instructing callers to allow two hours from specimen collection prior to results inquiries, followed by three targeted menu options for patient result inquiries, add-on or specimen availability questions, or direct connection to Central Processing staff. Operator routing logic was concurrently refined by reducing keyword triggers associated with incorrect transfers. Three study periods were evaluated: pre-intervention, post-intervention, and four months post-intervention. Daily call volume and average+/-standard deviation (SD) percent and type of calls directed to call center staff were measured. Results A total of 57 operational days were analyzed, including 25 pre-intervention days, 18 post-intervention days, and 14 days at four months post-intervention, representing 8,352 total inbound provider phone calls. Across all study periods, average (+/-SD) daily call volume was 147+/-42 calls. During the pre-intervention period (n=3,596 total calls), 24.1+/-9.8% of calls received per day met the criteria for the four predefined categories. During the post-intervention period (n=2,905 calls) and the four months post-intervention period (n= 1,851 calls), 8.4+/-5.6% and 15.6+/-7.9 % of calls received met the same criteria. In the pre-intervention period (n=3,596 calls), calls categorized as requests for transfer or operator misdirect, questions about testing for non-central clinical laboratory tests, calls requesting results or confirmation of specimen receipt, and questions regarding test add-on eligibility or specimen availability were 8.3%, 10.0%, 10.6%, and 8.2%, respectively. Corresponding category percentages during the post-intervention period (n=2905 calls) were 1.8%, 5.1%, 4.2%, and 4.3%, and at four months post-intervention (n=1851 calls) were 3.5%, 8.8%, 6.3%, and 4.9%, respectively. Conclusion Systematic measurement of provider phone call volume and call type distribution in a high-volume clinical laboratory call center enabled targeted redesign of phone triage options, including menu restructuring, standardized messaging, and refinement of operator routing logic. The volume of calls routed to call center staff was significantly reduced after redesigning the phone triage options and improvement was sustained four months post-intervention.

Abstract Background The Enhanced Liver Fibrosis test (ELF™; Siemens Healthineers) is an FDA-authorized serum biomarker test for assessing metabolic dysfunction-associated steatotic liver disease (MASLD). The ELF score is calculated from hyaluronic acid (HA), amino-terminal pro-peptide of type III collagen (PIIINP), and tissue inhibitor of matrix metalloproteinase 1 (TIMP-1), and has demonstrated prognostic value for MASLD disease progression and liver-related events. This study was aimed at verifying manufacturer’s performance characteristics for ELF including precision, accuracy, reportable range, analyte stability and common interferences. Methods HA, PIIINP, and TIMP-1 testing were performed, and ELF scores calculated using the Siemens ADVIA Centaur XP. Within-run precision was evaluated using 10 replicates of three Siemens ELF QC levels in a single run; between-run precision used 5 replicates across 4 days (n=20). Analytical measuring range (AMR) was verified by testing five Siemens Master Curve Material (MCM) levels in triplicate plus a dilution of level 1. Accuracy was assessed by linear regression analyses of blinded comparisons with three external laboratories (Labs 1–3) using de-identified residual samples from physician-ordered NASH-FibroTest™ (Mayo Clinic Laboratories, BioPredictive). Lab 1 comparison included ELF scores only. Serum on-gel stability (serum separator tube, SST™) was assessed (2 days ambient, 7 days refrigerated). Biotin, lipemia, and icterus interference thresholds were verified by spiking samples (n=3) with 10% saline control or interferent. For hemolysis studies, serum pools were divided into saline control or lab-produced hemolysate (10%) to create five increasing hemolysate concentrations per CLSI EP-07 and analyzed using EP Evaluator. Agreement within 10% for HA, PIIINP and TIMP-1 values, or within 0.3 units for ELF was considered acceptable for method comparison, stability and interference studies. Results Within- and between-run precision were consistent with manufacturer’s claims for all control levels with coefficients of variation <=5%. Manufacturer AMRs are 3.00–1000.00 ng/mL (HA), 1.00–150.00 ng/mL (PIIINP), and 5.0–1300.0 ng/mL (TIMP-1). MCM material was used to verify 4.00–691.10 ng/mL (HA), 1.70–103.60 ng/mL (PIIINP), and 6.6–1022.6 ng/mL (TIMP-1). Agreement within +/-0.3 ELF score was observed in 25/25, 36/39, and 3/3 samples compared with Labs 1, 2, and 3, respectively. Linear regression analysis yielded y = 0.99x + 0.32 and R2=0.99 for the combined data. For Lab 2, results agreed within 10% for 27/39 HA, 5/39 PIIINP, and 29/39 TIMP-1. For Lab 3, 3/3 samples agreed within 10% for all components. On-gel stability showed 10/10 samples within ±10% for components and ±0.3 for ELF score at both 2-day ambient and 7-day refrigerated conditions. Biotin interference thresholds were verified at 3500 ng/L (HA/TIMP-1), 65 ng/L (PIIINP), and 150 ng/L (ELF score); icterus and lipemia thresholds were verified as I=90 and L=2000, respectively, using Roche serum indices (Roche Diagnostics, Inc.). A hemolysis index (Roche) threshold of 100 was established. Conclusion Overall, manufacturer’s performance specifications were verified. ELF score comparisons between laboratories demonstrated excellent agreement, even when individual components had differences >10%. Hemolysis studies performed using patient-derived hemolysate demonstrated significant interference from hemolysis at an H-index of 100 which was a novel and notable finding for this testing.

Abstract Background Monitoring patient retesting and result corrections, or revision rates, following quality control (QC) failures in a high-volume clinical chemistry laboratory helps inform QC practices. Our laboratory uses rule-based statistical QC (via mean and standard deviation) to detect deviations from stable operation. Patient rechecks are performed systematically after QC failures to determine if results need to be corrected/revised based on each test’s unique revision criteria. Criteria for revising results are established for each test based on total allowable error and reflect clinically significant deviations in test results. The aim of this study was to calculate the recheck and revision rates over the life of two sets of chemistry analyzers. Methods A retrospective review of patient recheck and revision events was performed for the Central Clinical Laboratory (Mayo Clinic Rochester) from 2015–2025. The laboratory performed approximately 110 tests on combination chemistry/immunoassay lines replacing cobas6000 (two combination c701, c502, c602, e602 lines) equipment with cobas8000 (three combination c701, e801 lines) platforms between October 28, 2019 to February 6, 2020 (Roche Diagnostics, Indianapolis, IN). Average annual test volume was determined from the lab information system. Recheck rates were calculated as the proportion of samples retested following QC failure relative to total testing volume. Revision rates were calculated as the proportion of repeated results exceeding revision limits relative to total patient rechecks. These metrics were tallied annually and subsequently averaged across five years to compare c6000 and c8000 analyzer performance during the first two years of service (new) and the final two years of service (old). 2-tailed student t-test was calculated with p<0.05 considered significant. Results The c6000 instruments were in service 2015-2019 with average annual test volume=3.1M and c8000 instruments from 2020-2025 with 3.4M average annual volume excluding 2020. During these time periods c6000 had average (standard deviation) annual recheck rate of 0.21(0.08) % and c8000 = 0.34(0.03)% (p=0.01). C6000 increased from 0.13% in 2015 to 0.33% in 2019 while c8000 varied with no clear pattern over time. Revision rates for c6000 were 8.5(2.0)% and c8000 were 5.1(1.1)% (p=0.01). The first two years of service (new) for c6000 compared to its last 2 years of service (old) had recheck rates of new=0.14(0.01)% and old=0.30(0.06)% (p=0.06) and revision rates of new=8.3(1.5)% and old=9.1(3.4)% (p=0.8). The c8000 had recheck rates of new=0.33(0.02)% and old=0.33(0.05)% (p=0.9) and revision rates of new=5.0(0.2)% and old=5.7(1.9)% (p=0.7). Conclusion Recheck rates increased over the life of c6000 instruments and remained elevated and steady during the life of c8000 instruments. Revision rates decreased for c8000 instruments compared to c6000 and did not correlate with instrument age. This pattern may reflect increased testing volume, test menu complexity, and operational workflow changes associated with expansion from two to three chemistry lines. Monitoring recheck and revision rate trends longitudinally may provide an objective measure to support proactive maintenance strategies and inform capital replacement planning in high-volume laboratories.

Abstract Background Acute promyelocytic leukemia (APL) is a distinct subtype of acute myeloid leukemia (AML) with PML::RARA fusion and a high risk of early mortality without treatment. Prompt recognition and treatment are critical for improving outcomes. Although most APLs demonstrate classic morphology, severe cytopenia makes APL identification on the peripheral blood smear (PBS) challenging. Additionally, the microgranular variant may mimic other myeloid neoplasms. CellaVision, an automated digital microscopy platform widely used for PBS review and differential counts, does not distinguish APL blasts from non-APL blasts, limiting early recognition. We developed and validated a deep learning model to automatically subclassify PB cells and identify APL blasts from CellaVision images to enable real-time detection of APL blasts in a clinical laboratory. Methods A purposive sample of 36,938 curated single-cell images from 9,364 unique cases captured on CellaVision DM1200 and DI-60 systems were annotated by experts. Fourteen morphologic classes were defined: APL blasts, non-APL blasts, promyelocytes, myelocytes/metamyelocytes, mature myeloid cells, monocytes, lymphocytes, plasma cells, basophils, eosinophils, large platelets, nucleated red cells and artefact. The training and validation datasets comprised 32,840 and 4,098 images, with mature myeloid cells and lymphocytes comprising nearly 40%, APL blasts 6.1%, non-APL blasts 5.6% and promyelocytes 0.4% of images. A pretrained vision transformer backbone (DINOv3 ViT-B/16) was fine-tuned using a multi-view attention-based multiple-instance learning framework. Each cell image was processed through four complementary views, including full-field and multiscale center crops. An attention pooling head aggregated view-level embeddings for classification. Training used cross-entropy loss with label smoothing and layer-wise learning rate decay. Independent testing included 27 cases, including 8 non-APL AML, 6 APL, 3 myelodysplastic syndrome, 2 myeloma, 5 left-shift, and 3 normal cases, with one APL case scanned in triplicate for repeatability. Performance was evaluated using cell-level accuracy and Jensen–Shannon divergence (JSD) between predicted and reference distributions. False-positive rates were assessed on 1,000 randomly selected routine clinical cases. Results Across the 27-case independent test cohort, the mean accuracy at the cell-level was 89.0%. JSD values were low (<0.10) in all but one case, indicating close agreement between predicted and reference distributions. The model identified blasts in all non-APL AML cases. It detected APL blasts in all but one microgranular variant, in which atypical APL blasts were classified as non-APL blasts. Among 21 non-APL cases, two had misclassified as APL blasts: two plasma cells in myeloma and one blast in non-APL AML. Triplicate analysis of one APL case demonstrated consistent performance at cell-level (accuracy 0.79 ± 0.01 and JSD 0.082 ± 0.019). In the 1,000-case screening cohort (inference time 8 seconds/case), only two cells from two separate cases were falsely labeled as APL blasts (0.2% false-positive rate). Conclusion We developed a deep learning system capable of accurately subclassifying PB cells and identifying APL blasts from routine digital microscopy images. The model demonstrated strong concordance with expert review and low false-positive rate. Integrating the model into hematology workflows enables rapid, automated morphologic triage of suspected APL, facilitating earlier diagnosis and treatment, and may improve efficiency and standardization in morphologic assessment.