Background: In the context of fixed-duration (FD) Bruton's tyrosine kinase inhibitor (BTKi) and B-Cell Lymphoma 2 inhibitor (BCL2i) combination therapies for chronic lymphocytic leukemia (CLL), several key questions remain unresolved. These include optimal BTKi selection with venetoclax, especially in patients with substantial comorbidities, the therapeutic implications for high-risk populations, and the prognostic significance of undetectable measurable residual disease (uMRD).

Methods: A panel of 9 hematologists from both academic and community hospitals in the Campania, a region of southern Italy with an estimated population of six million, convened to develop a series of consensus statements. The process initially involved identifying the key domains critical to decision-making, which included: (1) patient-specific comorbidities, with particular emphasis on cardiovascular and renal conditions; (2) genetic risk profiles; and (3) logistical considerations (e.g. proximity to healthcare facilities, availability of caregivers). The panel developed 12 clinically relevant statements supported by a systematic review of the literature, conducted through an iterative process using PubMed and Embase databases. This review adhered to the PICO (Population, Intervention, Comparison, Outcome) framework to ensure a structured and comprehensive selection. Finally, a modified Delphi process was utilized to achieve consensus on the statements. During a one-day in-person meeting, participants anonymously rated their level of agreement using an online voting platform that employed a four-point Likert scale. A subsequent voting session was conducted to assess response consistency, and if consensus—defined as at least 75% agreement among voters—was not initially reached, a second round of voting was performed to facilitate convergence of opinions. Notably, the responses of Delphi panel's consensus were compared with those of ChatGPT-4, an AI-based platform.

Results: Overall, over 75% agreement, corresponding to Category 1 evidence, was achieved for ten out of twelve statements. Among these, seven were especially informative for guiding the selection of the most appropriate BTKi to combine with BCL2i across diverse patient populations.

Herein is a summary of these key consensus statements:

- Unmutated IGHV fit patients under 70 years old with wild-type TP53 should receive FD BCL2i–BTKi therapy as the preferred treatment option.

- Venetoclax plus first-generation BTKi (ibrutinib) FD combinations may not be recommended for elderly patients over 70 years who are unfit, regardless of the specific comorbidity.

- FD regimens that include first-generation BTK inhibitors, such as ibrutinib, should be avoided in medically unfit patients with low to intermediate cardiovascular risk.

- In such patients, second-generation BTKis—already preferred over ibrutinib in continuous therapy—are recommended as part of FD BCL2i–BTKi regimens.

- In FD regimens combining BTK and BCL2 inhibitors, MRD status exhibits lower sensitivity in predicting PFS compared to BCL2i–anti-CD20 antibody combinations.

- For patients lacking caregiver support or with limited access to healthcare facilities, second-generation BTKis administered as continuous therapy are preferred over FD regimens that necessitate intensive monitoring during the initial months.

- Despite the emergence of FD regimens continuous therapy remains the preferred first-line approach for TP53-mutated CLL, despite the emergence of FD regimens.

We subsequently compared the outcomes of the Delphi consensus with responses generated by an AI-based panel utilizing ChatGPT-4. The overall concordance rate was 66.6%, with a Cohen's kappa coefficient of 0.143, indicating a low level of agreement.

Conclusions: This consensus underscores the critical importance of adopting a personalized approach to optimize the selection of BTKis within venetoclax-based FD therapies. Specifically, for elderly patients—regardless of the presence of cardiovascular comorbidities—second-generation BTKis, such as acalabrutinib, should be prioritized in venetoclax-based FD regimens. Our study is pioneering in its comparative analysis of Delphi-derived expert consensus and responses generated by ChatGPT-4. The observed discrepancies underscore the continued superiority of human expertise in capturing nuanced, experiential factors that AI responses may fail to recognize.

This content is only available as a PDF.
Sign in via your Institution