The characteristics of DLBCL patients and healthy donors
Healthy donors (n=130) and patients with DLBCL (n=349) confirmed by previous histopathological analysis were included in the study. A full list of clinical characteristics of healthy donors and DLBCL patients were listed in Supplementary Table 1. It showed that DLBCL patients and healthy donors had similar age, gender percentage, white blood cell count (WBC), absolute neutrophil count (ANC), platelet count (PLT) and albumin/globulin ratio (AGR). However, the absolute monocyte count (AMC) and RDW in DLBCL patients were significantly higher than that in healthy controls; and the absolute lymphocyte count (ALC), hemoglobin (Hb), albumin (ALB) and globulin (GLB) in DLBCL patients were significantly lower than healthy donors. There were 174 patients (49.9%) treated with R-CHOP, and 175 patients (50.1%) treated with CHOP only. The subgroups of patients’ Ann Arbor tumor stage were 89 (25.5%) in stageⅠ, 86 (24.6%) in stage Ⅱ, 61 (17.5%) in stage Ⅲ and 113 (32.4%) in stage Ⅳ. There were no statistical differences in the age, gender, and other clinicopathological parameters between the training set and the testing set (Supplementary Table 2).
Cut-off values of RDW and PLT at diagnosis for survival analysis
RDW, PLT and Hb are three common parameters in blood routine test. Our initial purpose was to investigate their differences and significances for prognosis in patients with DLBCL, but eventually we only studied the prognosis value of RDW and PLT because of the Hb difference between male and female.
Using ROC analysis and calculating the Youden index (specificity+sensitivity–1), the optimal cutoff values chosen for RDW and PLT were 14.35% and 126.5×109/L respectively in the training set (Supplementary Figure 1). After applying the values on analysis of the whole cohort, DLBCL patients were able to be classified into high-level and low-level groups, where 93 (26.64%) patients fell in with higher RDW group and 44 (14.43%) patients with lower PLT group.
Association of RDW and PLT with other clinical pathological factors
Linear correlation analysis showed that higher RDW level was associated with higher NLR, lower ALB and lower Hb; while lower PLT was correlated directly with lower WBC, but was not correlated with NLR, ALB or Hb (Supplementary Figure 2)
Further analysis showed that, the value of RDW>14.35% was significantly correlated with a poorer ECOG-PS (P<0.001), more extranodal sites of disease (P=0.002), presence of B symptoms (P=0.011), and higher Ann Arbor stage (P<0.001) as well as higher LDH level(P<0.001) and IPI score (P<0.001). However, we found no statistical significance between age and gender with RDW level. We then found that, patients with PLT≦126.5×109/L were significantly correlated with higher Ann Arbor stage (P=0.003), more extranodal sites of disease(P=0.021), higher LDH (P=0.013) level and presence of B symptoms (P=0.033). There were no statistical correlation observed between low PLT with age (P = 0.602) and gender (P =0.726). In addition, ECOG PS (P=0.096) and IPI (P=0.061) had only borderline significance (Table 1).
Levels of RDW and PLT at diagnosis and clinical outcomes
At the time of statistics and analysis in our study, the median follow-up time was 21.3 months (range: 0.80-126.93). During follow-up, there were 134 (38.4%) patients presented disease recurrence, progress or death, of which 79 (22.6%) died. In the training set, the survival rate of patients with higher RDW was significantly worse than patients with lower RDW (5-year OS: 43%vs 69%; 5-year PFS: 29%vs 53%)(Supplementary Figure 3a,3b), and the patients with lower PLT showed significant worse PFS than the patients with higher levels (5-year PFS: 30% vs 49%)(Supplementary Figure 3d), but the overall survival wasn’t significantly different (P=0.074) (Supplementary Figure 3c). Similar results were observed in the testing set and the whole cohort set (Supplementary Figure 3e-3l). In addition, in order to explore whether different chemotherapy regimens affect the evaluation efficacy of the level of RDW and PLT, we divided the patients into two groups, one group treated with R-CHOP regimen and the other group treated with CHOP regimen. Kaplan-Meier analysis showed poor OS and PFS in patients with high RDW (P=0.021 for OS and P=0.039 for PFS) and low PLT (P=0.001 for OS, P<0.001 for PFS) levels in the R-CHOP cohort. Patients with higher RDW and lower PLT in CHOP treated cohort had poorer OS (P=0.001 for RDW, P=0.045 for PLT), but the results of PFS were not statistically significant (Figure1).
We further assessed the prognostic value of RDW and PLT in the IPI subgroup. Kaplan-Meier analysis showed that, for patients with IPI score of 0-2, the RDW and PLT levels may not distinguish those with favorable outcomes from those with adverse outcomes (data not shown). However, in patients with IPI scores 3-5, the RDW and PLT levels were able to further risk-stratify patients into high-risk and low-risk groups. In R-CHOP cohort, the patients with lower PLT had significantly poorer OS (P=0.003) and PFS (P=0.013); and in high level of RDW patients, OS (P=0.014) was significantly reduced (Figure 2), similar results were also showed in the whole cohort and CHOP cohort (Supplementary Figure 4).
High RDW and Low PLT at diagnosis showed poor prognostic factors
To investigate the association between RDW and PLT levels with patients’ clinical outcomes, we performed the Cox proportional risk model. Table 2 and Table 3 summarized the results of univariate and multivariate analysis for factors influencing OS and PFS in all DLBCL patients. Univariate Cox proportional analysis revealed that old age, advanced Ann Arbor stage, poor ECOG PS, elevated LDH, B symptoms, more extranodal sites of disease, higher IPI score, higher RDW and lower PLT were all predictors of DLBCL patients for OS and PFS (Table 2). To explore whether RDW and PLT were independent prognostic factors of DLBCL patients, we performed a multivariate analysis, including age, tumor Ann Arbor stage, ECOG PS, LDH, extranodal sites, B symptoms, IPI, RDW and PLT. Interestingly, our results showed that older age (P<0.001), advanced Ann Arbor stage (P=0.037), higher RDW (P=0.003) and lower PLT (P=0.046) were independent prognostic factors for OS. In another hand for PFS, only older age (P<0.001), advanced Ann Arbor stage (P=0.002) and higher RDW (P=0.002) were independent prognostic factors (Table 3).
We further performed multivariate analysis by applying above indicators to R-CHOP cohort and CHOP cohort. Surprisingly, we found that elevated RDW was an independent prognostic factor (P=0.012) in CHOP cohort, and depressed PLT was an independent prognostic factor (P=0.003) in R-CHOP cohort for OS. However, RDW wasn’t an independent prognostic factor for PFS either in R-CHOP cohort or in CHOP cohort, whereas PLT was an independent prognostic factor (P=0.003) in R-CHOP cohort but not in CHOP cohort (Table 4).
Development of a modified IPI by adding both RDW and PLT
From multivariate analysis, there are clearly four independent prognostic factors for OS in all patients cohort. We then used the four clinical parameters to construct a new adjusted IPI model, age equaled to two points; RDW, PLT and Ann Arbor stage equaled to one point respectively[14]. Three risk categories were generated: low (0-1 points), intermediate (2-3 points) and high (4-5 points).
Based on the risk stratification model, the results showed that patients assigned to the low-risk group had good outcomes (5-year OS: 83%, 5-year PFS: 62%) and high-risk patients had very poor outcomes (5-year OS: 9%, 5-year PFS: 0%, Figure 3a,b) in all patients cohort. The similar results were observed in the R-CHOP (n =174) cohort (Figure 3c,d) and CHOP cohort(n =175)(Figure 3e,f). To strengthen the results from the multivariate analysis, we conducted a Harrell’s C statistics analysis. The c-index of the IPI prognostic model for OS was 0.744 for patients treated with CHOP, 0.709 for patients treated with R-CHOP, 0.725 for all DLBCL patients, and 0.763, 0.718, 0,743 in NCCN-IPI prognostic model. When the factors of RDW and PLT values were added, the predictive power was increased in both IPI and NCCN-IPI prognostic model. And the c-index of the adjusted IPI in the three cohorts was 0.753, 0.732 and 0.748 (Table 5).