The Chronic Lymphocytic Leukemia Comorbidity Index (CLL-CI): A Three-Factor Comorbidity Model

Abstract

Purpose: Comorbid medical conditions define a subset of patients with chronic lymphocytic leukemia (CLL) with poor outcomes. However, which comorbidities are most predictive remains understudied.

Experimental design: We conducted a retrospective analysis from 10 academic centers to ascertain the relative importance of comorbidities assessed by the cumulative illness rating scale (CIRS). The influence of specific comorbidities on event-free survival (EFS) was assessed in this derivation dataset using random survival forests to construct a CLL-specific comorbidity index (CLL-CI). Cox models were then fit to this dataset and to a single-center, independent validation dataset.

Results: The derivation and validation sets comprised 570 patients (59% receiving Bruton tyrosine kinase inhibitor, BTKi) and 167 patients (50% receiving BTKi), respectively. Of the 14 CIRS organ systems, three had a strong and stable influence on EFS: any vascular, moderate/severe endocrine, moderate/severe upper gastrointestinal comorbidity. These were combined to create the CLL-CI score, which was categorized into 3 risk groups. In the derivation dataset, the median EFS values were 58, 33, and 20 months in the low, intermediate, and high-risk groups, correspondingly. Two-year overall survival (OS) rates were 96%, 91%, and 82%. In the validation dataset, median EFS values were 81, 40, and 23 months (two-year OS rates 97%/92%/88%), correspondingly. Adjusting for prognostic factors, CLL-CI was significantly associated with EFS in patients treated with either chemo-immunotherapy or with BTKi in each of our 2 datasets.

Conclusions: The CLL-CI is a simplified, CLL-specific comorbidity index that can be easily applied in clinical practice and correlates with survival in CLL.

Figure 1.. Construction of the CLL-CI index.

(A, B) Variable ranks by VI and MD, as computed from RSF modeling of EFS, for 100 random sub-samples (each with n=380) of the derivation dataset. (C, D) Frequency of top 3 rankings (among all CIRS classes) according to VI or MD for the 100 random sub-samples. CIRS class abbreviations: ‘Endo’ = Endocrine / metabolic; ‘GU’ = Genitourinary (non-renal); ‘HEENT’ = Head, Eye, Ear, Nose, and Throat; ‘HTN’ = Hypertension; ‘LGI’ = Lower gastrointestinal; ‘MSK’ = Musculoskeletal / integumentary; ‘Neuro’ = Neurological; ‘Psych’ = Psychiatric / behavioral; ‘Resp’ = Respiratory; ‘UGI’ = Upper gastrointestinal; ‘Vasc’ = Vascular / hematological

Figure 2.. CLL-CI predicts survival in patients with CLL (derivation dataset, N=570).

(A, C) Kaplan-Meier curves by CLL-CI category and log-rank test p-value for EFS and OS when evaluating all patients. (B, D) Multivariable Cox model adjusted (for 4 prognostic factors) curves by CLL-CI category for EFS and OS on all patients. (E-H) Cox model adjusted EFS curves by CLL-CI category among patients receiving CIT (E), ibrutinib (F), treated in the frontline setting (G) and in the R/R setting (H).

Figure 3.. CLL-CI predicts survival in patients with CLL (validation dataset, N=167).

(A, C) Kaplan-Meier curves by CLL-CI category and log-rank test p-value for EFS and OS when evaluating all patients in the validation set. (B, D) Multivariable Cox model adjusted (for 4 prognostic factors) curves by CLL-CI category for EFS and OS on all patients.