Skip to main content
NCBI home page
JAMA Network logoLink to JAMA Network
. 2019 Jun 26;155(8):890–898. doi: 10.1001/jamadermatol.2019.0621

Development and Initial Validation of Calculated Tumor Area as a Prognostic Tool in Cutaneous Malignant Melanoma

Gerald Saldanha 1,✉, Jeremy Yarrow 1, Somaia Elsheikh 2, Marie O’Riordan 3, Hussein Uraiby 3, Mark Bamford 3
PMCID: PMC6596334  PMID: 31241720

Key Points

Question

What is the prognostic value of calculated tumor area—a novel, simple microscopic feature and a 2-dimensional surrogate for tumor size—and is it a possible upgrade for the 1-dimensional criterion standard measurement, Breslow thickness?

Findings

In this cohort study of 1239 patients with melanoma, calculated tumor area had independent prognostic value with greater relative importance than Breslow thickness. Stratification of melanoma into groups based on calculated tumor area had better prognostic value than T category, which is based on Breslow thickness.

Meaning

These findings suggest that calculated tumor area should be prioritized for investigation to verify its prognostic value and to assess its applicability and acceptability in routine practice.

This cohort study investigates the precision and prognostic value of calculated tumor area among patients with cutaneous malignant melanoma.

Abstract

Importance

Breslow thickness is a 1-dimensional surrogate prognostic feature for tumor size, yet tissue sections have 2 dimensions. Therefore, a 2-dimensional feature, calculated tumor area (CTA), was devised.

Objective

To determine CTA precision and prognostic value.

Design, Setting, and Participants

This retrospective cohort of patients with cutaneous melanoma presented to the Leicester and Nottingham National Health Service hospital trusts in the United Kingdom. Eligible patients in the Leicester development sample had available primary tumor tissue; a diagnosis from January 1, 2004, through December 31, 2011; invasive disease; and Leicestershire residency. Patients in the Nottingham validation sample had an anonymized spreadsheet with primary melanoma diagnosed from January 1, 2003, through December 31, 2005, or from January 1, 2008, through December 31, 2010. From a starting population of 1463 patients in both data sets, a total of 224 (15.3%) were excluded to yield a study population of 1239. Data were analyzed from April 30, 2018, through January 10, 2019.

Intervention

An observational analysis of the prognostic value of CTA in patients with cutaneous melanoma.

Main Outcomes and Measures

Independent association of CTA with melanoma-specific survival and confounding effect of CTA on Breslow thickness in survival analysis.

Results

A total of 1239 patients with melanoma were assessed, including 649 (52.4%) women, with a median age of 60 years (interquartile range, 47-71 years). An intraclass correlation coefficient for CTA on 13 cases was 0.99. In 918 patients in the Leicester cohort, CTA was an independent prognostic factor in Cox proportional hazards regression models after adjusting for Breslow thickness, age, sex, ulcer, mitotic rate, and microsatellites (hazard ratio [HR], 1.87; 95% CI, 1.49-2.34; P < .001). Validation in 321 patients in the Nottingham cohort showed an HR of 1.55 (95% CI, 1.15-2.09; P = .005) and in the combined 1239 cases, an HR of 1.70 (95% CI, 1.43-2.03; P < .001). Breslow thickness was significant in multivariable analysis only when CTA was not in the model. The relative importance of CTA was shown by its retention in all 100 bootstrap multivariable models with backward selection, whereas Breslow thickness was retained in only 53. Melanomas stratified by CTA showed wider separation of survival curves than those stratified by Breslow thickness using the American Joint Committee on Cancer Staging Manual, 8th Edition (HRs, 1.00 to 41.46 vs 1.00 to 36.95, respectively), and the model with CTA categories had a Bayesian information criterion difference of 13.9 compared with T category, indicating substantially better fit. This model had a Harrell C index of 83.7%, and bootstrap analysis showed little evidence of model optimism, with a corrected calibration slope of 0.99.

Conclusions and Relevance

This study provides a novel microscopic feature, CTA, with evidence of its independent prognostic value. This evidence suggests that CTA should be a priority for further study.

Introduction

Cutaneous melanoma is a skin cancer1,2 with a poor prognosis when disease is advanced, making it important that patients with primary melanoma are accurately stratified for risk of progression. The criterion standard for prognostic stratification is the American Joint Committee on Cancer Staging Manual, 8th Edition (AJCC8), staging. The T category is determined by Breslow thickness and ulceration, with Breslow thickness being the cornerstone since documentation in 19703 and subsequent validation.4,5 Its prognostic value is likely to be associated with its surrogacy for tumor size, which correlates with risk of metastasis, as exemplified by breast cancer.6 At present, active pursuit of molecular prognostic biomarkers such as the DecisionDx-Melanoma test (Castle Biosciences, Inc)7,8 have the advantage of a rational basis founded on tumor biology, but these tests can be expensive and require specialist equipment or laborious protocols. Therefore, we focused on microscopic biomarkers requiring no more than the hematoxylin-eosin (H&E)–stained sections used for diagnosis.

Breslow thickness is measured in 1 dimension, but tissue sections have 2 dimensions, a fact that we sought to exploit. Rashed et al9 and Saldanha et al10 devised a feature called Breslow density that, in combination with Breslow thickness, created a composite feature called the targeted burden score (TBS). We have now devised a newer feature, the calculated tumor area (CTA), which yields an area in square millimeters of invasive melanoma cells on the same microscopic section used for measurement of Breslow thickness. Calculated tumor area was designed for speed and simplicity so that any pathologist could measure it prospectively during routine practice. We hypothesize that CTA improves on Breslow density and TBS because Breslow density measurement is limited to a 2-mm-wide window spanning the position of Breslow thickness measurement, whereas CTA assesses invasive cells across the entirety of the same slide. The purpose of the present study was to investigate CTA’s precision and prognostic value after adjustment for confounding and compare it with TBS and Breslow thickness.

Methods

Patients

Tissue samples were identified from 1104 patients with primary melanoma diagnosed at University Hospitals of Leicester and 359 patients at Nottingham University Hospitals in the United Kingdom. In the Leicester development sample, eligibility criteria consisted of primary tumor tissue availability; diagnosis from January 1, 2004, through December 31, 2011; the presence of invasive disease; and residency in Leicestershire. Samples were assessed sequentially by pathologic accession. For patients with multiple primary melanomas, the highest-stage melanoma or, if the same stage, the first diagnosed melanoma was included. Within the Nottingham validation sample, data were in an anonymized spreadsheet for patients diagnosed from January 1, 2003, through December 31, 2005, or from January 1, 2008, through December 31, 2010. Sufficient data were provided to determine stage and melanoma-specific survival (MSS). From 1463 patients in both data sets, 224 (15.3%) were excluded, yielding 1239 patients. Excluded cases are described in eFigure 1 in the Supplement. Tissue use was subject to National Health Service research ethics committee approval. The need for informed consent was waived because the study required access only to archived diagnostic tissue and associated clinical data, the study had no direct patient involvement, and data provided to researchers was in coded form with no patient identifying information. This study adhered to Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) guidelines11 (eTable 1 and eFigure 2 in the Supplement).

CTA Biomarker and Scoring

Calculated tumor area represents an estimated cross-sectional area (in square millimeters) of invasive melanoma cells in a transverse section containing the full tumor breadth measured on the same section used for determining Breslow thickness. Typically, the lesion’s breadth was present in a single transverse section. When the transverse section required division into multiple pieces to fit into a tissue block, cases were used if CTA calculation could be reconstructed from the separate pieces. If invasive melanoma cells were present at the peripheral surgical margin, CTA could not be estimated except when a partial biopsy and subsequent full resection permitted CTA measurement from both samples. Measurement of CTA entailed creation of a virtual box of arbitrary size containing all invasive melanoma cells on the section. The box dimension was estimated from microscope objective lens diameters or fractions thereof (Figure 1A). Whether the box extended beyond the limits of the tissue section was irrelevant (Figure 1B). The proportion of the entire virtual box (not just the dermis) occupied by invasive melanoma cells was estimated, and CTA value was calculated from the box height × box width × proportion divided by 100, which yielded an estimate in square millimeters.

Figure 1. Calculated Tumor Area (CTA) Scoring Method.

Figure 1.

Open in a new tab

Two stylized examples of melanoma CTA scoring are shown. A, The box dimension was estimated from microscope objective lens diameters or fractions thereof. B, The box is extended beyond the limits of the tissue section. The gray dashed lines show the CTA box, and the gray dashed circles show objective lens fields used for box measurement.

Statistical Analysis

Data were analyzed from April 30, 2018, through January 10, 2019. Statistical analyses were performed in R, version 3.3.112 and RStudio, version 1.0.143.13 P < .05 was considered significant, and tests were 2-tailed. Interobserver agreement was assessed using intraclass correlation coefficients for multiple raters14 using the Irr package.15 Baseline statistics for numerical variables consisted of median and interquartile range (IQR) with counts and percentages for categorical variables. Analysis of variables against 3 CTA categories used the Kruskal-Wallis test or χ2 test. Comparison of the Leicester and Nottingham cohorts used the Mann-Whitney test or χ2 test with continuity correction. For survival analysis, the outcome was MSS. An event was death due to melanoma, with censoring if death was due to another cause or if the patient remained alive. The Survival16 and Survminer packages17 were used for Kaplan-Meier plots and Cox proportional hazards regression models. The proportionality assumption was checked with plots of scaled Schoenfeld residuals against transformed time and a goodness-of-fit test. Proportionality was not violated.

Modeling Strategy

Split-sample validation was based on hospital site (Leicester for development and Nottingham for validation). This internal-external validation approach is reported to be superior to random splitting.18 The development data set was used for model specification, entailing investigation of CTA transformations for entry into Cox proportional hazards regression models and to assess CTA confounding. Findings were validated using the Nottingham cohort. Final model coefficients were estimated after merging the data sets to maximize power and minimize overfitting. The prognostic value of CTA was compared with that of TBS by comparing univariable Bayesian information criterion (BIC) and by entry together in a multivariable Cox proportional hazards regression model. Variable relative importance was assessed by comparing χ2 test minus degrees of freedom in a multivariable Cox proportional hazards regression model; by assessing the frequency of variable retention after backward stepwise selection applied to each multivariable Cox proportional hazards regression model in 100 bootstrap samples; and by comparing hazard ratios (HRs) of variables after transformation to the same scale of measurement. Classification of melanomas by using T category was compared with classification by CTA strata by inspecting Kaplan-Meier plots, HRs, and cross-tabulation. A bootstrap approach was used to estimate external validity of the Cox proportional hazards regression model using CTA strata, an efficient way to minimize model overfitting or optimism,18,19 taking 100 bootstrap samples to estimate a coefficient shrinkage factor. The rms package20 was used for bootstrap analysis and to check variable importance, and BIC was used to compare models.21

Analyses were prespecified to step 10 in eFigure 2 in the Supplement. Once CTA’s importance emerged, we proceeded to complete steps 10 and 11.

Results

CTA Scoring and Patient Characteristics

Two observers (G.S. and J.Y.) independently scored 13 primary melanoma samples. The intraclass correlation coefficient was 0.99, indicating almost perfect agreement (eFigure 3 in the Supplement). Photomicrographs of CTAs are shown in eFigure 4 in the Supplement. Characteristics of the 1239 patients (590 men [47.6%] and 649 women [52.4%]; median age, 60 years [IQR, 47-71 years]) are shown in Table 1. Median Breslow thickness was 0.9 mm (IQR, 0.5-2.0 mm), reflecting generally thin cases in an incident UK population. The Nottingham cohort was categorized by mitotic rate; therefore, for uniformity, the Leicester cohort was also categorized (absent, 0 mitoses/mm2; nonbrisk, 1-3 mitoses/mm2; and brisk, ≥4 mitoses/mm2). No information was provided on the cut points for the Nottingham cohort. Calculated tumor area was significantly associated with covariables. Patient characteristics for the Leicester and Nottingham data sets are shown in eTable 2 in the Supplement. A significant difference in mitotic rate was probably related to the arbitrary way that nonbrisk and brisk groups were distinguished at each hospital. The Nottingham cohort had longer follow-up (median, 143.0 [IQR, 92.0-154.0] vs 71.5 [IQR, 61.0-84.0] months; P < .001) because most patients were diagnosed from 2003 through 2004 compared with more uniform distribution of cases from 2004 through 2011 for the Leicester cohort. The Nottingham cohort was significantly younger (median age, 56 [IQR, 43-69] vs 61 [IQR, 48-72] years; P = .002).

Table 1. Baseline Features of Study Patientsa.

Feature All (n = 1239) Patients Stratified by CTA P Value
Low (n = 406) Medium (n = 412) High (n = 421)
CTA, median (IQR), mm2 1.3 (0.2-6.4) NA NA NA NA
Male sex 590 (47.6) 197 (48.5) 195 (47.3) 198 (47.0) .90
Age, median (IQR), y 60.0 (47.0-71.0) 54.5 (41.0-66.8) 58.00 (45.8-68.0) 66.00 (55.0-76.0) <.001
Breslow thickness, median (IQR), mm 0.9 (0.5-2.0) 0.5 (0.4-0.6) 0.9 (0.7-1.2) 2.8 (1.8-4.8) <.001
Ulcer present 198 (16.0) 4 (1.0) 21 (5.1) 173 (41.1) <.001
Mitosis category
Absent 505 (40.8) 356 (87.7) 141 (34.2) 8 (1.9) <.001
Nonbrisk 504 (40.7) 48 (11.8) 246 (59.7) 210 (49.9)
Brisk 230 (18.6) 2 (0.5) 25 (6.1) 203 (48.2)
Microsatellites present 23 (1.9) 0 3 (0.7) 20 (4.8) <.001
Follow-up time, median (IQR), mo 77 (63-113) 79 (67-128) 80.50 (65.75-125.25) 71.00 (40.00-90.00) <.001
AJCC8 category
IA 675 (54.5) 395 (97.3) 266 (64.6) 14 (3.3) <.001
IB 237 (19.1) 8 (2.0) 125 (30.3) 104 (24.7)
IIA 124 (10.0) 3 (0.7) 13 (3.2) 108 (25.7)
IIB 94 (7.6) 0 3 (0.7) 91 (21.6)
IIC 86 (6.9) 0 2 (0.5) 84 (20.0)
III 23 (1.9) 0 3 (0.7) 20 (4.8)
Open in a new tab

Abbreviations: AJCC8, American Joint Committee on Cancer, version 8; CTA, calculated tumor area; IQR, interquartile range; NA, not applicable.

a

Low indicates less than 0.35 mm2; medium, 0.35 to 3.48 mm2; and high, greater than 3.48 mm2. Unless otherwise indicated, data are expressed as number (percentage) of patients.

Assessment of CTA’s Prognostic Validity

Using discovery cases, the optimal way to specify CTA was assessed by entry into a Cox proportional hazards regression model with common transformations. Models were compared by using BIC (eTable 3 in the Supplement). A change in BIC of greater than 10 indicated very strong evidence against the model with higher BIC value.21 A natural logarithm CTA (ln-CTA) had the lowest BIC (95.0 lower than the next transformation), providing very strong evidence that this factor should be entered into statistical models.

We assessed whether CTA was a valid prognostic feature after adjusting for confounding using multivariable Cox proportional hazards regression by entering ln-CTA with sex, age, Breslow thickness, ulcer, mitotic count, and microscopic satellites (Table 2, with extended data given in eTable 4 in the Supplement). Calculated tumor area had independent prognostic value using the Leicester data set (HR, 1.87; 95% CI, 1.49-2.34; P < .001). Breslow thickness was not statistically significant with CTA in the model, but when it was removed, Breslow thickness became statistically significant (HR, 1.04; 95% CI, 1.04-1.11; P = .001), revealing that Breslow thickness was substantially confounded by CTA. We confirmed CTA’s prognostic value using the Nottingham data set (HR, 1.55; 95% CI, 1.15-2.09; P = .005). With merged data sets, CTA had an adjusted HR of 1.70 (95% CI, 1.43-2.03; P < .001). Breslow thickness remained nonsignificant.

Table 2. Prognostic Value of CTA With Adjustment for Confounding.

Variable Data Set
Leicester (n = 918) Nottingham (n = 321) Merged (n = 1239)
HR (95% CI) P Value HR (95% CI) P Value HR (95% CI) P Value
ln-CTA 1.87 (1.49-2.34) <.001 1.55 (1.15-2.09) .005 1.70 (1.43-2.03) <.001
Sex
Male 1.25 (0.86-1.80) .24 1.91 (1.01-3.61) .047 1.42 (1.04-1.95) .03
Female 1 [Reference] NA 1 [Reference] NA 1 [Reference] NA
Age 1.01 (1.00-1.02) .17 1.02 (1.00-1.04) .047 1.01 (1.00-1.02) .02
Breslow thickness 0.95 (0.89-1.02) .15 0.95 (0.83-1.09) .48 0.96 (0.91-1.02) .23
Ulcer
Absent 1 [Reference] NA 1 [Reference] NA 1 [Reference] NA
Present 1.75 (1.13-2.72) .01 1.59 (0.69-3.64) .27 1.59 (1.08-2.32) .02
Mitotic count
Absent 1 [Reference] NA 1 [Reference] NA 1 [Reference] NA
Nonbrisk 0.89 (0.37-2.12) .79 1.11 (0.35-3.47) .86 1.05 (0.53-2.09) .89
Brisk 1.23 (0.48-3.18) .66 1.19 (0.29-4.94) .80 1.39 (0.64-3.04) .41
Microsatellites
Absent 1 [Reference] NA 1 [Reference] NA 1 [Reference] NA
Present 1.29 (0.65-2.56) .46 2.39 (0.95-6.04) .06 1.73 (1.00-2.98) .049
Open in a new tab

Abbreviations: CTA, calculated tumor area; HR, hazard ratio; ln-CTA, natural logarithm CTA; NA, not applicable.

We assessed whether CTA was an improvement on the microscopic features described previously, TBS and Breslow density,9,22 which are cross-tabulated in eTable 5 in the Supplement. Targeted burden score was entered into Cox proportional hazards regression models with a natural logarithm transformation, as done previously.22 The BIC of a CTA-only model was 2054.3; BIC of TBS only, 2061.1 (difference, 6.8), showing strong evidence for the CTA model.21 In addition, a multivariable model was created in which both CTA and TBS were entered. Calculated tumor area had an HR of 1.51 (95% CI, 1.18-1.94; P = .001), whereas TBS had an HR of 1.44 (95% CI, 0.98-2.11; P = .07). Overall, these data provide evidence that CTA is superior to TBS. A Cox proportional hazards regression model of CTA and Breslow density was also fitted, and Breslow density was not significant (eTable 6 in the Supplement).

Comparison of CTA and Breslow Thickness as Prognostic Features

These data suggest that CTA may be a better estimate of dermal invasive cells than Breslow thickness. We therefore sought to further compare them. First, we looked at estimates of relative importance in the Cox proportional hazards regression model (Figure 2A), with CTA being substantially more important than other recognized prognostic factors (importance, 33.98). The same model was fitted without CTA (Figure 2B). The confounding effect of CTA on Breslow thickness is seen (with ln-CTA, 0.42; without ln-CTA, 22.53). As a second check on relative variable importance, we took 100 bootstrap samples from our 1239 cases and, for each sample, performed backward stepwise variable selection. Variable retention per bootstrap sample was used to indicate relative importance. Calculated tumor area was retained in all 100 bootstrap sample models, whereas Breslow thickness was retained in only 53, sex in 85, ulcer in 82, age in 81, microsatellitosis in 63, and mitotic rate in 41 (Figure 2C). This finding supports CTA as the most influential of these features. To compare the magnitude of association for CTA and Breslow thickness with MSS, they were normalized to the same scale by dividing each by its IQR, with entry together into a Cox proportional hazards regression model (the ln-CTA transformation was used). Thus, the HR represents a change from a typical low value (25th percentile) to a typical high value (75th percentile), making HRs meaningful and comparable.23 Calculated tumor area had an HR of 12.95 (95% CI, 7.93-21.14; P < .001), whereas Breslow thickness had an HR of 0.95 (95% CI, 0.87-1.03; P = .18). Overall, these findings suggest that CTA captures the variability within melanoma outcome better than Breslow thickness.

Figure 2. Relative Importance of Calculated Tumor Area (CTA) in a Cox Proportional Hazards Regression Model.

Figure 2.

Open in a new tab

Variable relative importance was measured by χ2 value minus degrees of freedom for a full model (A) and without natural logarithm CTA (ln-CTA) (B). Among retained variables after backward stepwise selection in each of 100 bootstrap sample-derived models (C), the variables of interest, CTA and Breslow thickness (BT), are highlighted in orange in panels A and B. MR indicates mitotic rate.

Comparison of Prognostic Strata Defined by CTA and AJCC8 T Category

We investigated how well primary melanoma prognostic groups defined by CTA strata compared with standard AJCC8 T-category groups defined by Breslow thickness. First, we inspected Kaplan-Meier plots of ln-CTA deciles to identify natural cut points (Figure 3A). We identified deciles 1 to 6 as low risk (CTA ≤2.4 mm2), deciles 7 and 8 as the next risk group (>2.4 mm2 to ≤9.4 mm2), decile 9 as the next risk group (>9.4 mm2 to ≤27.6 mm2), and decile 10 (CTA >27.6 mm2) as the highest risk group. We rounded these values to the nearest 0.5 mm2 so that CTA cut points of 2.5, 9.5, and 27.5 mm2 defined 4 strata for comparison with criterion standard AJCC8 stages T1 to T4. The number of patients in CTA-defined groups included 750 in T1, 243 in T2, 121 in T3, and 125 in T4 compared with AJCC8-defined groups of 675 in T1, 271 in T2, 154 in T3, and 139 in T4. Wider overall separation of CTA-defined groups (HRs, 1.00-41.46) was found compared with AJCC8 T category (HRs 1.00-36.95) (Figure 3B). A cross-tabulation revealed AJCC8 T categories were dispersed across CTA risk groups, suggesting that CTA revealed intrastage risk heterogeneity (Figure 3C). Examples of discrepant CTA and T category are shown in eFigure 5 in the Supplement. The Kaplan-Meier plots for CTA strata and AJCC8 T category–defined prognostic groups are shown in Figure 3D and E. Furthermore, prognostic groups determined by CTA had a lower BIC than T category (difference, 13.9), indicating very strong evidence against the model containing the T category.21 Last, we estimated how optimistic the model based on CTA stratification would be because of overfitting if generalized to new samples and assessed by bootstrap analysis.24 Model calibration was estimated by regression slope, which is 1.00 by design in the original model and 0.99 after optimism correction, indicating only trivial coefficient shrinkage was necessary to correct for optimism or overfitting. Thus, the Harrell C index, measuring discrimination, was 83.7% in the original model and 83.6% when corrected for optimism. These findings suggest that the size of our sample indicates that risk groups based on CTA would generalize well to a similar population.

Figure 3. Comparison of Calculated Tumor Area (CTA)–Defined Risk Strata and American Joint Committee on Cancer, Version 8 (AJCC8) T-Category Strata.

Figure 3.

Open in a new tab

BT indicates Breslow thickness; HR, hazard ratio; and MSS, melanoma-specific survival.

Discussion

We devised the CTA, a simple prognostic feature consisting of a visual estimate of dermal area occupied by invasive melanoma cells in the section where Breslow thickness is measured. We demonstrate reliable measurement and describe its development and initial validation as a prognostic feature by demonstrating evidence of prognostic superiority to Breslow thickness and proposing that strata defined by CTA may be more useful than those defined by Breslow thickness. These initial findings require external validation.

Calculated tumor area has practicality because it is quick to measure (in our hands, typically <1 minute) and requires only the same H&E stain used for diagnosis, making it simple and inexpensive. Notably, although new molecular biomarkers are being actively pursued, purely H&E-based microscopic biomarkers remain underinvestigated, and yet they represent a rich vein of enquiry. Calculated tumor area is semiquantitative and might benefit from using image analysis. However, we believe gains in estimation and accuracy are offset by the difficulty of translating this approach to histopathologists, who are used to making semiquantitative judgments using only H&E stains and for whom a purely H&E-based approach would be acceptable and easily translatable to any pathology department in the world. Furthermore, being based on H&E assessment, CTA has no commercial or technological barrier to uptake. If digital pathology ever becomes the global standard for pathologists, then an image analysis–based approach may be preferable.

Tumor size is important because it estimates invasive cancer cell burden. Breslow thickness is likely a surrogate for invasive cancer cell burden while remaining simple to measure and quantitative. We may reasonably assume that something more representative of invasive cancer cell burden would outperform Breslow thickness for prognostic importance. However, such a new measurement has to balance practicality vs complexity. Calculated tumor area strikes this balance well, being a simple advancement beyond Breslow thickness while retaining practicality. Evidence that CTA is an advancement is shown by the way it confounds Breslow thickness and by measures of relative variable importance. This outcome might be expected because Breslow thickness is somewhat entailed in CTA, given that the CTA box height depends on Breslow thickness. However, by addition of breadth and proportion, CTA would be expected to offer an improvement, as supported by the retention of CTA in all 100 bootstrap sample models. Each of these bootstrap samples and resulting Cox proportional hazards regression models is unique and thus subject to sampling variation. That CTA, but not Breslow thickness, was resilient to this variation testifies to its importance. If CTA were to be validated, a back transformation from the logarithm scale would be required for clinical application.

Implications for risk stratification suggested by our findings are that many cases of relatively thin melanomas would be upgraded from T1 to CTA risk group 2, and even more T2 melanomas would be downgraded for meager tumor area to CTA risk group 1. These findings need to be validated, but for centers that perform sentinel lymph node biopsies, this outcome could have a substantial effect on eligibility. The way that CTA might be used in clinical practice if externally validated needs to be established. The most extreme scenario is a replacement of Breslow thickness, but this eventuality would require compelling evidence from multiple independent sources. Other possibilities include use as a staging adjunct, perhaps with a validated prognostic model built using AJCC8 and CTA from which a nomogram or online calculator could be created.

Other studies have attempted to move beyond 1-dimensional Breslow thickness. In his original description of thickness, Breslow3 introduced the idea of tumor volume as a prognostic feature, combining maximal diameter and thickness. Tumor volume was found to be prognostically superior to Breslow thickness in 35 patients,25 whereas a study of 122 cases revealed that tumor volume was linked to a higher risk of recurrence in patients with low Breslow thickness.26 A study of 108 patients revealed prognostic value for volume, but it was not as important as Breslow thickness.27 These studies were all fairly small.

Generalizing biomarker findings to new data sets can be problematic. To mitigate this problem, we investigated a biomarker with a rational basis, analyzed a large number of cases from 2 different hospitals, and used bootstrap analysis for testing CTA’s ability to stratify melanoma cases, enabling use of the full data set to maximize variable estimate precision while still addressing model overfitting.

Limitations

This study is, to our knowledge, the first description of CTA. As such, this study can only be regarded as proof of concept, and several issues need to be resolved before the metric would be ready for general implementation. These issues include confirmation that the measurement protocol and interobserver agreement can be replicated by others. We used a split-sample approach to assess CTA’s prognostic validity. The 2 data sets were generally comparable, although mitotic rate was different between centers. Systematic differences between Leicester and Nottingham may exist, such as case mix, specimen handling, and histopathologist reporting practices. True external validation by independent researchers from different centers is needed; even then, the most practical way to translate CTA into clinical care still needs to be established.

Conclusions

This study describes the development and initial validation of a proposed histologic feature, CTA. We demonstrate evidence of its independent prognostic value. Its simplicity, speed, and low cost make it a priority for further study to assess its validity and to determine how it might best be translated into clinical care.

Supplement.

eFigure 1. Flowchart Showing Patient Selection

eFigure 2. Flowchart Illustrating Statistical Analyses

eFigure 3. Interobserver Agreement for CTA Scoring

eFigure 4. Photomicrographs With Examples of CTAs

eFigure 5. Examples of CTA and T Category Discrepancy

eTable 1. Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) Guideline Checklist

eTable 2. Baseline Features of Leicester and Nottingham Melanomas

eTable 3. BIC Values for Cox Proportional Hazards Regression Model With Various CTA Transformations, Measured in 918 Melanomas

eTable 4. Crude and Adjusted Cox Proportional Hazards Regression Models

eTable 5. Cross-tabulation of CTA Quartiles and BD and TBS Quartiles

eTable 6. Cox Proportional Hazards Regression Model With CTA and BD

Click here for additional data file. (1.2MB, pdf)

References

  • 1.Cancer Research UK Skin cancer statistics. http://www.cancerresearchuk.org/health-professional/cancer-statistics/statistics-by-cancer-type/skin-cancer#heading-Zero. Published 2017. Accessed April 11, 2018.
  • 2.Glazer AM, Winkelmann RR, Farberg AS, Rigel DS. Analysis of trends in US melanoma incidence and mortality. JAMA Dermatol. 2017;153(2):225-226. doi: 10.1001/jamadermatol.2016.4512 [DOI] [PubMed] [Google Scholar]
  • 3.Breslow A. Thickness, cross-sectional areas and depth of invasion in the prognosis of cutaneous melanoma. Ann Surg. 1970;172(5):902-908. doi: 10.1097/00000658-197011000-00017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Balch CM, Gershenwald JE, Soong SJ, et al. Final version of 2009 AJCC melanoma staging and classification. J Clin Oncol. 2009;27(36):6199-6206. doi: 10.1200/JCO.2009.23.4799 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Gershenwald JE, Scolyer RA, Hess KR, et al. Melanoma staging: evidence-based changes in the American Joint Committee on Cancer Eighth Edition Cancer Staging Manual. CA Cancer J Clin. 2017;67(6):472-492. doi: 10.3322/caac.21409 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Heimann R, Hellman S. Aging, progression, and phenotype in breast cancer. J Clin Oncol. 1998;16(8):2686-2692. doi: 10.1200/JCO.1998.16.8.2686 [DOI] [PubMed] [Google Scholar]
  • 7.Gerami P, Cook RW, Wilkinson J, et al. Development of a prognostic genetic signature to predict the metastatic risk associated with cutaneous melanoma. Clin Cancer Res. 2015;21(1):175-183. doi: 10.1158/1078-0432.CCR-13-3316 [DOI] [PubMed] [Google Scholar]
  • 8.Cook RW, Middlebrook B, Wilkinson J, et al. Analytic validity of DecisionDx-Melanoma, a gene expression profile test for determining metastatic risk in melanoma patients. Diagn Pathol. 2018;13(1):13. doi: 10.1186/s13000-018-0690-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Rashed H, Flatman K, Bamford M, Teo KW, Saldanha G. Breslow density is a novel prognostic feature in cutaneous malignant melanoma. Histopathology. 2017;70(2):264-272. doi: 10.1111/his.13060 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Saldanha G, Flatman K, Teo KW, Bamford M. A novel numerical scoring system for melanoma tumor–infiltrating lymphocytes has better prognostic value than standard scoring. Am J Surg Pathol. 2017;41(7):906-914. doi: 10.1097/PAS.0000000000000848 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.McShane LM, Altman DG, Sauerbrei W, Taube SE, Gion M, Clark GM; Statistics Subcommittee of the NCI-EORTC Working Group on Cancer Diagnostics . Reporting recommendations for tumor marker prognostic studies. J Clin Oncol. 2005;23(36):9067-9072. doi: 10.1200/JCO.2004.01.0454 [DOI] [PubMed] [Google Scholar]
  • 12.R: a language and environment for statistical computing [computer program]. Vienna, Austria: R Foundation for Statistical Computing; 2016.
  • 13.RStudio: integrated development for R [computer program]. Boston, MA: RStudio, Inc; 2015.
  • 14.Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull. 1979;86(2):420-428. doi: 10.1037/0033-2909.86.2.420 [DOI] [PubMed] [Google Scholar]
  • 15.Irr: various coefficients of interrater reliability and agreement, version 0.84. https://CRAN.R-project.org/package=irr. Published January 26, 2019. Accessed May 3, 2019.
  • 16.Therneau T. A package for survival analysis, version 2.38. http://CRAN.R-project.org/package=survival. Published April 1, 2019. Accessed May 3, 2019.
  • 17.Kassambra A, Kosinsji M Survminer: drawing survival curves using “ggplo2.” https://CRAN.R-project.org/package=survminer. Published August 4, 2018. Accessed May 3, 2019.
  • 18.Steyerberg EW, Harrell FE Jr. Prediction models need appropriate internal, internal-external, and external validation. J Clin Epidemiol. 2016;69:245-247. doi: 10.1016/j.jclinepi.2015.04.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Babyak MA. What you see may not be what you get: a brief, nontechnical introduction to overfitting in regression-type models. Psychosom Med. 2004;66(3):411-421. [DOI] [PubMed] [Google Scholar]
  • 20.Harrell FE. Rms: regression modeling strategies, version 5.1-3.1. https://CRAN.R-project.org/package=rms. Published April 22, 2019. Accessed X.
  • 21.Raftery A. Bayesian model selection in social research. Sociol Methodol. 1995;25:111-195. doi: 10.2307/271063 [DOI] [Google Scholar]
  • 22.Saldanha G, Yarrow J, Pancholi J, et al. Breslow density is a novel prognostic feature that adds value to melanoma staging. Am J Surg Pathol. 2018;42(6):715-725. doi: 10.1097/PAS.0000000000001034 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Steyerberg EW. Coding of categorical and continuous predictors In: Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating. New York, NY: Springer Science+Business Media, LLC; 2009. [Google Scholar]
  • 24.Steyerberg EW. Overfitting and optimism in prediction models In: Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating. New York, NY: Springer Science+Business Media, LLC; 2009. [Google Scholar]
  • 25.Walton RG, Kim J, Velasco C, Swetter SM. Tumor volume: an adjunct prognostic factor in cutaneous melanoma. Cutis. 2014;94(5):226-230. [PubMed] [Google Scholar]
  • 26.Voss B, Wilop S, Jonas S, et al. Tumor volume as a prognostic factor in resectable malignant melanoma. Dermatology. 2014;228(1):66-70. doi: 10.1159/000356121 [DOI] [PubMed] [Google Scholar]
  • 27.Friedman RJ, Rigel DS, Kopf AW, et al. Volume of malignant melanoma is superior to thickness as a prognostic indicator: preliminary observation. Dermatol Clin. 1991;9(4):643-648. doi: 10.1016/S0733-8635(18)30368-1 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement.

eFigure 1. Flowchart Showing Patient Selection

eFigure 2. Flowchart Illustrating Statistical Analyses

eFigure 3. Interobserver Agreement for CTA Scoring

eFigure 4. Photomicrographs With Examples of CTAs

eFigure 5. Examples of CTA and T Category Discrepancy

eTable 1. Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) Guideline Checklist

eTable 2. Baseline Features of Leicester and Nottingham Melanomas

eTable 3. BIC Values for Cox Proportional Hazards Regression Model With Various CTA Transformations, Measured in 918 Melanomas

eTable 4. Crude and Adjusted Cox Proportional Hazards Regression Models

eTable 5. Cross-tabulation of CTA Quartiles and BD and TBS Quartiles

eTable 6. Cox Proportional Hazards Regression Model With CTA and BD

Click here for additional data file. (1.2MB, pdf)

Articles from JAMA Dermatology are provided here courtesy of American Medical Association

RESOURCES