There were 643 patients with at least 1 incident invasive cutaneous melanoma in the 42-county NC ascertainment area in 2000. Twenty-eight cases (4.4%) were excluded because of missing Breslow thickness or street address information; this included patients diagnosed as having metastatic melanoma for whom no primary tumor was identified. Clinical and sociodemographic characteristics of the remaining 615 patients are given in .
Clinical and Sociodemographic Characteristics of the Study Populationa
Two hundred seventy-seven distinct diagnosing providers were identified. Only 15 providers diagnosed the melanomas in at least 1% (range, 0.2%–2.9%) of patients. All cases were diagnosed by providers in NC. Ninety-nine percent of patients traveled less than 120 miles to reach their diagnosing providers; the remaining 1% traveled between 233 and 386 miles. Patients were mapped to street address, and distance to diagnosing provider was visually examined (). Although most patients who traveled long distances were from the same region, their melanomas were diagnosed by different providers at different institutions, and no pattern could be identified. Within a given region, there was substantial variability in distance, with some patients traveling short distances to reach their diagnosing providers and other patients traveling much longer distances.
Distance traveled to diagnosing provider. Each data point represents the location of an incident case of melanoma. Points are color coded according to distance traveled to diagnosing provider.
In the bivariate analysis, Breslow thickness was statistically significantly associated with distance to diagnosing provider (). For distances not exceeding 120 miles, each 1-mile increase in distance corresponded with a 0.6% increase in the mean Breslow thickness (P=.003). In other words, each 10-mile increase in distance corresponded with a 6% increase in Breslow thickness (). After dichotomizing distance at 15 miles (75th percentile), patients who traveled more than 15 miles had 20% thicker tumors on average than patients who traveled 0 to 15 miles (P=.02).
Predictors of Breslow Thickness at Diagnosis
Figure 2 Bivariate relationship between Breslow thickness and distance traveled to diagnosing provider. The natural log of Breslow thickness is plotted against the distance traveled for a given case. The line represents the lowess, a locally weighted regression (more ...)
Consistent with historical evidence,1,30–32
Breslow thickness was associated with age at diagnosis (). The relationship between age at diagnosis and Breslow thickness was nonlinear, so age was categorized as 0 to 50 years (245 cases), 51 to 80 years (329 cases), or older than 80 years (41 cases). In the bivariate analysis, patients aged 51 to 80 years averaged 19% thicker tumors than patients aged 0 to 50 years (P
=.02), and patients older than 80 years averaged 109% thicker tumors than patients aged 0 to 50 years (P
<.001). Sex and primary tumor site were unassociated with Breslow thickness (P
Poverty rate was statistically significantly associated with Breslow thickness in the bivariate analysis; for every 1% increase in census tract poverty rate, Breslow thickness also increased by 1% (P=.04). No association between Breslow thickness and rurality could be identified using the Office of Management and Budget or US Department of Agriculture classifications (P>.05). However, when patients were stratified as rural vs metropolitan, the effect of distance to diagnosing provider on Breslow thickness seemed greater for cases from rural areas compared with cases from metropolitan areas. Every 10-mile increase in distance corresponded with a 10% increase in Breslow thickness (P=.06) for cases from rural counties compared with a 5% increase in Breslow thickness (P=.03) for cases from metropolitan counties.
The median Breslow thickness for cases diagnosed by dermatologists (0.5mm)was statistically significantly less than the median Breslow thickness for cases diagnosed by surgeons (1.04 mm) or by other providers (0.62 mm) (P<.001). When the supply of dermatologists was examined using the density of dermatologists per 100 000 residents in the county, there was no association between Breslow thickness and dermatologist supply (P>.05). Similarly, there was no association between the dichotomous dermatologist present or absent variable and Breslow thickness (P>.05). However, using the absolute number of dermatologists, Breslow thickness decreased by 0.9% for every additional dermatologist in the county (P=.004).
There was no statistically significant correlation between any of the sociodemographic factors (distance, poverty, rurality, and dermatologist supply), so all were included in the multivariate analysis. Because provider specialty cannot directly affect Breslow thickness, provider specialty was not included in the multivariate model. After adjusting for other factors, only age and distance to diagnosing provider were statistically significantly associated with Breslow thickness (). Because estimates of some variables can be unstable when the number of variables in the model is high relative to the number of observations, the final model did not include gender, rurality, and primary tumor site. Despite removal of these statistically nonsignificant variables, poverty rate and absolute number of dermatologists were not statistically significantly associated with Breslow thickness. Age remained statistically significantly associated with Breslow thickness in the multivariate analysis: patients aged 51 to 80 years had 16% thicker tumors than patients aged 0 to 50 years (P=.04), and patients older than 80 years had 103% thicker tumors than patients aged 0 to 50 years (P<.001). Similarly, distance to diagnosing provider was statistically significant with each 10-mile increase in distance associated with a 6%increase in Breslow thickness (P=.009). Even when the analysis was limited to tumors less than 2.0-mm thick, Breslow thickness increased by 5% for every 10-mile increase in distance (P=.002). Further exploration was performed to identify predictors of distance to diagnosing provider. Age, sex, and primary tumor site were unassociated with distance to diagnosing provider (P>.05) (). Although there was a statistically significant difference in distance traveled according to the specialty of the provider, the difference was too small to be clinically relevant: compared with patients whose melanomas were diagnosed by dermatologists, patients whose melanomas were diagnosed by surgeons traveled on average 1.3 miles farther (P=.03). The difference in distance to diagnosing provider between patients whose melanomas were diagnosed by dermatologists and those whose melanomas were diagnosed by nonsurgeon and nondermatologist providers was not statistically significant. The difference in distance to diagnosing provider based on poverty rate was also too small to be clinically relevant: for every 1%increase in poverty rate, distance decreased by 0.1 miles (P=.01).
Predictors of Distance to Diagnosing Provider
Patients from rural counties traveled a modest 2.4 miles farther on average than patients from metropolitan counties (P=.001). Using the US Department of Agriculture classifications, distance to diagnosing provider was inversely related to the size of the town-dwelling population of the county (). Compared with patients from metropolitan areas, patients from rural areas were also older (mean age, 58.2 vs 53.7 years, P=.007) and were more likely to live in poverty (12.3% vs 9.1%, P<.001). There were no statistically significant differences in patient sex or provider specialty between cases from rural areas and those from metropolitan areas.
Patients from counties with at least 1 dermatologist traveled on average 8.3 miles less than patients from counties with no dermatologist (P<.001). This association was independent of the specialty of the actual diagnosing physician. In other words, the presence of a dermatologist resulted in a shorter mean distance, even for patients whose melanomas were not actually diagnosed by a dermatologist, suggesting that the presence of a dermatologist does not directly affect distance to diagnosing provider but rather is a marker of an increased supply of local health care resources. To further explore this idea, the dermatologist variable was replaced with other measures of physician supply (number of primary care physicians, number of non–primary care physicians, and total number of physicians). Because they were correlated, only 1 provider supply variable was included in the model at a time. The relationships between each variable and distance to diagnosing provider were similar and substantial, and the magnitudes of the effects of the other coefficients in the model were stable regardless of which measure of provider supply was used.