Conditional survival to assess prognosis in patients with chronic lymphocytic leukemia

Chronic lymphocytic leukemia (CLL) is a clinically heterogeneous hematologic malignancy with variable outcomes [1, 2]. While some patients may have a prolonged survival without needing treatment, others experience a rapidly fatal disease course despite receiving highly effective therapies [3‐5]. Improving patient management and treatment strategies requires reliable biomarkers to predict prognosis, disease progression, and treatment responses.

Advancements in understanding the molecular mechanisms and detailed clinical characterization have led to the identification of numerous prognostic and predictive biomarkers that complement the classic clinical staging classifications [6, 7]. Cytogenetic and molecular genetic aberrations like deletions in chromosomes 11q and 17p, TP53 mutations, and the immunoglobulin heavy chain variable gene (IGHV) mutation status play a crucial role in prognosis estimation and treatment respons [8, 9]. Composite prognostic scores like the CLL International Prognostic Index (CLL-IPI) integrate genetic, biochemical, and clinical parameters to predict survival differences. However, these models have limitations as they fail to consider how prognosis may change over time [10‐13]. From a patient’s perspective, the probability of surviving another t years when she/he has already survived s years might be more relevant than a static prediction. Conditional survival (CS) analysis, which takes into account how long an individual has already survived, offers dynamic prognostic information about changes in survival probability over time. For many cancers, CS is reported to increase over time [14]. In contrast, CLL has shown remarkable stable survival estimates even over a 10-year period following diagnosis. However, it is unclear whether this stability applies to all clinical stages or risk groups [15].

To address these knowledge gaps, we performed CS analyses in CLL patients with different risk profiles to improve prognosis estimation in patient subgroups by accounting for years already survived [16]. As treatment options evolve, incorporating dynamic prognostic information becomes increasingly crucial for guiding clinical practice and improving patient outcomes in CLL.

Accurate long-term prognosis in CLL is a challenge. Existing prognostic models, such as the CLL-IPI, are limited as they provide predictions based on a single time point, typically at diagnosis, without taking into account years already survived. This limitation hinders effective disease management and appropriate counseling of CLL patients. Previous analyses of prognostic models in CLL have highlighted this issue, raising uncertainties about which models can be reliably used in clinical practice to predict long-term outcomes [15].

In this study, we present a systematic analysis of absolute CS in CLL patients based on data from a non-study cohort comprising individuals diagnosed between 1984 and 2021 at two university medical centers. Our goal was to investigate survival estimates over time, stratified across different CLL patient subgroups. With a mean post-diagnostic follow-up of 7.3 years, we observed a constant 5-year CS of approximately 75% for the entire patient cohort, demonstrating stable survival over a period of up to 10 years.

Notably, CS remained remarkably stable regardless of age at diagnosis, although patients older than 65 years exhibited approximately 20% lower CS likelihood. These findings are consistent with a Canadian study that demonstrated stable CS up to 5 years after diagnosis [19]. Data from the USA and the Netherlands noted very slight decreases over time [20, 21]. However, it should be noted that the analysis in these studies examined relative CS from an epidemiological perspective, whereas we emphasized the patient-relevant perspective and considered absolute CS. These data suggest that the probability of surviving additional 5 years remains at 75% or slightly below over the disease course, indicating that CLL patients face a constant risk of death with each additional year of survival [22]. This is in contrast to the patterns observed in most hematologic and solid malignancies that are potentially curable [23]. Some aggressive diseases (e.g., pancreatic cancer, malignant melanoma) are associated with increasing CS [14, 17]. For other entities at early stages or with a tendency to recur (e.g., prostate or breast cancer), CS increases slightly over time or remains stable [19, 20, 23]. Constant CS comparable to CLL has been shown for multiple myeloma [24]. Common to both entities is the lack of curative treatment options and the goal of remission maintenance. CLL remains incurable to date with a steady risk of infection, autoimmune complications, secondary malignancies, and conversion to high-grade B-cell lymphoma (Richter’s transformation). It is unclear whether the availability of highly effective targeted treatment options (e.g., BTKis, venetoclax, novel CD20 antibodies) might influence these results.

Previous analyses have mostly not included the clinical and biological heterogeneity of CLL. Because such factors are available for a large proportion of patients in our cohort, we were able to stratify patients by known risk parameters including the composite prognostic index CLL-IPI. For IGHV mutation status, TP53 deficiency, and the CLL-IPI, our CS analyses revealed a clinically meaningful and significant separation of subgroups. IGHV mutation status and TP53 deficiency have known prognostic value and impact on the choice of targeted therapy [25, 26]. Their importance is also emphasized by the weighting in the CLL-IPI scoring system. The much less favorable prognosis of patients with an unmutated IGHV locus and the marked deterioration over time (CS decreases by approximately 30% over 10 years) not seen in IGHV-mutated patients reflect the heterogeneity of CLL and the fundamental biological differences of the disease associated with the IGHV mutation status. While patients with del(17)(p13) had a similarly poor prognosis at diagnosis followed by worsening in CS over time, the small number of patients and possible acquisition of TP53 deficiency during disease progression could obscur prognostic trends over time.

Not surprisingly, the composite prognostic score CLL-IPI, whose “static prognostic significance” we can excellently reproduce here in a “real-world” cohort outside clinical trials, shows a similar prognostic separation of CS over time with a 5-year survival rate ranging from a stable 95% (CLL-IPI = 0) to a decline to 25% (CLL-IPI = 2 + 3) over the 10-year observation period.

Patients in need for treatment over the observation period were less likely to survive, whereas 5-year CS increased slightly over time for untreated patients. This finding aligns with a recent study from the USA, which reported an increase in 5-year CS for untreated CLL patients aged ≥ 66 years based on linked surveillance, epidemiology, and end results-Medicare data [27]. However, given the heterogeneity of the patient cohort, the length of the observation period, and the therapeutic advances that have led to substantial changes in treatment regimens over time, meaningful conclusions are difficult.

In patients with CLL and other cancers, comorbidity is associated with shorter survival [28‐31]. In CLL, comorbidity has been shown to be an independent predictor of outcome, and different types of comorbidities are associated with increased overall mortality and particularly higher CLL-related mortality [32]. While the extent of comorbidities also had a significant impact on prognosis in our cohort, as previously reported for CLL, CS probability did not show significant changes over the 10-year period when stratified by CIRS score, using a cutoff commonly used in clinical trials to identify patients with relevant comorbidity burden (≤ 6 vs. > 6). This indicates that the prognostic significance of comorbidities remains similarly relevant and does not decline with increasing disease duration. This might be connected to a significant interaction between comorbidities and CLL treatment (in terms of treatment options and treatment tolerance) as previously demonstrated [29].

Thus, in order to profoundly investigate the influence of comorbidities on prognosis and CS in CLL patients in more detail, it is relevant to identify the causes of mortality (CLL related vs. unrelated), as CLL-related deaths also contribute significantly to increased mortality in patients with a high burden of comorbidities. This is mainly due to the fact that increased comorbidities are associated with a reduced chance of sufficient disease control [29]. However, the documentation of causes of death was only very incomplete in our registry-based dataset and therefore does not allow any analyses in this regard. This clearly demonstrates that inclusion and documentation of comorbidities and, in particular, causes of death in cancer registries, are essential for a meaningful prognosis assessment at time of diagnosis and for dynamic CS assessment over the disease course.

The study has several other limitations. First, the limited number of patients restricts detailed subgroup analyses and statistical power. Additionally, molecular parameters were only available for a subset of patients, potentially limiting the scope of the findings. Second, no reassessment of prognostic parameters, such as clinical stage or genetic changes, was performed. The extended 35-year period of diagnosis and follow-up, with data collected independently in two cohorts, may introduce variability in the analyses. In addition, patient and disease characteristics of patients from the two university hospitals comprising the cohort may not be fully comparable, and there is a risk of a bias towards patients with higher complication rates and more comorbidities.

In addition, at later time points in the patient observation period, there were newer treatments available that were not approved at the beginning. Thus, the effects of treatment remain largely unclear because of the fundamentally different types of therapies administered. Stratification by type of therapy or by time windows encompassing different modes of disease management (e.g., pre-Rituximab era or post 2014 introduction of BTKis) was not possible because of the cohort size and limited follow-up. As targeted therapies have become increasingly important in treatment regimens in recent years, their impact on current prognostic models remains to be determined. Finally, no information is available on the specific causes of death, which precludes conclusions about the reasons for persistent excess mortality in the entire cohort and increasing mortality in high-risk patients.