Strengths and pitfalls of NNT and NNH in early breast cancer escalation trials

Background: The management of patients with high-risk early breast cancer (eBC) requires balancing between clinical benefit and toxicity. Number Needed to Treat (NNT) and Number Needed to Harm (NNH) may provide a clinically intuitive framework to assess benefit–risk profiles, but their use in eBC has not been systematically explored. Methods: We analysed 10 phase II–III randomized clinical trials (RCTs) of treatment escalation strategies in patients with mid to high-risk eBC. To ensure consistency and cross-trial comparison, invasive disease-free survival (IDFS), disease-free survival (DFS), event-free survival (EFS), and overall survival (OS) were extracted at 3 and 5 years (y). Safety outcomes included any-grade (G) and G ≥ 3 adverse events (AEs), G5 AEs, and treatment discontinuation (TD). Absolute risk reduction (ARR) and absolute risk increase (ARI) were used to calculate NNT and NNH, respectively, reported with 95% CIs. Results: In patients with the HER2+ eBC, KATHERINE showed the most favorable efficacy, with NNTs of 9 and 8 for IDFS at 3 and 5 years, respectively, and a NNT of 27 for OS at 5 years. APHINITY reported the most favorable safety profile, with the highest NNHs for G ≥ 3 AEs (15) and TD (80). In patients with TNBC, CREATE-X reported the lowest NNTs for DFS (7 at both 3 and 5 years). The most favorable results in OS were observed in OlympiA (NNT 26) and KEYNOTE-522 (NNT 20). Regarding safety, KEYNOTE-522 achieved the most favorable NNH for G ≥ 3 AEs (26), while OlympiA showed the best NNH for TD (16). In patients with HR+/HER2− eBC, OlympiA showed the strongest efficacy results, with the lowest NNTs for IDFS at 3 (10) and 5 years (9), as well as for OS at 5 years (26). The most favorable NNHs were reported for G ≥ 3 AEs in OlympiA (8) and for TD in monarchE (19). Conclusion: NNT and NNH provide intuitive metrics to communicate benefit–risk trade-offs in high-risk eBC, although their variability across trials, subgroups, and timepoints requires cautious interpretation.