Skip to content
DraconiticResearch
Search Sources API
226,742 studies31 sourcesUpdated 46 min ago
Peer-reviewedCC BY-NC-ND 4.0BMC Medical Informatics and Decision Making

Construction and validation of a differential diagnosis model for multiple myeloma and lymphoma based on XGBoost machine learning: an explaining analysis study of SHAP

  1. 1The Third Affiliated Hospital of Guangzhou University of Chinese Medicine, Guangzhou, 510378 China
  2. 2Guangzhou University of Chinese Medicine, Guangzhou, 510006 China

Abstract

Background: Multiple myeloma (MM) and lymphoma can share overlapping clinical features but require different diagnostic and treatment strategies. Machine learning (ML) may identify discriminative patterns from routinely available clinical data and support rapid triage.

Methods: This retrospective study included patients from the Third Affiliated Hospital of Guangzhou University of Chinese Medicine between 2015 and 2025. Data from 2015 to 2024 (n = 2,759) were used for model development, and the 2025 cohort (n = 1,011) served as an independent temporal external validation set. Feature selection combined univariate analysis, collinearity assessment, extreme gradient boosting (XGBoost), least absolute shrinkage and selection operator (LASSO), and Boruta, yielding 14 core features. Nine ML models were compared.

Results: XGBoost showed the best validation performance, with an area under the receiver operating characteristic curve (AUC) of 0.964, accuracy of 0.904, sensitivity of 0.913, specificity of 0.896, positive predictive value of 0.889, and negative predictive value (NPV) of 0.918. In the held-out internal test set, the final model achieved an AUC of 0.947 and an accuracy of 0.890. External validation yielded an AUC of 0.713, sensitivity of 0.400, specificity of 0.768, and NPV of 0.876. Shapley Additive Explanations (SHAP) identified age, mean corpuscular hemoglobin, red cell distribution width coefficient of variation, platelet-to-lymphocyte ratio, and mean corpuscular volume as the leading contributors to model predictions.

Conclusions: The XGBoost model demonstrated good discriminatory ability in distinguishing multiple myeloma from lymphoma by integrating 14 routinely available variables: age, WBC, MONO, EO, BASO%, RBC, HGB, HCT, MCV, MCH, RDW-CV, PLT, LYMPH/MONO, and PLT/LYMPH. This multivariable framework may serve as an adjunctive tool to support the differential diagnosis of these two hematological malignancies.

Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1186/s12911-026-03752-2.

Related

  1. Construction and validation of an identification model for frailty in maintenance hemodialysis patients using multiple machine learning algorithms: a cross-sectional study of retrospectively collected data
    Jiao Wang, Ping Yu, Wenjuan Sun +7
    Annals of medicinePeer-reviewedCC BY-NCFull textGeneral medicineEurope PMC
  2. Differential Diagnosis of Early-Stage Pulmonary Sarcoidosis With Normal Serum Angiotensin-Converting Enzyme Activity Using Peripheral Blood Indicators: Machine Learning Models Development and Validation
    Lixiuyun Wang, Yafei Tang, Jun Ma +5
    JMIR Medical InformaticsPeer-reviewedCC BY 4.0Full textGeneral medicinePMC
  3. Relative Importance of Positive and Potentially Traumatic Childhood Experiences in Explaining Multiple Sclerosis Symptoms
    C. Obenauf, A. Vazquez, G. Owens
    medRxivPreprintCC BYFull textNeurologymedRxiv
  4. High-Throughput Proteomics in Lymphangioleiomyomatosis: Premelanosome Protein as a Diagnostic Biomarker, Construction of a Diagnosis Score, and Evidence of Neutrophil Involvement
    Rémi Diesler, Tiegang Han, Wendy Steagall +17
    ChestPeer-reviewedLicence not stated3 citedPhysiologyHAL
  5. Epidemiology of Syphilis, Diagnosis, Differential Diagnosis and Treatment: Up-to-Date Knowledge
    Journal of clinical medicinePeer-reviewedCC BYFull textGeneral medicineEurope PMC
BibTeX
@article{su2026construction,
  title = {{Construction and validation of a differential diagnosis model for multiple myeloma and lymphoma based on XGBoost machine learning: an explaining analysis study of SHAP}},
  author = {Jing Su and Songyao Qi and Tao Peng and Zilin Chen and Yuanxin Zhao and Peiqi Liang and Zhuoyan Chen and Ziwen Xia and Hongguang Yang},
  journal = {BMC Medical Informatics and Decision Making},
  year = {2026},
  month = aug,
  doi = {10.1186/s12911-026-03752-2},
  url = {https://doi.org/10.1186/s12911-026-03752-2},
  note = {CC BY-NC-ND 4.0},
}
RIS
TY  - JOUR
TI  - Construction and validation of a differential diagnosis model for multiple myeloma and lymphoma based on XGBoost machine learning: an explaining analysis study of SHAP
AU  - Jing Su
AU  - Songyao Qi
AU  - Tao Peng
AU  - Zilin Chen
AU  - Yuanxin Zhao
AU  - Peiqi Liang
AU  - Zhuoyan Chen
AU  - Ziwen Xia
AU  - Hongguang Yang
PY  - 2026
DA  - 2026/08/12
JO  - BMC Medical Informatics and Decision Making
DO  - 10.1186/s12911-026-03752-2
UR  - https://doi.org/10.1186/s12911-026-03752-2
AB  - Background: Multiple myeloma (MM) and lymphoma can share overlapping clinical features but require different diagnostic and treatment strategies. Machine learning (ML) may identify discriminative patterns from routinely available clinical data and support rapid triage.  Methods: This retrospective study included patients from the Third Affiliated Hospital of Guangzhou University of Chinese Medicine between 2015 and 2025. Data from 2015 to 2024 (n = 2,759) were used for model development, and the 2025 cohort (n = 1,011) served as an independent temporal external validation set. Feature selection combined univariate analysis, collinearity assessment, extreme gradient boosting (XGBoost), least absolute shrinkage and selection operator (LASSO), and Boruta, yielding 14 core features. Nine ML models were compared.  Results: XGBoost showed the best validation performance, with an area under the receiver operating characteristic curve (AUC) of 0.964, accuracy of 0.904, sensitivity of 0.913, specificity of 0.896, positive predictive value of 0.889, and negative predictive value (NPV) of 0.918. In the held-out internal test set, the final model achieved an AUC of 0.947 and an accuracy of 0.890. External validation yielded an AUC of 0.713, sensitivity of 0.400, specificity of 0.768, and NPV of 0.876. Shapley Additive Explanations (SHAP) identified age, mean corpuscular hemoglobin, red cell distribution width coefficient of variation, platelet-to-lymphocyte ratio, and mean corpuscular volume as the leading contributors to model predictions.  Conclusions: The XGBoost model demonstrated good discriminatory ability in distinguishing multiple myeloma from lymphoma by integrating 14 routinely available variables: age, WBC, MONO, EO, BASO%, RBC, HGB, HCT, MCV, MCH, RDW-CV, PLT, LYMPH/MONO, and PLT/LYMPH. This multivariable framework may serve as an adjunctive tool to support the differential diagnosis of these two hematological malignancies.  Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1186/s12911-026-03752-2.
SN  - 1472-6947
ER  -