Discordance of pathological thin melanoma thickness and T stage in SEER registry: impacts on clinical management and research directions

Background Ultrathin melanoma was previously demonstrated to have higher risk for melanoma-specific mortality using SEER database. However, these guideline-changing conclusions has been recently challenged by miscoding of thickness. This present study was performed to assess the prognosis of thin and ultrathin melanoma using only surgically-treated, pathologically confirmed and after removal of discordant cases. Methods Melanoma patients from SEER database who were initially diagnosed with histologically confirmed and surgically treated melanoma from 1998 to 2012 were included. Subjects with discordance between T stage and tumor thickness were excluded. Kaplan-Meier curves, log-rank test and multivariate Cox proportional hazards regression models were used. Results 55,754 patients met the strict inclusion criteria, but 16 (0.02%) and 803 (1.4%) patients were removed due to T0 stage and discordance between T stage and thickness, respectively. Therefore, 54,935 patients entered the analyses, among which 52,751 were LN negative and 2,184 were LN positive. In either overall or LN-negative patients, a straightforward dose-effect relationship of larger thickness with increasing mortality was observed. In contrast, in LN positive patients, the T1 subgroup demonstrated a similar survival with tumors in T2 mm subgroup. Multivariable analysis revealed same pattern, and significant interaction between T stage and LN involvement was found. Further categorizing T1 melanoma into 10 equal 0.10 mm increments demonstrated an unexpected “N”-shaped pattern of mortality in overall and LN negative ultrathin melanoma but not in LN positive melanoma. Conclusions No difference in mortality was observed in T1-3 tumors with LN involvement. External and independent validation studies are warranted.

(tumor thickness less than 1.00 mm) [3].Knowledge about disease patterns and outcomes of thin melanomas will lead to better clinical managements and benefit melanoma patients.
Large population-based cancer registries are useful tools for determining cancer outcomes.By systematically collecting, storing and reporting data on patients with specific cancer of interest, cancer registry data can provide insightful findings to guide clinical practice.Surveillance, Epidemiology and End Results (SEER) is one of the largest cancer registries worldwide and has been used frequently in thin melanoma researches [4].
Based on melanoma SEER data from 1998 to 2008, Sanjay et al [5] discovered that tumor thickness ≤0.50 mm was a marker for poor prognosis in the setting of positive lymph node (LN) metastasis.Even in thin melanoma without positive LN status, Shoshana et al [6] identified that the 10-year risk of death was higher for thin melanoma with tumor thickness 0.01-0.25 mm than those with tumor thickness from 0.26-0.50mm.Prognosis pattern worsened with tumor thickness only starting from 0.51 mm and this unexpected pattern could not be explained by ulceration.
However, all these potentially guideline-changing findings have been challenged by significant proportion of miscoding of melanoma thickness in SEER registry, especially in thin melanoma (tumor thickness ≤ 1.00 mm).After re-examination of tumor thickness in one SEER region using pathology reports, decimal point placement was found to be the most common error in thin melanoma.After correction, 96% of original ultrathin-related death and 100% of ultrathin related positive LN status was determined to be miscoded [7].
With such significant miscoding error in thin melanoma in the SEER database, the previous findings of thin melanoma could be artificial and misleading.In order to get the most reliable subset of melanoma SEER data, firstly, we established strict inclusion/exclusion criteria to only get surgically treated, pathological confirmed melanoma entries.Secondly, parameters coded as unknown (including missing data) are treated as a separate category in the present analyses instead of "negative findings", which was taken for granted and routinely performed by previous melanoma SEER studies.Thirdly, internal consistency checks were performed for every included melanoma entry and inconsistent entries were removed.
Our hypothesis was that loose inclusion/exclusion criteria, improper categorization of unknown parameters and data internal inconsistency might be greatly biased previous findings and mislead clinical practice.In the present study, based on the most reliable data possible, we tested whether thin and ultrathin melanoma still indicated poor prognosis, especially in different lymph node subgroups.

Baseline characteristics
55,754 patients met the strict inclusion criteria.16 patients were excluded for T0 stage (unknown primary site) and 803 patients were further excluded for discrepancy between database derived T stage and manual categorization of tumor thickness according to AJCC criteria.In total, 54,935 patients entered the analyses, among which 52,751 were LN negative and 2,184 were LN positive.
Baseline characteristics were shown in Table 1.The included patients were dominantly white and around half were male.The median follow-up time was 48 months.Categories of LN involvement significantly differed with respect to T stage, sex, age, race, marital status, ulceration, mitogenic status and number of LNs dissected.Patients with LN-positive disease tended to have thicker primary tumors and more LNs examined than those with LNnegative disease.Expectedly, no Tis patients had LN involvement.In patients with positive LN status, the rate of primary melanoma with ulceration and positive mitogenic status was higher.It is clinically reasonable to demonstrated significantly higher number of LN dissected in LN positive patients.There was no significant difference in year of diagnosis in patients with LN-negative or LN-positive disease.The rate of LN positive patients in unmarried patients' group (7%, 733/10,523) is 1.5 times higher than in married patients' group (5%, 1,365/27,376).

Impact of T stage on melanoma specific mortality under positive/negative LN status
Overall and separate crude Kaplan-Meier curves for LN positive and LN negative patients stratified by T stage are provided in Figure 1.In either overall or LNnegative patients, a positive correlation of larger thickness with increasing mortality was observed.In contrast, in LN positive patients, the T1 subgroup demonstrated a similar survival (median 43 months) with tumors in T2 (median 40 months) subgroup, and the median survival in T3 and T4 subgroups were 34 and 25 months, respectively.
In multivariable analyses controlling for age, sex, year of diagnosis, marital status, ulceration, mitogenic status, lactic dehydrogenase (LDH) elevation and number of LN dissected, there was significant interaction between T stage and LN involvement (P < 0.001) using the likelihood ratio test, which compared the models with and without interaction terms.Each interaction term was also statistically significant in Wald test (all P < 0.001).In addition, the potential confounders were all independently related to melanoma specific mortality.
Figure 2 showed graphically the multi-adjusted hazard ratio of melanoma specific mortality in different combinations of LN involvement and T stage using T1N0M0 as reference.LN-positive tumors demonstrated worse prognosis than LN-negative tumors generally.The slope of increase in HR is sharper in LN negative than in LN positive subgroup.

Impact of tumor thickness on melanoma specific mortality under positive/negative LN status in ultrathin melanoma
In order to account for potential residual confounding in the categorization of T1 stage, we further categorized T1 melanoma into 10 equal 0.10 mm increments (0.01-0.10 mm, 0.11-0.20 mm, …, and 0.91-1.00mm).Melanoma specific morality was then compared in ultrathin melanoma, namely Tis and T1 subgroups (Table 3, Figure 3).
Within T1, the distribution of tumor thickness was generally even.Among these ultrathin Tis and T1 subcategories, we did not find a similar pattern of worse prognosis with greater thickness.The thinnest Tis subgroup had the best survival compared to 0.01-0.10mm subgroup (HR 0.21, 95%CI 0.13-0.34,p < 0.001).However, an unexpected trend of decrease in HR was observed in the next two subgroups with a thickness of 0.11-0.30mm, and reached another bottom at 0.31-0.40mm and 0.41-0.50mm subgroups, which were significantly lower than the reference group.Subsequently, a gradual increase in HR was observed in the last 5 subgroups without statistically significant difference.A similar pattern was observed in LN negative melanomas.In LN positive melanomas, no obvious difference in hazard of melanoma specific mortality was observed with a limited number of subjects in each subgroup.The pattern could not be clearly explained by ulceration status (Supplementary Table 1).Re-categorization into 4 equal 0.25 mm increments revealed similar patterns (Supplementary Table 2).Dichotomizing T1 into 0.01-0.50mm and 0.51-1.00mm also did not show difference in mortality in LN positive patients.(Supplementary Table 3).The number of LN-negative melanoma patients was 52751 and the number of LN-positive patients was 2184.HR: hazard ratio; CI: confidence interval.* significant at 0.05 level.# significant at 0.025 level after Borferroni correction to adjust for multiple comparison.

DISCUSSION
Although the traditional view of malignant progression is that cancer gains metastatic ability through an accumulation of mutations as they grow to a large size, recent studies have suggested that for some tumors, the acquisition of metastatic potential may occur early in cancer development, even in the absence of detectable primary tumors.It was previously reported that very small tumor size with intensive lymph node involvement was related to worse survival in breast cancer [8], colon cancer [9] and prostate cancer [10].In melanoma, Sanjay et al [5] discovered that in LN positive melanomas, patients with tumor thickness ≤0.50 mm had higher mortality than those with tumor thickness 0.51-1.00mm or 1.01-2.00mm.More unexpectedly, Shoshana et al [6] recently identified that in LN negative melanoma, patients with tumor thickness 0.01-0.25 mm had higher mortality than those with tumor thickness 0.26-0.50mm.In our study, we did demonstrate the effect modification by LN involvement in the effect of T stage on melanoma specific mortality.In LN positive melanoma, T stage became a less important predictor of mortality than in the LN negative setting.However, in LN positive setting, we did not find a worse prognosis in patients with T1 melanoma than those with higher T stage.
Residual confounding remains an important issue when categorizing continuous variables into categorical variables.Both Sanjay et al [5] and Shoshana et al [6] further categorized T1 stage into smaller subgroups according to tumor thickness.Repeating their analyses yielded different results.More specifically, when dichotomizing T1 into 0.01-0.50mm and 0.51-1.00mm like what Sanjay et al [5] did, our data did not show a higher mortality in 0.01-0.50mm subgroup compared to 0.51-1.00mm and T2 subgroups.A potential explanation is the difference in study population and misclassification bias.Sanjay et al included 6,134 LN positive subjects in 1998-2008 database, while we only could include 2,184 LN positive subjects in 1998-2012 database based on our stricter inclusion criteria which only got surgically treated and pathological confirmed melanoma entries, minimizing misclassification.More intensive categorization into 4 or 10 categories did not identify difference in mortality in LN positive melanoma either.
Re-categorizing T1 into 4 categories mimicking Shoshana et al's method [6] showed similar results with Shoshana et al.However, Shoshana did not take Tis into account when looking the relationship of subgroups in T1 and mortality in LN negative melanoma.When adding the subgroup of Tis, the sharp increase in mortality from Tis to 0.01-0.25 mm made the results even harder to interpret.In view of the large sample size, we further categorized T1 into 10 categories.The analyses together with Tis subgroup confirmed the same pattern that the mortality increased sharply from Tis to 0.01-0.10mm, then decreased to bottom at 0.31-0.50mm, and then increased afterwards.Although some P values were statistically significant due to large sample size, the "N"-shaped pattern of mortality within a very small range of thickness was not biologically plausible.
Recently, Gimotty et al [7] reported significant proportion of miscoding of melanoma thickness in SEER registry, especially in thin melanoma (tumor thickness ≤1.00 mm).After re-examination of tumor thickness in one SEER region using pathology reports, decimal point placement was found to be the most common error in thin melanoma.After correction, 96% of original ultrathinrelated death and 100% of ultrathin related positive LN status were determined as miscoded [7].Similar coding errors in very small tumor size were proved in prostate cancer [11].Furthermore, the miscoding of tumor thickness was also observed in our analyses.Notably, 803 subjects were excluded because of discrepancy between database derived T stage and manual categorization of tumor thickness according to AJCC criteria.Among them, 119 patients were recorded to have a thickness of 0 mm while T1-T4.Contrarily, 485 patients recorded as Tis had a thickness of more than 0.01 mm, most of which (436, 90%) of which were in the 0.01-0.10mm category.
Although our internal consistency checks had removed some of the coding errors, we have reasons to believe that still considerable coding errors exist even in those concordant entries.As in the code for tumor thickness, "1" but not "0.01"representing 0.01 mm, miscoding was easier to occur in ultrathin melanoma.Based on our analyses and previous literatures, we believe coding error or errors in the source document should explain part of the unexpected pattern of mortality in ultrathin melanoma.Therefore, a more standardized process should be established for SEER data collection and training specialized personal for SEER data entry is highly recommended at each SEER data contributing set [12,13].In addition, it might be useful to

Figure 1 :
Figure 1: Kaplan Meier survival curves stratified by T stage.(A) Overall number of patients: 54,935, (B) lymph node (LN)negative number of patients: 52,751, and (C) LN -positive number of patients: 2,184 from the Surveillance, Epidemiology and End Results database were stratified into 5 categories based on AJCC T staging.

Figure 2 :
Figure 2: Adjusted hazard ratios for melanoma specific mortality in different combinations of LN involvement and T stage.Hazard ratios were adjusted for age, sex, race, year of diagnosis, marital status, number of LN dissected, ulceration, mitogenic status and LDH elevation.Left half: LN-negative melanoma, number of patients: 52,751; Right half: LN-positive melanoma, number of patients: 2,184.

Figure 3 :
Figure 3: Adjusted hazard ratios for melanoma specific mortality in different combinations of LN involvement and thickness in ultrathin melanoma.(A) Under negative LN involvement, including Tis and 10 equal 0.10 mm increments, using 0.01-0.10mm group as reference group; Number of patients: Tis+T1, 45,595 patients; (B) Under positive LN involvement, including 10 equal 0.10 mm increments, using 0.01-0.10mm group as reference group; Number of patients: T1, 296 patients.Hazard ratios were adjusted for age, sex, race, year of diagnosis, marital status, number of LN dissected, ulceration, mitogenic status and LDH elevation.

Table 3 : Impact of tumor thickness on melanoma specific mortality in ultrathin melanoma
HR: hazard ratio; CI: confidence interval.* significant at 0.05 level, reference group T1 0.01-0.10mm group.# significant at 0.025 level after Borferroni correction of 10 to adjust for multiple comparison, reference group T1 0.01-0.10mm group.