Prognostic false-positivity of the sentinel node in melanoma

Arising from: Thomas MJ (2008) Prognostic false-positivity of the sentinel node in melanoma. Nat Clin Pract Oncol 5: 18–23 [doi:10.1038/ncponc1014]

We would like to take the opportunity to comment on the Review article by Thomas.¹ Thomas has calculated that the incidence of false positivity in the sentinel lymph node (SLN) group reported by Morton et al.² is 24%. Using a different calculation method we have produced an alternative interpretation of the MSLT-1 subgroup analysis results. The randomized design of MSLT-1 trial and its large sample size allow us to estimate the number of patients in the observation arm who were sentinel node positive at the time of randomization, and who subsequently died from melanoma. We have used these estimates to calculate a theoretical risk ratio that compares the survival of patients with positive sentinel nodes in the biopsy and observation groups. The steps are as follows:

• 16% of the biopsy group (764 patients) were SLN positive²

• Therefore, we would expect 16% of 500 patients of the observation group to have been SLN-positive, i.e. 80 patients

• Of the SLN-negative patients in the biopsy group, 62 (9.7%) died from melanoma²

• Therefore, we would expect 9.7% of the SLN-negative patients in the observation group to have died from melanoma, i.e. 40 of 420 patients

• In the whole observation group, 69 died from melanoma,² which means that 29 (i.e. 69–40) who died were patients who would have been SLN-positive at randomization

• Thus, we estimate that for the observation group, 29 of 80 (36%) SLN-positive patients died from melanoma

• In the biopsy group, 32 of 122 (26.2%) SLN-positive patients died from melanoma²

• Thus, the crude theoretical risk ratio is (32/122)/(29/80) = 0.72

• The crude risk ratio from the comparison made by Morton and coauthors² is (32/122)/(38/78) = 0.54

• The hazard ratio reported² was 0.51, which is very close to the crude relative risk of 0.54 that we calculated

This 'theoretical' analysis shows a non-significant benefit in favor of immediate CLND in SLN-positive patients indicating that the comparison presented by Morton et al.² was substantially biased and that any putative survival benefit was probably substantially overestimated. Inclusion of the false positive cases as calculated by Thomas¹ complements our own calculations, and is consistent with our findings, which indicate that false positivity was the main cause of bias.

Morton and coauthors³ have asserted that the statement by Thomas that false negative results were not included in their analysis was incorrect. They refer to figure 3B and page 1,312 in their article.² The first comparison was between subgroup 1 (122 sentinel-node positive patients and 32 related deaths) versus subgroup 3 (78 observation group patients who developed clinical lymph node metastases, and 38 related deaths). Subgroup 1 did not include patients who were false-negative or related deaths whereas subgroup 3, although making no allowance for false-negative patients, did include nine deaths which, according to our calculations, would have occurred in false-negative patients. The true number of deaths, therefore, in the SLN-positive patients in the observation group would have been 29 rather than 38. This is the key factor in the difference between our calculated risk ratio and that presented in the MSLT-1 study. The second comparison, according to Morton and coauthors³ showed that the survival comparison between subgroup 2 (composed of subgroup 1 plus the known 26 false-negative patients and their 14 related deaths) and subgroup 3, was also significant. This comparison, however, was again unbalanced because no equivalent proportion of false-negative patients had been included in subgroup 3.

It should be emphasized that the key result of MSLT-1 was the lack of a significant difference in melanoma-specific survival between the two randomized groups. Considerations of the subgroup analysis and the likely effects of false positivity and false negativity are necessarily of secondary importance but our calculations demonstrate the inherent risk of bias in inappropriate subgroup analyses and do not support the claim of a survival benefit for immediate lymphadenectomy in patients with positive sentinel nodes.