Six hundred seventy patients were included in this analysis. Of these, 334 patients were alive as of May 2007. The median follow-up time of these patients was 7.8 years (range, 1.4 to 10.6 years). The median OS time is 7.2 years (95% CI, 6.0 years to not reached). RFS was assessed in 667 patients. Of these, 408 patients have experienced relapse thus far. The median RFS time was 3.1 years (95% CI, 2.4 to 3.7 years).
Baseline patient characteristics are listed in . At baseline, the median serum S100B value was 0.08 μg/L (range, 0.02 to 1.54 μg/L; n = 670). The higher the S100B level was, the higher was the risk of relapse and death. At baseline ( and ), 582 patients (87%) had an S100B level less than 0.15 μg/L (normal), and 88 patients (13%) had an S100B level ≥ 0.15 μg/L (high).
Fig 1. (A) Overall survival for E1694 patients (n = 670) with baseline S100B level of ≥ 0.15 μg/L (high) compared with less than 0.15 μg/L (normal). (B) Recurrence-free survival for E1694 patients (n = 667) with baseline S100B level of (more ...)
In the univariate analysis (log-rank test), a baseline S100B level ≥ 0.15 μg/L significantly correlated with OS (P = .010), although it correlated less significantly with RFS (P = .062). and display the Kaplan-Meier plots of OS and RFS by the baseline S100B level (normal v high).
In developing a multivariate model, first, univariate analyses were carried out using baseline S100B (< v ≥ 0.15 μg/L), sex (male v female), ECOG PS (0 v 1), pigmentation of primary melanoma (no v yes), ulceration (no v yes), size of primary melanoma (< v ≥ 1.2 cm), Clark level (I, II, III, IV, or V), Breslow depth (< v ≥ 3.28 mm), lymph node involvement at random assignment (0, 1, 2-3, or 4+), AJCC stage (II v III), age (< v ≥ 51 years), and treatment arm (HDI v GMK) for OS and RFS. Note that the categoric variables, as defined in , and the continuous variables (age, size, and Breslow depth) were dichotomized using the median value. Baseline S100B (P = .010), node involvement (P < .001), age (P = .009), PS (P = .044), stage (P < .001), and ulceration (P = .007) proved to be significant prognostic factors for OS. Treatment was not significant (P = .312) for OS; however, in patients with a high baseline S100B level (≥ 0.15 μg/L), patients treated with HDI experienced a longer survival (P = .028).
The multivariate Cox model for OS was selected in the following way. Initially, all six significant covariates and treatment were included in the model; the least significant covariate in this model was stage (P = .795), and stage was eliminated. In the model with five significant covariates and treatment, the least significant covariate was PS (P = .227), and PS was eliminated. In the next model with four covariates and treatment, age was the least significant covariate (P = .088) and was eliminated. This process led to the final model summarized in . The Cox model for OS included the baseline S100B level, node involvement, ulceration, and treatment. S100B was significantly associated with OS (P = .043) after adjusting for the other significant prognostic factors.
Multivariate Cox Model for Baseline S100B, Treatment, Ulceration, and Lymph Node Status in E1694
In univariate analyses for RFS, ulceration (P < .001), size (P = .026), stage (P = .003), and node involvement (< 0.001) were significant, and S100B (P = .062) and treatment (P = .106) were marginally significant. Patients with a baseline S100B level of ≥ 0.15 μg/L experienced a longer RFS when treated with HDI (P = .018). A similar final model selection process as described for the final model for OS was used for the RFS model. shows the final multivariate model for OS and RFS.
There were four time points when serum samples were collected and assayed for S100B; these were at baseline (n = 670), weeks 4 to 6 (n = 615), weeks 12 to 14 (n = 562), and weeks 48 to 52 (n = 395). These S100B values were also evaluated. The baseline proportion of patients with an S100B level ≥ 0.15 μg/L was 13%. Subsequently, this proportion increased to 21%, 20%, and 20% at weeks 4 to 6, 12 to 14, and 48 to 52. Of the 670 patients, 611 survived at least 1 year. To adjust for a possible lead time bias, only the 611 patients who survived at least 1 year were included in evaluating S100B values collected over 1 year. In these 611 patients, the changes in S100B values over time were categorized into the following three groups: group A = 378 patients (62%) with an S100B value of less than 0.15 μg/L at baseline and any of the later time points; group B = 71 patients (12%) with a baseline S100B value ≥ 0.15 μg/L; and group C = 162 patients (26%) with an S100B value of less than 0.15 μg/L at baseline that then increased to ≥ 0.15 μg/L at any one of the later time points. The median OS has not been reached for group A (95% CI, 7.0 years to not reached). The median OS was 7.8 years (95% CI, 4.7 years to not reached) for group B and 7.0 years (95% CI, 4.5 to not reached) for group C. The OS distribution was marginally significantly different in these three groups (P = .099 by log-rank test). The comparison of OS between groups A and C was significant (P = .048). Of the 611 patients, 608 had RFS assessed. The median RFS was 4.6 years (95% CI, 3.5 to 6.0 years) for group A, 4.3 years (95% CI, 1.5 to 7.0 years) for group B, and 2.4 years (95% CI, 1.7 to 5.0 years) for group C. The RFS distribution was significantly different in these three groups (P = .058). The comparison of RFS between groups A and C was significant (P = .017). and display the Kaplan-Meier plots for OS and RFS by the three groups.
(A) Overall survival and (B) for recurrence-free survival for E1694 patients by S100B trend (normal, high, or changing from normal to high at later time points).
Subset analyses were conducted to evaluate the survival prognostic value of S100B at later time points. For weeks 4 to 6, to adjust for the possibility of lead time bias, only patients who survived at least 4 weeks were included. In the Cox model including both baseline and week 4 to 6 values as covariates, only the week 4 to 6 S100B level was significant (P = .039; hazard ratio [HR] = 1.33). Similarly, including patients who survived at least 12 weeks, the S100B level at weeks 12 to 14 was significant (P = .001; HR = 1.62) after adjusting for the baseline S100B level. Including patients who survived at least 48 weeks, the S100B level at week 48+ was a significant predictor (P < .001; HR = 2.21) after adjusting for the baseline level. S100B level at week 48+ remains significant (P = .002; HR = 2.10) when adjusted for nodal status, ulceration, treatment, and baseline S100B value in the Cox model. For RFS, subset analyses evaluating the later time point S100B levels, while adjusting for baseline values, indicated that these are significant predictors of relapse (P = .001 for weeks 4 to 6, P ≤ .001 for weeks 12 to 14, and P = .051 for week 48).
Given the larger portion of data missing for the last week 48+ measurement (266 of 661 patients; 40%), we used the S100B data from the first three time points (baseline, weeks 4 to 6, and weeks 12 to 14) as time-varying covariates in the Cox regression model while adjusting for the other significant prognostic factors (node involvement, ulceration, and treatment). S100B level proved to be a significant predictor of both death and relapse. HR for death was 1.44 (95% CI, 1.06 to 1.95; P = .0210), and HR for relapse was 1.70 (95% CI, 1.21 to 1.92; P < .001). lists these results.
Multivariate Cox Model for S100B Values (at baseline, weeks 4 to 6, and weeks 12 to 14 as time-varying covariate), Treatment, Ulceration, and Lymph Node Status in E1694