# Risk prediction model for thrombosis in leukemia patients: a systematic review

## TL;DR

This review evaluates risk prediction models for thrombosis in leukemia patients, finding them methodologically limited and not yet ready for routine clinical use.

## Contribution

The study systematically reviews and assesses the methodological quality of existing thrombosis risk prediction models in leukemia.

## Key findings

- 14 studies with 16 models were identified, showing AUC/C-index between 0.641 and 0.917.
- Most models had high risk of bias and lacked external validation.
- Key predictors included D-dimer levels, platelet count, and central venous catheter use.

## Abstract

Thrombosis represents a significant complication in leukemia patients, associated with treatment interruption and reduced survival outcomes. Although multiple risk prediction models have been developed, their methodological quality and applicability remain uncertain. This review aims to evaluate existing risk prediction models for thrombosis in patients with leukemia. We conducted comprehensive literature searches across nine databases from the inception to August 4, 2025. Two reviewers independently performed study selection, data extraction, and quality assessment using the CHARMS checklist and PROBAST tool. Of 1825 initially identified records, 14 studies comprising 16 prediction models were included. Development cohorts ranged from 102 to 1252 participants. Model discrimination measured by AUC/C-index varied between 0.641-0.917. Internal validation was performed in nine studies, while only one conducted external validation. Key predictors included central venous catheter placement, prior history of thrombosis, D-dimer levels, platelet count, white blood cell count, the Eastern Cooperative Oncology Group (ECOG) score, chemotherapy/radiotherapy, comorbidities, type of leukemia, use of hemostatic drugs, and age. All studies were rated high risk of bias, and five raised major concerns regarding applicability. Sensitivity analyses excluding chronic leukemia studies, excluding non-English publications and excluding dissertations yielded consistent overarching conclusions. In summary, current models often report moderate to good apparent discrimination, but are limited by methodological shortcomings and inadequate validation. All models should be considered exploratory and not ready for routine clinical use without prospective external validation. Future research should prioritize prospective, multicenter cohorts with standardized outcome adjudication and rigorous internal/external validation across diverse leukemia subtypes.