Impact of Hemoglobin Level As a Continuous Variable in Calculating Prognosis in MDS: Incorporation into LSC4 and IPSS-R

INTRODUCTION: Myelodysplastic syndromes (MDS) are a group of hematopoietic stem-cell disorders, with heterogenous prognosis. The Revised International Prognostic Scoring System (IPSS-R) (Greenberg et al., Blood, 2012) is the standard prognostic scoring system that uses several clinical and cytogenetic criteria as categorical parameters to predict prognosis of patients with newly diagnosed MDS. A recent study compared the performance of IPSS-R with a panel comprised of 4 stemness-related genes: LAPTM4B, NGFRAP1, EMP1, and CPXM1 (LSC4) and showed a significant improvement of survival classification performance when focusing on gene expression panel (Wang et al., Blood Advances, 2020). Despite the recent upgrade, the performance of none of these markers is sufficient enough and there is still need for improvement.

METHODS: In order to evaluate the predictive power of the LSC4 and develop novel models integrating RNA-sequencing data with clinical variables, we evaluated bone marrow samples from 56 independent MDS patients prior to any therapy. Patient samples were collected using institutional guidelines. BM mononuclear cells (MNCs) were enriched by Ficoll (GE Healthcare, Chicago, IL) protocol, following manufacture’s guidance. BM CD34+ cells were enriched using magnetic cell separation (MACS) and CD34+ magnetic beads (Miltenyi Biotech, Germany). RNA from sorted BM CD34+ cells was isolated using the TRIzol RNA isolation kit (Fisher Scientific, Waltham, MA) followed by RNA-Seq library construction. FASTQ files were processed in TopHat2 using the default options. We implemented cox proportional hazard model to create the survival panel and time dependent AUCs was used in order to evaluate the performance of the model. We divided the patients into high versus low groups using optimum cutoff method and used Kaplan-Meier estimator to show the survival behavior in each group.

RESULTS: In this work we first validated the LSC4 panel and compared it with the IPSS-R on our data set. Furthermore, in order to improve the classifier performance, we calculated the correlation of clinical features with survival status. We observed that Hemoglobin (Hgb), as a continuous variable, has a significant effect on predicting the overall survival and it can significantly improve the survival classification performance when combined with the IPSS-R or LSC4 scores (HR = 0.63, 95% CI 0.44 – 0.89, P <0.001 when Hgb combines with IPSS-R and HR = 0.67, 95% CI 0.48 – 0.93, P<0.001 when Hgb combines with LSC4). Our proposed models show an enhanced time AUC performance when focusing on survival status beyond 20 months of follow up compare to IPSS-R and LSC4 scores (Figure 1a). We have also observed that combining Hgb with LSC4 or IPSS-R significantly improves the confidence interval for long term survivals (Figure 1b). In fact, in these panels, unlike the IPSS-R panel, we treat Hgb as a continuous variable and we believe factorial consideration of Hgb can underestimates the true effect of this feature on survival. We also showed a distinct separation between two survival curves when we used new proposed panels amongst all patients (Figures 1c and 1d).

CONCLUSIONS: In this work, while providing further validation for LSC4 model, we showed that Hemoglobin, as a continuous variable, has a significant prognostic effect when integrated with either IPSS-R or LSC4. Hereby, we observed that increasing Hgb even by a single unit can improve patient’s overall survival.