Sentinel node biopsy in cutaneous melanoma: time for consensus to better inform patient choice

In this edition of the British Journal of Dermatology we see stark evidence of the polarization that bedevils the state of sentinel node biopsy (SNB) in melanoma.1, 2 Couple this with a fact almost universally acknowledged, that doctors are confused by statistics,3 and it is little wonder that we find it difficult to help patients make informed decisions about their treatment.

The 10-year analysis of SNB data from Multicenter Selective Lymphadenectomy Trial (MSLT)-I are now available,4 and it is time to reach consensus in management. Melanoma is a dangerous disease and patients deserve better than to have their management determined by personal conviction or possible gain, or informed by spin. Sladden et al.1 claim the evidence does not justify SNB as the gold standard of care, while Faries et al.2 (for the MSLT group) tell us that ‘failure to recommend this technique […] raises important ethical questions that cannot be ignored.’

There are four main controversies surrounding the MSLT-1 data: the conflation of SNB staging with perceived benefit to patients; the claimed survival benefit of intervention (SNB) over observation in those with nodal metastases; the possibility of overdiagnosing nodal micrometastases (nodes that contain melanoma cells but that would not have progressed even if the node had not been removed); and finally the claim that intervention significantly improves disease-free survival.

Biopsy-based staging of intermediate or thick primary melanoma provides important prognostic information

SNB does provide prognostic information, but this is not in dispute. The issue is whether SNB can be justified for staging if there is otherwise little benefit to the patient. We already have noninvasive means of staging.

SNB has emerged as a reliable technique for identifying melanoma micrometastases in the ‘sentinel node’ – the first draining node in the lymphatic chain. Moreover, the presence of micrometastases in the sentinel node has proven biologically relevant in melanoma, so that it is an accurate predictor of survival in a group of patients; those with evidence of micrometastatic disease do significantly worse than those without (although still not as badly as you might expect).4, 5 Many clinical trials have been designed around SNB staging, and this is likely to remain a prerequisite for entry to future trials. It is also likely to determine whether or not a patient will receive adjuvant therapy, so its importance will inevitably grow, not diminish, with time.

But how much more accurate is node status than clinical staging when it comes to discussing individual patient survival? There are, after all, a host of confounding factors that influence outcome for an individual, including race, age, sex, tumour site, concomitant illness and therapy. It is worth remembering that around 60% of node-positive patients with intermediate-thickness melanoma will survive, and 15–20% of those with negative nodes will still die from melanoma.4, 5

When it comes to evaluating SNB in this respect, it certainly provides accurate nodal mapping, but it is doubtful that it adds much value over clinical staging for most patients, especially given the added cost of anaesthesia and an acknowledged 5–10% surgical morbidity.4 Ironically, it is those with thicker melanomas (> 4 mm) who may gain most from the increased predictive value of nodal over clinical staging,6 a group who are not generally offered SNB.

MSLT-I provides no evidence of improved melanoma-specific survival associated with sentinel node biopsy and elective lymph node clearance

The registered primary outcome of MSLT-17 was to determine whether wide excision of the primary with intraoperative mapping followed by selective lymphadenectomy would effectively prolong overall survival, compared with wide excision of the primary melanoma alone. It did not: there was no treatment-related difference in the 10-year melanoma-specific survival in the overall study population.4 There was, in fact, no significant treatment-related difference in intermediate-thickness melanoma (1·2–3·5 mm) (P = 0·18), or in any other subgroup.4

This is where the statistics begin to get murky. The study was powered adequately, but as no overall treatment-related difference was found, the authors argue that analysis should focus instead only on those 20% of patients where benefit was most likely (those with occult metastases at the time of the primary excision). They do three controversial things. Firstly, they exclude all but intermediate-thickness melanoma from further analysis – data on thin melanomas are deemed exploratory, and data on thick melanomas are excluded on the basis that they proved nonsignificant…

Secondly, they apply an obscure ‘accelerated failure time latent subgroup model’ (which has not been independently validated) to their post hoc analysis. Somewhat surprisingly, the treatment effect on melanoma-specific death from intermediate-thickness melanoma in SNB is resurrected from a nonsignificant P = 0·18 to a significant P = 0·05. It is not clear how this happened. Although the latent subgroup model should better reflect the magnitude of any treatment effect in those with positive SNB, it is difficult to see how the P-value (if correctly calibrated) could change to such an extent.8 Indeed, a different latent subgroup model applied to distant disease-free survival in the third interim analysis of the MLST-I trial yielded P = 0·91.9

Thirdly, and finally, they look at 10-year melanoma-specific survival of patients with node-positive SNB and those with nodal relapse in the observation group – something they describe as an analysis of biopsy-based management in patients with intermediate-thickness melanoma and nodal metastases. It sounds respectable enough, but by any rigorous standard this postrandomization subgroup analysis is invalid; it is in no way clear that these are equivalent groups and it therefore breaks the randomization required for unbiased comparison. It would certainly have been thrown out by any regulator had it been applied to a pharmaceutical trial.

And this brings us to the possibility of bias and the problem of overdiagnosing nodal micrometastases – the existence, or not, of the so-called prognostic false positives.

The possibility of overdiagnosing nodal micrometastases

A sentinel node is reported as positive even if only one malignant naevocyte is seen in either the nodal parenchyma or afferent lymphatic vessels, and even if it is identifiable only by immunohistochemistry.4 So for the study, a single cell and an extracapsular metastatic deposit are both treated the same way – yet they may not be biologically equivalent. For example, it may be the case that not every melanoma cell inevitably progresses to nodal disease. This is important, because if even some of them were to regress if left in situ, or not to progress, then the MSLT-I outcomes would be biased in favour of intervention. SNB would be overdiagnosing patients with nodal disease, which, in fact, would never have progressed. Inevitably, these patients would do better. We need to consider this possibility, even though Faries et al. state that there is no evidence from MLST-I that overdiagnosis occurs.2

The cumulative incidence of nodal metastases at 10 years (taking into account censoring) was 19·5% in the observation group and 21·9% for SNB.4 It is true that this difference is nonsignificant, but lack of evidence is not the same as evidence for lack of overdiagnosis. The actual proportions with nodal metastases are 17·4% (87/500) and 19·9% [(122 + 31)/770], respectively, and the ratio of the proportions is 1·142 – this means that there were 14% more patients with nodal metastases found in the SNB arm. (Proportions are more relevant than Kaplan–Meier estimates because, for this purpose, we are not interested in those who might have developed positive nodes had they not died of something else first, but in those who were actually found to have positive nodes during follow-up.)

If these additional 14% were those patients with occult positive nodes (detectable by SNB) at the outset whose disease would have either regressed or not progressed during the study, which is plausible, then we should allow for 14% more ‘node-positive’ patients in the observation arm – of whom none would have progressed either. This changes the 10-year survival for the observation group from 41·5% (as reported) to 48·8% [100 − (100–41·5)/1·142]. This figure is much closer to the 56·0% survival for SNB (hazard ratio 0·81, 95% confidence interval 0·56–1·17), rendering the outcomes between observation and SNB nonsignificant (P = 0·26).

Sentinel node biopsy followed by completion lymphadenectomy (intervention) significantly improves disease-free survival

The claim that SNB prolongs disease-free survival is disingenuous; the apparent improvement is an artefact that arises because nodes containing melanoma in SNB are not counted as diseased, whereas any positive node in the observation group is. How did they get away with this? When nodal disease is excluded as a site of first metastasis, there is no improvement in real disease-free survival (P = 0·34).10, 11

But it is precisely this, the artefactual increase in disease-free survival, that gives SNB a perceived clinical advantage over observation. Patients who remain disease free after SNB and elective clearance believe they are recurrence free. By contrast, those equivalent 20% of patients under observation will suffer the psychological and physical trauma of nodal relapse. This becomes increasingly important as the likelihood of a positive node increases: again, patients with thicker melanomas (> 3·5 mm) stand to gain more here from SNB, as well as from improved surgical morbidity and better locoregional disease control following elective clearance lymphadenectomy.6, 12

In summary

It is hard to justify SNB for staging alone – except in the context of a clinical trial. However, this will become more important with the advent of effective new primary and adjuvant therapies.

SNB has the distinct advantage over observation of reducing the fear of nodal disease – patients appear less traumatized by sentinel node involvement than with nodal relapse or recurrence. The likelihood of a positive node ranges from 5% to 45% in melanomas from 1 mm to 4 mm.13 Although artefactual, any increase in perceived disease-free survival with SNB will be of real value to those with occult nodal disease from the outset. These patients will also benefit from early lymphadenectomy with consequent reduction in surgical morbidity and, possibly, improved locoregional disease control.6, 12 These are compelling reasons for early intervention and are likely to be important drivers for patients consenting to SNB.

There is no evidence of harm associated with SNB (excepting surgical morbidity for those 80% of patients who did not require it,4 and the likely cost to the health service), but neither is there convincing evidence of improved survival with biopsy-based management, even for those with intermediate-thickness melanoma and whatever the claim.2, 4

In an editorial in the Annals of Surgery, Daniel Coit, a colleague and friend of the late Dr Morton, who initiated the MSLT-I trial, writes critically and compassionately of the data, ‘we owe Dr Morton and his memory the respect of our honest appraisal of the extraordinary facts his trial has provided. We need to let these critically important data speak to us freely so that we can draw reasonable and rational conclusions about the role and the limitations of SLNB in the initial management of our patients with melanoma.’14 In light of this and in the spirit of openness, the MSLT-I group should consider making available individual patient data for independent reanalysis, as is increasingly required for trials of medicines.15

It is time for dermatologists and oncologists, surgeons and statisticians, to reach consensus on what this study does and does not provide by way of benefit to the patient. If that can be achieved there will have been one big step forward in providing informed consent for SNB in patients with melanoma.

Conflicts of interest

None declared.