Validation of Four Prognostic Models for Metastatic Posterior Uveal Melanoma in a Danish Cohort

Abstract

Purpose: The purpose of this study was to validate the four prognostic models in patients with metastatic posterior uveal melanoma: (I) the American Joint Committee on Cancer (AJCC) staging system, (II) the Helsinki University Hospital Working Formulation (HUHWF), and two nomograms (III) the Valpione-nomogram, and (IV) the Mariani-nomogram.

Methods: One hundred fifty-two patients with metastatic posterior uveal melanoma were retrospectively included. Dictated by data availability, five subcohorts were established: AJCC (n = 152), HUHWF (n = 93), Valpione-nomogram (n = 92), Mariani-nomogram (n = 68), and a complete dataset subcohort (n = 64). The predictive performance was evaluated with time-dependent Brier-score, calibration plots, receiver operating characteristic (ROC) curves, and the global Harrell's C-index.

Results: The 6-month area under the ROC curve (AUC) was between 0.83 and 0.87 for, respectively, the Mariani-nomogram, the HUHWF, and the Valpione-nomogram, and was 0.77 for the AJCC staging system. The 24-month AUC was 0.81 for the Mariani-nomogram, compared with 0.79 (Valpione-nomogram), 0.74 (HUHWF), and 0.64 (AJCC). The C-index was 0.69 for the Mariani-nomogram, 0.71 for the Valpione-nomogram, and 0.73 for HUHWF (not calculated for AJCC). The accuracy of the prediction models, represented by the Brier score, was after 6 months and 24 months: 0.08 and 0.13 for the Mariani-nomogram, 0.10 and 0.17 for the Valpione-nomogram, 0.10 and 0.14 for the HUHWF, and 0.15 and 0.14 for the AJCC staging system.

Conclusions: All four models demonstrated an acceptable predictive performance at 6 and 24 months. The Mariani-nomogram appears to perform well across most metrics, often showing the highest values for AUC and the lowest Brier scores.

Figure 1.

Flow chart showing the process of inclusion and exclusion of patients into the validation cohorts for each of the four prognostic models. AJCC, American Joint Committee on Cancer; ALP, alkaline phosphatase; DFI, disease-free interval; HUHWF, Helsinki University Hospital Working Formulation; LDH, lactate dehydrogenase; LDLM, largest diameter of the largest metastatic lesions; MRI, magnetic resonance imaging; PS, performance status.

Figure 2.

Kaplan-Meier curves of overall survival from the date of metastatic diagnosis, stratified by (A) lactase dehydrogenase level (LDH) levels as the fraction of upper normal limit (UNL; n = 105), and (B) alkaline phosphatase level (ALP) as a function of upper normal limit (UNL; n = 103). Increasing LDH and ALP was associated with poorer survival outcomes (pair-wise log-rank test: LDH ≤ 1.0 × UNL versus LDH 1.1-2-0 × UNL [P = 0.023], LDH ≤ 1.0 × UNL versus LDH < 2.0 × UNL [P < 0.001], LDH 1.1-2-0 × UNL versus LDH < 2.0 × UNL [P < 0.001], ALP ≤ 1.0 × UNL versus ALP 1.1-2-0 × UNL [P = 0.023], ALP ≤ 1.0 × UNL versus ALP < 2.0 × UNL [P < 0.001], and ALP 1.1-2-0 × UNL versus ALP < 2.0 × UNL [P < 0.001]). ALP, alkaline phosphatase; LDH, lactate dehydrogenase; UNL, upper normal limit.

Figure 3.

Calibration plots of predicted survival at 6 and 24 months for the four prognostic models (calculated on the main analysis subcohorts). The diagonal lines represent a perfect prognostic model. A point above the diagonal line indicates that the patients exhibit longer survival times than predicted by the model. Conversely, a point below the diagonal line indicates that the patients die earlier than the model predicts. Bars indicate the 95% confidence interval for the observed risk. AJCC, American Joint Committee on Cancer; HUHWF, Helsinki University Hospital Working Formulation.

Figure 4.

Six-month and 24-month ROC curves for the AJCC, the HUHWF, the Valpione-nomogram, and the Mariani-nomogram. AJCC, American Joint Committee on Cancer; AUC, area under the curve; HUHWF, Helsinki University Hospital Working Formulation; ROC, receiver operation characteristic.