Paperonium

← Back to the plain-language case

Registered primary outcome is overall survival; the final report uses melanoma-specific survival without comment

N Engl J Med (2014) Ā· doi:10.1056/NEJMoa1310460 Ā· PMC4058881 Ā· NCT00275496

Summary

Every version of the ClinicalTrials.gov record for MSLT-I (NCT00275496), from the first on 2006-01-10 to the current one dated 2015-09-01, gives one primary outcome: whether sentinel-node biopsy with selective lymphadenectomy "will effectively prolong overall survival", with a 10-year time frame added in 2009. The final trial report states instead that "The primary end point was melanoma-specific survival (survival until death from melanoma)." The paper does not mention the registered outcome or explain the difference, and it reports no overall-survival estimate, hazard ratio or curve. The only all-cause mortality information in the paper is a set of crude death counts in the Fig. 1B flow diagram. The trial was registered about 12 years after it started, so the registry entry may not reflect the original protocol. Even so, the registered primary outcome had been public for eight years before publication and the paper does not reconcile the two.

What was registered

ClinicalTrials.gov has 10 versions of the record. All of them postdate the study start (1993-11) and the end of enrollment (2002). The primary outcome measure text is the same, character for character, in all 10:

To determine whether wide excision of the primary with intraoperative lymphatic mapping (LM) followed by selective lymphadenectomy will effectively prolong overall survival compared to wide excision of the primary melanoma alone.

Version Date Primary outcome What changed in Outcome Measures
v0 2006-01-10 (first submitted; first posted 2006-01-12) text above, no time frame first registration
v1 2006-01-20 unchanged secondary only: "Sentinel node metastases in WEX arm; Incidence of nodal metastases" became "Incidence of Sentinel Node Metastases (biopys) vs clinical metastases (observation)"
v2, v3 2006-11-29, 2007-11-02 unchanged none (Study Status, References)
v4 2009-04-30 unchanged; time frame "10 years" added "10 years" added to primary and secondary; "biopys" corrected to "biopsy"
v5, v6 2011-11-07, 2012-07-30 unchanged none (v6 records primary completion 2012-06, actual)
v7 2012-11-19 (last version before publication) unchanged, "10 years" none (Study Status, References)
v8, v9 2014-05-13, 2015-09-01 unchanged, "10 years" none (Study Status only)

The registered secondary outcomes in v4 to v9 are: "Disease-free survival; Incidence, timing, and anatomic distribution of distant metastases; Morbidity of procedures; Significance of TA90 levels; Incidence of Sentinel Node Metastases (biopsy) vs clinical metastases (observation); Accuracy of LM" (time frame "10 years"). Melanoma-specific survival does not appear as a primary or secondary outcome in any version.

Other registry dates: study start 1993-11; primary completion 2012-06 (actual); last update posted 2015-09-02. The current record lists this paper (PMID 24521106) as a "derived" citation and three "result" citations: the 2006 third interim report (N Engl J Med 2006 Sep 28, PMID 17005948); a 2010 Ann Surg Oncol paper on morbidity and length of stay after early versus delayed completion lymphadenectomy (PMID 20614193); and a 2012 Ann Surg Oncol paper on metastasectomy for distant metastases in MSLT-I patients (PMID 22648554). The two Ann Surg Oncol citations first appear in v7. The registry records masking as None (open label) in every version.

What the paper reports

The primary end point, in Methods, Statistical Analysis (first paragraph):

The primary end point was melanoma-specific survival (survival until death from melanoma). The secondary end points, described previously,11,12 included disease-free survival, survival with tumor-positive or tumor-negative sentinel nodes, and the incidence of sentinel-node metastases, as compared with the incidence of clinically detected nodal metastases.

References 11 and 12 are the 2006 third interim report in the same journal and a 1999 Ann Surg report on the accuracy of lymphatic mapping (PMC1420894). Neither is a protocol or a statistical analysis plan.

The registration appears once, at the end of the abstract:

(Funded by the National Cancer Institute, National Institutes of Health, and the Australia and New Zealand Melanoma Trials Group; ClinicalTrials.gov number, NCT00275496.)

The main results for the paper's own primary end point (abstract; Results, Survival Rates):

No significant treatment-related difference in the 10-year melanoma-specific survival rate was seen in the overall study population

hazard ratio for death from melanoma in the biopsy group, 0.84; 95% confidence interval [CI], 0.64 to 1.09; P = 0.18

This is for patients with intermediate-thickness melanomas (1.20 to 3.50 mm), the primary study group.

The analysis methods name only melanoma-specific and disease-free survival (Methods, Statistical Analysis, third and fourth paragraphs):

The Kaplan–Meier method was used to estimate mean 5-year and 10-year rates of melanoma-specific and disease-free survival.

Comparisons of melanoma-specific and disease-free survival were based on data from patients who underwent their assigned treatment.

The Discussion (fourth paragraph) refers to survival without saying which kind:

We did not see a significant survival advantage for all patients with intermediate-thickness melanomas.

The design changes the paper does disclose are a larger sample size ("After the second interim analysis, the sample size for the intermediate-thickness group was increased to 1200 patients") and the latent-subgroup method ("In post hoc analyses, a latent-subgroup analytic method was developed"). Both are in Methods, Statistical Analysis.

What the paper does not contain: searches of the full text as archived in PMC (abstract, body, table, figure legends) for "overall survival", "all-cause", "any cause" and "amend" find nothing. Every use of "protocol" refers to the per-protocol analysis. The one use of "registered" refers to enrollment ("Enrollment closed in 2002, after 2001 patients had been registered."). The Supplementary Appendix archived with the PMC manuscript (16 pages) has no overall-survival analysis and no statement about end points or protocol amendments. Its survival analyses are all melanoma-specific, disease-free or distant disease-free.

All quotations above were checked verbatim with src/show.py PMC4058881 --grep, and each full quotation occurs exactly once in the archived text. The Fig. 3 legend repeats the definition "survival until death from melanoma". Registry quotations were checked against the archived version records and the current API v2 record.

The discrepancy

Outcome Registered (version, date) In the paper Status
Overall survival (death from any cause), 10 years Primary in every version, v0 2006-01-10 to v9 2015-09-01 (time frame "10 years" from v4 2009-04-30; v7 2012-11-19 is the last version before publication) Not named as an end point. No estimate, hazard ratio or curve. Fig. 1B gives only crude counts of deaths by cause not reported
Melanoma-specific survival, 10 years In no version, as either primary or secondary "The primary end point was melanoma-specific survival" added (as primary)
Time frame of the primary outcome 10 years (from v4 2009-04-30) 10-year rates matched
Disease-free survival, 10 years Secondary, all versions (time frame "10 years" from v4 2009-04-30) Secondary. The abstract's conclusions cite it first as a treatment benefit matched
Incidence of sentinel-node metastases vs clinical metastases Secondary, v1 onward Secondary ("the incidence of sentinel-node metastases, as compared with the incidence of clinically detected nodal metastases") matched
Distant metastases (incidence, timing, distribution) Secondary, all versions Reported as distant disease-free survival in the node-positive subgroup (Results; Fig. S2) matched (as distant disease-free survival)
Survival with tumor-positive or tumor-negative sentinel nodes In no version Secondary added (as secondary)

The core finding is two rows: the registered primary outcome (overall survival) is not reported, and a different measure (melanoma-specific survival) is reported as the primary end point. Neither the substitution nor the reason for it is stated in the paper.

Figures

Timeline of MSLT-I registration, registry versions and publications

Figure 1. Timeline of the trial and its registry record. The trial started in 1993-11 (registry; the paper says it "commenced in 1994") and enrolled 2001 patients until 2002. It was first registered on 2006-01-10, about 12 years after the start. Every registry version therefore postdates the study start and the end of enrollment. Diamonds mark the versions that created or edited the Outcome Measures module: v0 (primary outcome overall survival), v1 (secondary wording only) and v4 (time frame "10 years" added to the unchanged primary). Open circles mark versions that changed other modules. The hatched band, bounded by the dashed line, is the period after primary completion (2012-06). The four versions in it did not touch outcomes: v6 (2012-07-30, Study Status; status set to Completed and primary completion 2012-06 recorded as actual), v7 (2012-11-19, Study Status and References; the last version before publication), and v8 (2014-05-13) and v9 (2015-09-01), Study Status only. The final report (2014-02-13) names melanoma-specific survival as its primary end point. Sources: ClinicalTrials.gov record history; the paper as archived in PMC; the registry's reference to the 2006 interim report.

Registered versus reported primary outcome

Figure 2. Registered primary outcome versus the primary end point in the paper. Left, in blue: the registry's primary outcome as of v7 (2012-11-19), the last version before publication, worded the same since v0 (2006-01-10). Right, in orange: the paper's stated primary end point. The two use the same 10-year time frame but measure different things: overall survival counts deaths from any cause, while melanoma-specific survival counts only deaths from melanoma. The paper gives no overall-survival result. The nearest data are the crude counts of deaths by cause in the Fig. 1B flow diagram, which are not time-to-event estimates and do not account for losses to follow-up. Melanoma-specific survival appears in no registry version. Its result in the primary study group is null (hazard ratio 0.84, 95% CI 0.64 to 1.09, P = 0.18). Disease-free survival, a registered secondary outcome, is reported as a secondary end point.

Why it matters

Overall survival and melanoma-specific survival differ in more than wording. Melanoma-specific survival treats deaths from other causes as censored, so a treatment effect on non-melanoma mortality cannot appear in it. It also depends on cause-of-death attribution, which may not be blinded in an open surgical trial (the registry records masking: None in every version). For a comparison of an extra surgical procedure (sentinel-node biopsy, with immediate completion lymphadenectomy when positive) against observation, all-cause mortality is the outcome the registry named, and it captures any harm or benefit regardless of cause.

What the paper reports for the registered primary outcome: no estimate. No overall-survival rate, hazard ratio, P value or Kaplan–Meier curve appears in the abstract, main text, Table 1, figure legends or the archived Supplementary Appendix. The only related information is in the Fig. 1B flow diagram (image), which lists deaths by cause among patients included in follow-up. For intermediate-thickness melanomas: biopsy group 133 died from melanoma and 47 from other causes, 180 of 805 in all (22.4%); observation group 103 and 25, 128 of 522 (24.5%). For thick melanomas: biopsy 68 and 19, 87 of 185 (47.0%); observation 43 and 9, 52 of 126 (41.3%). These are crude proportions with unequal losses to follow-up (120 and 68 in the intermediate groups; 18 and 9 in the thick groups). They are not a substitute for the registered time-to-event analysis, and this write-up draws no inference about treatment effect from them. They show that the paper did record deaths from other causes, so an overall-survival analysis was possible from the trial data.

The abstract's positive conclusions rest on outcomes other than the registered primary: disease-free survival (a registered secondary), melanoma-specific and distant disease-free survival in the node-positive subgroup (a comparison the paper itself says raises "concern about ascertainment bias"), and a post hoc latent-subgroup analysis. The paper's own primary end point is stated openly as null in the whole primary study group. A reader comparing the paper with the registry cannot tell from the paper whether the registered overall-survival comparison was done or what it showed.

Explanations considered

Status

As of 2026-10-23:

Suggested correction

The authors and journal could add a correction or a note to the article that:

  1. States that the ClinicalTrials.gov record (NCT00275496, first registered 2006-01-10) lists overall survival at 10 years as the primary outcome, and that the paper's primary end point is melanoma-specific survival.
  2. Says when and why the primary end point was set as melanoma-specific survival. If the original pre-2006 protocol named melanoma-specific survival, the correction could say so and make that protocol available.
  3. Reports the registered primary outcome: 10-year overall survival by randomized group in the intermediate-thickness (primary study) group, with the Kaplan–Meier estimates, hazard ratio, 95% CI and P value, analyzed the same way as the melanoma-specific survival results (and by intention to treat).
  4. If the registry entry does not reflect the protocol, the sponsor could update ClinicalTrials.gov with a dated note explaining the difference, so the record and the publication agree.