AJCC indicates American Joint Committee on Cancer; HNSCC, head and neck squamous cell carcinoma; HPV, human papillomavirus; M0, without distant metastases; and M1, with distant metastases.
The number of cases per year is defined in the Results subsection Facility Volume Thresholds.
The number of cases per year is defined in the Results subsection Facility Volume Thresholds. Moderate-volume facilities (MVFs) were the reference for comparison. CoC indicates Commission on Cancer; Nonoropharyngeal HNSCC, nonoropharyngeal head and neck squamous cell carcinoma; HPV, human papillomavirus; HVF, high-volume facility; LVF, low-volume facility; oropharyngeal HNSCC, oropharyngeal head and neck squamous cell carcinoma. Full Cox regressions may be found in eTables 6, 9, 12, and 15 in the Supplement.
eTable 1. Survival for 10 Groups With an Approximately Equal Number of Events
eTable 2. Survival for Low-, Moderate-, and High-volume Facilities
eTable 3. Facility Characteristics Stratified by Facility Volume for the Eligible Cohort
eTable 4. Patient Demographics, Facility Characteristics, and Oncologic Factors Stratified by Facility Volume for the Whole Cohort
eTable 5. Patient Demographics, Facility Characteristics, and Oncologic Factors Stratified by Facility Volume for the All Patients Treated at 1 CoC Facility
eTable 6. Multivariate Cox Regression For Survival for All Patients and All Patients Treated at 1 CoC Facility
eTable 7. Localized Stage (I/II) Patient Demographics, Facility Characteristics, and Oncologic Factors Stratified by Facility Volume
eTable 8. Advanced Stage (III/IV) Patient Demographics, Facility Characteristics, and Oncologic Factors Stratified by Facility Volume
eTable 9. Multivariate Cox Regression Stratified by Localized/Advanced Status
eTable 10. Nondistant Metastatic Patient Demographics, Facility Characteristics, and Oncologic Factors Stratified by Facility Volume
eTable 11. Distant Metastatic Patient Demographics, Facility Characteristics, and Oncologic Factors Stratified by Facility Volume
eTable 12. Multivariate Cox Regression Stratified by M Status
eTable 13. Nonoropharyngeal HNSCC Demographics, Facility Characteristics, and Oncologic Factors Stratified by Facility Volume
eTable 14. 2010-2014 Oropharyngeal HNSCC With Known HPV Status Demographics, Facility Characteristics, and Oncologic Factors Stratified by Facility Volume
eTable 15. Multivariate Cox Regression Stratified by Nonoropharyngeal HNSCC and Oropharyngeal HNSCC With Known HPV Status)
Customize your JAMA Network experience by selecting one or more topics from the list below.
Torabi SJ, Benchetrit L, Kuo Yu P, et al. Prognostic Case Volume Thresholds in Patients With Head and Neck Squamous Cell Carcinoma. JAMA Otolaryngol Head Neck Surg. 2019;145(8):708–715. doi:10.1001/jamaoto.2019.1187
Are prognostic thresholds identifiable for facility case volume in the treatment of patients with head and neck squamous cell carcinoma?
In this US nationally based cohort study of 250 229 patients at 1229 facilities with head and neck squamous cell carcinoma, improvements in patient survival were reported at moderate-volume facilities (>54 to ≤165 cases per year). Additional improvements in survival were reported at high-volume facilities (>165 cases per year).
Facility case volume thresholds that may support the use of quality benchmarks for treatment of patients with head and neck squamous cell carcinoma.
Though described as an important prognostic indicator, facility case volume thresholds for patients with head and neck squamous cell carcinoma (HNSCC) have not been previously developed to date.
To identify prognostic case volume thresholds of facilities that manage HNSCC.
Design, Setting, and Participants
Retrospective analysis of 351 052 HNSCC cases reported from January 1, 2004, through December 31, 2014, by Commission of Cancer–accredited cancer centers from the US National Cancer Database. Data were analyzed from August 1, 2018, to April 5, 2019.
Treatment of HNSCC at facilities with varying case volumes.
Main Outcomes and Measures
Using all-cause mortality outcomes among adult patients with HNSCC, 10 groups with increasing facility case volume were created and thresholds were identified where group survival differed compared with each of the 2 preceding groups (univariate log-rank analysis). Groups were collapsed at these thresholds and the prognostic value was confirmed using multivariable Cox regression. Prognostic meaning of these thresholds was assessed in subgroups by category (localized [I/II] and advanced [III/IV]), without metastasis (M0), with metastasis (M1), and anatomic subsites (nonoropharyngeal HNSCC and oropharyngeal HNSCC with known human papillomavirus status).
Of 250 229 eligible patients treated at 1229 facilities in the United States, there were 185 316 (74.1%) men and 64 913 (25.9%) women and the mean (SD) age was 62.8 (12.1) years. Three case volume thresholds were identified (low: ≤54 cases per year; moderate: >54 to ≤165 cases per year; and high: >165 cases per year). Compared with the moderate-volume group, multivariate analysis found that treatment at low-volume facilities (LVFs) was associated with a higher risk of mortality (hazard ratio [HR], 1.09; 99% CI, 1.07-1.11), whereas treatment at high-volume facilities (HVFs) was associated with a lower risk of mortality (HR, 0.92; 99% CI, 0.89-0.94). Subgroup analysis with Bonferroni correction revealed that only the moderate- vs low- threshold had meaningful differences in outcomes in localized stage (I/II) cancers, (LVFs vs moderate-volume facilities [MVFs]: HR, 1.09 [99% CI, 1.05-1.13]; HVF vs MVF: HR, 0.95 [99% CI, 0.90-1.00]), whereas both thresholds were meaningful in advanced stage (III/IV) cancers (LVF vs MVF: HR, 1.09 [99% CI, 1.06-1.12]; HVF vs MVF: HR, 0.91 [99% CI, 0.88-0.94]). Survival differed by prognostic thresholds for both M0 (LVF vs MVF: HR, 1.09 [99% CI, 1.07-1.12]; HVF vs MVF: HR, 0.91 [99% CI, 0.89-0.94]) and nonoropharyngeal HNSCC (LVF vs MVF: HR, 1.10 [99% CI, 1.07-1.13]; HVF vs MVF: HR, 0.93 [99% CI, 0.90-0.97]) site cases, but not for M1 (LVF vs MVF: HR, 1.00 [99% CI, 0.92-1.09]; HVF vs MVF: HR, 0.94 [99% CI, 0.83-1.07]) or oropharyngeal HNSCC cases (when controlling for human papillomavirus status) (LVF vs MVF: HR, 1.10 [99% CI, 0.99-1.23]; HVF vs MVF: HR, 1.07 [99% CI, 0.94-1.22]).
Conclusions and Relevance
Higher volume facility threshold results appear to be associated with increases in survival rates for patients treated for HNSCC at MVFs or HVFs compared with LVFs, which suggests that these thresholds may be used as quality markers.
Head and neck cancer incidence is estimated at 64 690 cases per year in the United States and resulted in more than 13 000 deaths reported in 2018.1 Because of disease heterogeneity and intricate anatomy, treatment depends on highly specialized clinicians and advanced multidisciplinary teams to coordinate surgery, radiotherapy, and chemotherapy. The National Comprehensive Cancer Network guidelines specify that “outcomes are improved when patients with head and neck cancers are treated in high-volume centers.”2 and a rapidly expanding body of literature3-17 supports this statement.
Treatment at a high-volume facility (HVF) was reported to be associated with improved long-term survival in patients with locally advanced head and neck squamous cell carcinoma (HNSCC), including patients undergoing definitive radiotherapy3 or laryngectomy,4 elderly patients with surgically treated laryngeal cancer,5 and patients with localized laryngeal cancer.6,7 Care at HVFs was also reported to be associated with improvements in short-term mortality for patients with surgically treated head and neck cancers.8,9 In practice, HVFs reported better health care professional adherence to radiotherapy protocols for advanced-stage HNSCC,10 and to guidelines for treatment of elderly patients with laryngeal cancer.11 Cases at HVFs were associated with fewer delays in initiation of radiotherapy treatment for laryngeal cancer12 and decreased prolongation of radiotherapy treatment duration of hypopharyngeal and oropharyngeal cancers.13,14 In addition, high volume was independently associated with reduced rates of positive surgical margins in HNSCC and salivary malignancies.15-17
However, case volume definitions span a wide range and lack specific thresholds. Low-volume facilities (LVFs) are defined as facilities with fewer than 1 to 20 cases per year, and HVFs manage a minimum of 3 to 66 cases per year,3,5-16 making generalizability or standardization for policy or clinical decision making particularly challenging. Research toward establishing volume thresholds associated with survival outcomes across all HNSCC subsites and treatments is lacking. We aimed to identify prognostic thresholds for facility case volume in the treatment of HNSCC.
Data were obtained from the National Cancer Database (NCDB) from January 1, 2004, through December 31, 2014. Data were analyzed from August 1, 2018, to April 5, 2019. This database collaboration between the American College of Surgeons and the Commission on Cancer (CoC) collected data from more than 1500 US hospitals, representing more than 70% of all new diagnoses of cancer as previously described.18 The NCDB has been used in a number of studies of head and neck cancers.19-22 This study was exempt from institutional review by the Yale Human Investigation Committee because of use of deidentified data.
The study population included patients of all ages with squamous cell carcinoma at any head and neck subsite, identified by histology codes 8070-8073 from the International Classification of Disease for Oncology, Third Edition (ICD-O-3). Patients were categorized using the ICD-O-3 topologic site codes by subsites: oral cavity codes C00.0-C00.9 (lip), C02.0-C02.9 (all tongue sites excluding base of tongue), C03.0-C03.9 (gums), C04.0-C04.9 (floor of mouth), C05.0-C05.9 (palate), and C06.0-C06.9 (other and unspecified parts of mouth); oropharynx codes C01.9 (base of tongue), C09.0-C09.9 (tonsil), and C10.0-C10.9 (oropharynx); parotid and other major salivary glands codes C07.9 (parotid) and C08.0-C08.9 (other and unspecified major salivary glands); nasopharynx codes C11.0-C11.9; hypopharynx codes C12.9 (pyriform sinus) and C13.0-13.9 (hypopharynx); ill-defined areas in oral cavity and pharynx codes (C14.0-C14.8); sinonasal tract codes C30.0-C30.1 (nasal cavity) and C31.0-C31.9 (accessory sinuses); and larynx codes C32.0-C32.9. Patients with unknown or unclear (ie, classified as stage IV, not otherwise specified) American Joint Committee on Cancer (AJCC) clinical staging, missing vital status, unknown facility location and type, or missing follow-up were excluded (Figure 1). In subgroup analyses, patients were also excluded for unknown metastasis in metastatic (M1) and nonmetastatic (M0) subgroups, for an AJCC clinical stage of 0 in localized or advanced stage subgroups, or for unknown, unclear (human papillomavirus [HPV]–positive [risk and type not stated]), or low-risk HPV status in the oropharyngeal HNSCC subgroup.
Insurance was categorized as government (Medicaid, Medicare, or other government insurance), private (private insurance or managed care), or uninsured patients and patients with unknown insurance status. Facilities were categorized into academic facilities (academic or research programs including National Cancer Institute–designated comprehensive cancer centers) and nonacademic facilities (community cancer programs, comprehensive community cancer programs, and integrated network cancer programs). Facility location was grouped by region: North East (New England and Middle Atlantic); South (South Atlantic); Central (East North Central, East South Central, West North Central, and East North Central); and West (Mountain and Pacific). Income quartile was reported as the median household income from the patient’s area of residence, based on the 2012 American Community Survey.23 For subgroup analysis on oropharyngeal HNSCC, HPV status was assigned as either HPV-negative cases including those diagnosed without HPV-disease or HPV-positive cases including those with non-16–, non-18–high-risk HPV, HPV-16, HPV-18, HPV-16 and HPV-18, and high-risk HPV, not otherwise specified.
Mean annual facility volume was calculated before applying exclusion criteria to accurately characterize the values. Number of cases per facility was summed and divided by the number of years since the institution initially entered data in the NCDB, which may have been as early as 2004, up to a maximum of 11 years. Since the NCDB collected all cancer cases from its contributing hospitals, these values represented actual case volume.18
After applying exclusion criteria, patients were divided into 10 groups according to increasing facility volume (eTable 1 in the Supplement). These groups were determined by setting an approximately equal number of events (deaths) per group before application of exclusion criteria. Each group’s survival was compared with the preceding 2 groups via univariate log-rank analysis. Whenever a group had significantly increased survival compared with both of the preceding groups, a threshold was created. A certain number of thresholds were not set a priori. We used the aforementioned methods as a pragmatic way to minimize differences within groups and maximize differences between groups, while maintaining a large sample size in each of the initial volume groupings. Groups were then collapsed according to the identified thresholds and compared with one another via univariate log-rank analysis.
Age, sex, race/ethnicity, insurance, income, facility type, facility location, Charlson-Deyo comorbidity condition score, year of diagnosis, primary tumor site, clinical stage, tumor grade, facility type, and facility locations were compared by facility volume using χ2 analyses. Univariate survival analyses comparing the prognostic groups were performed via Kaplan-Meier 2-tailed log-rank tests. Thresholds were validated using the multivariate Cox proportional hazards regression model controlling for variables. Human papillomavirus was also analyzed within the oropharyngeal HNSCC subgroup. Effect size and 99% CIs were used where appropriate to express the magnitude of the difference and precision of the estimate. Because the NCDB may receive multiple records for the same patient treated by multiple institutions, it assigned the facility based on recency of patient contact or the most complete record; thus, thresholds were validated among patients reportedly treated at only 1 CoC facility. Subgroup analyses were also performed for patients with (M1) and without (M0) distant metastases, localized stage (I/II) and advanced stage (III/IV) cancers, and nonoropharyngeal HNSCC sites and oropharyngeal HNSCC sites with known HPV status. Data analyses was performed using SPSS, version 25.0 (IBM Corporation). Statistical significance was set at P < .00122 after a Bonferroni correction adjusting for the number of comparisons in a single Cox regression was applied.
Of 351 052 HNSCC cases identified from 1234 facilities, facility volume ranged from 0.13 to 342.18 cases per year, with a median of 15.36 (interquartile range [IQR], 8.44-29.82) cases and a mean (SD) of 26.82 (36.51) cases. After exclusions, 250 229 cases from 1229 facilities remained, ranging from 0.50 to 342.18 cases per year, with a median of 15.36 (IQR, 8.52-29.82) cases and a mean (SD) of 26.93 (36.54) cases.
Patients were allocated into 10 groups as described in the Methods section (eTable 1 in the Supplement). In univariate analysis, group 6, which contained 11 facilities that treated from 54.45 cases to fewer than 81.64 cases per year, had a meaningfully improved mean (SE) survival of 6.45 (0.04) years compared with group 5 (40.00 to <54.45 cases per year; mean [SE], 6.01 [0.04] years) and group 4 (31.09 to <40.00 cases per year; mean [SE], 6.02 [0.04] years). We found similar survival difference between group 9 (165.45-342.18 cases per year; mean [SE], 6.58 [0.04] cases per year) and groups 8 (121.09 to <165.45 cases per year; 6.23 [0.04] cases per year) and 7 (81.64 to <121.09 cases per year; 6.19 [0.04] cases per year). Thus, thresholds were created between groups 5 and 6 and between groups 8 and 9, and were collapsed accordingly.
Collapsing categories left 3 facility-volume groups. After exclusion criteria, LVFs (treating <54.45 cases per year) contained a total of 1099 facilities (of 1104 LVFs reported in the database) that treated 144 016 eligible patients. The LVFs treated a median number of 13.71 (IQR, 7.82-23.55) cases per year and the mean (SD) was 17.06 (11.94) cases per year. Moderate-volume facilities (MVFs) (managing from 54.45 to <165.45 cases per year) contained 111 facilities that treated 76 460 patients. The MVFs managed a median of 81.64 (IQR, 63.18-109.36) cases per year and a mean (SD) of 89.70 (29.63) cases per year. The HVFs (managing ≥165.45 cases per year) contained 19 facilities that treated 29 753 cases per year. The HVFs treated a median of 216.82 (IQR, 193.45-272.33) cases per year and a mean (SD) of 231.10 (52.96) cases per year. As shown in Figure 2, a smaller proportion of patients were treated at LVFs in 2014 compared with 2004 (54.2% vs 60.1%), whereas the proportion of patients treated in MVFs (32.1% vs 28.4%) and HVFs (13.7% vs 11.5%) increased during this time. Survival differences were found between the groups. Mean (SE) survival was 6.01 (0.02) years in the LVF group, 6.31 (0.02) years in the MVF group, and 6.58 (0.04) years in the HVF group. Univariate survival curves for these groups are presented in Figure 3. Among the 3 volume groups, no differences were noted in facility location; notably, no HVFs were located in the West. A large percentage of HVFs (≥47.5%) and MVFs (70.3%) were academic, whereas most LVFs (87.8%) were nonacademic.
The LVFs cared for older patients (≥64 years: LVF, 45.5%; MVF, 39.4%; and HVF, 39.1%), patients from a zip code with a lower socioeconomic status (≥$63 000 median income: LVF, 25.7%; MVF, 29.9%; and HVF, 27.4%), and patients with government insurance (LVF, 55.8%; MVF, 51.8%; and HVF, 48.2%), as opposed to private insurance. After controlling for these characteristics in a multivariate Cox proportional hazards model, treatment at LVFs was associated with higher risk of mortality vs treatment at MVFs (HR, 1.09; 99% CI, 1.07-1.11). Treatment at HVFs was associated with a lower risk of mortality vs MVFs (HR, 0.92; 99% CI, 0.89-0.94) (Figure 4). When looking specifically at cases treated at only 1 CoC facility (n = 204 537), the thresholds held (LVF vs MVF: HR, 1.09 [99% CI, 1.07-1.12] and HVF vs MVF: HR, 0.90 [99% CI, 0.87-0.93]) (Figure 5).
We identified 93 865 patients with localized stage cancer. In univariate analysis, we found meaningful survival difference based on facility volume in patients with localized stage disease. The 5-year survival (SE) in HVFs was 65.8% (0.6%); in MVFs, 64.4% (0.3%); and in LVFs, 60.6% (0.2%) (Figure 5A). The LVFs were independently associated with decreased survival compared with MVFs after Bonferroni correction (HR, 1.09; 99% CI, 1.05-1.13). However, no survival difference was found between HVFs and MVFs after Bonferroni correction (HR, 0.95; 99% CI, 0.90-1.00) (Figure 4).
We identified 148 066 patients with advanced stage cancer. The MVFs and HVFs had a higher proportion of stage IV cases compared with LVFs (LVF, 66.8%; MVF, 72.6%; and HVF, 71.5%). Univariate analysis identified a meaningful survival difference based on facility volume in this cohort. The 5-year survival (SE) in HVFs was 50.7% (0.4%); in MVFs, 46.5% (0.3%); and in LVFs, 43.5% (0.2%) (Figure 5B). After Bonferroni correction, LVFs were associated with decreases in survival compared with MVFs (HR, 1.09; 99% CI, 1.06-1.12), and HVFs were associated with increased survival compared with MVFs (HR, 0.91; 99% CI, 0.88-0.94) (Figure 4).
We identified 236 172 patients with M0 cancer. Univariate analysis revealed meaningful survival differences based on facility volume in the M0 cohort. The 5-year survival (SE) was 57.9% (0.3%) in HVFs; 54.8% (0.2%) in MVFs; and 52.5% (0.2%) in LVFs (Figure 5C). These prognostic thresholds were also significant on multivariate analysis in these patients. Compared with MVFs, treatment at LVFs was independently associated with decreased survival (HR, 1.09; 99% CI, 1.07-1.12), while treatment at HVFs was associated with increased survival (HR, 0.91; 99% CI, 0.89-0.94) (Figure 4).
We identified 7490 patients with M1 cancer. The 5-year survival (SE) was 12.4% (1.5%) in HVFs, 11.7% (0.8%) in MVFs, and 11.7% (0.5%) in LVFs (Figure 5D). Multivariate analysis also showed no difference in survival based on the thresholds before Bonferroni correction (LVF vs MVF: HR, 1.00 [99% CI, 0.92-1.09] and HVF vs MVF: HR, 0.94 [99% CI, 0.83-1.06]) (Figure 4).
We identified 172 638 patients with nonoropharyngeal HNSCC as well as 16 350 patients with oropharyngeal HNSCC with known HPV status. After excluding cases with unknown HPV status, HVFs were more likely to treat HPV-positive patients compared with MVFs and LVFs within the oropharyngeal HNSCC cohort (HPV-positive: HVF, 72.5%; MVF, 62.8%; and LVF, 55.6%). As in previous analyses, the thresholds held for the nonoropharyngeal HNSCC cohort (LVF vs MVF: HR, 1.10 [99% CI, 1.07-1.13] and HVF vs MVF: HR, 0.93 [99% CI, 0.90-0.97]) (Figure 4). However, after controlling for HPV status in the oropharyngeal HNSCC cohort, we found HPV to be the main driver for survival, not facility volume. Being HPV-positive carried a HR of 0.39 (99% CI, 0.36-0.43) compared with being HPV-negative. Whereas LVFs were associated with decreased survival compared with MVFs before Bonferroni correction, no difference was seen after correction (HR, 1.10; 99% CI, 0.99-1.23). The HVFs were also not associated with improved survival compared with MVFs (HR, 1.07; 99% CI, 0.94-1.22) (Figure 4).
Multiple studies3-17 have reported the prognostic association of facility case volume on HNSCC outcomes and guideline adherence. To our knowledge, this is the first study to determine facility case volume thresholds for all HNSCCs. We identified 2 potential prognostic thresholds for outcomes at approximately 54 cases per year and approximately 165 cases per year. Multivariate analysis confirmed that our thresholds remained predictive after adjustment for many potentially confounding factors. This remained true in a subgroup of patients treated at only 1 CoC facility; thus, it may be used as a quality measure for hospital selection. Whereas the mean survival increase was modest (approximately 6 months mean survival difference between HVF and LVF), HNSCC affects more than 60 000 patients a year,1 and 6 months represents a substantial amount of time across all patients. Our findings were especially relevant given the increasing treatment of patients at MVFs and HVFs during the last decade and the current transition of the US health care system from fee-for-service to value-based reimbursement.24
The mechanism by which facility case volume is associated with improvements in patient survival has been the subject of review articles.25,26 It may be that high hospital volume translates to a higher volume per clinician, which may be especially important for oncologist,27,28 radiologist,29 surgeon,30-32 and radiation oncologist33-35 expertise in diagnosis and management of many cancers. In addition, such facilities may have more subspecialization, which has been shown with more accurate diagnoses of breast cancer by specialized breast pathologists.36 The HVFs may also have the resources to create clinical care pathways37 and effective multidisciplinary teams and tumor boards that may help in decision making38 and evidence-based guideline adherence.10,39 The HVFs may be better equipped with ancillary services such as dieticians and social services, which may independently improve outcomes and quality of life.40-42 Regardless of the specific factors, evidence suggests that facility volume is a useful quality marker. While exact volumes have proven previously elusive, this study helped define specific threshold cutoffs, which can be used to facilitate prognostication and to improve HNSCC care.
Our findings were consistent among a variety of subgroups performed to better characterize patients who were more likely to benefit from HVF referral. Subgroup analysis revealed that only the MVF vs LVF thresholds were associated with significant differences in outcomes among the patients with localized stage (I/II) cancer, whereas both thresholds were significant among the patients with advanced stage (III/IV) cancer. This outcome suggests that patients with more complex issues should be treated at highly experienced institutions, whereas patients with less complex issues may not need the same level of specialty care. While we found facility volume-based survival differences for patients with M0 cancer, none was identified for patients with M1 HNSCC. This may be indicative of the worse prognosis facing patients with metastases.43 Since these patients are often treated with palliative intent,43 our analysis suggested that this group need not travel long distances for treatment at HVFs. Granted, we were unable to assess quality-of-life outcomes in this analysis.
Whereas HSNCC facility volume thresholds may exist, we found no association with survival within an oropharyngeal HNSCC cohort. The HPV positivity was a main factor associated with improved survival, consistent with the literature.44,45 However, a previous Radiation Therapy Oncology Group randomized trial10 secondary analysis of 471 patients with known HPV-status (including 266 patients with oropharyngeal HNSCC) reported an overall survival benefit in HVFs, even when controlling for HPV status (LVF vs HVF: HR, 1.91; 95% CI, 1.37-2.65). Thus, our results may reflect inadequate power because of limited sample size, shorter follow-up compared with other subgroups (HPV data collection for NCDB began in 2010), and the missing association of patients with unknown HPV status. The HVFs were also more likely to treat HPV-positive patients within our analysis, though it is important to note that in a post hoc analysis of all oropharyngeal HNSCC cases from 2010 through 2014 (without excluding cases with unknown or low-risk HPV status), it was noted that unknown or low-risk HPV cases were more likely to be in the LVF group compared with the MVF and HVF groups (unknown or low risk HPV: HVF, 45.3%; MVF, 52.2%; and LVF, 63.0%), indicating that perhaps HPV reporting or HPV testing within LVFs may be lower compared with HVFs, not necessarily that they are less likely to see these patients. Differences in HPV testing and reporting may also have precluded us from observing a volume-based survival difference. Future analyses may, indeed, show a survival difference. However, the results of our study support regionalization of patients with HNSCC, especially those with advanced-stage, M0 disease.
Nonetheless, it may not be feasible to regionalize HNSCC cancer care to 130 MVFs and HVFs. As indicated in our study, LVFs tended to care for larger proportions of low-income patients with government insurance and increased age. Referral to HVFs, likely more centralized institutions, may result in further financial, transportation, and emotional hardships. Solely focusing on care regionalization may risk neglecting efforts aimed at improving care structures in LVFs. In addition, negative perception of LVF outcomes may discourage continuous reporting of case volume geared to the creation of cancer databases necessary for clinical outcome studies.46 Thus, concurrent processes and initiatives aimed at mitigating the survival gap between LVFs and HVFs is imperative. Initiatives, such as virtual tumor boards in small centers, to provide high-quality, multidisciplinary expertise at institutions lacking the necessary personnel for in-house implementation may help reduce such disparities.47,48
Our study had several strengths. We identified thresholds without a priori assumptions, allowing us to find 2 points of inflection in survival. Our high-volume threshold was set at approximately 165 cases per year; a value not only high, but which encompassed only 19 facilities in the database. Given the complexity of HNSCC, including multiple anatomic subsites and a myriad of surgical procedures and medical treatment variations, it was possible that HVF expertise was achieved only after treating substantially large cohorts of patients and may explain the high 165 cases per year threshold.
There were many study limitations. We included all subsites and stages with no consistent clinical guidelines, and we did not control for treatment modality in the multivariate model. In addition, our database was limited by the absence of certain important variables, such as patient smoking status, because of missing data or complete absence from the database. Furthermore, patients seen at multiple CoC facilities were arbitrarily assigned to a single facility within the database based on recency of patient contact or the most complete treatment records. Though we attempted to mitigate this limitation with a subgroup analysis of patients reportedly treated at only 1 CoC facility, this analysis was also limited since patients may have sought treatment at a non-CoC facility not captured by the NCDB. Identification of only 130 MVFs and HVFs suggested possible confounding bias; those seeking care at these centers may have a higher level of engagement and stronger support systems associated with better compliance and utilization of ancillary services.
We proposed novel prognostic thresholds for HNSCC facility case volume. Independent of multiple oncologic and system factors, patient survival appeared to be improved at centers that treated approximately 54 or more cases per year and further improved at centers that treated approximately 165 or more cases per year. Our thresholds appear to support regionalization of care to HVFs to improve patient outcomes.
Corresponding Author: Benjamin L. Judson, MD, Section of Otolaryngology, Department of Surgery, Yale University School of Medicine, 330 Cedar St, PO Box 208062, New Haven, CT 06520-8062 (email@example.com).
Published Online: June 13, 2019. doi:10.1001/jamaoto.2019.1187
Author Contributions: Mr Torabi and Dr Judson had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Torabi, Kuo Yu, Cheraghlou, Tate, Judson.
Acquisition, analysis, or interpretation of data: Torabi, Benchetrit, Savoca, Tate, Judson.
Drafting of the manuscript: Torabi, Benchetrit, Savoca.
Critical revision of the manuscript for important intellectual content: Benchetrit, Kuo Yu, Cheraghlou, Tate, Judson.
Statistical analysis: Torabi, Cheraghlou.
Obtained funding: Torabi.
Administrative, technical, or material support: Judson.
Supervision: Kuo Yu, Savoca, Tate, Judson.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study was supported by a grant T35HL007649 (Dr Torabi) from the National Heart, Lung, and Blood Institute.
Role of the Funder/Sponsor: The National Heart, Lung, and Blood Institute had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer: The data used in the study are derived from a deidentified NCDB file. The American College of Surgeons and the Commission on Cancer have not verified and are not responsible for the analytic or statistical methodology employed, or the conclusions drawn from these data by the investigator.