Anzeige
Erschienen in:
13.08.2024 | Melanoma
Using Nomograms Wisely: Predicting Sentinel Node Positivity in Melanoma
verfasst von: Priscila Rojas-Garcia, MD, Bryan Ma, MD, Eva Lindell Jonsson, MD, PhD, Olivia Genereux, MD, Gregory McKinnon, MD, Thomas Brenn, MD, Golpira Elmi Assadzadeh, PhD, Claire Temple-Oberle, MD, MSc
Erschienen in: Annals of Surgical Oncology | Ausgabe 12/2024
Einloggen, um Zugang zu erhaltenAbstract
Background
Four externally validated sentinel node biopsy (SNB) prediction nomograms exist for malignant melanoma that each incorporate different clinical and histopathologic variables, which can result in substantially different risk estimations for the same patient. We demonstrate this variability by using hypothetical melanoma cases.
Methods
We compared the MSKCC and MIA calculators. Using a random number generator, 300 hypothetical thin melanoma “patients” were created with varying age, tumor thickness, Clark level, location on the body, ulceration, melanoma subtype, mitosis, and lymphovascular invasion (LVI). The chi-square test was used to detect statistically significant differences in risk estimations between nomograms. Multivariate linear regression was used to determine the most relevant contributing pathologic features in cases where the predictions diverged by > 10%.
Results
Of 300 randomly generated cases, 164 were deleted as their clinical scenarios were unlikely. The MSKCC nomogram generally calculated a lower risk than the MIA (p < 0.001). The highest risk score attained for any “patient” using MSKCC calculator was 15% achieved in one of 136 patients (0.7%), whereas using the MIA nomogram, 58 of 136 patients (43%, p < 0.001) had predicted risk >15%. Regression analysis on patients with >10% difference between nomograms revealed LVI (26, p < 0.001), mitosis (14, p < 0.001), and melanoma subtype (8, p < 0.001) were the factors with high coefficients within MIA that were not present in MSKCC.
Conclusions
Nomograms are useful tools when predicting SNB risk but provide risk outputs that are quite sensitive to included predictors.
