VTE risk models in multiple myeloma show low predictive accuracy, pooled AUC below 0.7

Photo by ThisisEngineering / Unsplash

PeerJ Published June 1, 2026 Medically reviewed June 1, 2026 Study authors: Yang Jin, Pu You, Li Yuqin, Li Xuelian, Jiang Xia, Zou Liping PubMed ↗ DOI ↗ By Dr. Julia Lee, PhD · Oncology, Genomics & Drug Development

Key Takeaway

Interpret existing VTE risk models in multiple myeloma cautiously due to low AUC and high bias risk.

This systematic review and meta-analysis of 14 studies assessed VTE risk prediction models in patients with multiple myeloma. The pooled VTE incidence across studies was 7.9% (95% CI 6.2-10.1%). However, the predictive performance of existing models was low, with combined area under the curve (AUCs) ranging from 0.57 to 0.68, indicating poor discrimination.

All included studies were at high risk of bias, particularly in the outcome and analysis domains. Only two studies used the Hosmer-Lemeshow test for model calibration, and all models were presented in formula form, which may limit their practical use in clinical settings.

The authors note that existing VTE risk models for multiple myeloma patients show low predictive performance (pooled AUC < 0.7) and limited clinical utility. Clinicians should interpret these models cautiously, as their ability to accurately stratify VTE risk is suboptimal. Further research with robust methodology and external validation is needed before these models can be recommended for routine practice.

Study Details

Study typeMeta analysis

EvidenceLevel 1

PublishedJan 2026

PMID42222482

View Original Abstract ↓

BACKGROUND: Risk prediction models help identify multiple myeloma (MM) patients at high risk of venous thromboembolism (VTE) and guide clinical decisions. However, their applicability and accuracy remain unclear. This study aims to systematically review existing VTE risk models in MM patients. METHODS: We systematically searched PubMed, Embase, Cochrane Library, and Web of Science for studies on VTE risk prediction models in patients with MM, up to March 31, 2026. Two investigators independently screened the literature, extracted data, and assessed the risk of bias and applicability of the included studies using the PROBAST tool. Data analysis was performed using the "meta" and "metafor" packages in R software. RESULTS: A total of 14 studies on VTE risk prediction models in MM patients were included, involving the development and/or validation of seven risk assessment tools. Meta-analysis showed that the overall VTE incidence in MM patients was 7.9% (95% CI [6.2-10.1%]). The combined area under the curve (AUCs) of the seven tools ranged from 0.57 to 0.68, with the IMPEDE VTE and IMPEDED VTE scores showing the best performance. Only two studies used the Hosmer-Lemeshow test for model calibration, and all studies presented the models in formula form. All included studies were at high risk of bias, mainly in the outcome and analysis domains. CONCLUSION: Existing VTE risk models for MM patients show low predictive performance (pooled AUC < 0.7) and limited clinical use. Future research should focus on model updating and external validation to improve accuracy and applicability. PROSPERO Registration number ID: CRD420251024346.