Improved Risk Stratification of Smoldering Multiple Myeloma (SMM) Using Trajectory Data in the Pangea 2.0 Model: A Multicenter Study in 1,431 Participants

Background

The 20/2/20 model is the current gold standard to stratify smoldering multiple myeloma (SMM) patients at baseline into three subgroups (low, intermediate, and high) according to the risk of progression based on the free light chain ratio (FLCr), M-protein concentration, and percentage of bone marrow (BM) plasma cells (PC). Evolving patterns that may alter the risk of progression are not considered in this static model. We previously proposed the PANGEA model that allows for personalized risk prediction using FLCr, M-protein, creatinine, age, hemoglobin trajectory, and optionally BM PC. We developed an improved PANGEA 2.0 model that includes trajectory modeling of these biomarkers to capture evolving patterns and improve predictions of MM progression.

Methods
We conducted a retrospective review of clinical data from 1,431 participants diagnosed with SMM at 4 international sites (Dana-Farber Cancer Institute, Boston, US, n = 737; National and Kapodistrian University of Athens, Greece, n = 379; University College London, UK, n = 97; and University of Navarra, Spain, n = 218). Dana-Farber participants comprised a training cohort to identify biomarker trajectories and to develop the PANGEA 2.0 model. The model was validated on two international cohorts: validation cohort 1 included patients from Greece and the UK (n = 476) and validation cohort 2 included patients from Spain (n = 218). The longitudinal data collected from 2018-2024 included current values and historical trajectories of age, M-protein, FLCr, creatinine, and hemoglobin, as well as BM PC (optional).

We used a systematic grid search with 5-fold cross-validation to determine optimal trajectory definitions for M-protein, FLCr, creatinine, and hemoglobin. For each, we evaluated seven binary trajectory definitions based on average increase over time (slopes) or recent increase from the previous visit on absolute or relative (%) increase scales, with varying thresholds and time periods. We used Cox regression models to create PANGEA 2.0 risk prediction models including the optimal trajectory variables. We compared the PANGEA 2.0 trajectory model with BM data to the 20/2/20 score at the last available time point by predictive accuracy (C-statistics) in the validation cohorts.

Results
Median follow-up of the training cohort was 3.5 years (IQR: 1.2 – 7.0 years), with a median of 5 visits per patient (IQR: 2 – 9 visits). Median age was 67 years, 53% were female, and 68%, 21%, and 12% had low, intermediate, and high-risk SMM at baseline per the 20/2/20 model. Thus far, 227 (19%) patients progressed to overt MM with a median time-to-progression of 3 years (IQR: 1.1 – 6.1 years).

The BM PANGEA trajectory model improved predictions of SMM patients’ progression risk with C-statistics of 0.86, 0.83, and 0.72 in the training cohort and validation cohorts 1 and 2 respectively, improving on the 20/2/20 model defined at the latest available time point (C-statistic: 0.77, 0.76, and 0.71). Importantly, in 33 (25%) cases of MM progressors who had increasing biomarker trajectories in validation cohort 1, the PANGEA 2.0 model accurately identified an increased risk of progression within 2 years while the 20/2/20 model classified them as low-risk (n=10) or intermediate-risk (n=23). In validation cohort 2, in 4 (44%) cases of MM progressors with increasing biomarker trajectories, PANGEA 2.0 accurately identified high-risk of progression while 20/2/20 classified them as intermediate-risk.

Conclusion

We developed the PANGEA 2.0 trajectory model to predict progression risk in SMM. In a large-scale, multicenter cohort with longitudinal follow-up, we demonstrated that adding trajectory information improved SMM risk prediction compared to the 20/2/20 model, particularly for patients with evolving biomarker values. We advocate adding these trajectories to 20/2/20 in a collaborative international study.