Artificial intelligence for risk assessment and outcome prediction in malignant haematology

Risk stratification is the backbone of clinical decision-making for patients with haematological malignancies. Models currently used in routine clinical practice often rely on static rules, conventional statistics and expert consensus. While pragmatic and accessible, these models fail to capture the complex biological interactions and patient heterogeneity, especially as data dimensionality grows given the availability of more comprehensive molecular analysis. Artificial intelligence (AI), particularly machine learning (ML), offers the potential to process such high-dimensional data, uncover latent patterns and provide dynamic, individualized risk predictions. In this review, we introduce haematologists to core concepts and limitations of AI-based risk stratification. We examine studies in myeloid and lymphoid neoplasms developing AI models that challenge current standard models and identify novel disease subtypes and risk markers. While these models offer unique potential for data-driven personalized decision support, several challenges remain to routine applicability such as training cohort composition, validation and bias mitigation highlighting an urgent need for interdisciplinary collaboration and dialogue with regulatory agencies to ensure robust and safe model deployment. As data dimensionality, digital infrastructure and regulatory frameworks evolve, AI-guided risk stratification is poised to complement and potentially reshape clinical practice in malignant haematology requiring haematologists to be aware of potential capabilities and pitfalls.