CLL TIM risk model predicts infection and treatment need in newly diagnosed chronic lymphocytic leukaemia

Prof Carsten Niemann speaks to ecancer about the first prospective, multicenter validation of the Chronic Lymphocytic Leukemia Treatment-Infection Model (CLL-TIM), a machine-learning algorithm designed to predict risk of severe infection or need for treatment within two years in newly diagnosed, asymptomatic chronic lymphocytic leukemia (CLL) patients.

Conducted within the PreVent-ACaLL trial across nine international sites, this study evaluates the model’s generalizability and reliability in a real-world prospective setting.

Prof Niemann says that patients were stratified into four groups based on risk and prediction confidence (HR-HC, HR-LC, LR-LC, LR-HC).

Results demonstrate strong generalisation, with the high-risk/high-confidence (HR-HC) group showing significantly worse infection-free and treatment-free survival compared with low-risk groups, including lower composite infection/treatment-free survival (54% vs up to 95.9%).

The model also showed reliable confidence calibration, as HR-HC patients had worse outcomes than HR-LC patients, supporting the predictive value of model confidence scores.

Importantly, HR-HC patients also had lower overall survival and significantly reduced infection-free and treatment-free survival, despite the model not being trained on survival outcomes.

These findings support the clinical utility of CLL-TIM as a decision-support tool to guide early risk stratification, inform infection prophylaxis strategies, and improve monitoring intensity in high-risk CLL patients.