Knowledge of sentinel lymph node status improves accuracy when predicting melanoma mortality and selecting patients for adjuvant immunotherapy

In a recent letter, Dr Dixon et al.1 stated that predicting the likelihood of melanoma mortality can assist patient selection for adjuvant immunotherapy. We fully agree with this statement. However, we have concerns about the statistical validity of the approach used to support their proposition that knowledge of sentinel lymph node (SLN) status obtained by SLN biopsy (SLNB) does not significantly improve predictions of melanoma mortality. Assessing data that we had previously analysed and published,2 Dixon et al. observed that combining SLN status with clinical and pathological predictors (CAPP) increased the C-statistic for overall survival by ‘only’ 3%, and stated that overlapping confidence intervals (CIs) indicated an absence of statistical significance. In fact, a C-statistic increase of 3% is generally regarded as substantial, and overlapping CIs do not necessarily indicate a lack of statistical significance. When interpreting CI comparisons, it is recommended that p ≤ 0.05 be assigned when the 95% CIs overlap by no more than half the average margin of error.3 In the figures produced by Dixon et al., the 95% CIs for CAPP and CAPP+SLNB only marginally overlapped, thus demonstrating a statistically significant incremental value of including SLN status over a model with CAPP alone. This is convincingly confirmed by re-analysing the actual data reported by El Sharouni et al.; the Wald-type p-values were 0.023 and 0.018 for comparisons using the MIA and Dutch cohorts respectively.4 These statistically significant C-statistic improvements reflect both higher sensitivity (identification of more high-risk melanoma patients (HRMPs)) and higher specificity (identification of more low-risk patients). Therefore, we disagree with the suggestion by Dixon et al. that replacing SLN status with CAPP-based algorithms would lead to more accurate detection of HRMPs. Also unsubstantiated is the claim that ‘510 patients would need to undergo SLNB to predict one more HRMP compared with using CAPP alone’. This number was apparently determined by multiplying the net benefit difference between the CAPP and CAPP+SLNB models to identify patients who would experience recurrence within 3 years as reported by El Sharouni et al.2 and the estimated 2-year recurrence rate in a HRMP cohort from von Schuckmann et al.5 However, the definitions of HRMPs and the follow-up periods were completely different in the two studies, invalidating direct comparisons. The former study defined HRMPs as those who would recur within 3 years of their primary diagnosis, while the latter considered HRMPs as all those with AJCC T1b-T4b melanomas, and reported 2-year recurrence data. Finally, it was asserted that ‘SLNB is now recognised as a considerably less precise technique for predicting melanoma survival and identifying patients suitable for adjuvant drug therapy’. This assertion is not supported by the available data. El Sharouni et al. analysed two large, independent datasets, and demonstrated that the inclusion of SLN status resulted in substantial, statistically significant improvements in the prediction of overall survival, melanoma-specific survival and recurrence-free survival. Thus, the recommendation by Dixon et al. that the focus should be exclusively on CAPP when selecting patients for clinical trials of adjuvant systemic therapy is problematic. Neglecting SLN status when assessing eligibility for such trials would result in the inclusion of numerous low-risk patients (due to low specificity) while potentially excluding some high-risk patients (due to low sensitivity). Selecting the most appropriate melanoma patients to receive adjuvant immunotherapy depends on predicting each patient's melanoma-specific mortality risk. Given the high cost of immunotherapy and the possibility of serious side-effects, as well as its potential benefits, achieving the most accurate risk prediction is clearly important. At present, this requires SLN staging, rather than reliance on less accurate predictions provided by CAPP alone. SNL is supported by Melanoma Institute Australia. RAS is supported by an Australian National Health and Medical Research Council (NHMRC) Investigator Grant (2022/GNT2018514). SNL and MAES have nothing to disclose. AHRV has received an honorarium from Novartis. RAS has received fees for professional services from MetaOptima Technology Inc., F. Hoffmann-La Roche Ltd, Evaxion, Provectus Biopharmaceuticals, Qbiotics, Novartis, Merck Sharp & Dohme, NeraCare, AMGEN, Bristol-Myers Squibb, Myriad Genetics and GlaxoSmithKline. JFT has received honoraria from BMS Australia, MSD Australia, GSK and Provectus Biopharmaceuticals, and travel and conference support from GSK, Provectus Biopharmaceuticals and Novartis. Data sharing is not applicable to this article as no datasets were generated during the current study.