Background:The increasing prevalence of antifungal-resistant Aspergillus isolates underscores the need for rapid antifungal susceptibility testing. In this proof-of-concept study, we evaluated whether artificial intelligence (AI) can enable rapid antifungal susceptibility classification of Aspergillus species using routine matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS). Methods:Models were developed using MALDI-TOF MS spectra for 240 isolates (Aspergillus flavus N=126, A. fumigatus N=75, Aspergillus lentulus N=39). Their performance in predicting susceptibility to four antifungal agents (voriconazole, posaconazole, itraconazole, and amphotericin B) was evaluated using a validation set comprising A. flavus (N=24), A. fumigatus (N=24), and A. lentulus (N=22), based on categorical agreement (CA), very major error (VME), major error (ME), sensitivity, and specificity. Results:Although spectrum-level evaluation showed excellent analytical discrimination, isolate-level clinical evaluation demonstrated limited performance across seven species-antifungal agent combinations. Across these combinations, the AI models achieved a mean CA of 56.0%, with VME and ME rates of 53.3% and 45.2%, respectively. Sensitivity ranged from 0% to 100% and specificity from 18.2% to 87.5%, indicating inconsistent susceptibility classification performance across species-antifungal agent combinations. Conclusions:AI-assisted analysis of routine MALDI-TOF MS spectra showed the feasibility of classifying antifungal susceptibility in Aspergillus spp. However, its clinical performance across species-antifungal agent combinations was limited. Further optimization, evaluation with larger multicenter datasets, and external validation are required before considering clinical adoption.
Background:Platelet concentrates from apheresis donations are indispensable in transfusion medicine. Integration of pathogen reduction (PR) technologies into platelet manufacturing requires adjustment of apheresis devices to ensure consistent platelet yields while minimizing donation time. We investigated the suitability of Sysmex XN1000 (Sysmex Corporation, Kobe, Japan) in blood bank mode (BBM) for measuring platelet concentrates to optimize the yield-scaling factor (YSF) of apheresis devices, thereby standardizing target yields and reducing donation time. Methods:We compared measurements with the Sysmex XN530 and XN1000 BBM and evaluated three Trima Accel (Terumo, Tokyo, Japan) YSF setups (YSF 0.90/XN530, YSF 0.90/XN1000, and YSF 1.05/XN1000). Differences in yield deviation and donation time were assessed using linear mixed-effects models adjusted for target yield with robust SEs. Results:We analyzed 4,383 platelet apheresis procedures from 1,573 donors across setups in large-volume (LV) and dual-storage (DS) collections. Compared with YSF 0.90/XN530, YSF 0.90/XN1000 and YSF 1.05/XN1000 significantly reduced yield deviation by 0.26 and 0.41×1011 platelets/unit (95% confidence interval [CI] 0.18-0.34 and 0.35-0.48, respectively) in LV collections and by 0.34 and 0.24×1011 platelets/unit (95% CI 0.24-0.45 and 0.01-0.47, respectively) in DS collections. Compared with YSF 0.90/XN530, YSF 1.05/XN1000 reduced the donation time by 8.09 min (6.93-9.26) and 9.88 min (6.81-12.94) for LV and DS collections, respectively. Conclusions:YSF 0.90/XN1000 and YSF 1.05/XN1000 reduced yield deviation and donation time than did YSF 0.90/XN530, supporting the use of XN1000 BBM to calibrate Trima Accel apheresis devices and can help in standardizing target yields while reducing donation time in PR.
Background:Circulating tumor DNA (ctDNA) analysis enables real‑time assessment of the tumor burden and genomic complexity in lymphomas. However, real‑world evidence across lymphoma subtypes is limited. Methods:We analyzed cell‑free DNA (cfDNA) and ctDNA data from 336 consecutive patients with newly diagnosed Hodgkin or non-Hodgkin lymphoma in 2022 and evaluated their prognostic impact in diffuse large B‑cell lymphoma (DLBCL). Results:We detected somatic alterations in 248 of 336 patients (73.8%). DLBCL and follicular lymphoma showed the highest variant prevalences and ctDNA burdens. Epigenetic regulators, including KMT2D, CREBBP, TET2, and HIST1H1E, constituted the dominant class of genes with recurrent alterations. Plasma variant profiles closely mirrored publicly available, tissue‑based next-generation sequencing datasets. The baseline ctDNA burden correlated with adverse clinical features, and ctDNA positivity was associated with failure to achieve complete remission. In DLBCL, elevated cfDNA (top quartile) and a high International Prognostic Index (IPI) were independently associated with shorter overall and progression‑free survival. However, the total variant count per patient was not significantly associated with survival after adjustment. Conclusions:Baseline plasma cfDNA and ctDNA assessments are feasible in routine practice and recapitulate tissue-variant landscapes. Elevated cfDNA concentrations-but not the total variant count-were independently associated with survival in DLBCL, providing prognostic information beyond the IPI and supporting integration of plasma-based biomarkers into multiparameter risk models. Gene‑level ctDNA associations should be regarded as exploratory and hypothesis‑generating.
Newborn screening using tandem mass spectrometry requires complex post-analytical workflows, including QC review, patient report generation, and cutoff reassessment, which are repetitive, time-consuming, and error-prone. We developed a Python-based clinical decision support system with a graphical user interface, integrating four functionalities: QC and patient result evaluation with deterministic flagging, report drafting using hybrid rule-based and retrieval-augmented generation-enabled large language models, cutoff analysis, and a source-attributed reference chatbot. All artificial intelligence (AI)-generated outputs required mandatory specialist review before release. Implementation reduced the mean per-batch post-analytical processing time from 50.5±14.8 min to 24.3±8.8 min and the cutoff evaluation time from 9.2±1.9 hr to 63.6±13.6 min. Across 20 synthetic cases representing 10 disorders, five specialists identified no major errors affecting clinical interpretation. Repeated generation from identical inputs (50 drafts) produced stable diagnostic conclusions, and all source-attributed statements cited correct sources, with 56 of 60 fully covered by the cited excerpt. Deterministic QC flagging, patient result flagging, and cutoff calculations were fully concordant with manual review. This human-in-the-loop system demonstrates the feasibility and efficiency of controlled AI integration into clinical laboratory workflows, whereas diagnostic performance requires further validation using confirmed-positive specimens and human-comparator benchmarks.
Background:Digital morphology (DM) analyzers are increasingly used for white blood cell (WBC) differential counting in routine hematology laboratories, necessitating tailored external quality assessment (EQA) schemes. We evaluated EQA-relevant practical considerations for DM analyzer-based WBC differentials. Methods:Fifteen clinical laboratories participated in a multicenter EQA simulation using five centrally prepared, unstained peripheral blood smear samples. Institution-level post-classification results were analyzed as the primary outcomes. Exploratory analyses were performed using pre-classification and manual microscopic differentials. Examiner-level variability within institutions was evaluated, along with the distributions of WBC-classified and unclassified images following examiner verification. Results:Most participating laboratories used DM analyzers primarily for screening WBC differentials, with predefined criteria for manual microscopic reviews, particularly when abnormal cell populations were noted. Inter-institutional and examiner-level variability were most pronounced for morphologically similar cell classes. Post-classification results were generally comparable with those of manual differentials across most cell classes. The proportion of unclassified images (including artifacts) increased, suggesting that operational and sample-related factors influenced the interpretation of the EQA results. Conclusions:We identified key practical considerations for the EQA of DM analyzer-based WBC differential counting. An appropriate EQA design should incorporate clinically relevant abnormal cell populations and challenging sample types; data interpretation should extend beyond standard deviation index-based metrics to address abnormal cell detection. The overall concordance between post-classification results and manual differentials supports the importance of periodic verification, and slide quality should be considered when implementing DM analyzer-based EQA programs.
Background:Although measurable residual disease (MRD) is an established prognostic marker in pediatric acute myeloid leukemia (AML), its dynamic prognostic value during treatment and its integration with the KIT variant status in core-binding factor (CBF)-AML remain incompletely defined. We evaluated the prognostic impact of sequential MRD and KIT variants in pediatric CBF-AML. Methods:We retrospectively analyzed 112 children with CBF-AML. We assessed MRD in bone marrow samples collected at diagnosis, the end of induction (EOI), the end of second consolidation (EO2C), and the end of consolidation (EOC) using quantitative reverse transcription PCR with TaqMan assays targeting two fusion transcripts. Results:In RUNX1::RUNX1T1 AML, a ≥ 3-log reduction (LR) in MRD at EO2C or a ≥ 4-LR at EOC was associated with favorable outcomes, regardless of the MRD at EOI. Conversely, RUNX1::RUNX1T1 AML with an MRD of <4-LR and a normalized copy number of >500 at EOC was associated with extremely poor survival. CBFB::MYH11 AML with an MRD at EOI of ≥ 2-LR and an MRD of ≥ 4-LR at EOC was associated with better 5-yr overall survival and relapse-free survival (RFS), respectively. RUNX1::RUNX1T1 AML with a KIT D816 variant showed significantly lower 5-yr RFS (P =0.017). In patients with CBF-AML harboring a KIT variant, those who achieved MRD of ≥ 4-LR at EOC had significantly more favorable outcomes. Conclusions:Our findings highlight the importance of sequential MRD monitoring in pediatric patients with RUNX1::RUNX1T1 AML and show that integrating the KIT variant status with MRD assessment provides additional prognostic value during treatment.
Background:Hb variants are often incidentally detected during HbA1c testing and can cause interference. We evaluated the ability of the ARKRAY HA-8180V HPLC system operating in fast mode (FM) to identify Hb variants, particularly HbE, in samples submitted for HbA1c testing based on characteristic retention times (RTs). Methods:Chromatograms of samples tested for HbA1c were analyzed; those showing predefined abnormal chromatographic patterns suggestive of Hb variants were further evaluated using capillary electrophoresis (CE) and β- and α-globin gene sequencing. RTs were recorded. Results:Of 4,888 chromatograms analyzed, 588 (12.0%) were abnormal. Among these, 564 (95.9%) showed an abnormal peak at an RT of 23-25 s and were confirmed to contain HbE via CE. Identified Hb phenotypes included EA (N=526), compound heterozygotes [Constant Spring (CS) EA, N=2; CSEE, N=1], EE (N=27), EFA (N=3), and EF (N=5). Other variants included Hb CSA2ABart'sH, A2FA, A2FABart's, and A2F. Rare variants, Hb Hope (N=12) and Hb Korle-Bu (N=1), and two uncharacterized variants were also detected. Hb Hope interfered with HbA1c measurements, and all CSA2ABart'sH samples produced falsely low HbA1c values. Artificially decreased HbA1c was detected in 21/27 (77.8%) HbEE samples. ARKRAY HA-8180V FM analysis yielded reproducible RTs for HbE, achieving 100% sensitivity, specificity, and accuracy in the validation cohort. Conclusions:HbE and thalassemia variants can interfere with HbA1c quantification. ARKRAY HA-8180V FM chromatogram review allows reliable detection of HbE and prevents misinterpretation of HbA1c results.
Background:In the 5th edition of the WHO Classification of Hematopoietic Neoplasms (WHO-HAEM5) and the International Consensus Classification (ICC), acute myeloid leukemia (AML) diagnostic criteria were revised to incorporate genetic abnormalities, such as variants in myelodysplasia-related (MR) genes. As the implications of these changes on AML reclassification in Korean patients had not yet been fully evaluated, we comprehensively investigated them in this large-scale, multicenter study. Methods:We retrospectively analyzed data from 2,668 patients with AML aged ≥ 18 yrs from seven institutions who were originally diagnosed according to WHO-HAEM4R criteria. Clinical, cytogenetic, and targeted next-generation sequencing data (including MR genes and TP53 variants) were collected and analyzed. All cases were reassessed according to both WHO-HAEM5 and ICC criteria. Results:Overall, 28.5% of patients harbored at least one MR gene variant. Among the MR genes, variants in RUNX1 and ASXL1 were most frequently detected. Compared with the original diagnoses, 20.3% and 29.9% of patients were reclassified to different categories according to the WHO-HAEM5 and ICC criteria, respectively. Discordance between WHO-HAEM5 and ICC classifications was 13.2%, which is largely attributed to the inclusion of "AMLs with TP53" variants in the ICC. AML-MR was associated with older age, male predominance, and poorer survival compared with AML, not otherwise specified. Patients with recurrent fusion genes largely retained their original classification. Conclusions:Incorporating MR gene variant data and TP53-variant status in diagnosis influenced AML classification and risk stratification in our cohort, highlighting the potential importance of comprehensive molecular profiling in AML characterization.
Background:Analytical performance specifications (APSs) are essential for quality management in medical laboratories. Considering the discrepancies between laboratory performance and existing guidelines, lack of consideration of concentration-dependent variability, and absence of recommendations for certain parameters, we aimed to define APSs for internal use in biochemistry, hemostasis, and hematology based on external quality assessment (EQA) peer group data and evaluate their suitability for intermediate precision assessment versus biological variation (BV)-based APSs. Methods:EQA-based allowable CV (CVallowable) was estimated from pooled CVs derived from EQA peer group results. Allowable bias, expanded measurement uncertainty, and total allowable error were calculated. CVallowable targets were assessed using intermediate precision data from different analytical systems within a laboratory group and compared with BV-based APSs. Concordance between theoretical specifications derived from mathematical models and observed analytical performance (AP) was evaluated using (i) the proportion of internal QC CVs meeting the predefined CVallowable across laboratories and (ii) observed analytical imprecision expressed as a percentage of the allowable imprecision budget. Results:APSs were established for 110 biochemical analytes, 23 hemostasis parameters, and 28 hematology parameters across different concentration ranges. EQA-derived APSs agreed with observed intermediate precision for 102 biochemical, 14 hemostasis, and 24 hematology parameters. BV-based APSs showed agreement for only 49 biochemical, two hemostasis, and 11 hematology parameters, while overly restrictive goals or lack of agreement were observed for several analytes. Conclusions:APSs derived from pooled EQA peer group data provide realistic and technically achievable intermediate precision targets consistent with current AP and thus can complement BV-based specifications.
Background:With developments in artificial intelligence, patient-based real-time quality control (PBRTQC) has advanced. However, its application to semiquantitative tests remains unreported. Taking urine protein (URP), a semiquantitative parameter in routine urinalysis, as an example, we explored the application of PBRTQC in semiquantitative tests. Methods:We assessed the correlation between URP and the quantitative parameter, urine total protein (UTP). Measurement results for both analytes were classified into five grades to calculate hierarchical differences (HDs). Three HD-based algorithms were established: the moving rate of inconsistency (MRI), moving average of HDs (MAHD), and moving average of absolute HDs (MAAD). Their performance was compared with that of the moving rate of positive results (MRP) using computer simulations to evaluate the detection of systematic errors (SEs) and random errors (REs). Anti-interference capability was evaluated by rearranging the sample sequence. Results:The MRP method detected SEs but failed to monitor REs. Compared with MRP, HD-based algorithms reduced the median number of patients affected before detection at critical errors by 83.1% for SEs and by 94.2% for REs. MRI, MAHD and MAAD showed significantly improved detection capabilities for SEs and, particularly, REs. Regarding anti-interference, rearranging the sample sequence significantly deteriorated MRP performance, whereas the HD-based algorithms remained unaffected and stable. Conclusions:The novel HD-based algorithms demonstrate superior error detection and anti-interference capabilities compared with MRP. This study provides an effective PBRTQC strategy for semiquantitative tests by leveraging related quantitative data.
Candida auris clade II isolates are generally more susceptible to antifungal agents and are less frequently associated with outbreaks than non-clade II isolates. We developed and validated a Fourier-transform infrared (FTIR) spectroscopy-based classifier (AurisC2-ID) to distinguish clade II from non-clade II C. auris isolates. In total, 106 C. auris isolates col-lected from 14 Korean hospitals, representing clades I and II, and 10 reference isolates from the Centers for Disease Control and Prevention and Food and Drug Administration Antimicrobial Resistance Isolate Bank, representing four clades (I-IV), were analyzed using FTIR spectroscopy (IR Biotyper; Bruker Daltonics, Bremen, Germany) as the training set for classifier development. The classifier was constructed using an artificial neural network al-gorithm following principal component analysis and was validated using 87 additional clini-cal isolates collected from nine Korean hospitals. The training set spectra showed clear separation between clade II and non-clade II isolates, with minor overlap between the two groups. During validation, all 31 clade II isolates were correctly classified as clade II, and the remaining 56 clade I isolates as non-clade II. These results demonstrate that AurisC2-ID can accurately distinguish clade II from non-clade II C. auris isolates and may serve as a useful tool for infection control in Korea.
Background:Commercial line immunoassays (LIAs) are used to detect myositis-specific and myositis-associated autoantibodies (MSAs/MAAs) in patients with suspected idiopathic inflammatory myopathy (IIM). However, performance consistency with LIAs remains challenging. We evaluated the appropriateness, clinical consistency, and diagnostic performance of MSA/MAA LIA testing in real-life clinical practice. Methods:Between January and December 2024, eight Italian laboratories performed 1,127 MSA/MAA LIA determinations (Euroline Myopathies 16 Ags plus cN-1A). LIA appropriateness was defined as concordance between the clinical information provided and the MSA/MAA request; clinical consistency was defined as agreement between the IIM diagnosis and LIA results; and diagnostic performance was defined as the reliability of LIA in confirming an IIM diagnosis. Results:MSA/MAA positivity was observed with 564/1,127 (50.1%) LIA requests. LIA appropriateness and consistency were 31.6% and 51.3%, respectively. Of the 1,127 LIA requests, 666 (59.1%) included a clinical diagnosis. The appropriateness and consistency of the 666 LIA reports were 53.4% and 86.8%, respectively. Among the 666 patients, 337 (50.6%) were diagnosed as having IIM, and 329 (49.4%) were diagnosed as having a non-IIM disease. In patients with IIM, 288/337 (85.4%) received true-positive MSA/MAA LIA results, whereas 105/329 (31.9%) of patients with non-IIM disease received false-positive results (P<0.001). Significant positive odds ratios were observed for anti-Jo-1, anti-MDA5, anti-PM-Scl-75/100, and MSA-associated anti-Ro52 antibodies (P<0.001). Conclusions:With increasing MSA/MAA requests for suspected IIM, LIA testing can help identify certain antibodies in real-life scenarios. The appropriateness of the request and the clinical information provided mainly influence the LIA diagnostic performance.
Background:Hepatocellular carcinoma (HCC) remains a leading cause of cancer-related mortality worldwide. To enhance early detection, the ASAP (age, sex, alpha-fetoprotein [AFP], protein induced by vitamin K absence or antagonist-II [PIVKA-II]) and GAAD (gender, age, AFP, des-gamma-carboxyprothrombin [DCP]/PIVKA-II) models were developed by integrating demographic data with serum biomarkers. We compared their performance in a Korean chronic liver disease cohort. Methods:We retrospectively analyzed data from 524 patients, including 132 with and 392 without HCC. AFP and PIVKA-II levels were measured using Abbott (ASAP) and Roche (GAAD) analyzers. Performance was assessed based on area under the ROC curve (AUROC) and optimal cutoff values for the overall cohort, etiologic subgroups (hepatitis B virus [HBV], hepatitis C virus [HCV], alcohol-related), and early-stage HCC (modified Union for International Cancer Control stage I or II). Results:In the overall cohort, both models demonstrated high, comparable performance (P =0.482). The ASAP model achieved an AUROC of 0.945 (sensitivity 81.8%, specificity 93.4%; cutoff 0.404), whereas the GAAD model yielded an AUROC of 0.950 (sensitivity 85.6%, specificity 93.6%; cutoff 1.34). No statistically significant differences were observed in etiologic subgroups or early-stage HCC (P =0.702), with AUROCs remaining high (0.911 for ASAP and 0.916 for GAAD). Conclusions:The ASAP and GAAD algorithms provide excellent and comparable diagnostic performance for detecting HCC, including in early-stage cases, regardless of etiology. Given Korea's high HBV prevalence and platform variability, these models serve as robust, non-invasive complementary tools for surveillance. This validation supports their clinical utility in the Korean population.
