INTRODUCTION

Chronic lymphocytic leukemia (CLL) has marked

clinical heterogeneity, ranging from indolent disease

to rapid progression, treatment failure, and premature

mortality. Established prognostic tools such as the

CLL-IPI may not capture complex interactions among

clinical, molecular, immunologic, and treatment-

related factors. AI/ML can integrate these

multidimensional data to support individualized

prediction.

AIM

To systematically review AI/ML models used to

predict survival and prognostic outcomes in

CLL and compare their performance with

conventional prognostic tools.

METHODOLOGY

PRISMA 2020-guided systematic review (2010-2026). Risk

of bias and applicability were assessed using PROBAST+AI

across participants/data sources, predictors, outcomes, and

analysis.

Records identified (n = 13)

Duplicates removed (n = 0)

Records screened (n = 13)

Excluded

(n = 7)

Full-text reports assessed (n = 6)

Excluded

(n = 1)

Studies included (n = 5)

Keywords: CLL • artificial intelligence • machine learning •

survival prediction • systematic review

RESULTS

• Treatment failure: AUC 0.95 vs 0.78 for CLL-IPI

• Infection risk: HR 20.47 vs 1.79 for CLL-IPI

• Survival clustering: OS p=0.016

• Composite/PFS outcomes: p<0.001

STUDY CHARACTERISTICS

STUDY MODEL OUTCOME KEY RESULT

Hoffma

nn

2023

ALPODS XAI +

logistic regression

Treatment

failure

AUC 0.95 vs 0.78;

p<0.0001

Mouaze

r

RSF / Decision

Tree / Cox

OS; time to

treatment

Risk scores; C-index

and time-AUC

2025

Parviz

2022

Ensemble ML

(+BL)

Death;

treatment;

infection;

composite

+BL > IPI; infection HR

20.47 vs 1.79

Coomb

es

2020

k-medoids PAM

clustering

OS; TTT; TTP OS p=0.016; TTT

p=0.004; TTP p=0.045

Ladyzy

nski

2022

Dynamic Bayesian

Network

60-month OS Simulated survival

aligned with registry

curves

OS: overall survival; TTT: time to first treatment; TTP: time to progression;

PFS: progression-free survival.

KEY OUTCOMES

Treatment failure • infection-related outcomes • overall

survival • time to first treatment • progression-free survival

CONCLUSIONS

AI/ML improved prediction across clinically important

CLL outcomes, with the greatest gains in treatment-

failure and infection risk. These findings support the

potential of multidimensional models to augment

conventional prognostic tools. Prospective multicenter

validation is required before routine clinical use.

PROBAST+AI ASSESSMENT

Risk of bias and applicability were reviewed across

participants/data sources, predictors, outcomes, and

analysis. Validation was predominantly internal (bootstrap,

train/test split, or cross-validation); one study used a single

cohort and one compared simulated survival with external

registries. No patient-based model had independent

external validation.

CLINICAL IMPLICATIONS

• AI/ML may improve individualized risk estimation beyond

CLL-IPI.

• Models using routinely available clinical variables may

be more feasible for implementation.

• Future studies should report calibration, decision-curve

analysis, fairness, and workflow impact.

• Multicenter prospective validation remains the essential

next step.

STUDY INFORMATION

Study type: Systematic review

Funding: No external funding reported

Conflicts of interest: None stated in supplied materials

Contact: Lakshmi Chandana Velaga, Northwest Medical

Center, Tucson, Arizona

INCLUDED STUDIES

Hoffmann (2023) • Mouazer (2025) • Parviz (2022) • Coombes (2020) • 

Ladyzynski (2022)