Molecular International Prognosis Scoring System for Myelodysplastic Syndromes

Background

Despite significant advances in our understanding of the genes recurrently mutated in myelodysplastic syndromes (MDS) and their clinical associations, current guidelines for risk stratification and treatment decisions do not consider gene mutations. Developed and validated in 3,675 representative MDS patients, we propose the Molecular International Prognosis Scoring System (IPSS-M) that considers clinical, cytogenetic and genetic parameters.

Methods

Diagnostic MDS samples from 2,957 patients with less than 20% blasts and white blood cell count below 13x10⁹/L were profiled for mutations in 156 driver genes (discovery cohort). Overall survival (OS) information was available in 95% of patients and leukemia-free survival (LFS) in 88%. Median follow-up was 3.8 years.

Correlative analysis between genetic alterations and outcomes was performed on LFS, OS and leukemic transformation. LFS was selected as the primary endpoint for IPSS-M model development. Candidate target risk variables consisted of blood counts, blasts, cytogenetics and gene mutations, while patient age, sex and MDS type (de novo or not) were treated as confounders. Feature selection was performed using complementary pair stability selection from L1-penalized Cox multivariable models. The relative weights of the selected variables were estimated using a robust Cox model adjusted for confounders. The model was validated in an independent cohort of 718 patients.

Results

We characterized 9,339 driver point mutations or short indels involving 124 genes across the 2,957 patients. We identified at least one gene mutation in 90% of patients, and 2 or more in 71%. We detected copy-number alterations and copy-neutral loss of heterozygosity in 40% and 10% of patients. Only 3% (n=101) of patients had chromosomal alterations without gene mutations.

The IPSS-M risk score was built as a continuous index, defined as a weighted sum of prognostic variables. The model consisted of i) hemoglobin, platelets and bone marrow blasts ii) IPSS-R cytogenetic category iii) 22 binary features derived from the presence of mutations in 21 predictive genes and iv) one feature representing the number of mutations from a group of 17 additional genes. To ease interpretation, a scaling was applied so that a score value of zero represented an average patient (i.e. a theoretical patient with mean values for all variables), while values of minus one, one or two corresponded to half, double or four-fold risk compared to the average patient, respectively (Fig 1A). The IPSS-M score resulted in improved discrimination compared to the IPSS-R score, across LFS, OS and leukemic transformation. We observed a 5 point increase in C-index for each endpoint in the discovery cohort and an increase of 3.4, 1.5 and 4.9 points in the validation cohort (Fig 1B).

A six risk category schema was defined based on score cutoffs: Very Low (14%, n=387), Low (32%, n=876), Moderate Low (11%, n=299), Moderate High (11%, n=284), High (14%, n=382) and Very High (18%, n=473). The IPSS-M risk categories demonstrated strong prognostic separation across all endpoints. Median LFS was 9.7, 6.0, 4.1, 2.3, 1.7 and 0.75 years from Very Low to Very High categories (Fig 1C).

Compared to the IPSS-R, IPSS-M categories (by merging Moderate Low and Moderate High as Moderate) resulted in re-stratification of 46% (n=1246) of patients (Fig 1D). Of these, 74% (n=926) were upstaged and 26% (n=320) were downstaged. All IPSS-R strata were affected by recategorization. More than half of patients from the IPSS-R Intermediate category shifted, among which 18% (n=56/310) were upstaged to IPSS-M Very High. Remarkably, 6% (n=98) of patients from IPSS-R Very Low / Low shifted to IPSS-M Very High / High. Highlighting the clinical relevance of this restratification, we found marked differences in outcomes between patients from different IPSS-M categories within each IPSS-R strata. Strikingly, the median LFS of IPSS-R Intermediate patients classified as IPSS-M Very High was 0.75 years (95% CI 0.63-1.1) while it was 6.5 years (95% CI 5.3-NR) for those classified as IPSS-M Low. Last, we showed that the IPSS-M clearly stratified patients with secondary or therapy-related disease (n=234).

Conclusions

The IPSS-M risk score is continuous, interpretable and reproducible. By including information from 38 gene mutations, it leads to improved discrimination compared to the IPSS-R score and restratifies nearly half of MDS patients.