Artificial Intelligence based Models for Screening of Hematologic Malignancies using Cell Population Data

Cell Population Data (CPD) provides various blood cell parameters that can be used for differential diagnosis. Data analytics using Machine Learning (ML) have been playing a pivotal role in revolutionizing medical diagnostics. This research presents a novel approach of using ML algorithms for screening hematologic malignancies using CPD. The data collection was done at Konkuk University Medical Center, Seoul. A total of (882 cases: 457 hematologic malignancy and 425 hematologic non-malignancy) were used for analysis. In our study, seven machine learning models, i.e., SGD, SVM, RF, DT, Linear model, Logistic regression, and ANN, were used. In order to measure the performance of our ML models, stratified 10-fold cross validation was performed, and metrics, such as accuracy, precision, recall, and AUC were used. We observed outstanding performance by the ANN model as compared to other ML models. The diagnostic ability of ANN achieved the highest accuracy, precision, recall, and AUC ± Standard Deviation as follows: 82.8%, 82.8%, 84.9%, and 93.5% ± 2.6 respectively. ANN algorithm based on CPD appeared to be an efficient aid for clinical laboratory screening of hematologic malignancies. Our results encourage further work of applying ML to wider field of clinical practice.

Automation of Harboe method for the measurement of plasma free hemoglobin

BACKGROUND

Although plasma free hemoglobin (fHb) test is important for assessing intravascular hemolysis, it is still dependent on the gold standard Harboe method using manual and labor-intensive spectrometric measurements at the wavelength of 380-415-450 nm. We established an automated fHb assay using a routine chemistry autoanalyzer that can be tuned to a wavelength of 380-416-450 nm.

METHODS

The linearity, precision, accuracy, correlation, and sample carryover of fHb measurement using TBA200FRneo method and manual Harboe method were evaluated, respectively. fHb values measured by manual Harboe method were compared with those measured by our new automated TBA200FRneo method.

RESULTS

fHb measurements were linear in the range of 0.05~38.75 µmol/L by TBA200FRneo and 0.05~9.69 µmol/L by manual Harboe method. Imprecision analysis (%CV) revealed 0.9~2.8% for TBA200FRneo method and 5.3~13.6% for the manual Harboe method. Comparison analysis showed 0.9986 of correlation coefficient (TBA200FRneo = 0.970 × Harboe + 0.12). In analytical accuracy analysis, the manual Harboe method revealed about 4 times higher average total error % (12.2%) than the TBA200FRneo automated method (2.8%). The sample carryover was -0.0016% in TBA200FRneo method and 0.0038% in Harboe method.

CONCLUSIONS

In the measurement of fHb, the automated TBA200FRneo method showed better performance than the conventional Harboe method. It is expected that the automated fHb assay using the routine chemistry analyzer can replace the gold standard Harboe method which is labor-intensive and need an independent spectrophotometry equipment.

Performance Evaluation of the QXDx BCR-ABL %IS Droplet Digital PCR Assay

Accurate detection of BCR-ABL fusion transcripts at and below molecular response (MR) 4 (0.01% International Scale [IS]) is required for disease monitoring in patients with chronic myeloid leukemia (CML). We evaluated the analytical performance of the QXDx BCR-ABL %IS (Bio-Rad, Hercules, CA, USA) droplet digital PCR (ddPCR) assay, which is the first commercially available ddPCR-based in vitro diagnostics product. In precision analysis, the %CV was 9.3% and 3.0%, with mean values of 0.031% IS and 9.4% IS, respectively. The assay was linear in the first order, ranging from 0.032% IS to 20% IS. The manufacturer-claimed limit of blank, limit of detection, and limit of quantification were verified successfully. There was a very strong correlation between the results of the QXDx BCR-ABL %IS ddPCR assay and the ipsogen BCR-ABL1 Mbcr IS-MMR (Qiagen, Hilden, Germany) real-time quantitative PCR assay (r=0.996). In conclusion, the QXDx BCR-ABL %IS ddPCR assay can provide reliable results for CML patients.

Benefits of VISION Max automated cross-matching in comparison with manual cross-matching: A multidimensional analysis

BACKGROUND

VISION Max (Ortho-Clinical Diagnostics, Raritan, NJ, USA) is a newly introduced automated blood bank system. Cross-matching (XM) is an important test confirming safety by simulating reaction between packed Red Blood Cells (RBCs) and patient blood in vitro before transfusion. We assessed the benefits of VISION Max automated XM (A-XM) in comparison with those of manual XM (M-XM) by using multidimensional analysis (cost-effectiveness and quality improvement).

MATERIALS AND METHODS

In a total of 327 tests (130 patients), results from A-XM and M-XM were compared. We assessed the concordance rate, risk priority number (RPN), turnaround time, hands-on time, and the costs of both methods. We further simulated their annual effects based on 37,937 XM tests in 2018.

RESULTS

The concordance rate between A-XM and M-XM was 97.9% (320/327, kappa = 0.83), and the seven discordant results were incompatible for transfusion in A-XM, while compatible for transfusion in M-XM. None of the results was incompatible for transfusion in A-XM, while compatible for transfusion in M-XM, meaning A-XM detect agglutination more sensitively and consequently provides a more safe result than M-XM. A-XM was estimated to have a 6.3-fold lower risk (229 vs. 1,435 RPN), shorter turnaround time (19.1 vs. 23.3 min, P < 0.0001), shorter hands-on time (1.1 vs. 5.3 min, P < 0.0001), and lower costs per single test than M-XM (1.44 vs. 2.70 USD). A-XM permitted annual savings of 46 million RPN, 15.1 months of daytime workers' labor, and 47,042 USD compared with M-XM.

CONCLUSION

This is the first attempt to implement A-XM using VISION Max. VISION Max A-XM appears to be a safe, practical, and reliable alternative for pre-transfusion workflow with the potential to improve quality and cost-effectiveness in the blood bank.

Circulating Biologically Active Adrenomedullin Predicts Organ Failure and Mortality in Sepsis

BACKGROUND

Sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection. Biologically active adrenomedullin (bio-ADM) is an emerging biomarker for sepsis. We explored whether bio-ADM concentration could predict severity, organ failure, and 30-day mortality in septic patients.

METHODS

In 215 septic patients (109 patients with sepsis; 106 patients with septic shock), bio-ADM concentration was measured at diagnosis of sepsis, using sphingotest bio-ADM (Sphingotec GmbH, Hennigsdorf, Germany) and analyzed in terms of sepsis severity, vasopressor use, and 30-day mortality. The number of organ failures, sequential (sepsis-related) organ failure assessment (SOFA) score, and 30-day mortality were compared according to bio-ADM quartiles.

RESULTS

Bio-ADM concentration was significantly higher in patients with septic shock, vasopressor use, and non-survivors than in patients with solitary sepsis, no vasopressor use, and survivors, respectively (all P<0.0001). Bio-ADM quartiles were associated with the number of organ failures (P<0.0001), as well as SOFA cardiovascular, renal, coagulation, and liver subscores (all P<0.05). The 30-day mortality rate showed a stepwise increase in each bio-ADM quartile (all P<0.0001). Bio-ADM concentration and SOFA score equally predicted the 30-day mortality (area under the curve: 0.827 vs 0.830).

CONCLUSIONS

Bio-ADM could serve as a useful and objective biomarker to predict severity, organ failure, and 30-day mortality in septic patients.

Performance evaluation of cobas HBV real-time PCR assay on Roche cobas 4800 System in comparison with COBAS AmpliPrep/COBAS TaqMan HBV Test

BACKGROUND

Hepatitis B virus (HBV) nucleic acid amplification testing (NAAT) is important for the diagnosis and management of HBV infection. We evaluated the analytical performance of the cobas HBV NAAT (Roche Diagnostics GmbH, Mannheim, Germany) on the cobas 4800 System in comparison with COBAS AmpliPrep/COBAS TaqMan HBV Test (CAP/CTM HBV).

METHODS

Precision was evaluated using three levels of cobas HBV/HCV/HIV-1 Control Kit, and linearity was evaluated across the anticipated measuring range (10.0-1.0×109 IU/mL) at seven levels using clinical samples. Detection capability, including limit of blank (LOB), limit of detection (LOD) and limit of quantitation (LOQ), was verified using the 4th WHO International Standard for HBV DNA for NAT (NIBSC code: 10/266). Correlation between the two systems was compared using 205 clinical samples (102 sera and 103 EDTA plasma).

RESULTS

Repeatability and total imprecision (coefficient of variation) ranged from 0.5% to 3.8% and from 0.5% to 3.5%, respectively. Linearity (coefficient of determination, R2) was 0.999. LOB, LOD and LOQ were all acceptable within the observed proportion rate (85%). Correlation was very high between the two systems in both serum and plasma samples (correlation coefficient [r]=0.995).

CONCLUSIONS

The new cobas HBV real-time PCR assay on the cobas 4800 System showed reliable analytical performances.

Strategy for performing treponemal tests in reverse-sequence algorithms of syphilis diagnosis

OBJECTIVES

In South Korea, automated T. pallidum Latex Agglutination (TPLA) based on turbidoimmunoassays and immunochromatographic assay (ICA) are widely used for syphilis diagnosis. However, there is sparse data on the validation of these assays in the reverse-sequence algorithm setting.

METHODS

We assessed 551 specimens submitted for syphilis testing. We compared varying reverse-sequence algorithms using combinations of the Cobas Syphilis EIA (Roche Diagnostics, Mannheim, Germany), Mediace TPLA (Sekisui Medical Co., Tokyo, Japan), TPPA (Fujirebio Inc., Tokyo, Japan), and SD Bioline ICA (Standard Diagnostic, Yongin, Korea). We also evaluated modified algorithms incorporating a cut off of high specificity for EIA and TPLA using receiver operating characteristic curves.

RESULTS

The agreement was almost perfect between EIA and TPLA (Kappa, 0.953) and strong between TPPA and ICA (Kappa, 0.887). Among TPPA positive and ICA negative specimens, 67% of the specimens were from individuals with syphilis histories. Compared to EIA/RPR/TPPA, the agreement with EIA/RPR/ICA, TPLA/RPR/TPPA and TPLA/RPR/ICA were almost perfect (Kappa, 0.930, 0.995 and 0.914, respectively). When a cut off of 95% specificity was applied, the number of TPPA tests could be reduced by 44% and 40% in EIA and TPLA, respectively.

CONCLUSIONS

TPLA showed almost perfect agreement with EIA and that it could be used in the site of EIA in a reverse sequence algorithm. ICA showed a lower detection rate than TPPA as a 2nd treponemal test and should be used with caution. With cut offs of higher specificity, more efficient reverse-sequence algorithms can be made possible.

Usefulness of Enhanced Liver Fibrosis, Glycosylation Isomer of Mac-2 Binding Protein, Galectin-3, and Soluble Suppression of Tumorigenicity 2 for Assessing Liver Fibrosis in Chronic Liver Diseases

BACKGROUND

Liver biopsies have been partially replaced by noninvasive methods for assessing liver fibrosis. We explored the usefulness of four novel biomarkers, enhanced liver fibrosis (ELF), glycosylation isomer of Mac-2 binding protein (M2BPGi), galectin-3, and soluble suppression of tumorigenicity 2 (sST2), in association with liver fibrosis.

METHODS

ELF, M2BPGi, galectin-3, and sST2 were assayed in 173 patients with chronic liver diseases. The results were analyzed according to fibrosis grade (F0/1, F2, and F3/4) by transient elastography (TE).

RESULTS

ELF, M2BPGi, galectin-3, and sST2 values differed significantly according to TE grade; ELF and M2BPGi values were higher in F2 and F3/4 than in F0/1 (P≤0.001, all), sST2 values were higher in F3/4 than in F0/1 and F2 (P<0.05), and galectin-3 values were higher in F3/4 than in F0/1 (P=0.0036). ELF and M2BPGi showed good TE fibrosis detection performance (area under the curves [AUC], 0.841 and 0.833 for ≥F2; and 0.837 and 0.808 for ≥F3). The sensitivity and specificity for predicting TE grade F≥2 were 84.1% and 76.7% for ELF and 63.6% and 91.5% for M2BPGi.

CONCLUSIONS

This is the first study to compare the liver fibrosis assessment of four novel biomarkers: ELF, M2BPGi, galectin-3, and sST2. The biomarkers varied significantly according to TE grade, and each biomarker showed a different trend. ELF and M2BPGi seem to have comparable good performance for detecting liver fibrosis.

Characteristics of Helicobacter pylori-seropositive subjects according to the stool antigen test findings: a prospective study

BACKGROUND/AIMS

In countries with a higher risk of gastric atrophic gastritis, noninvasive tests are helpful for a more reliable diagnosis of Helicobacter pylori infection. The aim of this study was to evaluate the characteristics of seropositive subjects according to their stool H. pylori antigen test, serum pepsinogen (PG) assay, and endoscopic findings.

METHODS

Consecutive subjects who visited Konkuk University Medical Center for upper gastrointestinal endoscopy for a regular check-up were included in a prospective setting if the serum anti-H. pylori immunoglobulin G assay was positive. A H. pylori antigen stool test was measured using a stool H. pylori antigen enzyme-linked immunosorbent assay kit on the same day as a serum PG assay and endoscopy.

RESULTS

Of 318 seropositive subjects, 256 (80.5%) showed positive stool test findings. Subjects with a negative stool test result showed lower serum PG I (p < 0.001) and PG II (p < 0.001) levels and higher PG I/II ratio (p < 0.001) than those with a positive stool test. Chronic atrophic gastritis was more common in the positive stool test group than the negative stool test group on endoscopic finding (p = 0.009). A higher serum PG I level (p = 0.001) and a lower serum PG I/II ratio (p = 0.001) were independent risk factors for the presence of H. pylori antigen in stool.

CONCLUSION

A high serum PG level denotes an ongoing current H. pylori infection with positive stool H. pylori antigen test findings. Seropositive subjects with increased gastric secreting ability tend to have H. pylori in their fecal material as reflected by a positive stool H. pylori antigen test finding.

Prognostic Role of High-sensitivity Cardiac Troponin I and Soluble Suppression of Tumorigenicity-2 in Surgical Intensive Care Unit Patients Undergoing Non-cardiac Surgery

Background

The prognostic utility of cardiac biomarkers, high-sensitivity cardiac troponin I (hs-cTnI) and soluble suppression of tumorigenicity-2 (sST2), in non-cardiac surgery is not well-defined. We evaluated hs-cTnI and sST2 as predictors of 30-day major adverse cardiac events (MACE) in patients admitted to the surgical intensive care unit (SICU) following major non-cardiac surgery.

Methods

hs-cTnI and sST2 concentrations were measured in 175 SICU patients immediately following surgery and for three days postoperatively. The results were analyzed in relation to 30-day MACE and were compared with the revised Goldman cardiac risk index (RCRI) score.

Results

Overall, 30-day MACE was observed in 16 (9.1%) patients. hs-cTnI and sST2 concentrations differed significantly between the two groups with and without 30-day MACE (P<0.05). The maximum concentration of sST2 was an independent predictor of 30-day MACE (odds ratio=1.016, P=0.008). The optimal cut-off values of hs-cTnI and sST2 for predicting 30-day MACE were 53.0 ng/L and 182.5 ng/mL, respectively. A combination of hs-cTnI and sST2 predicted 30-day MACE better than the RCRI score. Moreover, 30-day MACE was observed more frequently with increasing numbers of above-optimal cut-off hs-cTnI and sST2 values (P<0.0001). Reclassification analyses indicated that the addition of biomarkers to RCRI scores improved the prediction of 30-day MACE.

Conclusions

This study demonstrates the utility of hs-cTnI and sST2 in predicting 30-day MACE following non-cardiac surgery. Cardiac biomarkers would provide enhanced risk stratification in addition to clinical RCRI scores for patients undergoing major non-cardiac surgery.