Key Points
Question
What is the association between geospatial factors and treatment selection for patients with laryngeal cancer?
Findings
In this cohort study of 21 289 patients, patient demographic characteristics were associated with treatment selection after controlling for oncologic factors and accounting for county-level clustering. In addition, patients in counties least likely to provide surgical treatment experienced inferior survival.
Meaning
These findings suggest that nonclinical factors, including county of residence, are associated with laryngeal cancer treatment choice and may contribute to survival disparities.
Abstract
Importance
Guidelines for many head and neck cancers, especially laryngeal cancers, allow for multiple treatment options. Currently, inequitable provision of surgery may contribute to outcome disparities. However, the role of geospatial factors remains understudied.
Objective
To assess the association between US geospatial factors and treatment selection for patients with laryngeal cancer.
Design, Setting, and Participants
In this retrospective cohort study, patients diagnosed with laryngeal squamous cell carcinoma between January 1, 2004, and December 31, 2014, were identified from the Surveillance, Epidemiology, and End Results database. Adjusted odds ratios (aORs) for surgical treatment were generated from multivariable, hierarchical models to assess associations with oncologic, demographic, and county variables. Outlier US counties with the highest and lowest aORs were described. Data analysis was performed from April 29 to September 11, 2020.
Exposures
County of residence.
Main Outcomes and Measures
The aORs for surgical treatment were generated from multivariable, hierarchical models. Outlier counties with the highest and lowest aORs are described.
Results
The cohort includes 21 289 patients (mean [SD] age, 63.6 [11.2] years; 17 214 [80.9%] male) in 598 counties. Most counties had no otolaryngologist (365 [61.0%]) or radiation oncologist (434 [72.6%]). Surgery rates varied from 7.1% to 85.7% among counties with at least 10 cases. After oncologic variables were controlled for, factors independently associated with surgical treatment included patient age (aOR [95% CI], 0.94; 0.91-0.98 per 10 years), marital status (single versus married: aOR [95% CI], 0.87 [0.79-0.97]), and county social deprivation index (aOR [95% CI], 0.98 [0.97-1.00 per 5 points]) but not physician number (≥2 otolaryngologists: aOR [95% CI], 0.91 [0.75-1.11] vs ≥1 radiation oncologist: aOR [95% CI], 0.91; 0.75-1.11). The 5% of counties most likely to provide surgery (aOR, >1.23) were nearly all large metropolitan areas (2593 patients [93.3%]) and treated a disproportionately large number of patients (2778 [13.1%]). The 5% of counties least likely to provide surgery (aOR, <0.79) were also mostly large metropolitan areas (1676 patients [91.2%]) and treated a disproportionately large number of patients (1838 [8.6%]). Patients in counties least likely to provide surgery had inferior survival compared with those most likely to provide surgery (adjusted hazard ratio, 1.16; 95% CI, 1.00-1.35).
Conclusions and Relevance
These findings suggest that sociodemographic factors contribute to the wide variety in surgical treatment practices by county. The largest metropolitan counties were often outliers regarding their adjusted odds of surgical treatment. This finding is concerning for the counties least likely to provide surgery where survival is inferior.
This cohort study assesses the association between geospatial factors and treatment selection for patients with laryngeal cancer.
Introduction
Patients with a new diagnosis of head and neck cancers (HNCs) often have multiple viable treatment options available,1 and the treatment choice may be affected by the physicians locally available. For HNCs, selection of the optimal treatment to maximize survival and minimize treatment-related morbidity requires access to multiple disciplines of oncologic physicians, including surgeons, radiation oncologists, and medical oncologists. These discussions ideally occur in a multidisciplinary fashion with additional input from speech and language pathologists, dieticians, prosthodontists, and others.2,3,4 Many patients unfortunately have difficulty accessing multidisciplinary expertise because of socioeconomic limitations, geospatial barriers, and the local health system resources.5,6,7 As cancer treatments, including for HNCs, become increasingly centralized in high-volume centers, there is a need to better understand variation within and between regions.
Among HNCs, most laryngeal cancers will have similar oncologic outcomes whether treated by a primary surgical- or radiotherapy-based approach.8,9,10,11 For early-stage disease, this approach is typically a single modality, whereas advanced-stage disease requires multimodality treatment. Even among very locally advanced disease (T4a), for which guidelines recommend surgical treatment,1 as many as 64% are treated nonsurgically with chemoradiotherapy, resulting in inferior survival outcomes.12,13,14,15 Indeed, a trend toward nonsurgical treatment has emerged over time.9
Some of the variability in treatment choices may be justified by patient factors, such as comorbidities and preference, or by oncologic factors, such as the anatomical extent of the tumor and surgical access for transoral resections. However, other nonmedically justifiable factors may influence treatment choice, including geographic location, socioeconomic status, or physician biases. Existing evidence suggests several biases affect treatment decisions, resulting in worse functional and oncologic outcomes.16,17,18 In addition, some patients cannot easily access an otolaryngologist who can provide an informed opinion about surgical treatment options. By 2025, a 2500-physician deficiency in the otolaryngology workforce is anticipated,19 and even today, the available otolaryngologists are inequitably distributed.20 A lack of physicians in rural areas and competition for patients in urban areas may inappropriately bias patients’ treatment decisions, leading to suboptimal functional and oncologic outcomes and further worsen disparities.
Existing research on HNC disparities is challenging to act on because it often fails to identify modifiable risk factors, especially those within the control of physicians. This study examines treatment choices made by physicians and their patients with laryngeal cancer for whom both surgical and nonsurgical options are often viable. In addition, geospatial variation and clustering are rarely accounted for in statistical models, which can lead to biased effect estimates.21 This study aims to identify geospatial factors associated with treatment selection for laryngeal cancer in a nationally representative cohort, focusing on geographic access to otolaryngologists and other county-level effects.
Methods
Data Sources
Three data sets were merged by Federal Information Processing System codes at the county level. First, patient-level demographic and oncologic data were extracted from the Surveillance, Epidemiology, and End Results (SEER) registries.22 This data source has been extensively described and provides high-quality cancer registry data from a nationally representative sample collected from 18 registries located across the US. Second, county data, including the number of health care physicians by specialty, were extracted from the Area Health Resource Files (AHRF), maintained by the US Health Resources and Services Administration.23 The AHRF database compiles data from more than 50 sources, including health professional workforce data from the American Medical Association Master File. Third, county-level socioeconomic status was assessed using the Social Deprivation Index (SDI), a weighted composite measure of multiple social and demographic factors that are associated with health care access.24 The SDI compiles 7 area-level variables from the American Community Survey to generate an index of area deprivation, with higher scores reflecting worse deprivation. Because this study did not meet the definition of human subjects review for Washington University, it was exempt from institutional review board review. The study used deidentified, publicly available data, and informed consent was not applicable. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Cohort Selection
Histologically confirmed cases of laryngeal squamous cell carcinoma were identified from the SEER registry diagnosed from January 1, 2004, to December 31, 2014. Patients were excluded because of metastatic (M0) or unresectable disease (T4b), survival time listed as 0 months, or unknown treatment modality. In addition, cases in counties not listed in the AHRF database were removed, such as those living on Native American reservations. Data analysis was performed from April 29 to September 11, 2020.
Variable Definitions
Variables were categorized into oncologic, individual demographic, and county characteristics. Oncologic variables included year of diagnosis, overall stage, and treatment. Surgical treatment was considered any extirpative or ablative surgery of the primary site (excludes biopsies) or a lymph node dissection (excludes excisional biopsies or sentinel lymph node biopsies). Survival time was measured from the time of diagnosis. Cause of death was classified as attributable to the index cancer or any other cause using the SEER algorithm.25
Individual demographic variables included county of residence at the time of diagnosis, age, sex, race/ethnicity, marital status, and insurance status. Race and ethnicity were based on SEER classifications and were combined into White, Black, Hispanic (any race), and other. Marital status was categorized as married, single, divorced or separated, or widowed. Insurance status was categorized as private insurance, any government insurance, or uninsured.
Counties were described by state, urban-rural continuum code, socioeconomic status quantified by the SDI, population size, and number of hospital beds. The total number of physicians registered in the county and the number of otolaryngologists, radiation oncologists, and primary care physicians were extracted from the AHRF database. All population and workforce measures were extracted for 2010, which fell in the middle of the study period and had the most complete data because it was a census year. For the primary analysis, the number of otolaryngologists and radiation oncologists were dichotomized as 2 or more otolaryngologist and 1 or more radiation oncologists, both of which correspond to the 72nd percentile among included counties.
Statistical Analysis
The cohort was characterized using descriptive statistics. Key county-level variables are presented graphically on maps to demonstrate the geographic distributions. Multivariable, hierarchical logistic regression models were constructed to assess the association between the study variables and surgical treatment. Effects are reported as odds ratios (ORs) with 95% CIs. All models included a random intercept to account for county-level clustering and to assess unmeasured county-level effects. An empty model was created with only the random intercept (model 1) to assess the intraclass correlation, a measure of the proportion of the variation in surgical treatment explained by patients’ location within a specific county.26 Univariate models were used to evaluate crude associations. Next, multivariable models were built using only county-level variables (model 2), then only patient-level variables (model 3), then a combined model with all variables eligible (model 4). Multivariable model results are reported with adjusted ORs (aORs) and 95% CIs. Variables were eligible for inclusion in the multivariable model based on clinical reasoning and their crude association with surgical treatment. Linearity of the model predictors was confirmed graphically. Model fit was assessed by the Akaike information criterion.
Interaction effects were assessed between SDI and insurance status with urban-rural setting (model 5) because those with access limitations attributable to socioeconomic or insurance may be more affected in rural areas where geographic access is also limited. No significant interaction effects were observed; therefore, model 4 was used as the final model in the subsequent analysis.
To assess the variation attributable to patients’ county of residence, the random intercepts associated with each county are presented on a caterpillar plot as aORs with 95% CIs. These aORs represent the association of the patient’s odds of surgical treatment with their location in a specific county after accounting for other covariates. The 95% CIs were calculated using bootstrapping with 10 000 repeats. The characteristics of counties with the highest and lowest 5% for their odds of surgical treatment were summarized with descriptive statistics. Crude cancer-specific 5-year survival with 95% CIs was calculated for each category of outlier counties using Kaplan-Meier methods. Adjusted survival estimates were produced from Cox proportional hazards regression models, adjusting for stage, year of diagnosis, age, sex, race/ethnicity, marital status, SDI, and urban-rural setting,
Several sensitivity analyses were performed to ensure the stability of the findings. The number of each physician type (otolaryngologist and radiation oncologist) was assessed as dichotomized, linear continuous, log-transformed continuous, and mean with adjacent counties. Next, proxy measures for county-level socioeconomic status were investigated as alternatives to the SDI. Last, the analysis was repeated on subsets of the data that may have differential effects, specifically stage (early and advanced) and urban-rural setting (metropolitan, urban, and rural). Modestly higher intraclass correlation coefficient values were seen for early stage (3.0%) and rural setting (3.2%).
Results
Cohort Description
A total of 21 289 patients (mean [SD] age, 63.6 [11.2] years; 17 214 [80.9%] male) met the inclusion criteria. Exclusions included age younger than 18 years (n = 10); survival time less than 1 month (n = 364); stage IVb, IVc, or unknown cancer (n = 3251); and unknown whether surgery was performed (n = 189). The demographic characteristics of the cohort are given in Table 1. Patients in the cohort resided across 12 states and 598 counties, representing 24% of the states and 19.9% of the counties in the US. A total of 17 718 patients (83.2%) lived in a metropolitan area, with a median of 17 otolaryngologists (range, 0-318) and 6 radiation oncologists (range, 0-143) in each county. However, 365 counties (61.0%) had no otolaryngologists and 434 (72.6%) had no radiation oncologists. The geographic distribution of surgical treatment, SDI, and number of physicians is displayed in the eFigure in the Supplement. Overall, 8369 patients (39.3%) were treated surgically. However, this number varied by county from 14 of 196 (7.1%) to 14 of 16 (85.7%) (after excluding those with <10 cases).
Table 1. Cohort Summary and Crude Associations with Surgical Treatmenta.
| All (N = 21 289) | Surgical treatment | |||
|---|---|---|---|---|
| No (n = 12 920) | Yes (n = 8369) | Crude OR (95% CI) | ||
| County-level variables | ||||
| Urban-rural setting | ||||
| Metropolitan | 17 718 (83.2) | 10 740 (83.1) | 6978 (83.4) | 1 [Reference] |
| Rural | 503 (2.4) | 310 (2.4) | 193 (2.3) | 1.00 (0.82-1.21) |
| Urban | 3068 (14.4) | 1870 (14.5) | 1198 (14.3) | 1.02 (0.92-1.12) |
| Hospital beds, median (IQR), No. | 1399 (240-2736) | 1398 (237-3736) | 1405 (240-3738) | 1.00 (1.00-1.00) |
| Population, median (IQR) | 513 657 (98 764-1 202 362) | 501 226 (97 265-1 202 362) | 514 453 (100 157-1 202 362) | 1.00 (1.00-1.00) |
| Smoking patients, mean (SD), % | 19.7 (6.0) | 19.7 (6.1) | 19.5 (6.0) | 0.98 (0.94-1.01) |
| Below poverty level, mean (SD), % | 15.0 (5.6) | 15.1 (5.6) | 14.8 (5.6) | 0.96 (0.92-0.99) |
| Median income, mean (SD), $ | 66 844 (17 001) | 66 559 (16 973) | 67 284 (17 037) | 1.02 (1.00-1.05) |
| Social Deprivation Index, mean (SD) | 57.1 (28.1) | 57.7 (27.9) | 56.1 (28.4) | 0.99 (0.98-0.99) |
| Otolaryngologists | ||||
| Total No., median (IQR) | 17.0 (2-43) | 17.0 (2-43) | 17.0 (2-43) | NA |
| ≥2, No. (%) | 5649 (26.5) | 3467 (26.8) | 2182 (26.1) | 0.99 (0.91-1.08) |
| Radiation oncologists | ||||
| Total No., median (IQR) | 6.0 (1-22) | 6.0 (1-21) | 6.0 (1-24) | NA |
| ≥1 | 6393 (30.0) | 3939 (30.5) | 2454 (29.3) | 0.97 (0.89-1.05) |
| Patient-level variables | ||||
| Age, mean (SD), y | 63.6 (11.2) | 63.8 (11.1) | 63.3 (11.4) | 0.96 (0.94-0.98) |
| Male sex | 17 214 (80.9) | 10 353 (80.1) | 6861 (82.0) | 1.12 (1.05-1.21) |
| Race | ||||
| White | 15 615 (73.6) | 9488 (73.7) | 6127 (73.5) | 1 [Reference] |
| Black | 3187 (15.0) | 2040 (15.8) | 1147 (13.8) | 0.88 (0.81-0.96) |
| Hispanic | 1674 (7.9) | 957 (7.4) | 717 (8.6) | 1.18 (1.06-1.31) |
| Otherb | 733 (3.5) | 392 (3.0) | 341 (4.1) | 1.26 (1.07-1.48) |
| Marital status | ||||
| Married | 11 144 (54.7) | 6560 (53.1) | 4584 (57.2) | 1 [Reference] |
| Separated | 3140 (15.4) | 2023 (16.4) | 1117 (13.9) | 0.79 (0.73-0.86) |
| Single | 4058 (19.9) | 2472 (20.0) | 1586 (19.8) | 0.92 (0.86-1.00) |
| Widowed | 2032 (10.0) | 1301 (10.5) | 731 (9.1) | 0.81 (0.73-0.89) |
| Insurance | ||||
| Private | 11 757 (76.4) | 7231 (76.5) | 4526 (76.3) | 1 [Reference] |
| Uninsured | 832 (5.4) | 516 (5.5) | 316 (5.3) | 1.01 (0.87-1.17) |
| Any government | 2798 (18.2) | 1708 (18.1) | 1090 (18.4) | 1.03 (0.95-1.12) |
| Year of diagnosis, mean (SD) | 2009.0 (3.1) | 2009.1 (3.1) | 2008.9 (3.1) | 0.99 (0.98-0.99) |
| Stage | ||||
| I | 8191 (38.5) | 4572 (35.4) | 3619 (43.2) | 1 [Reference] |
| II | 3814 (17.9) | 2703 (20.9) | 1111 (13.3) | 0.52 (0.48-0.56) |
| III | 3949 (18.5) | 2730 (21.1) | 1219 (14.6) | 0.57 (0.52-0.62) |
| Iva | 5335 (25.1) | 2915 (22.6) | 2420 (28.9) | 1.06 (0.98-1.13) |
| T classification | ||||
| T1 | 8761 (41.2) | 4958 (38.4) | 3803 (45.4) | 1 [Reference] |
| T2 | 5310 (24.9) | 3823 (29.6) | 1487 (17.8) | 0.51 (0.47-0.54) |
| T3 | 4265 (20.0) | 2940 (22.8) | 1325 (15.8) | 0.59 (0.55-0.64) |
| T4a | 2953 (13.9) | 1199 (9.3) | 1754 (21.0) | 1.94 (1.78-2.11) |
Abbreviations: IQR, interquartile range; NA, not applicable; OR, odds ratio.
Data are presented as number (percentage) unless otherwise indicated. County variables are derived from 2010 data. For continuous variables, ORs are reported per individual unit change except as follows: per 10 beds for hospital beds, 10 000 residents for population, 5% for smoking and poverty rate, 5 points for Social Deprivation Index, $10 000 for median income, and 10 years for age.
Other includes American Indian, Alaskan Native, Asian, Pacific Islander, and unknown.
Factors Associated With Surgical Treatment
Compared with patients treated nonsurgically, those receiving surgical treatment were more likely to be younger (63.3 vs 63.8 years; OR per year, 0.96; 95% CI, 0.94-0.98) and male (6861 [82.0%] vs 10 353 [80.1%]; OR, 1.12; 95% CI, 1.05-1.21). Compared with White patients, Black patients were less likely to receive surgery (OR, 0.88; 95% CI, 0.81-0.96), whereas patients of Hispanic ethnicity of any race were more likely to receive surgery (OR, 1.18; 95% CI, 1.06-1.31). Unmarried patients were less likely to receive surgery than married patients (OR, 0.79; 95% CI, 0.92-0.81 for separated or divorced, single, and widowed).
Oncologic factors, specifically overall stage and tumor classification, were also associated with surgical treatment. Specifically, compared with patients with stage I cancer, patients with stage II (OR, 0.52; 95% CI, 0.48-0.56) and stage III (OR, 0.57; 95% CI, 0.52-0.62) cancer were less likely to receive surgery, whereas patients with patients with stage IVa received surgery at a similar rate. Similarly, compared with patients with T1 disease, patients with T2 (OR, 0.51; 95% CI, 0.47-0.54) and T3 (OR, 0.59; 95% CI, 0.55-0.64) disease were less likely to receive surgery, whereas those with T4a disease received surgery at a similar rate (OR, 1.94; 95% CI, 1.78-2.11).
Patients receiving surgery were more likely to live in counties with less poverty (14.8% vs 15.1%; OR per 5% of impoverished counties, 0.96; 95% CI, 0.92-0.99), high median income ($67 284 vs $66 559; OR per $10 000 median income, 1.02; 95% CI, 1.00-1.05), and lower SDI (56.1 vs 67.7; OR, 0.99; 95% CI, 0.98-0.99). The median number of otolaryngologists and radiation oncologists for patients receiving surgical and nonsurgical treatment were equal (17 otolaryngologists and 6 radiation oncologists). In addition, no differences were found in the proportion of counties with fewer than 2 otolaryngologists or less than 1 radiation oncologist.
Results of Multivariable County-Level, Patient-Level, and Combined Models
The interclass correlation coefficient for the models was 2.1%, which can be interpreted as 2.1% of the variation in likelihood of surgical treatment is accounted for at the county level. The model fit for the patient-level model was superior to the county-level model (Akaike information criterion, 18 677 vs 28 438) and did not improve further with the combination of patient and county variables.
The county-level model (Table 2) included SDI, 2 or more otolaryngologists, and more than 1 radiation oncologist. A higher SDI tended to be associated with a modest reduction in the odds of surgical treatment (aOR, 0.98; 95% CI, 0.97-1.00). The number of physicians was not associated with surgical treatment (≥2 otolaryngologists: aOR, 1.05; 95% CI, 0.86-1.28; ≥1 radiation oncologist: aOR, 0.91; 95% CI, 0.75-1.11).
Table 2. Adjusted Odds of Surgical Treatment: Results of Multivariable Hierarchical Models.
| Adjusted odds ratio (95% CI) | |||
|---|---|---|---|
| County model | Patient model | Combined model | |
| County-level variables | |||
| Poverty level (per 5%) | 1.00 (0.93-1.08) | NA | NA |
| Median income (per $10 000) | 0.98 (0.93-1.03) | NA | NA |
| Social Deprivation Index (per 5 points) | 0.98 (0.97-1.00) | NA | 0.98 (0.97-1.00) |
| Otolaryngologists ≥2 | 1.04 (0.89-1.22) | NA | 1.05 (0.86-1.28) |
| Radiation oncologists ≥1 | 0.90 (0.76-1.05) | NA | 0.91 (0.75-1.11) |
| Patient-level variables | |||
| Age (per 10 y) | NA | 0.94 (0.91-0.98) | 0.94 (0.91-0.98) |
| Male sex | NA | 1.05 (0.96-1.15) | 1.05 (0.96-1.15) |
| Race | |||
| Black | NA | 0.88 (0.79-0.98) | 0.90 (0.81-1.01) |
| Hispanic | NA | 1.11 (0.97-1.27) | 1.13 (0.98-1.29) |
| Other | NA | 1.19 (0.98-1.45) | 1.19 (0.98-1.45) |
| Marital status (reference: married) | |||
| Divorced or separated | NA | 0.80 (0.72-0.88) | 0.80 (0.72-0.88) |
| Single | NA | 0.87 (0.79-0.95) | 0.87 (0.79-0.95) |
| Widowed | NA | 0.85 (0.75-0.97) | 0.85 (0.75-0.97) |
| Insurance (reference: private) | |||
| Uninsured | NA | 0.93 (0.79-1.10) | 0.94 (0.80-1.11) |
| Any Medicaid | NA | 1.03 (0.93-1.14) | 1.04 (0.94-1.14) |
| Year | NA | 1.02 (1.00-1.03) | 1.02 (1.00-1.03) |
| Stage (reference: stage I) | |||
| Stage II | NA | 0.68 (0.52-0.91) | 0.69 (0.52-0.91) |
| Stage III | NA | 0.57 (0.45-0.73) | 0.57 (0.45-0.73) |
| Stage IVa | NA | 0.48 (0.38-0.61) | 0.48 (0.38-0.61) |
| Tumor class (reference: T1) | |||
| T2 | NA | 0.73 (0.56-0.95) | 0.72 (0.56-0.94) |
| T3 | NA | 0.96 (0.76-1.21) | 0.96 (0.76-1.21) |
| T4a | NA | 3.79 (2.97-4.85) | 3.79 (2.96-4.85) |
In the patient-level model, the largest effect sizes were seen for oncologic variables (stage and tumor class), but demographic factors also remained independently associated with treatment, including age (aOR, 0.94; 95% CI, 0.91-0.98), Black race (aOR, 0.88; 95% CI, 0.79-0.98), and marital status (divorced or single: aOR, 0.80; 95% CI, 0.72-0.88); single: aOR, 0.87; 95% CI, 0.79-0.95; widowed: aOR, 0.85; 95% CI, 0.75-0.97). Sex (aOR, 1.05; 95% CI, 0.96-1.15) and insurance (uninsured: aOR, 0.93; 95% CI, 0.79-1.10; any Medicaid: aOR, 1.03; 95% CI, 0.93-1.14) were no longer associated with treatment after controlling for other variables (Table 2). In the combined model, similar results were seen. However, the association between Black race and reduced likelihood for surgery (aOR, 0.90; 95% CI, 0.81-1.01) was no longer statistically significant after accounting for county-level factors (Table 2).
Individual County Effects
The ORs associated with each county are summarized by a waterfall plot (Figure 1). The counties in the tails of the plot represent outliers with clinically meaningful effect sizes (aOR as low as 0.57 and as high as 1.77 compared with the median county). Counties with aORs below the fifth percentile (aOR, 0.79) and above the 95% percentile (aOR, 1.23) were considered outliers and further described (Figure 2). Twelve of these counties had aORs for the use of surgery that were statistically significantly different from the median county odds (eTable 1 in the Supplement). The outlier counties represent a disproportionately large number of patients; the 5% of counties with the highest odds of surgical treatment represent 2778 patients (13.1% of the cohort), and the 5% of counties with the lowest odds represent 1838 patients (8.6%). Compared with other counties, outlier counties, in either direction, tended to be metropolitan, with lower smoking rates, less poverty, and higher median income (Table 3). Comparing high and low outlier counties, those least likely to provide surgery were less often metropolitan and had fewer hospital beds, smaller populations, higher income, lower SDI, fewer otolaryngologists (median, 11 vs 30) and radiation oncologists (median, 6 vs 13), and inferior survival (adjusted hazard ratio, 1.16; 95% CI, 1.00-1.35). Several large metropolitan areas were among those most and least likely to provide surgery. For example, the counties that included Seattle and Trenton had significantly low adjusted odds of surgery (aOR 0.57 and 0.65, respectively), whereas the counties that included Santa Barbara, New Haven, Savanah, Honolulu, Salt Lake City, and Tacoma had high odds of surgery (aOR 1.50, 1.41, 1.77, 1,51, 1,76, and 1.39, respectively). In some cases, adjacent counties were outliers in opposite directions, such as Kitsap and Pierce counties in Washington (aOR 0.57 vs 1.39), or Morris and Somerset counties in New Jersey (aOR 1.29 vs 0.76).
Figure 1. Association of County With Odds of Surgical Treatment.
Adjusted odds ratios (aORs) are presented for each county with vertical bars representing 95% CIs. This aOR represents the odds of receiving surgical treatment attributable to the patient’s county residence relative to the median county after controlling for other patient- and county-level covariates. Counties with 95% CIs that exclude the 1 (horizontal black dotted line) are indicated with black dots and darker gray vertical bars. The blue horizontal dotted lines mark the 5th and 95th percentile aOR. Counties with aORs that fall outside these lines are considered outliers.
Figure 2. Map of Outlier Counties Regarding Adjusted Odds of Surgical Treatment.
Counties with adjusted odds ratios (aORs) below the 5th percentile have the lowest odds of surgical treatment, whereas those above the 95th percentile have the highest odds.
Table 3. Description of Patients With Laryngeal Cancer by Adjusted County Odds of Surgical Treatmenta.
| Grouped by county-level likelihood to provide surgery | |||
|---|---|---|---|
| Most likely counties (n = 30 counties; n = 2778 patients) | Medium likelihood counties (n = 526 counties; n = 16 654 patients) | Least likely counties (n = 30 counties; n = 1838 patients) | |
| County-level variables | |||
| Urban-rural setting | |||
| Metropolitan | 2593 (93.3) | 13443 (80.7) | 1676 (91.2) |
| Rural | 19 (0.7) | 452 (2.7) | 22 (1.2) |
| Urban | 166 (6.0) | 2759 (16.6) | 140 (7.6) |
| Hospital beds, median (IQR) | 2769 (1399-3258) | 1295 (196-3980) | 1069 (457-2151) |
| Population, median (IQR) | 862 477 (423 895-1 029 655) | 440 171 (72 155-1 520 271) | 366 513 (131 613-688 078) |
| Smokers, mean (SD), % | 18.0 (4.5) | 20.1 (6.3) | 18.5 (5.4) |
| Below poverty level, mean (SD), % | 14.0 (4.9) | 15.3 (5.6) | 13.6 (6.0) |
| Median income, mean (SD) | 70 172 (17 250) | 65 604 (16 639) | 73 268 (17 674) |
| Social Deprivation Index, mean (SD) | 59.0 (26.3) | 57.6 (28.7) | 49.7 (24.1) |
| Otolaryngologists, median (IQR) | 33 (13-42) | 15 (1-44) | 11 (3-19) |
| Radiation oncologists, median (IQR) | 13 (5-20) | 5 (0-25) | 6 (2-11) |
| Patient-level variables | |||
| Age, mean (SD), y | 64.6 (11.4) | 63.5 (11.2) | 63.2 (11.2) |
| Female sex | 523 (18.8) | 3193 (19.2) | 356 (19.4) |
| Race | |||
| White | 1987 (72.2) | 12 259 (73.8) | 1353 (73.7) |
| Black | 348 (12.6) | 2476 (14.9) | 361 (19.7) |
| Hispanic | 244 (8.9) | 1342 (8.1) | 87 (4.7) |
| Otherb | 174 (6.3) | 524 (3.2) | 35 (1.9) |
| Marital status | |||
| Married | 1452 (54.4) | 8759 (54.9) | 922 (53.1) |
| Separated | 395 (14.8) | 2475 (15.5) | 267 (15.4) |
| Single | 516 (19.3) | 3169 (19.9) | 371 (21.4) |
| Widowed | 307 (11.5) | 1549 (9.7) | 175 (10.1) |
| Insurance | |||
| Private | 1500 (76.4) | 9199 (76.2) | 1058 (78.0) |
| Uninsured | 113 (5.8) | 618 (5.1) | 101 (7.4) |
| Any government | 350 (17.8) | 2250 (18.6) | 197 (14.5) |
| Stage | |||
| I | 1078 (38.8) | 6387 (38.4) | 720 (39.2) |
| II | 518 (18.6) | 2983 (17.9) | 308 (16.8) |
| III | 505 (18.2) | 3088 (18.5) | 351 (19.1) |
| IVa | 677 (24.4) | 4196 (25.2) | 459 (25.0) |
| 5-Year survival, % (95% CI) | |||
| Crude | 70.2 (68.2-72.2) | 70.5 (69.7-71.3) | 68.6 (66.1-71.1) |
| Adjustedc | 74.6 (72.2-77.1) | 74.7 (73.5-75.8) | 71.1 (68.0-74.3)d |
Abbreviation: IQR, interquartile range.
Data are presented as number (percentage) of patients unless otherwise indicated.
Other includes American Indian, Alaskan Native, Asian, Pacific Islander, and unknown.
Adjusted for age, sex, race/ethnicity, marital status, insurance, year of diagnosis, stage, Social Deprivation Index, and urban-rural setting.
Results from the Cox proportional hazards regression model show inferior survival in counties least likely to provide surgery compared with those most likely to provide surgery (adjusted hazard ratio, 1.16; 95% CI, 1.00-1.35).
Discussion
This cohort study combined multiple national data sources to better understand variation in the treatment choice among more than 21 000 patients with laryngeal cancer. Wide geographic variation in practices was observed, with the rate of surgical treatment varying from 7% to 86% among counties. Defining the ideal treatment for laryngeal cancer has eluded creators of national guidelines, and the variation observed in the results of the current study reflect this ambiguity. Nevertheless, the inferior survival observed in counties least likely to provide surgery raises concern. In addition, the variation associated with nonclinical factors is unlikely to be medically justifiable and may contribute to outcome disparities.
Establishing an ideal proportion of the patients with laryngeal cancer who should be treated surgically is beyond the scope of this study, but the observed rate of surgical treatment nationally was 39.3% of patients with laryngeal cancer. At a county level, after controlling for other oncologic and demographic differences, there remained clinically meaningful and often statistically significant differences in the odds of surgical treatment based on individual counties. Counties with the highest odds of surgical treatment were nearly all large metropolitan areas (93.3%) with high numbers of physicians and more social deprivation. Despite the increased likelihood for surgery in these counties, the cancer-specific survival was similar to other counties with mean odds of surgical treatment.
Meanwhile, counties least likely to provide surgical treatment after controlling for other factors had lower cancer-specific survival. Similar to the counties where surgical treatment was most likely, these counties were very often metropolitan areas (91.2%). However, the counties least likely to treat surgically had fewer hospital beds, fewer head and neck cancer specialists (otolaryngologists and radiation oncologists), and less social deprivation compared with other counties. These results build on existing literature that suggests the increase in nonsurgical treatment of laryngeal cancer over time may contribute to decreases in population-level survival.9,13,27 The current study suggests that patients in counties where nonsurgical treatments predominate may highlight where geographically these outcomes arise.
The largest cities in this data set tended to be outliers regarding their use of surgical treatment, in one direction or the other. The presence of such significant variation after controlling for oncologic differences suggests this variation may not be clinically justifiable, especially in such close geographic proximity.
Associations between physician density and oncologic outcomes have been demonstrated for some cancer sites,28 whereas no correlation has been found in others.29 In this study, the number of otolaryngologists and radiation oncologists was not associated with treatment choice. One possible explanation is the high geographic correlation among specialties within counties. Therefore, patients with many otolaryngologists in their county also have many radiation oncologists, even though 60% of counties have no physicians of either specialty. The maldistribution of the medical workforce, including otolaryngologists, has been described previously.19,20 Interestingly, more than 2 otolaryngologists in a county has been associated with inferior overall survival for patients with HNC30; however, that study’s effect size was modest, and the study did not account for geospatial data clustering or assess cancer-specific survival.
Limitations
This study has important limitations, and it is not intended to establish the appropriateness of surgery for an individual or within a health care system. Important factors informing treatment decisions were not available in this data set, including patients’ comorbidities, substance use, and preferences and health systems’ expertise and resources. These unmeasured variables may account for the observed county-level differences. The characterization of outlier counties should be considered descriptive only because this was added post hoc based on the initial findings. In addition, these outliers describe patients’ county of residence and may not reflect the institutions based in those counties. This study was intentionally limited to laryngeal cancer, so these results may not apply to other HNC sites. Human papillomavirus–associated oropharyngeal cancer, in particular, represents an important and increasing subgroup of HNCs,31,32 for which multiple treatment options provide similar outcomes.33,34 Future studies addressing geospatial differences in oropharyngeal cancer would require data sources that accurately classify human papillomavirus association and include geospatial information, which are currently limited.
Conclusions
In the context of efforts to regionalize cancer care, including HNC,35,36 this study adds important insight into geographic treatment variation, especially the inferior survival observed in counties least likely to provide surgery. The study refutes a simplistic hypothesis in which regions with less medical care overuse 1 treatment modality and thus experience inferior outcomes. Instead, substantial differences in treatment preference exist among counties with large, high-volume centers, whereas smaller counties tend to provide surgical treatment at average levels. This finding may be associated with long-standing institutional preferences and expertise with a particular treatment modality. There is likely benefit to patients when physicians and institutions use modalities for which they have the most experience. However, historical practices may have to evolve if additional research confirms that regional preferences contribute to inferior outcomes.
eTable. List of Outlier Counties and Adjusted Odds Ratios for Surgical Treatment
eFigure. Geographic Distribution of Key Study Variables
References
- 1.National Comprehensive Cancer Network. NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines)—Head and Neck Cancer. Version 1. 2016. Accessed July 20, 2017. https://oralcancerfoundation.org/wp-content/uploads/2016/09/head-and-neck.pdf
- 2.Lewis CM, Nurgalieva Z, Sturgis EM, Lai SY, Weber RS. Improving patient outcomes through multidisciplinary treatment planning conference. Head Neck. 2016;38(suppl 1):E1820-E1825. doi: 10.1002/hed.24325 [DOI] [PubMed] [Google Scholar]
- 3.Patil RD, Meinzen-Derr JK, Hendricks BL, Patil YJ. Improving access and timeliness of care for veterans with head and neck squamous cell carcinoma: a multidisciplinary team’s approach. Laryngoscope. 2016;126(3):627-631. doi: 10.1002/lary.25528 [DOI] [PubMed] [Google Scholar]
- 4.Starmer H, Sanguineti G, Marur S, Gourin CG. Multidisciplinary head and neck cancer clinic and adherence with speech pathology. Laryngoscope. 2011;121(10):2131-2135. doi: 10.1002/lary.21746 [DOI] [PubMed] [Google Scholar]
- 5.de Moor JS, Virgo KS, Li C, et al. Access to cancer care and general medical care services among cancer survivors in the United States: an analysis of 2011 Medical Expenditure Panel Survey data. Public Health Rep. 2016;131(6):783-790. doi: 10.1177/0033354916675852 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Massa ST, Liebendorfer AP, Zevallos JP, Mazul AL. Distance traveled to head and neck cancer provider: a measure of socioeconomic status and access. Otolaryngol Head Neck Surg. 2020;162(2):193-203. doi: 10.1177/0194599819892015 [DOI] [PubMed] [Google Scholar]
- 7.Walker BB, Schuurman N, Auluck A, Lear SA, Rosin M. Socioeconomic disparities in head and neck cancer patients’ access to cancer treatment centers. Rural Remote Health. 2017;17(3):4210. doi: 10.22605/RRH4210 [DOI] [PubMed] [Google Scholar]
- 8.Timme DW, Jonnalagadda S, Patel R, Rao K, Robbins KT. Treatment selection for T3/T4a laryngeal cancer: chemoradiation versus primary surgery. Ann Otol Rhinol Laryngol. 2015;124(11):845-851. doi: 10.1177/0003489415588130 [DOI] [PubMed] [Google Scholar]
- 9.Chen AY, Fedewa S, Zhu J. Temporal trends in the treatment of early- and advanced-stage laryngeal cancer in the United States, 1985-2007. Arch Otolaryngol Head Neck Surg. 2011;137(10):1017-1024. doi: 10.1001/archoto.2011.171 [DOI] [PubMed] [Google Scholar]
- 10.Forastiere AA, Ismaila N, Lewin JS, et al. Use of larynx-preservation strategies in the treatment of laryngeal cancer: American Society of Clinical Oncology clinical practice guideline update. J Clin Oncol. 2018;36(11):1143-1169. doi: 10.1200/JCO.2017.75.7385 [DOI] [PubMed] [Google Scholar]
- 11.Pfister DG, Laurie SA, Weinstein GS, et al. ; American Society of Clinical Oncology . American Society of Clinical Oncology clinical practice guideline for the use of larynx-preservation strategies in the treatment of laryngeal cancer. J Clin Oncol. 2006;24(22):3693-3704. doi: 10.1200/JCO.2006.07.4559 [DOI] [PubMed] [Google Scholar]
- 12.Khoueir N, Matar N, Farah C, Francis E, Tabchy B, Haddad A. Survival of T4aN0 and T3N+ laryngeal cancer patients: a retrospective institutional study and systematic review. Am J Otolaryngol. 2015;36(6):755-762. doi: 10.1016/j.amjoto.2015.07.009 [DOI] [PubMed] [Google Scholar]
- 13.Grover S, Swisher-McClure S, Mitra N, et al. Total laryngectomy versus larynx preservation for T4a larynx cancer: patterns of care and survival outcomes. Int J Radiat Oncol Biol Phys. 2015;92(3):594-601. doi: 10.1016/j.ijrobp.2015.03.004 [DOI] [PubMed] [Google Scholar]
- 14.Dziegielewski PT, O’Connell DA, Klein M, et al. Primary total laryngectomy versus organ preservation for T3/T4a laryngeal cancer: a population-based analysis of survival. J Otolaryngol Head Neck Surg. 2012;41(suppl 1):S56-S64. [PubMed] [Google Scholar]
- 15.Massa ST, Osazuwa-Peters N, Franco J, Ward GW, Walker RJ. Survival after refusal of surgical treatment for locally advanced laryngeal cancer. Oral Oncol. 2017;71:34-40. doi: 10.1016/j.oraloncology.2017.05.019 [DOI] [PubMed] [Google Scholar]
- 16.Dovidio JF, Fiske ST. Under the radar: how unexamined biases in decision-making processes in clinical interactions can contribute to health care disparities. Am J Public Health. 2012;102(5):945-952. doi: 10.2105/AJPH.2011.300601 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Ringstrom MJ, Christian J, Bush ML, Levy JE, Huang B, Gal TJ. Travel distance: impact on stage of presentation and treatment choices in head and neck cancer. Am J Otolaryngol. 2018;39(5):575-581. doi: 10.1016/j.amjoto.2018.06.020 [DOI] [PubMed] [Google Scholar]
- 18.Derks W, de Leeuw JRJ, Hordijk GJ, Winnubst JA. Reasons for non-standard treatment in elderly patients with advanced head and neck cancer. Eur Arch Otorhinolaryngol. 2005;262(1):21-26. doi: 10.1007/s00405-004-0744-x [DOI] [PubMed] [Google Scholar]
- 19.Kim JSC, Cooper RA, Kennedy DW. Otolaryngology-head and neck surgery physician work force issues: an analysis for future specialty planning. Otolaryngol Head Neck Surg. 2012;146(2):196-202. doi: 10.1177/0194599811433977 [DOI] [PubMed] [Google Scholar]
- 20.Lango MN, Handorf E, Arjmand E. The geographic distribution of the otolaryngology workforce in the United States. Laryngoscope. 2017;127(1):95-101. doi: 10.1002/lary.26188 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Greenland S. Principles of multilevel modelling. Int J Epidemiol. 2000;29(1):158-167. doi: 10.1093/ije/29.1.158 [DOI] [PubMed] [Google Scholar]
- 22.Surveillance, Epidemiology, and End Results (SEER) Program. National Cancer Institute, DCCPS, Surveillance Research Program, Surveillance Systems Branch; 2014. Accessed April 28, 2020. http://www.seer.cancer.gov
- 23.Area Health Resources Files . Accessed August 25, 2020. https://data.hrsa.gov/topics/health-workforce/ahrf
- 24.Butler DC, Petterson S, Phillips RL, Bazemore AW. Measures of social deprivation that predict health care access and need within a rational area of primary care service delivery. Health Serv Res. 2013;48(2 pt 1):539-559. doi: 10.1111/j.1475-6773.2012.01449.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Howlander N, Noone A, Krapcho M, Miller D, Bishop K. SEER Cancer Statistics Review, 1975-2014. National Cancer Institute. Accessed April 29, 2020. https://seer.cancer.gov/archive/csr/1975_2014
- 26.Nakagawa S, Johnson PCD, Schielzeth H. The coefficient of determination R2 and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded. J R Soc Interface. 2017;14(134):20170213. doi: 10.1098/rsif.2017.0213 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Silver CE, Beitler JJ, Shaha AR, Rinaldo A, Ferlito A. Current trends in initial management of laryngeal cancer: the declining use of open surgery. Eur Arch Otorhinolaryngol. 2009;266(9):1333-1352. doi: 10.1007/s00405-009-1028-2 [DOI] [PubMed] [Google Scholar]
- 28.Eide MJ, Weinstock MA, Clark MA. The association of physician-specialty density and melanoma prognosis in the United States, 1988 to 1993. J Am Acad Dermatol. 2009;60(1):51-58. doi: 10.1016/j.jaad.2008.08.040 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Backhus LM, Hayanga AJ, Au D, Zeliadt SB. The effect of provider density on lung cancer survival among blacks and whites in the United States. J Thorac Oncol. 2013;8(5):549-553. doi: 10.1097/JTO.0b013e318287c24c [DOI] [PubMed] [Google Scholar]
- 30.Pagedar NA, Davis AB, Sperry SM, Charlton ME, Lynch CF. Population analysis of socioeconomic status and otolaryngologist distribution on head and neck cancer outcomes. Head Neck. 2019;41(4):1046-1052. doi: 10.1002/hed.25521 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Pytynia KB, Dahlstrom KR, Sturgis EM. Epidemiology of HPV-associated oropharyngeal cancer. Oral Oncol. 2014;50(5):380-386. doi: 10.1016/j.oraloncology.2013.12.019 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Panwar A, Batra R, Lydiatt WM, Ganti AK. Human papilloma virus positive oropharyngeal squamous cell carcinoma: a growing epidemic. Cancer Treat Rev. 2014;40(2):215-219. doi: 10.1016/j.ctrv.2013.09.006 [DOI] [PubMed] [Google Scholar]
- 33.de Almeida JR, Byrd JK, Wu R, et al. A systematic review of transoral robotic surgery and radiotherapy for early oropharynx cancer: a systematic review. Laryngoscope. 2014;124(9):2096-2102. doi: 10.1002/lary.24712 [DOI] [PubMed] [Google Scholar]
- 34.Morisod B, Simon C. Meta-analysis on survival of patients treated with transoral surgery versus radiotherapy for early-stage squamous cell carcinoma of the oropharynx. Head Neck. 2016;38(S1)(suppl 1):E2143-E2150. doi: 10.1002/hed.23995 [DOI] [PubMed] [Google Scholar]
- 35.Bhattacharyya N, Abemayor E. Combined changing patterns of hospital utilization for head and neck cancer care: implications for future care. JAMA Otolaryngol HeadNeck Surg. 2013;139(10):1043-1048. doi: 10.1001/jamaoto.2013.4525 [DOI] [PubMed] [Google Scholar]
- 36.Bhattacharyya N, Abemayor E. Patterns of hospital utilization for head and neck cancer care: changing demographics. JAMA Otolaryngol Head Neck Surg. 2015;141(4):307-312. doi: 10.1001/jamaoto.2014.3603 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eTable. List of Outlier Counties and Adjusted Odds Ratios for Surgical Treatment
eFigure. Geographic Distribution of Key Study Variables
