Inconsistencies in Cutaneous Melanoma Staging Within SEER Registries

Abstract

This cohort study explores inconsistences in cutaneous melanoma staging within the Surveillance, Epidemiology, and End Results Program registries to determine if miscoding of staging variables related to cutaneous melanoma would result in inconsistencies in final tumor staging.

Public cancer databases, such as the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) Program, are increasingly used for cutaneous oncology research.¹ However, coding discrepancies among staging variables in SEER data may inaccurately estimate patient outcomes.² The objective of this cohort study was to explore inconsistences in cutaneous melanoma (CM) staging within the 18 population-based SEER registries for cancer incidence to determine if miscoding of staging variables related to CM would result in inconsistencies in final tumor staging.

Methods

The University Hospitals Institutional Review Board determined this study to be exempt from review and waived the requirement for informed consent owing to use of deidentified data. We identified patients in the SEER registries who were diagnosed with CM from 2004 through 2017, excluding cases diagnosed at autopsy. For each case, we recorded numeric stage at diagnosis and TNM (tumor, node, metastasis) components provided by SEER based on the predominant staging system during the year of diagnosis. We compared the TNM components to 7 supplementary staging variables within the SEER registries, assessing for consistency as follows: (1) T stage: Breslow depth, ulceration, and mitoses; (2) N stage: regional node positivity and clinical lymph node status; and (3) M stage: distant metastasis. The seventh Combined Summary Stage variable allowed for comparison with all 3 TNM components. Coding discrepancies between reported TNM stage and at least 1 of the 7 supplementary staging variables were recorded for each CM case (eTable in the Supplement). Cases with a stage-changing discrepancy (ie, resulting in stage increase or stage decrease from the official tumor stage) were defined as inconsistent. The primary outcome was proportion of inconsistent CM cases per year. We also recorded cases without a reported tumor stage. To assess trends in total cases and proportion of inconsistent cases, we performed Mann-Kendall and Cochrane-Armitage tests, as appropriate. Analyses were performed in SAS, version 9.4 (SAS Institute), and a 2-sided P < .05 was considered statistically significant.

Results

Among the 246 047 CMs diagnosed from 2004 through 2017, there were 6383 (2.6%) inconsistent cases stemming from 8309 coding discrepancies (Table). There were 27 155 additional cases without tumor stage provided. Most coding discrepancies (n = 7846; 94.4%) were stage changing, of which 3533 (45.0%) resulted in tumor stage increase and 4313 (55.0%) resulted in tumor stage decrease. There has been an increase in proportion of inconsistent cases over time (Kendall τ, 0.67; 95% CI, 0.54-0.77; P < .001), with a corresponding increase in CM incidence (Figure).

Table. Frequency and Source of Discrepancies in Tumor Stage Among Primary Cutaneous Melanomas Within SEER Registries, 2004-2017.

Years of diagnosis	No. of discrepancies for each supplementary staging variable							Total discrepancies by resulting stage changea
	Breslow depth (SSF1)b	Ulceration status (SSF2)b	Presence of mitoses (SSF7)b	Regional nodes positivec	Clinical lymph node status (SSF3)c	Other distant metastasisd	SEER combined summary stage 2000e	Not stage changing	Stage increase	Stage decrease
Introduction of AJCC 6th edition staging to SEER
2004	0	0	NA	20	4	0	129	13	118	22
2005	2	0	NA	12	7	0	107	12	101	15
2006	2	0	NA	12	8	0	111	12	110	11
2007	3	0	NA	15	6	0	106	9	105	16
2008	0	0	NA	25	13	0	111	16	109	24
2009	7	0	NA	25	22	0	115	30	109	30
Introduction of AJCC 7th edition staging to SEER
2010	29	356	355	20	28	0	118	23	121	762
2011	108	262	261	35	39	0	133	25	156	657
2012	134	217	217	60	91	0	243	51	304	607
2013	175	203	209	58	104	0	218	70	287	610
2014	131	170	175	61	113	0	240	103	282	505
2015	129	151	152	88	119	0	224	99	265	499
Introduction of UICC 7th edition staging to SEER
2016	0	105	713	72	38	3	202	0	811	322
2017	0	62	532	70	58	3	163	0	655	233
Totals	720	1526	2614	573	650	6	2220	463	3533	4313

Abbreviations: AJCC, American Joint Committee on Cancer; NA, not applicable; SEER, Surveillance, Epidemiology, and End Results Program; SSF, site-specific factor; UICC, Union for International Cancer Control.

^aStage-changing discrepancies are those that result in an increase or decrease to the tumor stage listed (eg, stage IIIB T4aN1bM0 tumor with site-specific factor 3 coded as “nodes negative on pathologic examination” would imply decrease to stage II).

^bMismatch with T (tumor) component of TNM cancer staging system.

^cMismatch with N (node) component of TNM cancer staging system.

^dMismatch with M (metastasis) component of TNM cancer staging system.

^eThe SEER Combined Summary Stage 2000 variable allowed for comparison of all 3 TNM components because it incorporates distant nodes/sites involved, regional by nodes involved, regional by extension only, regional by nodes involved and extension, regional not otherwise specified, and localized only.

Figure. Annual Trends in Incidence of Cutaneous Melanoma and Inconsistencies in Tumor Stage Coding.

AJCC indicates the American Joint Committee on Cancer; UICC, the Union for International Cancer Control.

Discussion

We identified a subset of CM cases within the SEER registries that had inconsistent staging data, and the majority of coding discrepancies resulted in either an increase or decrease from the American Joint Committee on Cancer or Union for International Cancer Control tumor stage at diagnosis, as provided by SEER.^3,4 We observed an increase in discrepancies in 2010, which corresponds with the introduction of the American Joint Committee on Cancer’s AJCC Cancer Staging Manual, 7th edition, and the variable for histologic mitoses. The increase in discrepancies related to ulceration status and mitoses may be explained by data coders being unfamiliar with the requirement of obtaining tumor mitoses data from the pathology report, or by clinicians reporting a T stage (eg, T1b) without specifying criteria for ulceration/mitoses that justify the stage reported. Regardless, the observed rate of inconsistent cases (2.6%) appears low for a database as large as SEER, and increasing and decreasing of tumor stage may counterbalance when aggregated in a study. The present study is limited by the inability to distinguish whether the TNM component or the supplementary staging variable is inaccurate when considering coding discrepancies.

Producing high-quality research requires checking data quality and completeness. Our hope is that the current findings will raise awareness among CM researchers regarding inconsistent coding of staging variables within SEER registries, which may inaccurately estimate outcomes. For example, analysis of survival outcomes for patients with stage T1b CMs receiving sentinel lymph node biopsy would need to account for cases with discrepancies between listed T1b stage and the variables for Breslow depth, ulceration, and mitoses (depending on year). Correcting these discrepancies requires careful consideration because removing all inconsistently coded cases could result in selection bias or reduced power, depending on one’s outcome of interest. The statistical literature provides advice for handling inconsistent data and mitigating biases that may arise from inappropriately accounting for inconsistent or missing staging data.⁵

Supplement.

eTable. Examples of coding discrepancies between reported TNM components and SEER supplementary staging variables for cutaneous melanoma