Machine Learning-Based Prediction of Hemoglobinopathies Using Complete Blood Count Data
Open accessAbstract
Background: Hemoglobinopathies, the most common inherited blood disorder, are frequently underdiagnosed. Early identification of carriers is important for genetic counseling of couples at risk. The aim of this study was to develop and validate a novel machine learning model on a multicenter data set, covering a wide spectrum of hemoglobinopathies based on routine complete blood count (CBC) testing. Methods: Hemoglobinopathy test results from 10 322 adults were extracted retrospectively from 8 Dutch laboratories. eXtreme Gradient Boosting (XGB) and logistic regression models were developed to differentiate negative from positive hemoglobinopathy cases, using 7 routine CBC parameters. External validation was conducted on a data set from an independent Dutch laboratory, with an additional external validation on a Spanish data set (n = 2629) specifically for differentiating thalassemia from iron deficiency anemia (IDA). Results: The XGB and logistic regression models achieved an area under the receiver operating characteristic (AUROC) of 0.88 and 0.84, respectively, in distinguishing negative from positive hemoglobinopathy cases in the independent external validation set. Subclass analysis showed that the XGB model reached an AUROC of 0.97 for β-thalassemia, 0.98 for α0-thalassemia, 0.95 for homozygous α+-thalassemia, 0.78 for heterozygous α+-thalassemia, and 0.94 for the structural hemoglobin variants Hemoglobin C, Hemoglobin D, Hemoglobin E. Both models attained AUROCs of 0.95 in differentiating IDA from thalassemia. Conclusions: Both the XGB and logistic regression model demonstrate high accuracy in predicting a broad range of hemoglobinopathies and are effective in differentiating hemoglobinopathies from IDA. Integration of these models into the laboratory information system facilitates automated hemoglobinopathy detection using routine CBC parameters.
AI snapshot
Written from the paper’s full text.
TL;DR
A multicenter retrospective study developed and validated XGBoost and logistic regression models using routine CBC parameters to detect a broad range of hemoglobinopathies in adults. Both models showed high discrimination, including differentiating thalassemia from iron deficiency anemia, with the logistic regression model offering a practical formula for laboratory implementation.
Key findings
- In independent external validation, the XGB model achieved an AUROC of 0.88 and logistic regression 0.84 for distinguishing negative from positive hemoglobinopathy cases.
- XGB subclass AUROCs were 0.97 for β-thalassemia, 0.98 for α⁰-thalassemia, 0.95 for homozygous α⁺-thalassemia, 0.78 for heterozygous α⁺-thalassemia, and 0.94 for structural hemoglobin variants Hb C, Hb D, and Hb E.
- Both models attained an AUROC of 0.95 for differentiating thalassemia from iron deficiency anemia in a Spanish external data set; for α-thalassemia vs IDA, XGB and logistic regression reached 0.90 and 0.89, and for β-thalassemia vs IDA both reached 0.97.
- In a screening simulation on 19,969 CBC results with 118 known positive cases, logistic regression had an AUROC of 0.98 vs 0.97 for XGB; at 99.8% specificity, logistic regression sensitivity was 57% vs 35% for XGB, and maximum MCC was 0.65 for logistic regression vs 0.51 for XGB.
Methods
- Retrospective multicenter study using hemoglobinopathy test results from 10,322 adults across 8 Dutch laboratories over 12 years (2011–2022); model development used 7 laboratories with a final data set of 5,908 cases (3,249 positive, 2,659 negative), split 80% training/tuning and 20% internal validation.
- XGBoost and logistic regression models were trained on 7 routine CBC parameters—Hb, MCV, MCH, MCHC, RDWCV, RBC, and platelets—using clinical interpretation of hemoglobinopathy test results as ground truth; external validation used Medlon data (n = 2,656; 1,004 positive, 1,652 negative) and a Spanish data set (n = 2,629; 1,370 thalassemia, 1,259 IDA).
- Performance was assessed with AUROC, AUPRC, and MCC, including subclass analyses and a screening simulation on 19,969 CBC results from Jeroen Bosch Hospital enriched with 68 positive cases.
Limitations
- The paper states the model was developed exclusively for adults, and that GAP-PCR/strip-assay methods detect approximately 90% of α-thalassemia mutations, so some “negative” cases may have undetected α-thalassemia, leading to model underperformance.
- The paper states that XGB is not advised for ICU patients because many false-positive XGB cases were severely anemic ICU patients with high MCV, likely mimicking sickle cell CBC results.
- Evident from the text: the screening evaluation classified “negative” cases as those without a diagnosed hemoglobinopathy, so this group may contain an unknown limited number of false-negative cases.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
10 of 16 cited sources checked · 5 supported · 0 flagged · accuracy not yet measured
- Premisecitation
The strip-assay methods can detect approximately 90% of α-thalassemia gene mutations.
“Secondly, the GAP-PCR was specifically designed to identify the common α-thalassemia deletions, and the strip-assay methods can detect approximately 90% of α-thalassemia gene mutations (8,37).”
Quote in the paperSubstantial claim1 of 2 sources checked- cites [8] Harteveld, 2010, Alpha-thalassaemia, Orphanet J Rare Dis source
The Molecular analysis section describes gap-PCR, sequencing and MLPA, but does not address strip-assay detection rates. Searches for strip assays and the approximately 90% figure found no passage supporting that claim. (LLM, conf=0.74)
Experimental — accuracy not yet measured - cites [37] Puehringer, 2007, Validation of a reverse-hybridization StripAssay for the simultaneous analysis of common alpha-thalassemia point mutations and deletions, Clin Chem Lab Med source
- Premisecitation
The GAP-PCR was specifically designed to identify the common α-thalassemia deletions.
“Secondly, the GAP-PCR was specifically designed to identify the common α-thalassemia deletions, and the strip-assay methods can detect approximately 90% of α-thalassemia gene mutations (8,37).”
Quote in the paperSubstantial claim1 of 2 sources checked - Premisecitation
Existing models are primarily designed using small single-center data sets lacking independent validation.
“Moreover, these models are primarily designed using small single-center data sets lacking independent validation (15,16,18,(33)(34)(35)(36).”
Quote in the paperSubstantial claim3 of 3 sources checked- cites [15] Laengsri, 2019, ThalPred: a web-based prediction tool for discriminating thalassemia trait and iron deficiency anemia, BMC Med Inform Decis Mak source
The Data collection section (page unavailable) describes a single laboratory source and a final dataset of 186 subjects; the Discussion calls the iron-deficiency-anemia sample small. But the Methods and Results report external validation and testing on an independent dataset, so this paper does not support the claim that its model lacked independent validation. (LLM, conf=0.87)
Experimental — accuracy not yet measured - cites [16] Fu, 2021, The TVGH-NYCU thal-classifier: development of a machine-learning classifier for differentiating thalassemia and non-thalassemia patients, Diagnostics source
- cites [18] Rustam, 2022, Prediction of β-thalassemia carriers using complete blood count features, Sci Rep source
The Dataset description identifies one institute’s database as the source, and Data splitting calls the dataset small and describes an internal train–test split. This supports the paper as an example of a model built on a limited dataset without reported independent validation, but it does not establish that existing models are *primarily* designed this way. The copy also does not explicitly identify the database as single-center. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
The majority of existing ML models or rule-based formulas specifically differentiate one distinct thalassemia (α or β) from IDA.
“The majority of these formulas specifically differentiate one distinct thalassemia (α or β) from IDA (e.g., [14][15][16][17][18][19].”
Quote in the paperSubstantial claim2 of 3 sources checked- cites [14] Urrechaga, 2017, Critical appraisal of discriminant formulas for distinguishing thalassemia from iron deficiency in patients with microcytic anemia, Clin Chem Lab Med source
- cites [15] Laengsri, 2019, ThalPred: a web-based prediction tool for discriminating thalassemia trait and iron deficiency anemia, BMC Med Inform Decis Mak source
The Methods (Models construction; page unavailable) describe ML models that distinguish thalassemia trait from IDA, so this paper supports that part of the claim as an example. The Methods (Data collection; page unavailable) classify patients as having thalassemia trait without specifying a single α- or β-thalassemia subtype. The paper also does not establish that a majority of existing formulas are subtype-specific. (LLM, conf=0.86)
Experimental — accuracy not yet measured - cites [16] Fu, 2021, The TVGH-NYCU thal-classifier: development of a machine-learning classifier for differentiating thalassemia and non-thalassemia patients, Diagnostics source
- Premisecitation
The majority of existing rule-based formulas and ML models are tailored to specific subcategories or subgoals, such as identifying one specific thalassemia or distinguishing these from IDA.
“The majority of these rule-based formulas and ML models are tailored to specific subcategories or subgoals, such as identifying one specific thalassemia (19,(23)(24)(25)(26) and distinguishing these from IDA (15,17,(27)(28)(29)(30)(31)(32).”
Quote in the paperSubstantial claim2 of 2 sources checked- cites [19] Phirom, 2022, DeepThal: a deep learning-based framework for the large-scale prediction of the α+-thalassemia trait using red blood cell parameters, J Clin Med source
- cites [15] Laengsri, 2019, ThalPred: a web-based prediction tool for discriminating thalassemia trait and iron deficiency anemia, BMC Med Inform Decis Mak source
- Findingown data
Positive hemoglobinopathy cases had significantly lower MCV than negative cases.
“These patterns included significantly lower MCV, MCH, and MCHC, and notably higher RDWCV and RBC levels than negative cases (Table 1).”
Quote in the paperSubstantial claim - Findingown data
The logistic regression model achieved an AUROC of 0.84 (±0.01) in distinguishing negative from positive hemoglobinopathy cases in the independent external validation set.
“The XGB and logistic regression models achieved an area under the receiver operating characteristic (AUROC) of 0.88 and 0.84, respectively, in distinguishing negative from positive hemoglobinopathy cases in the independent external validation set.”
- Also stated in: § Automated Detection of Hemoglobinopathies · plain
“On the independent external validation set, the XGB model achieved an AUROC of 0.88 (±0.01) and the logistic regression an AUROC of 0.84 (±0.01) (Fig. 2A and D).”
Quote in the paperSubstantial claim- Rests on an assumption: Medlon's comprehensive diagnostic evaluation, including Sanger sequencing of both α and β globin genes, provides accurate hemoglobinopathy classification for external validation.
- Also stated in: § Automated Detection of Hemoglobinopathies · plain
- Findingown data
The logistic regression model distinguished α-thalassemia from IDA with an AUROC of 0.89.
“Subclass analysis demonstrated that the XGB and logistic regression models distinguished α-thalassemia from IDA with an AUROC of 0.90 and 0.89, respectively.”
Quote in the paperSubstantial claim - Findingown data
Positive hemoglobinopathy cases had significantly lower MCHC than negative cases.
“These patterns included significantly lower MCV, MCH, and MCHC, and notably higher RDWCV and RBC levels than negative cases (Table 1).”
Quote in the paperNot a substantial claim - Findingown data
The XGB model attained an AUROC of 0.95 in differentiating IDA from thalassemia.
“Both models attained AUROCs of 0.95 in differentiating IDA from thalassemia.”
- Also stated in: § DIFFERENTIATING THALASSEMIA FROM IDA · plain
“Both the XGB and logistic regression model attained an AUROC of 0.95 (±0.01) in differentiating cases with IDA and thalassemia (Fig. 4A and D).”
Quote in the paperSubstantial claim - Also stated in: § DIFFERENTIATING THALASSEMIA FROM IDA · plain
- Findingown data
The XGB model achieved an internal validation AUROC of 0.90 (±0.01).
“The internal validation of the XGB model achieved an AUROC of 0.90 (±0.01) compared to an”
Quote in the paperSubstantial claim - Findingown data
The XGB model achieved an AUROC of 0.88 (±0.01) in distinguishing negative from positive hemoglobinopathy cases in the independent external validation set.
“The XGB and logistic regression models achieved an area under the receiver operating characteristic (AUROC) of 0.88 and 0.84, respectively, in distinguishing negative from positive hemoglobinopathy cases in the independent external validation set.”
- Also stated in: § Automated Detection of Hemoglobinopathies · plain
“On the independent external validation set, the XGB model achieved an AUROC of 0.88 (±0.01) and the logistic regression an AUROC of 0.84 (±0.01) (Fig. 2A and D).”
Quote in the paperSubstantial claim- Rests on an assumption: Medlon's comprehensive diagnostic evaluation, including Sanger sequencing of both α and β globin genes, provides accurate hemoglobinopathy classification for external validation.
- Also stated in: § Automated Detection of Hemoglobinopathies · plain
- Findingown data
The XGB model achieved an AUROC of 0.70 for Hb S heterozygotes.
“Subclass analysis on structural hemoglobin variants demonstrated AUROCs of 0.94 for Hb E, Hb C and Hb D homozygotes and heterozygotes, 0.91 for sickle cell anemia, and 0.70 for Hb S heterozygotes (Fig. 3C).”
Quote in the paperNot a substantial claim - Findingown data
The XGB model achieved an AUROC of 0.97 for β-thalassemia.
“Subclass analysis showed that the XGB model reached an AUROC of 0.97 for β-thalassemia, 0.98 for α 0 -thalassemia, 0.95 for homozygous α + -thalassemia, 0.78 for heterozygous α + -thalassemia, and 0.94 for the structural hemoglobin variants Hemoglobin C, Hemoglobin D, Hemoglobin E.”
- Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · plain
“In the case of β-thalassemia, the XGB model attained an AUROC of 0.97 (Fig. 3B).”
Quote in the paperSubstantial claim - Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · plain
- Findingown data
The XGB model achieved an AUROC of 0.94 for the structural hemoglobin variants Hb E, Hb C, and Hb D homozygotes and heterozygotes.
“Subclass analysis showed that the XGB model reached an AUROC of 0.97 for β-thalassemia, 0.98 for α 0 -thalassemia, 0.95 for homozygous α + -thalassemia, 0.78 for heterozygous α + -thalassemia, and 0.94 for the structural hemoglobin variants Hemoglobin C, Hemoglobin D, Hemoglobin E.”
- Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · boosted
“Subclass analysis on structural hemoglobin variants demonstrated AUROCs of 0.94 for Hb E, Hb C and Hb D homozygotes and heterozygotes, 0.91 for sickle cell anemia, and 0.70 for Hb S heterozygotes (Fig. 3C).”
Quote in the paperNot a substantial claim - Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · boosted
- Findingown data
The XGB model achieved an AUROC of 0.78 for α+ heterozygote.
“Subclass analysis showed that the XGB model reached an AUROC of 0.97 for β-thalassemia, 0.98 for α 0 -thalassemia, 0.95 for homozygous α + -thalassemia, 0.78 for heterozygous α + -thalassemia, and 0.94 for the structural hemoglobin variants Hemoglobin C, Hemoglobin D, Hemoglobin E.”
- Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · plain
“In a subclassification analysis of α-thalassemia, the XGB model achieved AUROC scores of 0.98 for α 0 -heterozygote, 0.95 for α + -homo-and compound heterozygote, and 0.78 for α + -heterozygote (Fig. 3A).”
Quote in the paperNot a substantial claim - Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · plain
- Findingown data
The logistic regression model achieved an internal validation AUROC of 0.86 (±0.01).
“AUROC of 0.86 (±0.01) of the logistic regression (online Supplemental Fig. 2).”
Quote in the paperSubstantial claim - Findingown data
Positive hemoglobinopathy cases had notably higher RDWCV than negative cases.
“These patterns included significantly lower MCV, MCH, and MCHC, and notably higher RDWCV and RBC levels than negative cases (Table 1).”
Quote in the paperNot a substantial claim - Findingown data
The XGB model achieved an AUROC of 0.98 for α0-heterozygote.
“Subclass analysis showed that the XGB model reached an AUROC of 0.97 for β-thalassemia, 0.98 for α 0 -thalassemia, 0.95 for homozygous α + -thalassemia, 0.78 for heterozygous α + -thalassemia, and 0.94 for the structural hemoglobin variants Hemoglobin C, Hemoglobin D, Hemoglobin E.”
- Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · plain
“In a subclassification analysis of α-thalassemia, the XGB model achieved AUROC scores of 0.98 for α 0 -heterozygote, 0.95 for α + -homo-and compound heterozygote, and 0.78 for α + -heterozygote (Fig. 3A).”
Quote in the paperNot a substantial claim - Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · plain
- Findingown data
A probability threshold of ≥0.80 for the XGB model is associated with a 95% specificity cutoff.
“A probability threshold of ≥0.80 for the XGB model and a probability threshold of ≥0.73 for the logistic regression are associated with a 95% specificity cutoff.”
Quote in the paperNot a substantial claim - Findingown data
The logistic regression model reached an AUROC of 0.97 for distinguishing β-thalassemia from IDA.
“Both models reached an AUROC of 0.97 for distinguishing β-thalassemia from IDA (Fig. 4B and E).”
Quote in the paperSubstantial claim - Findingown data
Positive hemoglobinopathy cases had notably higher RBC levels than negative cases.
“These patterns included significantly lower MCV, MCH, and MCHC, and notably higher RDWCV and RBC levels than negative cases (Table 1).”
Quote in the paperNot a substantial claim - Findingown data
The XGB model reached an AUROC of 0.97 for distinguishing β-thalassemia from IDA.
“Both models reached an AUROC of 0.97 for distinguishing β-thalassemia from IDA (Fig. 4B and E).”
Quote in the paperSubstantial claim - Findingown data
The logistic regression model's median predicted probability for IDA (0.56, IQR 0.45 to 0.65) was significantly lower than that for α-thalassemia (0.83, IQR 0.74 to 0.88).
“The median predicted probability scores of logistic regression for cases with an IDA (median = 0.56, IQR: 0.45 to 0.65) was significantly lower compared to α-thalassemia (median = 0.83, IQR: 0.74 to 0.88) and β-thalassemia cases (median = 0.94, IQR: 0.89 to 0.98) (Fig. 4F).”
Quote in the paperNot a substantial claim - Findingown data
The logistic regression model attained an AUROC of 0.95 in differentiating IDA from thalassemia.
“Both models attained AUROCs of 0.95 in differentiating IDA from thalassemia.”
- Also stated in: § DIFFERENTIATING THALASSEMIA FROM IDA · plain
“Both the XGB and logistic regression model attained an AUROC of 0.95 (±0.01) in differentiating cases with IDA and thalassemia (Fig. 4A and D).”
Quote in the paperSubstantial claim - Also stated in: § DIFFERENTIATING THALASSEMIA FROM IDA · plain
- Findingown data
The XGB model achieved an AUROC of 0.95 for α+ homozygote and compound heterozygote.
“Subclass analysis showed that the XGB model reached an AUROC of 0.97 for β-thalassemia, 0.98 for α 0 -thalassemia, 0.95 for homozygous α + -thalassemia, 0.78 for heterozygous α + -thalassemia, and 0.94 for the structural hemoglobin variants Hemoglobin C, Hemoglobin D, Hemoglobin E.”
- Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · plain
“In a subclassification analysis of α-thalassemia, the XGB model achieved AUROC scores of 0.98 for α 0 -heterozygote, 0.95 for α + -homo-and compound heterozygote, and 0.78 for α + -heterozygote (Fig. 3A).”
Quote in the paperNot a substantial claim - Also stated in: § SUBCLASS ANALYSIS OF THE MODEL ON SPECIFIC HEMOGLOBINOPATHIES · plain
- Findingown data
Pathology-specific classifiers showed no substantial improvement compared to the original positive-negative hemoglobinopathy classification model.
“However, the results indicated no substantial improvement of the pathology-specific models compared to the original positive-negative hemoglobinopathy classification model (online Supplemental Fig. 1).”
Quote in the paperSubstantial claim - Findingown data
The XGB model distinguished α-thalassemia from IDA with an AUROC of 0.90.
“Subclass analysis demonstrated that the XGB and logistic regression models distinguished α-thalassemia from IDA with an AUROC of 0.90 and 0.89, respectively.”
Quote in the paperSubstantial claim - Findingown data
Positive hemoglobinopathy cases had significantly lower MCH than negative cases.
“These patterns included significantly lower MCV, MCH, and MCHC, and notably higher RDWCV and RBC levels than negative cases (Table 1).”
Quote in the paperNot a substantial claim - Findingown data
A probability threshold of ≥0.73 for the logistic regression model is associated with a 95% specificity cutoff.
“A probability threshold of ≥0.80 for the XGB model and a probability threshold of ≥0.73 for the logistic regression are associated with a 95% specificity cutoff.”
Quote in the paperNot a substantial claim - Findingown data
The XGB model's median predicted probability for α-thalassemia (0.93, IQR 0.85 to 0.96) significantly surpassed that for IDA (0.54, IQR 0.36 to 0.75).
“Median predicted probability scores of the XGB model for α-thalassemia (median = 0.93, IQR : 0.85 to 0.96) and β-thalassemia cases (median = 0.98, IQR: 0.96 to 0.99) significantly surpassed that for cases with an IDA (median = 0.54, IQR: 0.36 to 0.75) (Fig. 4C).”
Quote in the paperSubstantial claim - Findingown data
The XGB model achieved an AUROC of 0.92 for combinations of hemoglobinopathies.
“For combinations of hemoglobinopathies, the XGB model reached an AUROC of 0.92 (Fig. 3D).”
Quote in the paperNot a substantial claim - Findingown data
The logistic regression model's median predicted probability for IDA (0.56, IQR 0.45 to 0.65) was significantly lower than that for β-thalassemia (0.94, IQR 0.89 to 0.98).
“The median predicted probability scores of logistic regression for cases with an IDA (median = 0.56, IQR: 0.45 to 0.65) was significantly lower compared to α-thalassemia (median = 0.83, IQR: 0.74 to 0.88) and β-thalassemia cases (median = 0.94, IQR: 0.89 to 0.98) (Fig. 4F).”
Quote in the paperNot a substantial claim - Findingown data
The XGB model achieved an AUROC of 0.91 for sickle cell anemia.
“Subclass analysis on structural hemoglobin variants demonstrated AUROCs of 0.94 for Hb E, Hb C and Hb D homozygotes and heterozygotes, 0.91 for sickle cell anemia, and 0.70 for Hb S heterozygotes (Fig. 3C).”
Quote in the paperNot a substantial claim - Findingown data
The XGB model's median predicted probability for β-thalassemia (0.98, IQR 0.96 to 0.99) significantly surpassed that for IDA (0.54, IQR 0.36 to 0.75).
“Median predicted probability scores of the XGB model for α-thalassemia (median = 0.93, IQR : 0.85 to 0.96) and β-thalassemia cases (median = 0.98, IQR: 0.96 to 0.99) significantly surpassed that for cases with an IDA (median = 0.54, IQR: 0.36 to 0.75) (Fig. 4C).”
Quote in the paperSubstantial claim - Findingown data
Incorporating sex as a parameter did not affect the models' performance.
“Since incorporating sex as a parameter did not affect the models' performance, it was excluded from our model.”
Quote in the paperSubstantial claim - Findingown data
The logistic regression model exhibited greater uncertainty at high precision levels (0.75 to 1.00).
“Notably, the logistic regression model exhibited greater uncertainty at high precision levels (0.75 to 1.00) (Fig. 5C and F).”
Quote in the paperNot a substantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingown data
The maximum MCC for the XGB model was 0.51 at a probability threshold of 0.92.
“The maximum MCC for the XGB model was 0.51 at a probability threshold of 0.92, whereas, for logistic regression, it was 0.65 at a threshold of 0.78.”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingown data
The maximum MCC for logistic regression was 0.65 at a threshold of 0.78.
“The maximum MCC for the XGB model was 0.51 at a probability threshold of 0.92, whereas, for logistic regression, it was 0.65 at a threshold of 0.78.”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingown data
The models showed robust performance across various hematological analyzers.
“The XGB and logistic regression models were developed using extensive multicenter data collected over a span of 12 years, showcasing robust performance across various hematological analyzers.”
Quote in the paperNot a substantial claim - Findingown data
The XGB model reached an AUROC of 0.97 (±0.01) on differentiating negative from positive cases in the screening evaluation.
“reached an AUROC of 0.97 (±0.01), and the logistic regression model 0.98 (±0.01) on differentiating negative from positive cases (Fig. 5A and D).”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingown data
The logistic regression model reached an AUROC of 0.98 (±0.01) on differentiating negative from positive cases in the screening evaluation.
“reached an AUROC of 0.97 (±0.01), and the logistic regression model 0.98 (±0.01) on differentiating negative from positive cases (Fig. 5A and D).”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingown data
The XGB model yielded a lower average precision (0.55 ±0.06) compared to logistic regression's (0.65 ±0.05).
“Precision-recall curves revealed that the XGB model yielded a lower average precision of 0.55 (±0.06) compared to logistic regression's of 0.65 (±0.05).”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingcitation
The XGB model outperformed all other rule-based formulas, including Jayabose, Janel, Green and King, and Shine and Lal, in distinguishing thalassemia from IDA in a Spanish population.
“Moreover, the efficacy of the XGB model in distinguishing thalassemia from IDA was substantiated in a Spanish population, where it outperformed all other rule-based formulas, including Jayabose, Janel, Green and King, and Shine and Lal (14) (Supplemental Table 5).”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [14] Urrechaga, 2017, Critical appraisal of discriminant formulas for distinguishing thalassemia from iron deficiency in patients with microcytic anemia, Clin Chem Lab Med source
- Findingcitation
The efficacy of the XGB model in distinguishing thalassemia from IDA was substantiated in a Spanish population.
“Moreover, the efficacy of the XGB model in distinguishing thalassemia from IDA was substantiated in a Spanish population, where it outperformed all other rule-based formulas, including Jayabose, Janel, Green and King, and Shine and Lal (14) (Supplemental Table 5).”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [14] Urrechaga, 2017, Critical appraisal of discriminant formulas for distinguishing thalassemia from iron deficiency in patients with microcytic anemia, Clin Chem Lab Med source
- Findingreasoning
Many of the false-positive cases in the XGB model were from ICU patients with severe anemia with high MCV.
“Moreover, many of the false-positive cases in the XGB model were from ICU patients with severe anemia with high MCV, likely mimicking sickle cell CBC results.”
Quote in the paperSubstantial claim - Findingnone given
In the screening analysis, there was only one sickle cell patient in the positive cases.
“In this screening analysis, there was only one sickle cell patient in the positive cases.”
Quote in the paperSubstantial claim - Findingown data
At a specificity of 99.8%, the XGB model had a sensitivity of 35%.
“At a specificity of 99.8%, the logistic regression model reached a sensitivity of 57%, whereas the XGB model had a sensitivity of 35% (Fig. 5B and E).”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingown data
At a specificity of 99.8%, the logistic regression model reached a sensitivity of 57%.
“At a specificity of 99.8%, the logistic regression model reached a sensitivity of 57%, whereas the XGB model had a sensitivity of 35% (Fig. 5B and E).”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingown data
A significant number of false-positive cases from the XGB model were severely anemic patients from the ICU.
“Given that a significant number of false-positive cases from the XGB model were severely anemic patients from the intensive care unit (ICU), we do not advise using this algorithm for ICU patients.”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Findingown data
The simulation of the case-finding capacities of the models in a screening-based approach showed satisfactory results.
“Our simulation of the case-finding capacities of our models in a screening-based approach showed satisfactory results.”
Quote in the paperSubstantial claim- Rests on an assumption: The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
- Study design and conductreasoning
The screening data set's negative cases may contain an unknown limited number of false-negative cases.
“Therefore, the category of”
Quote in the paperNot a substantial claim - Study design and conductown data
The integrity of the imputation process was validated through pre- and post-imputation distribution comparisons, kernel density estimation plots, mean and interquartile range assessments, and individual case evaluations.
“The integrity of the imputation process was validated through pre-and post-imputation distribution comparisons, conducted via kernel density estimation plots, mean and interquartile range assessments, and individual case evaluations.”
Quote in the paperSubstantial claim - Interpretationown data
An ML algorithm was developed and extensively validated using routine CBC parameters to accurately predict a broad range of hemoglobinopathies, including thalassemias, various hemoglobin variants, and their diverse combinations.
“In this study, an ML algorithm was developed and extensively validated using routine CBC parameters to accurately predict a broad range of hemoglobinopathies, including thalassemias, various hemoglobin variants, and their diverse combinations, showcasing the ML algorithm's wide-ranging applicability in medical diagnostics.”
- Also stated in: § Additional information · boosted
“In conclusion, this study effectively demonstrated the capability of an ML model to accurately predict a wide spectrum of hemoglobinopathies using routine CBC parameters.”
Quote in the paperSubstantial claim - Also stated in: § Additional information · boosted
- Study design and conductnone given
The XGB and logistic regression models were developed using extensive multicenter data collected over a span of 12 years.
“The XGB and logistic regression models were developed using extensive multicenter data collected over a span of 12 years, showcasing robust performance across various hematological analyzers.”
Quote in the paperSubstantial claim - Interpretationown data
The false-positive cases from ICU patients with severe anemia and high MCV likely mimic sickle cell CBC results.
“Moreover, many of the false-positive cases in the XGB model were from ICU patients with severe anemia with high MCV, likely mimicking sickle cell CBC results.”
Quote in the paperSubstantial claim - Interpretationreasoning
Leveraging a high PPV, the model can screen routine CBC tests to flag new potential hemoglobinopathy cases.
“Firstly, leveraging a high positive predictive value (PPV), the model can screen routine CBC tests to flag new potential hemoglobinopathy cases.”
Quote in the paperSubstantial claim- Rests on an assumption: The deployment of ML models necessitates selecting an optimal threshold that prioritizes sensitivity or specificity, and the positive or negative predictive value.
- Interpretationreasoning
The current model was exclusively developed and designed for adults.
“Firstly, the current model was exclusively developed and designed for adults, recognizing the substantial differences in CBC parameters between young children and adults.”
Quote in the paperSubstantial claim - Premisereasoning
Existing rule-based formulas and ML algorithms for identifying hemoglobinopathies have limitations.
“However, existing rule-based formulas and ML algorithms for identifying hemoglobinopathies have limitations.”
Quote in the paperSubstantial claim - Interpretationreasoning
These factors limit the applicability and reliability of many of these formulas and models when applied outside their training context.
“These factors limit the applicability and reliability of many of these formulas and models when applied outside their training context.”
Quote in the paperSubstantial claim - Interpretationreasoning
A logistic regression formula was concurrently developed to offer a more straightforward and feasible implementation in clinical care.
“To address the significant challenge many laboratories encounter in integrating ML algorithms into their laboratory information systems, a logistic regression formula was concurrently developed, offering a more straightforward and feasible implementation in clinical care.”
- Also stated in: § Additional information · plain
“Moreover, a logistic regression model was developed, providing a more practical approach for implementation.”
Quote in the paperSubstantial claim - Also stated in: § Additional information · plain
- Interpretationreasoning
The models presented in the study enable 2 potential clinical implementation strategies.
“Considering these factors, the models presented in our study enable 2 potential clinical implementation strategies.”
Quote in the paperSubstantial claim- Rests on an assumption: The deployment of ML models necessitates selecting an optimal threshold that prioritizes sensitivity or specificity, and the positive or negative predictive value.
- Interpretationreasoning
The inclusion of such false-negative cases in the model development data set leads to an underperformance of the model.
“The inclusion of such false-negative cases in the model development data set leads to an underperformance of the model.”
Quote in the paperSubstantial claim - Study design and conductnone given
Both models underwent thorough external validation.
“Furthermore, both models underwent thorough external validation.”
Quote in the paperSubstantial claim
- May combine two claims. 5
- The quote only partly states this claim. 1
- Classified as probably not a substantial claim. 6
May combine two claims.
PremisecitationExisting formulas were not designed to differentiate structural hemoglobin variants, combinations of multiple hemoglobinopathies, or cases with a hemoglobinopathy and concomitant iron deficiency.
“These formulas were not designed to differentiate structural hemoglobin variants, combinations of multiple hemoglobinopathies, or cases with a hemoglobinopathy and concomitant iron deficiency (20,21).”
Quote in the paperSubstantial claim1 of 2 sources checked- cites [20] Traeger-Synodinos, 2014, Advances in technologies for screening and diagnosis of hemoglobinopathies, Biomark Med source
- cites [21] Jameel, 2017, Differentiation of beta thalassemia trait from iron deficiency anemia by hematological indices, Pak J Med Sci source
The Introduction describes the indices and formulas as tools for differentiating iron deficiency anemia from β-thalassemia trait. The Discussion notes that coexisting iron deficiency can complicate diagnosis, but the paper does not establish that existing formulas were not designed for structural hemoglobin variants, multiple hemoglobinopathies, or all cases of concomitant iron deficiency. No page numbers are available in our copy. (LLM, conf=0.78)
Experimental — accuracy not yet measured
May combine two claims.
Interpretationown dataBoth the XGB and logistic regression models are effective in differentiating hemoglobinopathies from IDA.
“Both the XGB and logistic regression model demonstrate high accuracy in predicting a broad range of hemoglobinopathies and are effective in differentiating hemoglobinopathies from IDA.”
Quote in the paperSubstantial claim- Rests on an assumption: CBC changes seen in hemoglobinopathies can often mimic those of iron deficiency anemia (IDA), which can result in unnecessary iron supplementation.
May combine two claims. The quote only partly states this claim.
Study design and conductown dataSixty-eight positive cases were artificially added to enrich the screening data set, without preselection, ensuring no selection bias.
“These were all positive cases, without any preselection (ensuring no selection bias), that underwent hemoglobinopathy diagnostics at Jeroen Bosch Hospital between April 2022 and June 2023 and were not part of the original model development data set.”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationown dataBoth the XGB and logistic regression models demonstrate high accuracy in predicting a broad range of hemoglobinopathies.
“Both the XGB and logistic regression model demonstrate high accuracy in predicting a broad range of hemoglobinopathies and are effective in differentiating hemoglobinopathies from IDA.”
Quote in the paperSubstantial claim- Rests on an assumption: CBC changes seen in hemoglobinopathies can often mimic those of iron deficiency anemia (IDA), which can result in unnecessary iron supplementation.
Classified as probably not a substantial claim. May combine two claims.
Study design and conductown dataAUPRCs and MCCs were generated to account for the large class imbalance in the data set and assess real-world utility scenarios.
“Additionally, the area under the precision-recall curves (AUPRCs) and Matthews correlation coefficients (MCCs) were generated to account for the large class imbalance in the data set and assess real-world utility scenarios.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductreasoningThe TRIPOD checklist was used for increased transparency of the methodology.
“As this study used a multivariable XGB model, the Transparant Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) checklist was used for increased transparency of the methodology (online Supplemental”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S4 (based).
InterpretationreasoningIntegration of either of these models into the laboratory information system facilitates automated detection of hemoglobinopathies based on routine CBC parameters.
“Integration of either of these models into the laboratory information system facilitates automated detection of hemoglobinopathies based on routine CBC parameters.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
InterpretationreasoningEmploying its high NPV, the model offers a cost-effective approach by advising against the pursuit of expensive α-thalassemia DNA diagnostics for patients who have both a negative HPLC/CE result combined with a low probability of hemoglobinopathy according to the model.
“Secondly, employing its high negative predictive value (NPV), the model offers a cost-effective approach by advising against the pursuit of expensive α-thalassemia DNA diagnostics for patients who have both a negative HPLC/CE result combined with a low probability of hemoglobinopathy according to our model.”
Quote in the paperSubstantial claim- Rests on an assumption: The deployment of ML models necessitates selecting an optimal threshold that prioritizes sensitivity or specificity, and the positive or negative predictive value.
Classified as probably not a substantial claim.
Interpretationnone givenThe XGB model was also trained to identify sickle cell patients.
“This is likely due to the fact that the XGB model was also trained to identify sickle cell patients.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S4 (based).
InterpretationreasoningIntegration of these models into the laboratory information system facilitates automated hemoglobinopathy detection using routine CBC parameters.
“Integration of these models into the laboratory information system facilitates automated hemoglobinopathy detection using routine CBC parameters.”
Quote in the paperNot a substantial claim
The paper relies on this without support.
CBC changes seen in hemoglobinopathies can often mimic those of iron deficiency anemia (IDA), which can result in unnecessary iron supplementation.
“The CBC changes seen in hemoglobinopathies can often mimic those of iron deficiency anemia (IDA), which can result in unnecessary iron supplementation.”
The paper relies on this without support.
Medlon's comprehensive diagnostic evaluation, including Sanger sequencing of both α and β globin genes, provides accurate hemoglobinopathy classification for external validation.
“Medlon carries out a comprehensive diagnostic evaluation for each hemoglobinopathy diagnostic request, which encompasses Sanger sequencing of both α and β globin genes.”
The paper relies on this without support.
The general population of Jeroen Bosch Hospital used as negative cases in the screening evaluation are truly negative for hemoglobinopathies.
“distinguishing cases of the general population of Jeroen Bosch Hospital, considered negative, from known positive hemoglobinopathy cases.”
The paper relies on this without support.
The choice of threshold is dependent upon local requirements and preferences, related to additional workload, cost-effectiveness, and the prevalence of hemoglobinopathies, which varies substantially among endemic and non-endemic regions.
“This choice is dependent upon local requirements and preferences, related to additional workload, cost-effectiveness, and the prevalence of hemoglobinopathies, which varies substantially among endemic and non-endemic regions.”
The paper relies on this without support.
The deployment of ML models necessitates selecting an optimal threshold that prioritizes sensitivity or specificity, and the positive or negative predictive value.
“The deployment of ML models necessitates selecting an optimal threshold that prioritizes sensitivity or specificity, and the positive or negative predictive value.”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Hb E median exceeds its upper quartile
MinorThe reported MCH median is 24.7, but the upper end of its reported interquartile range is 24.2. A median cannot exceed the 75th percentile.
“Hb E homozygote/heterozygote 203 12.3 (11.0-13.4)b 24.7 (22.6-24.2) b”
“CBC parameters are presented as medians and interquartile ranges (±IQR) (Table 1 and online Supplemental Tables 2 and 4).”
Conflicting XGBoost package versions
MinorThe methods identify two different XGBoost versions for the model analyses: 1.7.3 and 0.24.
“The XGB model was trained using an eXtreme Gradient Boosting algorithm with the XGBoost package (1.7.3).”
“Analyses were carried out using Python 3.8 (Python Software Foundation), with the packages Numpy (1.20), Pandas (0.28), Sklearn (1.1.1),XGBoost (0.24), and SHAP (0.41).”
α0-heterozygote median exceeds its upper quartile
MinorThe reported MCH median is 21.2, but the upper end of its reported interquartile range is 20.9. A median cannot exceed the 75th percentile.
“α 0 -heterozygote 185 11.9 (11.1-12.7)21.2 (19.3-20.9)b”
“CBC parameters are presented as medians and interquartile ranges (±IQR) (Table 1 and online Supplemental Tables 2 and 4).”
Cites
Global epidemiology of haemoglobin disorders and derived service indicators
Modell, B. (2008) · Bulletin of the World Health Organization · doi:10.2471/blt.06.036673 · Cited by 1,932 · see citing works
Taher, A. T., Weatherall, D. J., & Cappellini, M. D. (2018) · The Lancet · doi:10.1016/s0140-6736(17)31822-6 · Cited by 807 · see citing works
Harteveld, C. L., & Higgs, D. R. (2010) · Orphanet Journal of Rare Diseases · doi:10.1186/1750-1172-5-13 · Cited by 508 · see citing works
Self-citationWorld Distribution, Population Genetics, and Health Burden of the HemoglobinopathiesPreprint
Williams, T. N., & Weatherall, D. J. (2012) · Cold Spring Harbor Perspectives in Medicine · doi:10.1101/cshperspect.a011692 · Cited by 448 · see citing works
DIFFERENTIATION OF IRON DEFICIENCY FROM THALASSÆMIA TRAIT
Mentzer, W. C. (1973) · The Lancet · doi:10.1016/s0140-6736(73)91446-3 · Cited by 306 · see citing works
LeDell, E., Petersen, M. L., & van der Laan, M. J. (2015) · Electronic Journal of Statistics · doi:10.1214/15-ejs1035 · Cited by 241 · see citing works
KOHNE, E. (2011) · Deutsches Ärzteblatt international · doi:10.3238/arztebl.2011.0532 · Cited by 234 · see citing works
DIFFERENTIATION OF IRON DEFICIENCY FROM THALASSÆMIA TRAIT BY ROUTINE BLOOD-COUNT
England, J. M., & Fraser, P. (1973) · The Lancet · doi:10.1016/s0140-6736(73)91878-3 · Cited by 232 · see citing works
Amendolia, S. R., Cossu, G. F., Ganadu, M. L., Golosio, B., Masala, G. L., & Mura, G. M. (2003) · Chemometrics and Intelligent Laboratory Systems · doi:10.1016/s0169-7439(03)00094-7 · Cited by 173 · see citing works
A STRATEGY TO DETECT β-THALASSÆMIA MINOR
Shine, I., & Lal, S. (1977) · The Lancet · doi:10.1016/s0140-6736(77)92128-6 · Cited by 157 · see citing works
Epidemiology of haemoglobin disorders in Europe: an overview
Modell, B., Darlison, M. W., Birgens, H., et al. (2007) · Scandinavian Journal of Clinical and Laboratory Investigation · doi:10.1080/00365510601046557 · Cited by 155 · see citing works
meeting, O. B. O. C. T. T. E. H. B. P., Traeger‐Synodinos, J., Harteveld, C. L., et al. (2014) · European Journal of Human Genetics · doi:10.1038/ejhg.2014.131 · Cited by 122 · see citing works
Self-citationThe hemoglobinopathies, molecular disease mechanisms and diagnostics
Harteveld, C. L., Achour, A., Arkesteijn, S. G. J., ter Huurne, J., Verschuren, M., Bisoen, S., Schaap, R., Vijfhuizen, L., el Idrissi, H., & Koopmann, T. T. (2022) · International Journal of Laboratory Hematology · doi:10.1111/ijlh.13885 · Cited by 108 · see citing works
Self-citationPopulation Screening for Hemoglobinopathies
Goonasekera, H. W. W., Paththinige, C. S., & Dissanayake, V. H. W. (2018) · Annual Review of Genomics and Human Genetics · doi:10.1146/annurev-genom-091416-035451 · Cited by 87 · see citing works
#262 Differentiating iron deficiency anemia from thalassemia minor by using an RDW-based index
Jayabose, S., Giamelli, J., LevondogluTugal, O., Sandoval, C., Ozkaynak, F. M., & Visintainer, P. (1999) · Journal of Pediatric Hematology/Oncology · doi:10.1097/00043426-199907000-00040 · Cited by 79 · see citing works
Differentiation of beta thalassemia trait from iron deficiency anemia by hematological indices
Jameel, T. A., Baig, M., Ahmed, I., Hussain, M. B., & Doghaim Alkhamaly, M. B. (2017) · Pakistan Journal of Medical Sciences · doi:10.12669/pjms.333.12098 · Cited by 79 · see citing works
Laengsri, V., Shoombuatong, W., Adirojananon, W., Nantasenamat, C., Prachayasittikul, V., & Nuchnoi, P. (2019) · BMC Medical Informatics and Decision Making · doi:10.1186/s12911-019-0929-2 · Cited by 66 · see citing works
Çil, B., Ayyıldız, H., & Tuncer, T. (2020) · Medical Hypotheses · doi:10.1016/j.mehy.2020.109611 · Cited by 54 · see citing works
Advances in technologies for screening and diagnosis of hemoglobinopathies
Traeger-Synodinos, J., & Harteveld, C. L. (2014) · Biomarkers in Medicine · doi:10.2217/bmm.13.103 · Cited by 53 · see citing works
Self-citationUrrechaga, E., & Hoffmann, J. J. M. L. (2017) · Clinical Chemistry and Laboratory Medicine (CCLM) · doi:10.1515/cclm-2016-0856 · Cited by 47 · see citing works
Self-citationPuehringer, H., Najmabadi, H., Law, H.-Y., Krugluger, W., Viprakasit, V., Pissard, S., Baysal, E., Taher, A., Farra, C., Al-Ali, A., et al. (2007) · Clinical Chemical Laboratory Medicine · doi:10.1515/cclm.2007.125 · Cited by 45 · see citing works
AlAgha, A. S., Faris, H., HAMMO, B. H., & Al-Zoubi, A. M. (2018) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2018.04.009 · Cited by 45 · see citing works
Classification ofβ-Thalassemia Carriers From Red Blood Cell Indices Using Ensemble Classifier
Sadiq, S., Khalid, M. U., Mui-zzud-din, et al. (2021) · IEEE Access · doi:10.1109/access.2021.3066782 · Cited by 39 · see citing works
Fu, Y.-K., Liu, H.-M., Lee, L.-H., Chen, Y.-J., Chien, S.-H., Lin, J.-S., Chen, W.-C., Cheng, M.-H., Lin, P.-H., Lai, J.-Y., et al. (2021) · Diagnostics · doi:10.3390/diagnostics11091725 · Cited by 38 · see citing works
Traeger-Synodinos, J., Harteveld, C., Old, J., et al. (2015) · European Journal of Human Genetics · doi:10.1038/ejhg.2015.39 · Cited by 35
Self-citationDas, R., Saleh, S., Nielsen, I. E., Kaviraj, A., Sharma, P., Dey, K., & Saha, S. (2022) · International Journal of Medical Informatics · doi:10.1016/j.ijmedinf.2022.104866 · Cited by 33 · see citing works
A decision support scheme for beta thalassemia and HbE carrier screening
Das, R., Datta, S., Kaviraj, A., Sanyal, S. N., Nielsen, P., Nielsen, I. E., Sharma, P., Sanyal, T., Dey, K., & Saha, S. (2020) · Journal of Advanced Research · doi:10.1016/j.jare.2020.04.005 · Cited by 29 · see citing works
Differential Diagnostics of Thalassemia Minor by Artificial Neural Networks Model
Barnhart-Magen, G., Gotlib, V. A., Marilus, R., et al. (2013) · Journal of Clinical Laboratory Analysis · doi:10.1002/jcla.21631 · Cited by 28 · see citing works
Prediction of $$\beta $$-Thalassemia carriers using complete blood count features
Rustam, F., Ashraf, I., Jabbar, S., Tutusaus, K., Mazas, C., Barrera, A. E. P., & de la Torre Diez, I. (2022) · Scientific Reports · doi:10.1038/s41598-022-22011-8 · Cited by 28 · see citing works
Feng, P., Li, Y., Liao, Z., Yao, Z., Lin, W., Xie, S., Hu, B., Huang, C., Liu, W., Xu, H., et al. (2021) · Clinica Chimica Acta · doi:10.1016/j.cca.2021.12.003 · Cited by 26 · see citing works
Zhang, F., Yang, J., Wang, Y., Cai, M., Ouyang, J., & Li, J. (2023) · Clinica Chimica Acta · doi:10.1016/j.cca.2023.117368 · Cited by 22 · see citing works
Machine Learning in Predicting Hemoglobin Variants
Borah, M. S., Bhuyan, B. P., Pathak, M. S., et al. (2018) · International Journal of Machine Learning and Computing · doi:10.18178/ijmlc.2018.8.2.677 · Cited by 21 · see citing works
Phirom, K., Charoenkwan, P., Shoombuatong, W., Charoenkwan, P., Sirichotiyakul, S., & Tongsong, T. (2022) · Journal of Clinical Medicine · doi:10.3390/jcm11216305 · Cited by 20 · see citing works
van Vliet, M. E., Kerkhoffs, J., Harteveld, C. L., et al. (2022) · European Journal of Human Genetics · doi:10.1038/s41431-022-01051-8 · Cited by 15 · see citing works
Self-citationThalassämie-Syndrome: Klinik und Diagnose
Huber, A., Ottiger, C., Risch, L., et al. (2004) · Swiss Medical Forum ‒ Schweizerisches Medizin-Forum · doi:10.4414/smf.2004.05311 · Cited by 12 · see citing works
Angastiniotis, M., Cannon, L., Antoniou, E., Brunetta, A. L., Constantinou, G., Knoll, E. M., Loukopoulos, D., Skafi, A., & Eleftheriou, A. (2021) · Thalassemia Reports · doi:10.4081/thal.2021.9803 · Cited by 11 · see citing works
Laengsri, V., Shoombuatong, W., Adirojananon, W., Nantasenamat, C., Prachayasittikul, V., & Nuchnoi, P. (2019) · BMC Medical Informatics and Decision Making · doi:10.1186/s12911-019-0977-7 · Cited by 3 · see citing works
Sadiq, S., Khalid, U., Mui-Zzud-Din, et al. (2021) · IEEE Access · doi:10.1109/access.2021.3071048
Flagged first, then by citations
Research works printed but not linked: 2
- Centers of Disease Control and Prevention, 2015, Hemoglobinopathies: current practices for screening, confirmation and follow-up
- Kulozik, 2010, Editorial: hämoglobinopathien nehmen zu, Dtsch Arztebl Int, 107, 63
Cited by
Sorted by citations
Coelho, H. G. B., Silva, F., Correia, M., & Rodrigues, P. M. (2025) · Journal of Clinical Medicine · doi:10.3390/jcm14238479 · Cited by 9 · see citing works
Machine Learning‐Based Prediction of β‐Thalassemia Trait Using Red Blood Cell Indices
Hataysal, E. P., & Körez, M. K. (2025) · International Journal of Laboratory Hematology · doi:10.1111/ijlh.70017 · Cited by 7 · see citing works
Abbas, M., Khan, M. B. S., Khan, A. H., Bilal, A., Algarni, A. M., & Sarwar, R. (2026) · Frontiers in Medicine · doi:10.3389/fmed.2025.1688645 · Cited by 8 · see citing works
Machine learning improves detection of alpha thalassemia carriers compared to clinical features
Mohammadi, E., Rastegar, M., Jamshidnezhad, A., & Azizi, A. (2025) · Scientific Reports · doi:10.1038/s41598-025-20605-6 · Cited by 5 · see citing works
Machine Learning for Discriminating Microcytic Hypochromic Anemia Based on Erythrocyte Parameters
Lv, J., Li, J., Ren, X., Huang, Q. Y., & Deng, S. (2025) · International Journal of Laboratory Hematology · doi:10.1111/ijlh.14524 · Cited by 5 · see citing works
Groen, J., de Haan, B. M., Overduin, R. J., et al. (2025) · Orphanet Journal of Rare Diseases · doi:10.1186/s13023-025-03537-2 · Cited by 4 · see citing works
Komninaka, V., Liossi, S., Tsagarakis, N. J., Fotopoulou, I., Oudatzis, G., Chaliori, I., Karela, C., Liakopoulos, D., Vasileiou, P., Papageorgiou, E. G., et al. (2025) · International Journal of Laboratory Hematology · doi:10.1111/ijlh.70033 · Cited by 4 · see citing works
Gui, Y., Chen, W., Su, Y., Qin, C., zhang, F., zhao, J., & Liu, Q. (2026) · Annals of Hematology · doi:10.1007/s00277-026-06860-1 · Cited by 7 · see citing works
Nadarajan, V. S., Sthaneshwar, P., Qi, L. J., et al. (2025) · Annals of Laboratory Medicine · doi:10.3343/alm.2025.0027 · Cited by 2 · see citing works
Yonathan, Y., Dewi, N. S., & Indrati, A. R. (2025) · International Journal of Laboratory Hematology · doi:10.1111/ijlh.14558 · Cited by 2 · see citing works
Diagnostik von Hämoglobinopathien: Integrierter Befund für mehr Klarheit
Piehler, A. P., & Gregor Hörmann (2024) · Trillium Diagnostik · doi:10.47184/td.2024.04.02 · Cited by 1 · see citing works
Hematology in community medical care
Nakamura, S., Maeda, Y., Hori, T., et al. (2025) · The Journal of Medical Investigation · doi:10.2152/jmi.72.21 · Cited by 1 · see citing works
Machine Learning‐Based Detection of HbS and HbC Carriers in the UK General Population
Christensen, F., Kılıç, D. K., Fuglkjær, A. D., et al. (2025) · eJHaem · doi:10.1002/jha2.70170 · Cited by 1 · see citing works
Basancelebi, M., Arslan, R. U., & Yapıcı, İ. Ş. (2026) · Black Sea Journal of Engineering and Science · doi:10.34248/bsengineering.1945642 · Cited by 1 · see citing works
Al-Samkari, H. T., Aboobacker, F., Andemariam, B., Eleftheriou, P., Ezzat, H. M., Farmakis, D. T., Hashmi, S. K., Kakkar, S., Maggio, A., Piga, A. G., et al. (2026) · The Lancet Haematology · doi:10.1016/s2352-3026(26)00197-3 · Cited by 1 · see citing works
Zhang, L., An, Z., Cui, L., Li, W., & Zhang, X. (2025) · Chinese Science Bulletin (Chinese Version) · doi:10.1360/tb-2024-1190 · Cited by 1 · see citing works
Shean, R. C., Deshmukh, N., Palmer, M., et al. (2025) · Hemoglobin · doi:10.1080/03630269.2025.2573384
Fan, J., Guo, X., Cao, S., Li, R., & Zhan, S. (2025) · Computer Methods and Programs in Biomedicine · doi:10.1016/j.cmpb.2025.109201
Jusman, Y., Savitri, M. W. S., Nur'aini, M. A., et al. (2025) · International Conference on Information and Communications Technology (ICOIACT) · doi:10.1109/icoiact67584.2025.11344990
Sahu, P., & Sukhchandan (2025) · International Conference on Evolutionary Computing and Mobile Sustainable Networks (ICECMSN) · doi:10.1109/icecmsn68058.2025.11382680
Review and Analysis of Deep Learning Techniques for Alpha Thalassemia Identification
Chopra, S., & Singh, M. (2025) · IEEE Uttar Pradesh Section Women in Engineering International Conference on Electrical, Electronics and Computer Engineering (UPWIECON) · doi:10.1109/upwiecon67212.2025.11390429
Guo, X., Cheng, W., Hao, X., Zhang, L., & Lei, L. V. (2026) · Diabetes Metabolic Syndrome and Obesity · doi:10.2147/dmso.s583656
Wang, X., Huang, J. A., Xi, J., et al. (2026) · Practical Laboratory Medicine · doi:10.1016/j.plabm.2026.e00526
Anitha, R., Mandapati, P., Kumar, V. V., et al. (2026) · Australian Journal of Electrical & Electronics Engineering · doi:10.1080/1448837x.2025.2611668
Tatu, T., Pornaram, P., Samran, W., Jaisutham, S., & Somwong, J. (2026) · Suranaree Journal of Science and Technology · doi:10.55766/sujst8272
AI-enabled multi-disease diagnosis using low-cost CBC reports
Bibi, M., Saeed, W. K., Jabeen, S., et al. (2026) · Frontiers in Imaging · doi:10.3389/fimag.2026.1754419
Liao, H., Chen, S. Y., Meng, Q., et al. (2026) · BMC Medicine · doi:10.1186/s12916-026-05036-y
Yaman, N., & Yavuz, U. (2026) · Black Sea Journal of Engineering and Science · doi:10.34248/bsengineering.1917276
Cai, A., Cai, Y., Wu, S., et al. (2026) · Research Square · doi:10.21203/rs.3.rs-10063729/v1
Frecuencia de hemoglobinopatías en pacientes con anemia en un hospital pediátrico en México
Guerrero-González, D. L., Sandoval-Cabrera, A., Santillán-Benítez, J. G., et al. (2026) · Revista Mexicana de Pediatría · doi:10.35366/123988
Zainulabdeen, F. H., & Bino, D. (2026) · MEC International Conference on Big Data and Smart City (ICBDSC) · doi:10.1109/icadaai64945.2026.11679279
Basancelebi, M., Yapıcı, İ. Ş., & Arslan, R. U. (2026) · Bioengineering · doi:10.3390/bioengineering13101118
Case-Based Morphological Diagnosis of Erythroid Disorders: From Microscopy to AI
Safdari, S. M., Bakhtiyaridovvombaygi, M., Jaafarian, A. S., Hosseini, S., Shams, M., Roshanzamir, F., Aliabad, G. M., & Narouei, F. (2026) · Springer eBooks · doi:10.1007/978-3-032-27768-8_1
- Authors
- Schipper, A., Rutten, M., van Gammeren, A., Harteveld, C. L., Urrechaga, E., Weerkamp, F., den Besten, G., Krabbe, J., Slomp, J., Schoonen, L., Broeren, M., van Wijnen, M., Huijskens, M. J. A. J., Koopmann, T., van Ginneken, B., Kusters, R., & Kurstjens, S.
- Affiliations
- Radboud University Nijmegen (NL)
- Laboratory of Clinical Chemistry and Laboratory Medicine, Amphia Hospital , Breda , the Netherlands
- Leiden University Medical Center (NL)
- Hospital de Galdakao (ES)
- Maasstad Ziekenhuis (NL)
- Laboratory of Clinical Chemistry and Laboratory Medicine, Isala Hospital , Zwolle , the Netherlands
- Medisch Spectrum Twente (NL)
- Máxima Medisch Centrum (NL)
- Laboratory of Clinical Chemistry and Laboratory Medicine, Meander Medical Center , Amersfoort , the Netherlands
- Zuyderland Medisch Centrum (NL)
- Diagnostic Image Analysis Group, Radboudumc , Nijmegen , the Netherlands
- Jeroen Bosch Ziekenhuis (NL)
- Corresponding author
- Kurstjens, Steef
- ORCID
- Venue
- Clinical Chemistry
- Publisher
- Oxford University Press (OUP)
- Published
- 2024-06-22
- Volume
- 70
- Issue
- 8
- Pages
- 1064-1075
- ISSN
- 0009-9147 (ISSN-L)
- 1530-8561
- OpenAlex
- W4399922956
- Cited by
- 33
- References
- 35
- Citations by year
- 2026: 16 · 2025: 16 · 2024: 1
- FWCI
- 7.54
- Type
- journal article
- Language
- en
- Open access
- Locations
- Clinical Chemistry (publishedVersion)
- PubMed (publishedVersion)
- Leiden Repository (Leiden University) (submittedVersion)
- Radboud Repository (Radboud University) (submittedVersion)
- University of Twente Research Information (submittedVersion)
- Topics
- Health Sciences › Medicine › Genetics › Hemoglobinopathies and Related Disorders
- Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Imaging for Blood Diseases
- Keywords
- hemoglobinopathies, complete blood count, machine learning prediction, thalassemia, iron deficiency anemia, beta-thalassemia, alpha-thalassemia, hemoglobin C, hemoglobin D, hemoglobin E, genetic counseling, carrier screening
- MeSH terms
- Machine Learning, Adult, Blood Cell Count, Female, Hemoglobinopathies, Humans, Male, Retrospective Studies, ROC Curve, Logistic Models
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-27.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-27.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
