Real-world patterns of treatment intensity and prognostic model performance in older AML: Insights from the french ALFA-PPP cohort

Abstract Background: In older patients (pts) with newly diagnosed (ND) acute myeloid leukemia (AML), treatment options have expanded significantly. While intensive chemotherapy (ICT) remains the standard of care for fit pts, venetoclax (VEN)-based low-intensity therapies (LIT) are increasingly used. Despite existing guidelines, the choice between ICT and LIT varies in real-life practice. Advances in understanding the prognostic role of disease genetics have led to risk models for pts receiving LIT, mostly validated in clinical trials. Their relevance in unselected real-world populations, however, remains uncertain. In this study, we analyzed patterns of treatment intensity and evaluated the performance of LIT risk models in a real-world cohort of older pts with ND-AML. Methods: We analyzed data from the ND-AML cohort of the prospective ALFA-PPP study (NCT04777916), initiated in 2022 to collect real-life clinical and biological data from 29 centers. Centralized genomic (50-gene NGS) and MRD assessments were planned for all pts. We report findings from the first 652 pts aged ≥60 years enrolled between April 2022 and June 2025. In those treated with LIT, we assessed the prognostic value of the 4-gene molecular risk signature (mPRS, PMID: 38113472), the ELN-2024 classification (PMID: 39133932), and the refined ELN (rELN)-2024 model (PMID: 39486083). Results: Among 652 pts, 315 were female and 337 male (median age 71y [60–94]; ECOG 0–2/3–4: 625/27; HCT-CI 0–2/≥3: 401/251). AML was de novo in 442 pts (68%), therapy-related in 105 (16%), post-MDS in 70 (11%), and post-MPN in 35 (5%). Among 591 evaluable pts, NPM1, FLT3, IDH1/2, sAML-like mutations (PMID: 35767897), and MR cytogenetic abnormalities were seen in 25%, 21%, 21%, 58%, and 28%, respectively. Of 652 pts, 346 (53%) received ICT (7+3 based regimen: n=187; CPX-351: n=80; other: n=7) and 306 (47%) VEN-based LIT (VEN+azacitidine: n=305; VEN+LDAC: n=1). ICT was used in 89%, 78%, 60%, and 8% of pts aged 60–65, 65–70, 70–75, and >75y, respectively. According to investigators, main reasons for choosing LIT were advanced age (n=253, 83%), comorbidities (n=79, 26%), and AML features (n=74, 24%). Analysis showed that LIT use was independently associated with older age (p<.001), ECOG>2 (p=0.02), HCT-CI≥3 (p<.001), post-MDS (p=0.001), post-MPN (p<.001), sAML-like mutations (p<.001), MR karyotype (p<.001), and TP53 mutations (p<.001). With median follow-up of 15.3m (IQR 7.2–23.8), 12-month OS was 73% (95% CI, 68–79) with ICT vs 46% (95% CI, 40–52) with LIT. Among 288 LIT pts with genetic data (129 female [45%], 159 male [55%]; median age, 77 years [60–94]; ECOG 0–2/3–4: 270/18; HCT-CI 0–2/≥3: 152/136), the mPRS classified 139 (48%) pts as favorable risk, 77 (27%) as intermediate risk, and 72 (25%) as adverse risk. According to ELN-2024, 148 (51%) pts were categorized as favorable, 79 (27%) as intermediate, and 61 (21%) as adverse. The rELN-2024 model reclassified 58 pts (20%) as favorable, 133 (46%) as intermediate, and 97 (34%) as adverse. Median OS was 10.1 months (95% CI, 7.7–12.0). Median OS by mPRS was 13.7 months (95% CI, 11.7–18.9), 10.0 (6.5–13.9), and 4.8 (3.5–7.4) for favorable, intermediate, and adverse-risk pts, respectively. According to ELN-2024, median OS was 13.1 months (11.7–15.5), 10.0 (6.5–13.9), and 4.3 (3.0–7.4), while for rELN-2024, it was 31.0 months (14.2–NA), 11.5 (8.9–14.8), and 4.4 (3.3–7.2), for the favorable, intermediate, and adverse risk group, respectively. Discriminative performance for OS showed a Harrell's c-index of 0.64 (95% CI, 0.58–0.70) for mPRS, 0.64 (0.58–0.70) for ELN-2024, and 0.70 (0.64–0.76) for rELN-2024. Compared to ELN-2024, mPRS showed no significant difference in discriminative ability (pairwise C-index difference via non-parametric bootstrap: ΔC-index = –0.01 [95% CI, –0.02–0.01]; p=0.49). rELN-2024 significantly outperformed both mPRS (ΔC-index = +0.06 [95% CI, 0.02–0.10]; p<.01) and ELN-2024 (ΔC-index = +0.06 [95% CI, 0.02–0.10]; p<.01). Conclusions: In this real-world cohort of ND-AML pts aged ≥60 years, both ICT and VEN-based LIT were commonly administered, with treatment choice guided by patient- and disease-related factors. ICT was associated with encouraging OS outcomes in older selected pts. Among the prognostic models evaluated in the LIT group, the rELN-2024 showed superior discriminative performance compared to mPRS and ELN-2024, most notably by better identifying favorable-risk pts with substantially prolonged OS.