Estimating Overdiagnosis of Melanoma Using Trends Among Black and White Patients in the US

Key Points

Question

What proportion of melanoma cases among White patients in the US reflects overdiagnosis?

Findings

This cohort study used incidence and mortality trends among Black patients in the US as proxies for changes in true disease burden and improvements in treatment. More than one-half of the melanomas diagnosed in White patients in 2014 were estimated to represent overdiagnosis.

Meaning

These findings suggest that although some of the current melanoma epidemic likely reflects rising true disease burden, more than half of this increase may be attributable to overdiagnosis.

This cohort study used trends for mortality due to melanoma among Black patients in the US as a marker to estimate melanoma incidence among White patients in the US.

Abstract

Importance

The incidence of cutaneous melanoma has been rising rapidly among White patients in the US; however, a commensurate increase in mortality due to melanoma has not been observed. These trends suggest overdiagnosis is occurring.

Objective

To quantify melanoma overdiagnosis among White patients compared with Black patients in the US.

Design, Setting, and Participants

This cohort study used joinpoint regression of Surveillance, Epidemiology, and End Results data from 1975 to 2014 to determine melanoma incidence and mortality trends among Black and White patients in the US. Using trends in mortality due to melanoma in Black patients as a marker for improvements in medical care, the expected mortality trends in White patients if medical care had not improved were estimated. This served as a marker for the change in true cancer occurrence. Overdiagnosis was calculated as the difference between observed incidence and estimated true cancer occurrence. Analyses were stratified by sex. Data were analyzed from September to December 2017.

Main Outcomes and Measures

Proportion of melanoma cases overdiagnosed among White patients in 2014.

Results

From 1975 to 2014, melanoma incidence increased approximately 4-fold in White women (incidence rate ratio [IRR], 4.01 [95% CI, 3.65-4.41]) and 6-fold in White men (IRR, 5.97 [95% CI, 5.47-6.52]), whereas it increased less than 25% in Black women (IRR, 1.21 [95% CI, 0.97-1.49]) and men (IRR, 1.17; [95% CI, 0.77-1.78]). Mortality due to melanoma decreased approximately 25% in Black women (morality rate ratio [MRR], 0.76 [95% CI, 0.63-0.90]) and men (MRR, 0.72 [95% CI, 0.62-0.84]), was stable in White women (MRR, 1.02 [95% CI, 0.96-1.09]), and increased almost 50% in White men (MRR, 1.49 [95% CI, 1.25-1.77]). Had medical care not improved, estimated mortality would have increased 60% in White women and more than doubled in White men. Based on these trends, an estimated 59% (95% CI, 45%-70%) of White women and 60% (95% CI, 32%-75%) of White men with melanoma were overdiagnosed in 2014.

Conclusions and Relevance

The discrepancies in incidence and mortality trends found in this cohort study suggest considerable overdiagnosis of melanoma occurring among White patients in the US.

Introduction

The reported incidence of cutaneous melanoma in the US has been increasing steadily.^1,2,3 In the mid-1970s, the Surveillance, Epidemiology, and End Results (SEER) Program began collecting nationally representative data on cancer incidence; by the mid-1990s, the reported incidence of melanoma had more than doubled. Some dermatologists argued that the increased incidence represented a true increase in cancer occurrence and called for more public awareness and screening skin examinations.^4,5 Others argued that the melanoma epidemic was more apparent than real⁶ and that increased awareness and screening might be the source of the epidemic.⁷ They noted that most of the incidence increase reflected the detection of early-stage, thin (<1.5 mm) melanomas and worried that the histopathological criterion standard for melanoma diagnosis was divorced from its biological behavior. More than 2 decades later, the incidence of melanoma continues to increase, now quadruple the rate of the mid-1970s. At present, nearly 90% of melanomas diagnosed in the US are thin (<1.0 mm), compared with 80% in 1990.³ The question of what proportion of this increase represents an increase in true disease burden vs overdiagnosis (the detection of cancers not destined to cause symptoms or death) remains unsettled.^8,9,10,11

To explore this question, we sought external standards to make judgments about true cancer occurrence and examined trends in incidence and mortality among Black patients in the US. Although the disease is much rarer in Black patients (1.7 per 100 000 population in 2014) than White patients (58.4 per 100 000 population in 2014) in the US, Black patients can also develop and die of melanoma. Black patients, however, are much less likely than White patients to be exposed to screening skin examinations.¹² Furthermore, although screening rates have increased over time among White patients, there has been little change in screening among Black patients.¹³ Using trends in mortality among Black patients in the US as a marker for improvements in medical care and expected mortality trends in White patients had medical care not improved as a marker for changes in true cancer occurrence, we estimated excess melanoma incidence among White patients in the US. These quantitative estimates could shed light on the scope of the problem of melanoma overdiagnosis.

Methods

Conceptual Overview

Because this cohort study was a retrospective observational study of publicly available data, it was exempted from institutional review board approval and informed consent by the University of Texas. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Overdiagnosis can be described as the difference between the apparent amount of cancer (observed incidence) and the true amount of clinically relevant disease (true cancer occurrence). Although data on observed incidence are readily available, data on true cancer occurrence are not. To make inferences about trends in true cancer occurrence in White patients, we sought a population subject to similar environmental exposures, but not subject to similar early detection efforts. We thus obtained melanoma incidence and mortality data on Black patients. We chose to use Black patients as the comparator population because they constitute an unscreened population for which data have been collected within SEER since 1975. Using a reference population of unscreened individuals is an established method for the quantification of cancer overdiagnosis.¹⁴

At the outset, we made 2 simplifying assumptions: (1) there is no overdiagnosis among Black patients, and (2) the observed incidence among White patients in 1975 was the true cancer occurrence at the time (ie, no overdiagnosis occurred in 1975). Both bias subsequent analyses toward the null, understating overdiagnosis.

As shown in Figure 1, to approximate trends in true cancer occurrence among White patients, we posited that the trend in true cancer occurrence among White patients is evident in melanoma mortality trends after correcting for improvements in medical care (which, as subsequently described, is inferred from incidence and mortality trends in Black patients). In simplest terms, if improved medical care was responsible for a 50% decline in mortality during a period in which White patients had stable mortality, this approach would posit the true cancer occurrence in White patients must have doubled. Failure to account for improvements in medical care would underestimate true cancer occurrence trends among White patients.

Figure 1. Conceptual Overview of Method Used to Draw Inferences for Melanoma Overdiagnosis Among White Patients in the US.

Trend Determination—Source Data and Joinpoint Regression

Annual data on melanoma incidence (both invasive and in situ) were obtained from the SEER program, which is functionally the cancer registry for the US.¹⁵ Melanoma incidence among Black patients is inherently unstable, both because the population is relatively small and has a low rate of melanoma occurrence. To increase the precision of this measurement, we incorporated data from additional SEER registries as they became available. SEER initially comprised 9 nationally representative population-based registries, which expanded to 13 registries in 1992 and 18 registries in 2000.

Annual data on melanoma mortality come from the National Vital Statistics System, the longstanding database providing cause of death information on all deaths in the US.¹⁶ All incidence and mortality data were age adjusted to the 2000 US standard population and were obtained using the SEER*Stat software.¹⁵

Observed trends in incidence and mortality were modeled using the National Cancer Institute’s Joinpoint regression program.¹⁷ A joinpoint refers to the inflection point of a changing trend; the Joinpoint program is the standard method used by SEER to determine changes in cancer trends.¹⁸ Our trend analyses used the same default joinpoint settings used by SEER (details regarding the joinpoint analysis and the year by year output are provided in eMethods 1 and 2 in the Supplement). Trends in incidence are expressed as incidence rate ratios (IRRs), and trends in mortality are expressed as mortality rate ratios (MRRs), comparing rates in 2014 with rates in 1975. All analyses were stratified by race and sex.

Statistical Analysis

Data were analyzed from September to December 2017. We used observed incidence and mortality trends in Black patients to infer improvements in medical care during the study period. In the setting of stable incidence, we posit that declining mortality would directly reflect improvements in medical care. Given the slight rise in incidence, however, using declining mortality directly would understate improvements in medical care. We thus inferred improvements in medical care as the ratio of observed mortality trends to the observed incidence trends in the Black population.

We then removed the effect of improvements in medical care from mortality trends in White patients by dividing the observed mortality trends by the inferred improvements in medical care. The resulting data address the question: What would the mortality trend among White patients be had medical care not improved? This serves as our proxy for the trend in true cancer occurrence in White patients. Again, true cancer occurrence and observed incidence are assumed equivalent in 1975. In subsequent years, the true cancer occurrence in White patients was obtained by applying the annual percentage change in the mortality trend in White patients that was expected had medical care not improved. The proportion of cases overdiagnosed in 2014 was calculated as the difference in observed incidence and true cancer occurrence trends from 1975 to 2014 divided by the observed incidence trend (details of this analysis are provided in eMethods 3 in the Supplement). Confidence intervals for the proportion of cases overdiagnosed in 2014 were calculated using parametric bootstrapping (described in eMethods 4 in the Supplement).

Results

Incidence and Mortality

Figure 2 illustrates that melanoma incidence has risen dramatically among White patients in the US. From 1975 to 2014, incidence increased roughly 4-fold in White women (IRR, 4.01 [95% CI, 3.65-4.41]) and 6-fold in White men (IRR, 5.97 [95% CI, 5.47-6.52]). Little corresponding change in incidence occurred among Black women (IRR, 1.21 [95% CI, 0.97-1.49]) or Black men (IRR, 1.17 [95% CI, 0.77-1.78]). During the same time interval, mortality due to melanoma was stable in White women (MRR, 1.02 [95% CI, 0.96-1.09]) and increased almost 50% in White men (MRR, 1.49 [95% CI, 1.25-1.77]). Melanoma mortality decreased approximately 25% both in Black women (MRR, 0.76 [95% CI, 0.63-0.90]) and Black men (MRR, 0.72 [95% CI, 0.62-0.84]).

Figure 2. Melanoma Incidence and Mortality Among White and Black Patients in the US, 1975 to 2014.

Incidence rate ratio (IRR) and mortality rate ratio (MRR) are based on National Cancer Institute joinpoint analysis. Blue lines represent joinpoint incidence trends. Small blue dots represent incidence per the Surveillance, Epidemiology, and End Results (SEER) database comprising 9 population-based registries; medium-sized blue dots represent incidence per the SEER database comprising 13 registries; and large blue dots represent incidence per the SEER database comprising 18 registries. Orange lines represent joinpoint mortality trends. Orange dots represent US mortality.

Estimate of Overdiagnosis of Melanoma

We used expected mortality trends in White patients had medical care not improved as a proxy for the change in true cancer occurrence in White patients. The Table details the 3 steps involved in this analysis. First, we inferred improvements in medical care from the changes in melanoma mortality in Black patients expected had incidence remained constant. This analysis suggests that improvements in medical care since 1975 are responsible for reducing melanoma mortality by more than one-third since 1975 (rate ratios, 0.63 [95% CI, 0.47-0.83] in women; 0.62 [95% CI, 0.40-0.97] in men). Second, we calculated the expected mortality trends had treatment not improved in White women (MRR, 1.63 [95% CI, 1.23-2.18]) and White men (MRR, 2.41 [95% CI, 1.48-4.04]).

Table. Calculations Used to Estimate Extent of Overdiagnosis.

Step	Data for 2014 vs 1975 (95% CI)
	Women	Men
1. Infer improvements in medical care from mortality trends in Black patients expected had incidence remained constant
MRR	0.76 (0.63-0.90)	0.72 (0.62-0.84)
IRR	1.21 (0.97-1.49)	1.17 (0.77-1.78)
Inferred improvement in medical care, MRR/IRR	0.63 (0.47-0.83)	0.62 (0.40-0.97)
2. Calculate expected mortality trends in White patients had medical care not improved
Observed MRR	1.02 (0.96-1.09)	1.49 (1.25-1.77)
Inferred improvement in medical care, MRR/IRR (from step 1)	0.63 (0.47-0.83)	0.62 (0.40-0.97)
Expected MRR had medical care not improved (proxy for change in true cancer occurrence), observed MRR/inferred improvement in care	1.63 (1.23-2.18)	2.41 (1.48-4.04)
3. Compare change in observed incidence with expected true cancer occurrence
Observed IRR	4.01 (3.65-4.41)	5.97 (5.47-6.52)
Change in inferred true cancer occurrence (from step 2)	1.63 (1.23-2.18)	2.41 (1.48-4.04)
Cancers overdiagnosed in 2014 (IRR − inferred true cancer occurrence)/IRR, %	59 (45-70)	60 (32-75)

Abbreviations: IRR, incidence rate ratio; MRR, morality rate ratio.

Finally, we used these mortality trends as a proxy for changes in true cancer occurrence, as shown both in the third step in the Table and Figure 3. Again, the difference between observed incidence and true cancer occurrence represents overdiagnosis. Using this approach, the proportion of cases overdiagnosed in 2014 was 59% (95% CI, 45%-70%) among White women and 60% (95% CI, 32%-75%) among White men.

Figure 3. Estimates of Observed Incidence and True Cancer Occurrence in White Patients in the US.

Blue lines represent observed incidence; orange lines represent true cancer occurrence. Shaded areas represent 95% CIs for trend. The difference between observed and true cancer occurrence represents overdiagnosis. IRR indicates incidence rate ratio.

Discussion

We used a unique approach to determine the contribution of changes in true disease burden to the increase in observed melanoma incidence among White patients in the US. Our findings suggest that there has been an increase in true cancer occurrence—particularly among White men, whose mortality rate increased by nearly 50% during the study period. The reasons for this increase are unknown but could be related to declines in skin immunosurveillance in the elderly, the accumulation of DNA damage over time, or differences in tumor biology. Nevertheless, we conclude that much of the current melanoma epidemic is indeed more apparent than real⁶ and that at least half of current melanoma diagnoses likely represents overdiagnosis. Furthermore, our findings are remarkably similar to those of a recent study from Australia that used a different method of assessing overdiagnosis and estimated that 58% of melanoma cases represented overdiagnosis in Australia.¹⁹

There may be concerns that comparing incidence and mortality rates between White and Black patients obscures potential biological differences in melanoma cases between groups. Indeed, the histological distribution of melanoma cases differs between groups: acral lentiginous melanoma is the dominant subtype of melanoma among Black patients in the US, whereas superficial spreading melanoma is the dominant subtype among White patients in the US.²⁰ However, our inclusion of incidence and mortality data for Black patients is limited to assessments about overall medical care improvements over time. Furthermore, our approach accommodates this concern by positing that changes in melanoma mortality among White patients—after correcting for the effect of medical care improvement—serve as a proxy for changes in true cancer incidence among White patients themselves.

We estimated the effect of improvements in medical care from the incidence and mortality trends in Black patients. Black patients experienced a small but steady decrease in mortality due to melanoma during the study period, a decrease that was remarkably similar in Black women and men. This decline clearly predates the advancements in melanoma treatment (targeted therapy and immunotherapy), which only became apparent in mortality data among White patients after 2013.³ The cause of this decline is unknown but may be related to improved access to medical care in general for Black patients in the US and better management of comorbid conditions (eg, hypertension, hyperlipidemia, and diabetes). In fact, our use of Black patients for the comparator population may have overestimated the general improvement in medical care among White patients, thus leading us to (1) overestimate their increase in true cancer occurrence and (2) underestimate the magnitude of overdiagnosis.

Although it is well known that cancer mortality is generally higher in Black patients than in White patients in the US, it is less well-known that the mortality for all cancer combined has been decreasing faster in Black patients for the past 3 decades.²¹ We suspect that the White patient population received the standard of care throughout this period, whereas the Black patient population experienced steadily increasing access to the standard of care over time. Again, to the extent that we have overestimated the effect of improved medical care, we have then underestimated overdiagnosis.

Some might point out, however, that we have not accounted for the possibility that screening is also reducing mortality in White patients. Were this the case, our estimates of true cancer occurrence would be rising faster. However, little evidence suggests that population-based melanoma screening affects melanoma mortality, and what little evidence there is suggests that it does not.^22,23,24,25

Clear evidence, however, suggests that melanoma screening affects the rate of melanoma detection. A quality initiative in the US reported a 2.5-fold increased detection associated with screening (94 vs 37 per 100 000 population), whereas the SCREEN (Skin Cancer Research to Provide Evidence for Effectiveness of Screening in Northern Germany) study reported more than a 6-fold increase (162 vs 24 per 100 000 population).^26,27 Our findings suggest that although there is some evidence of increasing true cancer occurrence (particularly among White men), much of the increased incidence of melanoma occurring among White patients represents overdiagnosis.

Developing estimates of overdiagnosis is important for clinicians to understand the scope of the problem, its potential harms, and its impact over time.^3,28 Overdiagnosis in the US has been quantified in breast, prostate, thyroid, and kidney cancers. An estimated 25% of breast cancers in women are overdiagnosed primarily owing to mammography screening.²⁹ In men, an estimated 60% of prostate-specific antigen–detected prostate cancers are overdiagnosed.²⁹ Approximately 45% of thyroid cancers and 50% of kidney cancers are overdiagnosed through increases in incidental detection driven by thyroid function testing and computed tomographic scans, respectively.^30,31

Limitations

This study has notable limitations. It is a study that includes only Black and White patients in the US, and therefore may not be generalizable to other populations. These estimates are based on populations and do not provide information on individual patients. Our analysis does not include more recent trends in overdiagnosis; we limited our analysis to 2014 because melanoma mortality among White patients declined sharply in 2013 due to the advent of effective therapy for metastatic melanoma.

Conclusions

The findings of this cohort study suggest that overdiagnosis of melanoma may be occurring among White patients in the US. As with clinicians in other subspecialties, dermatologists must grapple with the very real prospect that screening is a major source of the increased incidence of melanoma. Further study is needed to quantify the reservoir of asymptomatic, indolent disease by performing an “epidemiologic necropsy,”³² an autopsy study in which individuals not known to have melanoma during life have all moles biopsied to fully understand the contribution of medical care to the melanoma epidemic.

Supplement.

eMethods 1. Description of the Joinpoint Regression Method

eMethods 2. Incidence and Mortality Trends

eMethods 3. Estimating Underlying Incidence and Overdiagnosis

eMethods 4. Calculation of Confidence Intervals Using Parametric Bootstrapping