The time to first treatment is an independent predictor of overall survival in chronic lymphocytic leukemia

Approximately, 80% of chronic lymphocytic leukemia (CLL) patients are diagnosed in the asymptomatic phase and/or at an early stage of the disease. Although the majority of these patients show a low-risk profile and are managed with active observation,1 the time to the first treatment (TTFT) in these CLL patients appears quite heterogeneous. Historical clinical observations suggested that CLL patients experiencing early progression develop a persistent risk of short overall survival (OS).2 Similar findings indicating a correlation between early progression after therapy and short OS have been reported in other hematological malignancies.3-5 Here, we hypothesized that an early need for treatment could inform OS also in the CLL setting. Therefore, we investigated the inclusion of the duration of the therapy-free interval following diagnosis as additional prognosticator to refine the OS prediction of well-recognized risk factors in a wide multicenter cohort of newly diagnosed CLL patients on a watch-and-wait strategy. The REporting recommendations for tumor MARKer prognostic studies (REMARK) criteria were followed throughout this study.6 Further details in Supplementary Appendix. Distribution in the whole cohort of 3860 CLL cases of the clinical and biological variables validated in CLL-IPI7 and Brno-Barcelona8 score systems, including age, Rai staging, β2-microglobulin, immunoglobulin heavy-chain variable (IGHV) gene mutational status, del(11q) and del(17p), all tested at diagnosis/first presentation, are reported in Supplementary Table 1. Distribution of variables is as expected in a cohort of early-stage CLL.9 Sufficient information for evaluation of the CLL-IPI7 and Brno-Barcelona8 score systems were available in 2573 cases, which were included in the final analysis (Supplementary Figure 1). The median follow-up time was 6.8 years (interquartile range: 4.2–9.9 years). At univariate analyses, age ≥65 years, Rai stages I–II, abnormal β2M serum levels, del(11q), IGHVunmut status and del(17p) were all OS predictors both in univariable and multivariable analysis; this basic model provided an explained variation in mortality (R2) and a Harrell's C index of 37.7% and 72.6%, respectively (Supplementary Table 2). Patients experiencing therapy need within 6, 12, 18, 24, 30, 36, 42, 48, 54, and 60 months from diagnosis were defined as TTFT6-60 failures (TTFT6-60Fail). Cases remaining therapy-free within 6–60 months were categorized as TTFT6-60 achievers (TTFT6-60Achieve). OS for TTFT6-60Fail cases was defined as the time from TTFT6-60 failure to death or last follow-up. OS for patients with TTFT6-60Achieve was defined as the survival time from achieving TTFT6-60 until death or last follow-up. In each TTFT6-60 category, patients died before 6–60 months or with <6–60 months of follow-up were excluded from the analysis (Supplementary Figure 1). Cases requiring therapy within 6–60 months (TTFT6-60Fail) experienced a roughly three-time higher risk of dying (Hazard ratios (HRs) ranging from 2.6 to 3.2) as compared to patients with a longer therapy-free interval (TTFT6-60Achieve; reference group, HR = 1) in univariate Cox analyses (Supplementary Table 3). All the investigated cut-offs of TTFT, except the TTFT6 (p = .083), independently predicted the mortality incidence rate (HR ranging from 1.45 to 2.04) in a sequence of multivariable models including all the variables of the CLL-IPI and the Brno-Barcelona scores as potential confounders (Supplementary Table 4). In addition, the inclusion of the biomarker TTFT in the range from TTFT12 to TTFT42 significantly improved the prognostic accuracy of this risk stratification model compared to a model including the CLL-IPI and Brno-Barcelona scores variables alone (Supplementary Table 2), as witnessed by higher values for both R2 and Harrell's indices (values ranging from 38.6% to 40.4%, and from 72.9% to 73.8%, respectively; Supplementary Table 4) in models that include TTFT values. Conversely, TTFT at time points ranging from 48 to 60 months failed to add prognostic accuracy to the same model in terms of R2 and Harrel's indices (Supplementary Table 4). Remarkably, the zenith of the prognostic accuracy of TTFT was found at a cut-off value of 18 months (TTFT18; HR 2.04, 95% CI 1.52–2.75; R2 40.4%; Harrell's C index 73.8%), while at values of TTFT ranging from 24 to 60 months, a gradual reduction of both R2 (from 39.8% to 36.3%) and Harrell's C index (from 73.3% to 70.9%) was observed (Supplementary Figure 2). To establish whether the prognostic power of TTFT18Fail was dependent not only on the time of the therapy (i.e., less than 18 months) but also on treatment (i.e., treated/untreated), we performed univariate and multivariable Cox analyses by splitting cases of our cohort in those never-treated (never-treated TTFT18Achieve) from those treated but after 18 months (TTFT18Achieve) and comparing them with TTFT18Fail cases. A significantly shorter OS was accounted for in cases with TTFT18Fail (median 8 years) compared with treated TTFT18Achieve (median 13 years) and with never-treated TTFT18Achieve cases (median 22 years; Supplementary Figure 3). In this context, however, when tested by multivariable analysis (n = 2336), the treated TTFT18Achieve group lost its independent prognostic role with respect to the never-treated TTFT18Achieve group, while TTFT18Fail remained an independent prognostic factor of OS together with del(17p), del(11q), abnormal β2M serum level, IGHVunmut status, and age, but not Rai stage (Supplementary Table 5; R2, 42%; C index, 75%). In an additional multivariable model utilizing the same variables but combining the treated and never-treated TTFT18Achieve groups in a single group, TTFT18 remained again independent predictor of OS along with β2M, del(11q), del(17p), IGHV gene status and age, while the Rai stage lost its prognostic value (Supplementary Table 6). According to this multivariable model, we derived a novel survival risk score (SRS), which included TTFT18 as a new prognostic factor (SRSTTFT18), which utilizes the regression coefficients in predicting mortality to assign weights to the various variables independently associated with death (Supplementary Table 6). After calculating the assigned risk scores on an individual basis, patients were first grouped into four risk categories by quartile (Supplementary Figure 4); then, since the second quartile did not significantly differ from the third quartile, the two curves were collapsed together (Figure 1). Based on such stratification, 622 (26.6%) cases of the cohort were allocated in the low-risk group, 930 (39.8%) in the intermediate-risk group, and 784 (33.6%) in the high-risk group (SRSTTFT18 0 = low risk; score >0–37.8 = intermediate risk; score > 37.8 = high risk; Supplementary Table 6). The 5-year probability was respectively 97.5% for low-risk group (HR = 1, reference category), 91.8% (HR = 3.2, 95% CI 2.2–4.7, p < .001) for intermediate-risk group, and 77.7% (HR = 10.3, 95% CI 7.2–14.7, p < 0.001) for high-risk group (Figure 1). Notably, differently from the low-risk group, which did not cross the median timeline, the estimated median OS was 16 years (95%CI 10.7–21.6 years) and 9.9 years (95%CI 8.1–10.1 years) for intermediate- and high-risk cases, respectively (Figure 1). The Harrell's C index and the explained variation in mortality were consistently higher for scores including TTFT18 than those excluding this variable (72.6% vs. 73.8% and 37.7% vs. 40.4% for HC and R2, respectively). The significant improvement in the goodness of fit (expanded vs. reduced model, χ2 = 22.14 [1 df], p < .001) indicated that the inclusion of TTFT18 into the reduced model significantly increased the prognostic accuracy of the same model for predicting mortality. Herein, we demonstrated for the first time the independent prognostic value on mortality of the duration of TTFT ranging from 12 to 60 months in a sizeable cohort of newly diagnosed CLL patients, initially followed by a watch-and-wait strategy. We showed that patients who experienced an early therapy need within 6–60 months (TTFT6-60Fail) experienced roughly three-time higher risk of dying than those with a longer therapy-free interval (TTFT6-60Achieve). When risk factors included in the CLL-IPI2 and Brno-Barcelona score systems3 were jointly introduced into the same multivariable model with several TTFT thresholds, cases experiencing therapy need within 12 through 60 months, i.e., those labeled as TTFT12-60Fail were independently associated with the worst outcome. Notably, the inclusion into the basic model of the time-to-therapy starts thresholds significantly increased the prognostic accuracy of the same risk model, achieving the best performance at the time-point of 18 months. Nevertheless, Harrell's C indices exceeded the 70% threshold in all the investigated TTFT intervals, underscoring the prognostic utility of the first therapy-free length in addition to the treatment itself. The clinical impact of TTFT18 on OS prompted us to design a novel scoring system incorporating this variable (SRSTTFT18) to predict mortality in CLL. Several measures of predictive performance were considered, all validating the SRSTTFT18. First, a significant improvement in fitness favoring the expanded model (with the TTFT18 variable) versus the reduced model (without the TTFT18 variable) was demonstrated. Second, Harrell's C index was higher for the model including TTFT18, although the accepted 70% threshold was exceeded in both cases. These data demonstrated that a risk prediction model that included TTFT18 had better prognostic accuracy than models without this new prognostic variable. Our SRSTTFT18 included both unchanging variables, that is, IGHV mutational status and TTFT18, along with a changing variable (age) and other variables that may change over time (i.e., β2M, 17p, and 11q deletions) thus ultimately modifying the initial risk class assignment. In this context, a reiterative model, similar to the CIRI score,10 that redefines the risk at different time points, including those of TTFT calculation, would be more meaningful for the patients. In conclusion, the results of this study may aid the identification of early-stage progressive cases who allegedly will fail from conventional therapies and can in turn benefit from risk-adapted treatment approaches with chemo-free regimens. However, this new prognosticator could be reconsidered in the setting of patients up-front treated with new drugs. Fortunato Morabito, Valter Gattei, and Massimo Gentile designed the study. Fortunato Morabito, Giovanni Tripepi, Massimo Gentile, Sabrina Mezzatesta, performed statistical analysis; Fortunato Morabito, Luca Laurenti, Gianluigi Reda, Iolanda Vincelli, Ernesto Vigna, Antonella Bruzzese, Antonella Zucchetto, Riccardo Bomben, Francesco Di Raimondo, Jacopo Olivieri, Giovanni Del Poeta, Massimo Gentile, Manlio Ferrarini, Giovanni Tripepi, Davide Rossi, Gianluca Gaidano, Enrica Antonia Martino, Gilberto Fronza, Adalgisa Condoluci, Giovanna Cutrona, Francesca Romana Mauro, Riccardo Moia, Francesco Di Raimondo, Fiorella Ilariucci, Antonino Neri, and Francesco Zaja analyzed and interpreted data. Fortunato Morabito, Massimo Gentile, Manlio Ferrarini, Antonino Neri, and Valter Gattei wrote the manuscript; all authors gave final approval with the only exclusion of Giovanni Del Poeta (deceased). Associazione Italiana Ricerca sul Cancro (AIRC) Grant 5× 1000 n. 9980 (to Fortunato Morabito, Manlio Ferrarini, Antonino Neri); AIRC, Special Program Metastases (n. 21198) 5× 1000 to Gianluca Gaidano; AIRC IG-5506 (to Gilberto Fronza), IG-14326 (to Manlio Ferrarini), IG-21687 (to Valter Gattei); AIRC and Fondazione CaRiCal co-financed Multi-Unit Regional Grant 2014 no. 16695 to Fortunato Morabito; Progetto Ricerca Finalizzata Italian Ministry of Health, Rome, RF-2018-12 365 790 (to Antonella Zucchetto and Giovanna Cutrona) and PE-2016-02362756 (to Valter Gattei); Italy Compagnia S. Paolo, Turin, Italy, Project 2017.0526 (to Gilberto Fronza); partially funded by Italian Ministry of Health (Project 5× 1000, 2015 and 2016) and Current Research 2016 and 2023 (to Gilberto Fronza, Giovanna Cutrona, and Antonino Neri); Swiss Cancer League, ID 3746, 4395, 4660, and 4705, Bern, Switzerland; European Research Council (ERC) Consolidator Grant CLLCLONE, ID: 772051 (to Davide Rossi); Swiss National Science Foundation, ID 320030_169670/1 and 310030_192439, Berne, Switzerland; The Leukemia & Lymphoma Society, Translational Research Program, ID 6594-20, New York. The authors declare no conflicts of interest. The data that support the findings of this study are available from the corresponding author upon reasonable request. Data S1. Supplementary appendix Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.