Patient selection and clinical characteristics
A total number of 275 MM patients were enrolled with available lipids profiles and survival data and randomly divided them into the training cohort (n=138) and validation cohort (n=137) as shown in Figure 1. The baseline patient characteristics are summarized in Supplemental Table 1. The age of patients ranged from 24 to 84 years with a median age of 60 years. 164 patients (59.6%) were male and 111 (40.4%) patients were female. Approximately half of the patients had a serum monoclonal protein of IgG, while about 22.9% (63) patients with monoclonal protein of IgA and 22.1% (61) patients did not have records of monoclonal protein. 96 (34.9%) patients received bortezomib-based treatment regimens and 179 (65.1%) were treated with traditional regimens. Approximately 25% of cases achieved complete response (CR).
Identification independent prognostic features through survival analysis
The univariate and multivariate COX analyses were used to evaluate the individual prognostic values of the profiles. Univariate Cox analysis showed that lipids including ApoB, CHO, TG, LDH, and HDL were independent prognostic indicators (P<0.05). While multivariate analysis showed that variables including ApoB, TG, LDH, LDL, HDL and the ApoB/ApoA1 ratio were independent prognostic indicators (Figure 2, P<0.05).
In addition, we compared the OS between the low- and high- level of the lipids and apolipoproteins with Kaplan-Meier survival analysis respectively (Figure 3A-H). The cumulative OS rate of patients in the low HDL, CHO, TG, ApoB group was significantly lower than that of patients in the high group (P<0.05) (Figure 3 B, D, F, H). Besides, The OS rate of patients in the low LDH, β2MG group was significantly higher than that of patients in the high group (P<0.05) (Figure 3 E, G).
Construction and validation of lipid and apolipoprotein risk scores
The potential prognostic factors based on univariate and multivariate Cox analysis were considered to associate with OS. Then, the LASSO regression analysis was employed to construct a prognostic model using the seven prognostic factors (ApoB, TG, LDH, LDL, HDL and ApoB/ApoA1 ratio) in the training cohort. Based on the penalized maximum likelihood estimator of 1000 bootstrap replicates, a 6-prognostic factor model was established via the minimum criteria optimal λ value (Supplemental Figure1). The equation for the model was as follows:
Risk score=-0.75 X ApoB serum level + 0.53 X ApoB / ApoA1 ratio - 0.28 X TG serum level + 0.95 X LDH serum level + 0.26 X LDL serum level - 0.77 X HDL serum level.
The lipid profile risk score was generated for each patient according to the above formula. Patients were further grouped into high and low-risk groups according to the median threshold of the lipids profile risk scores based on the training cohort. The distribution of age, ApoB, TG, HDL, LDH, β2MG and ISS stage were significantly different between the two subgroups (Table 1).
The survival of MM patients is summarized in Figure 3. In the low-risk group, patients had a significantly longer OS time compared to the high-risk group (p<0.05) in the training cohort (Figure 4A). The AUCs for 1-, 3-, 5- and 10-year survival were 0.696, 0.705, 0.756 and 0.940, respectively (Figure 4C).
Furthermore, in validation cohort, the low-risk group had longer OS time (p<0.05) as well (Figure 4B). The AUCs for 1-, 3-, 5- and 10-year survival were 0.707, 0.734, 0.661 and 0.806 (Figure 4D). We analyzed the distribution of the lipid profile risk scores in patients with different survival outcomes using dot plots to compare the survival of subjects. Our data showed that the survival of patients in the low-risk group was higher than survival in the high-risk group (Figure 4E-F) in the two cohorts.
Univariate and multivariate Cox analysis
The univariate and multivariate Cox analyses for risk score and other prognostic values were performed in MM. The univariate analysis indicated that the risk score was an independent prognostic indicator for OS (Figure 5 A, C) in the training and validation cohorts. After adjusting other clinical confounding factors in multivariate Cox analysis, the risk score was still an independent prognostic factor for OS (Figure 5 B, D) in the two cohorts mentioned above.
Comparison of the prognostic factors and merged risk scores
The AUC of the risk score model for 5-year OS was 0.756 [95% CI: 0.662-0.850]. The AUCs for the level of β2-MG, ISS and DS stage models were 0.656 [95%CI: 0.562-0.750], 0.637 [95%CI: 0.536-0.738] and 0.641 [95% CI: 0.538-0.745] in the training cohort. However, there were no significant differences in prognostic accuracy was observed between the three models (Figure 6A). The AUC of the risk score model for 3-year OS was 0.661 [95% CI: 0.544-0.778]in the validation cohort (Figure 6B). The AUC of the risk score model for 10-year OS was 0.940[95% CI: 0.883-0.997] (Figure 6C), while the β2-MG, ISS and DS stage models were 0.545 [95%CI: 0.338-0.756], 0.519 [95%CI: 0.272-0.765] and 0.410 [95% CI: 0.273-0.582], which was significantly higher than the AUCs for ISS and DS stage models. The AUC of the risk score model for 10-year OS was 0.806 [95% CI: 0.689-0.923] (Figure 6D) in the validation cohorts. Heat maps were used to compare the clinical characteristics and level of serum lipid profiles. Patients with high-risk scores were associated with older age and a higher stage of disease (Figure 6E-F).
Furthermore, to generate a more accurate evaluation system, a nomogram was used to integrate the classic prognostic factors, the ISS stage (Figure 7A) combining two cohorts. The calibration plots showed good performance of the nomogram in predicting the 1, 3, 5, 10-year OS (Figure 7B). The AUC of merged score for 1, 3, 5, 10-year were 0.725 [95% CI: 0.624-0.827], 0.709 [95% CI: 0.642-0.777], 0.732 [95% CI: 0.660-0.803] and 0.857 [95% CI: 0.767-0.948], which was significantly higher than ISS stage, suggesting that the nomogram can enhance the OS prediction compare to the standard prognostic factor (Figure 7C-F).