Page 158 - 2019_02-Haematologica-web

P. Baliakas et al.
AB
Figure 1. Kaplan-Meier curves for time-to-first-treatment (TTFT). (A) In Binet A, B and C mutated chronic lymphocytic leukemia (M-CLL) patients and (B) Binet A, B and C unmutated chronic lymphocytic leukemia (U-CLL) patients.
son error and the significance level was set to P<0.017. TTFT was evaluated from the diagnostic date until the date of initial treat- ment; untreated cases were censored at the time of last follow up. Survival curves were constructed using the Kaplan-Meier method, and the log-rank test was applied to determine statistically signif- icant differences between survival proportions.
The univariable Cox regression model was used to determine the most important prognostic factors. Multivariable Cox regres- sion analysis was subsequently applied to evaluate the simultane- ous effect of the important variables on TTFT. For the multivari- able Cox analysis, we considered only those cases with available data for all the factors included in the model (n=918 for M-CLL and n=384 for U-CLL) on the grounds that imputing the values of the biomarkers could introduce substantial bias. The same rule was applied for the binary recursive partitioning. That said, no major differences were observed between the entire cohort and the proportion of cases included in the multivariable Cox analysis (Online Supplementary Table S3); this suggests that the patients included in the multivariable analysis were representative of the entire cohort.
The proportional hazard assumption was assessed for both the univariable and multivariable analysis. The stability of the multi- variable Cox model was internally validated using a bootstrapping procedure. Harrell’s C index was also calculated to further assess the discriminatory power of the multivariable analysis when: a) all important prognostic variables were included in the model; and b) the resultant prognostic index was included in the model as the sole prognostic variable (C=1 indicates perfect discrimination; C=0.5 equates to chance).40
Binary recursive partitioning, based on the development of con- ditional inference trees, was used to further validate the results of Cox regression analysis.41 The patients were initially divided in subgroups (terminal nodes) with different survival behavior. An amalgamation algorithm was subsequently applied to merge the terminal nodes that did not exhibit a significant difference in sur- vival. All tests were two-sided and P<0.05 was considered statis- tically significant. Statistical analyses were performed using the Statistica Software 10.0 (StatSoftInc, Tulsa, OK, USA) and R-3.2.2. Details are provided in the Online Supplementary Appendix.
Results
Prognostic impact of clinical staging within M-CLL and U-CLL
The median follow-up time for the entire cohort was 7.1 years (range, 0.1-33.1) with a median TTFT of 6.4 years (95%CI: 0.01-20.2 years) (Online Supplementary Figure S1). As expected, in both M-CLL and U-CLL, Binet A patients exhibited significantly longer TTFT compared to Binet B and C cases (Figure 1). Most likely, this reflects the sharp clinico-biological differences between Binet A and Binet B/C patients while also underscoring the current indications for treatment that broadly overlap with the cri- teria for Binet stage B and C assignment (Online Supplementary Tables S4 and S5). In keeping with previous reports, M-CLL cases (n=1364, 58%) exhibited significant- ly (P<0.0001) longer TTFT compared to U-CLL cases (n=1002, 42%) (Online Supplementary Figure S2A). Significant differences (P<0.0001) between M-CLL versus U-CLL remained when the analysis was restricted to Binet A cases (Online Supplementary Figure S2B).
Prompted by these findings, and also considering that Binet A cases predominated in both M-CLL and U-CLL (90% and 67%, respectively), we focused our attention on 1900 Binet A cases (M-CLL: n=1224; U-CLL: n=676). We refrained from investigating the impact of biomarkers within Binet B/C cases since the limited number of cases within each subgroup would not allow firm conclusions to be drawn.
Analysis for time-to-first-treatment within early stage mutated chronic lymphocytic leukemia
Univariable Cox regression analysis within M-CLL Binet A cases (n=1224) revealed that male sex, CD38 pos- itivity, +12, subset #2, TP53 abnormalities (TP53abn) and borderline IGHV gene germline identity (GI: 97-97.9%) were associated with significantly shorter TTFT (Table 1). Bordeline GI remained significant in the univariable analy- sis even when subset #2 cases were excluded. TP53abn,
362
haematologica | 2019; 104(2)