Patient characteristics
The characteristics of 1077 NPC patients were listed in Table 1. The median follow-up time was 54 months. During the long-term follow-up, 331 patients (30.7%) experienced disease progression, including 76 cases of locoregional recurrence (7.1%), 172 cases of distant metastasis (16.0%), 40 cases of both locoregional recurrence and distant metastasis (3.7%), and 147 deaths (13.6%). The 5-year PFS, DMFS, LRFS, and OS rates were 68.4%, 79.2%, 87.7% and 84.8%, respectively.
Pre- and post-treatment plasma EBV DNA levels at various time points
A total of 1077 patients received plasma EBV DNA tests at a single time point pre- and post-treatment (Figure 1), and 252 patients received plasma EBV DNA tests at 4 time points (pre-treatment, 3, 12, and 24 months post-treatment). The results showed that the positive rate and viral load of pre-treatment plasma EBV DNA were significantly higher than those at each time point post-treatment (Figure 2A and 2B). Among 252 patients receiving plasma EBV DNA tests at 4 time points pre- and post-treatment, post-treatment EBV-DNA positivity was significantly lower than pre-treatment, and post-treatment plasma EBV DNA seemed to show an increase over time (Figure 2C). As shown in Figure 2D, the changes of plasma EBV DNA load pre- and post-treatment were quite different among these 252 NPC patients. Some patients maintained plasma EBV DNA baseline negativity (0 copies/ml) at all 4 time points, some patients showed plasma EBV DNA above baseline at only a single time point (pre- or post-treatment), and some patients rebounded above the baseline level after treatment (single or multiple time points), while some patients maintained plasma EBV DNA positivity at all 4 time points.
Patients with pre-treatment plasma EBV DNA load ≥1500 copies/mL had worse 5-year PFS, DMFS, LRFS, and OS than those with plasma EBV DNA load <1500 copies/mL (all P<0.001). Patients with plasma EBV DNA levels >0 copies/mL at 3 different time points post-treatment (3, 12, and 24 months) had a significantly lower 5-year PFS, DMFS, LRFS, and OS than those with negative plasma EBV DNA (all P<0.001). There were also significant differences in 5-year PFS, DMFS, and LRFS between patients with EBV DNA levels >0 copies/mL and EBV DNA levels <0 copies/mL at 36 months post-treatment (all P<0.001, Additional file 4, Figure 3). However, it was difficult to compare OS between these two groups at 36 months post-treatment as there was only 1 death among these patients.
Dynamic changes in plasma EBV DNA at 4 time points pre- and post-treatment
The positive rate and viral load of plasma EBV DNA fluctuated markedly at the 4 time points pre- and post-treatments (Figure 2C and2D). To examine the prognostic value of dynamic changes in plasma EBV DNA, we divided the 252 patients receiving tests at all 4 time points (pre-treatment and 3, 12, and 24 months post-treatment) into 5 subgroups according to the dynamic change pattern observed in Figure 2D. Pattern 1 (Pre- & Post 0+) consisted of 82 patients (82/252, 32.5%) with persistently undetectable plasma EBV DNA at all 4 time points. Pattern 2 (Pre+ & Post 0+) included 104 patients (104/252, 41.3%) with positive detection pre-treatment but negative detection at all time points post-treatment. Pattern 3 (Pre- & Post 1+) included 18 patients (18/252, 7.2%) with negative detection pre-treatment and positive detection at one time point post-treatment. Pattern 4 (Pre+ & Post 1+) contained 17 patients (17/252, 6.7%) with positive detection pre-treatment and at one time point post-treatment. Pattern 5 (Pre-/+ & Post ≥2+) included 31 patients (31/252, 12.3%) with positive detection at ≥2 time points post-treatment and negative or positive detection pre-treatment (Table 2).
There were significant differences in the rates of disease progression, distant metastasis, locoregional relapse, and mortality among these 5 patterns (all P<0.001, Figure 4A-4D). Similarly, the 3-year PFS, DMFS, LRFS, and OS also differed significantly among these 5 patterns (all P<0.001, Figure 4E-4H). The patients of Pattern 1 had the best prognosis, followed by Patterns 2, 3 and 4, while the patients of Pattern 5 had the worst prognosis (Table 2, Figure 4).
Multivariate analysis using a Cox hazard ratio model
Multivariate analysis identified the pattern of dynamic changes in plasma EBV DNA (defined by 4 time points pre- and post-treatment) as an independent predictor of PFS, DMFS, LRFS, and OS in NPC patients (Table 3). Pattern 4 and Pattern 5 were independent risk factors for worse PFS, DMFS, LRFS, and OS (all P<0.05), however, Pattern 3 was an independent risk factor for poorer PFS and DMFS (all P<0.01).
ROC curve analysis
ROC analyses demonstrated that dynamic changes in plasma EBV DNA showed larger area under the curve (AUC) values than plasma EBV DNA level at any single time point for predicting NPC disease progression (AUC=0.805; 95%CI, 0.740-0.869), distant metastasis (AUC=0.804; 95%CI, 0.730-0.879), locoregional relapse (AUC=0.704; 95%CI, 0.606-0.802), and mortality (AUC=0.817; 95%CI, 0.726-0.908) (all P<0.001) (Figure 5).