Refining risk stratification in CMML: A comprehensive assessment of the IPSS-M and other molecularly informed models.

7020 Background: Multiple molecularly-informed models have been designed in hopes of improving prognostication in CMML. More recently the IPSS-M was introduced for MDS and showed improved prognostic accuracy over the IPSS-R. Here we aimed to analyze whether the applicability of the IPSS-M extends to CMML and provide a definitive comparison between all major molecularly-informed models. Methods: A total of 367 pts with CMML treated at Moffitt Cancer Center with annotated clinical and molecular data were analyzed. The mean IPSS-M score was used as reference. Correlative analysis between CPSS-Mol, Mayo Molecular (MMM), GFM and IPSS-M was performed. Time-to-event analyses were estimated using Kaplan–Meier. We used Cox regression for survival endpoints. Model discrimination was evaluated using Harrell’s C concordance index. Results: We identified 987 driver point mutations involving 21 genes across 367 pts. We identified at least 1 mutation in 96% of pts (n = 352). Median follow-up was 4.4 yrs (3.8-5.0). The most prevalent mts were TET2, SRSF2, ASXL1, RUNX1 and NRAS. Using the IPSS-M schema, pts were classified as very low (7%, n = 26), low (29%, n = 108), moderate low (21%, n = 76), moderate high (16%, n = 58), high (18%, n = 65) and very high-risk (9%, n = 34). The IPSS-M identified 6 risk categories with mOS of 5.0, 5.1, 3.9, 2.6, 1.7 and 1.1 yrs, ranging from VL to VH (p < .05) and a 4-yr cumulative incidence of AML evolution of 2%, 14%, 17%, 18%, 25% and 14% respectively (p < .05) CPSS-Mol and IPSS-M equally showed improved OS discrimination over the MMM and GFM as demonstrated by an 11% and 6% increase in c-index respectively (0.71 vs 0.60 and 0.65). Similarly, CPSS-Mol outperformed CPSS by 3% (0.71 vs 0.68), as did IPSS-M relative to IPSS-R (6%, 0.71 vs 0.65) A 4-to-4 mapping between CPSS-Mol and IPSS-M risk groups (by merging VL/L and H/VH into LR and HR), resulted in the restratification of 57% of pts (n = 202). 22% (n = 15) of LR CPSS-Mol pts were upstaged. 30% (n = 23) of HR pts were downstaged 87 pts (23.7%) developed AML: 34 by 1 (9.2%), 60 by 2 (16.3%) and 79 by 4 yrs (21.5%). There was significant segregation between IPSS-M and CPSS-Mol groups and AML risk (p < .05), but not MMM (p = .191). CPSS-Mol showed the highest accuracy, followed by IPSS-M, MMM and GFM (0.66 vs 0.63 vs 0.57 vs 0.52) We then analyzed the predictive value of the models in pts receiving frontline HMA (n = 146). IPSS-M retained its prognostic accuracy with significant segregation between strata (p < .01), as did CPSS-Mol (p < .01) and MMM (p = .01). Conclusions: The IPSS-M can be used reliably in pts with CMML and shows comparable prognostic accuracy to the CPSS-Mol. Both outperformed MMM and GFM. All models retained its predictive value in pts receiving frontline HMA. This carries particular importance in community settings where CMML tends to be misdiagnosed as MDS. In such instances use of the IPSS-M is unlikely to adversely impact outcomes when used to guide treatment decisions.