Commercial Claims–Based Comparison of Survival and Toxic Effects of Definitive Radiotherapy vs Primary Surgery in Patients With Oropharyngeal Squamous Cell Carcinoma

Key Points

Question  Is there a difference in survival and toxic effects in patients with oropharyngeal squamous cell carcinoma treated with definitive radiotherapy vs primary surgery?

Findings  In this study of 884 patients, no clinically meaningful differences in survival were found between patients treated with radiotherapy vs surgery. Because of the more common use of concurrent chemotherapy, radiotherapy significantly increased short- but not long-term gastrostomy dependence; however, other toxic effects were not different, and the payer- and patient-borne costs were large but not different between modalities.

Meaning  The findings suggest that primary radiotherapy is associated with increased risk of short-term gastrostomy use, but differences in survival were not clinically meaningful and other long-term toxic effects were not apparent.

Importance  Definitive radiotherapy (RT) and primary surgery (PS) are considered to be equally viable local therapy modalities for oropharyngeal squamous cell carcinoma (OPSCC). The comparative effectiveness of these therapies is often debated, and treatment decisions are based on a paucity of comparative data.

Objective  To examine the differences in overall survival and key toxic effects in patients with OPSCC treated with RT and PS.

Design, Setting, and Participants  This retrospective cohort analysis used the HealthCore Integrated Research Database to identify 884 patients diagnosed with OPSCC from January 1, 2007, to December 31, 2014. Patients were categorized as receiving definitive RT (with or without chemotherapy) or PS (with or without adjuvant RT or chemoradiotherapy). Administrative claims data were linked with state cancer registries from California, Connecticut, Georgia, Kentucky, New York, and Ohio. Data analysis was performed from February 29, 2016, to February 6, 2018.

Exposures  Definitive RT or PS.

Main Outcomes and Measures  Overall survival was analyzed using Cox proportional hazards regression. Risks of gastrostomy dependence, esophageal stricture, and osteoradionecrosis were determined through claims and analyzed using logistic regression.

Results  A total of 884 patients (608 [68.8%] in the RT group and 276 [31.2%] in the PS group; mean [SD] age, 61.5 [10.7] years; 727 [82.2%] male; 842 [95.3%] white) were included in this study. The 3-year overall survival was 76% among patients treated with RT and 81% among patients treated with PS (hazard ratio, 0.76; 95% CI, 0.54-1.01). On multivariable analysis, increasing age, female sex, and low income were associated with inferior survival; treatment type was not. Patients treated with RT were more likely to have gastrostomy dependence within the first year (391 [64.3%] vs 127 [46.0%]; adjusted OR, 0.57; 95% CI, 0.42-0.77). After treating chemotherapy as an effect modifier, there was no difference between modalities. Treatment type was not associated with esophageal stricture or osteoradionecrosis risk. Mean costs were approximately $100 000 for payers and $5000 for patients, with no adjusted differences between RT and PS.

Conclusions and Relevance  This study suggests that RT and PS are equally viable treatment options for OPSCC; therefore, local therapy decisions may be individualized to each patient. However, the frequent addition of chemotherapy was associated with increased gastrostomy dependence among patients undergoing RT, which may be relevant in clinical decision making.

During the past 20 years, there has been an increase in cases of oropharyngeal squamous cell carcinoma (OPSCC) attributable to human papillomavirus (HPV) infection. The prognosis of HPV-positive OPSCC is superior to that of nonvirally mediated disease, and the outcomes are so favorable that novel treatment approaches are now focused on reducing toxic effects rather than increasing efficacy.

Both primary surgery (PS) and definitive radiotherapy (RT) with or without chemotherapy are viable local therapies for OPSCC. Postoperative RT or chemoradiotherapy (CRT) are typically administered after surgery, with the exact indication determined by the pathologic findings. For patients treated with primary RT, CRT is the standard recommendation in eligible patients with locally advanced disease.¹ One potential advantage of initial surgery is, therefore, the possibility of avoiding concurrent chemotherapy. Significant technological innovations in PS and RT have led to continuously improving therapeutic ratios with these treatments. Transoral surgery, often performed with robotic assistance (transoral robotic surgery), has led to faster recovery times and less tissue destruction,² and intensity-modulated radiation therapy significantly improves xerostomia.³

No randomized clinical trials comparing PS and definitive RT have been published. Two meta-analyses^4,5 suggest similar survival outcomes between the 2 local therapy options, but they report differences in adverse effects, such as gastrostomy and tracheostomy tube placement, esophageal stenosis, hemorrhage, and fistula. Given the significant increase in the prevalence of OPSCC in an otherwise healthy population, it is imperative to define comparative outcomes of these dueling local therapies administered with assumed curative intent. In this study, we linked 6-state cancer registry data with commercially insured health plan member data and performed a comparative effectiveness analysis of PS vs definitive RT, focusing on survival, toxic effects (such as gastrostomy dependence), and cost.

The cohort was created by linking 6-state cancer registry information from California, Connecticut, Kentucky, Georgia, Ohio, and New York with claims data from the HealthCore Integrated Research Database (HIRD), which collects all claims associated with patients covered by commercial health plans from Anthem. The HIRD contains health plan eligibility files and fully adjudicated inpatient and outpatient medical claims from facilities and medical professionals. Pharmacy claims are limited to outpatient dispensing. Data from each registry were matched with claims from clients in HIRD by using a set of patient identifiers, specifically, first and last name, date of birth, sex, and Social Security numbers, and the final database was then merged into the analytic data set. This study was conducted in full compliance with relevant provisions of the Health Insurance Portability and Accountability Act. Researchers accessed the analytic data set as a limited data set, and no patients were contacted directly. The study was approved by an investigational review board oversight under the provisions of Privacy Rule 45 CFR §164.514(e). The key claims used in the data dictionary are given in the eTable in the Supplement. All data were deidentified, and consent was waived by each institutional review board associated with each state registry and by the University of Texas Southwestern Medical Center Institutional Review Board and the New England Institutional Review Board.

The eFigure in the Supplement details the inclusion and exclusion characteristics that defined the cohort composition. Patients were 18 years or older and had received a diagnosis of squamous cell carcinoma of the oropharynx from January 1, 2007, to December 31, 2014. Data analysis, including the development and compilation of the data set, was performed from February 29, 2016, to February 6, 2018. Continuous health plan coverage was required from 1 year before diagnosis to 1 year after diagnosis. The Surveillance, Epidemiology, and End Results (SEER) historical stage was used to define the patient’s stage, given the relatively high number of missing or unknown American Joint Committee on Cancer stages (n = 416) in the 6 state registries. Because some locally advanced primary tumors were coded as distant historical stage in the SEER classification of disease stage (eg, extension to larynx or hypopharynx, mandible, extrinsic tongue musculature, other bone, or hard palate), patients with this stage of disease were included in the analysis. The first treatment (surgery, RT, or chemotherapy) was required to have started within 6 months of diagnosis.

Claims for at least 20 RT fractions were required for patients to have been coded as receiving curative-intent RT within 12 weeks of starting treatment; shorter regimens may have been palliative in intent. Concurrent CRT was defined by a chemotherapy claim at some point between 3 weeks before starting RT until the end of RT. Chemotherapy administered before 3 weeks before RT was considered to be induction therapy unless the interval between the last chemotherapy claim and the beginning of RT was longer than 90 days, in which case patients were excluded from analysis. Adjuvant RT was assigned after surgery if the first RT claim was present within 90 days of surgery, and adjuvant CRT was assigned if chemotherapy was administered between 3 weeks before RT to the last RT claim. Patients were dichotomized into receipt of definitive RT (with or without chemotherapy) or PS.

Patients who underwent a diagnostic tonsillectomy without neck dissection were classified as receiving primary RT. Diagnostic lymph node excisions were not considered as PS; only neck dissections were considered to be a curative-intent operation. Patients were designated as having nontransoral surgery if there was a surgical code associated with any type of composite resection, mandible or jaw surgery (eg, mandibulotomy), or reconstruction or if the patient underwent a complete or total glossectomy. All other patients were assumed to have undergone transoral surgery, including patients with a robotic code.

Overall survival was calculated from the date of diagnosis to the date of death reported in state registries. All patients were analyzed for toxic effect end points (gastrostomy dependence, esophageal stricture, and osteoradionecrosis) using claims data from the beginning of treatment to 1 year from the end of treatment (the last day of RT or the PS date if no RT was given). Gastrostomy dependence was evaluated at 3 time points after the end of treatment: occurring at any time within 12 months, present 6 months or longer, and present at 12 months. Gastrostomy dependence required a code for placement of the tube plus the claim for enteral nutrition; the date of last use was the date of the last claim for nutrition. The presence of a clinically relevant early esophageal stricture was defined as a dilation code within the first year for surviving patients. The presence of a clinically relevant bone toxic effect (osteoradionecrosis) was coded using the method of Beadle et al.⁶ One year was chosen in the primary analysis for these latter 2 toxic effects because subsequent procedures may be more likely attributable to disease progression. For toxic effect analyses at 6 and 12 months, the denominators were patients alive at those time points. Plan- and patient-borne costs during the first year were calculated and compared between patients treated with RT and PS.

Patient, disease, and treatment characteristics were compared between definitive RT and PS using the χ² test for categorical variables, and the Mood median test was used to compare age groups. Overall survival was estimated using the Kaplan-Meier method. Univariable and multivariable Cox proportional hazards regression analyses were used to assess the association between the overall survival and patient, disease, and treatment characteristics. The Charlson-Deyo comorbidity score was calculated according to a previously validated algorithm in claims data.⁷ Variables with a univariable P ≤ .20 were entered into the multivariable regression model, without additional selection. Hazard ratios (HRs) and 95% CIs were reported. Univariable logistic regression and multivariable logistic regression were used to identify the indicators for toxic effect end points at 1 year. To further understand the mechanism of these toxic effects, multivariable regression models were developed with an interaction term between chemotherapy use and primary treatment. Multivariable cost modeling and mean cost differences were estimated using generalized linear models with γ distribution and log link function. All analyses were conducted using SAS Enterprise Guide, version 7.1 (SAS Institute Inc). Statistical significance was set at a 2-sided P = .05.

A total of 884 patients (608 [68.8%] in the RT group and 276 [31.2%] in the PS group; mean [SD] age, 61.5 [10.7] years; 727 [82.2%] male; 842 [95.3%] white) were included in this study. Patient characteristics are given in Table 1. The historical stage was localized in 186 cases (21.0%) and advanced in 681 cases (77.0%), with 17 cases (1.9%) having unknown or missing stages. Definitive RT was used in 608 patients (68.6%); of those patients treated with primary RT, RT alone was implemented in 57 (9.4%), induction chemotherapy in 105 (17.3%), and concurrent chemotherapy in 446 (73.4%). Of the patients treated with PS, 141 (51.1%) were treated with adjuvant CRT, 54 (19.6%) with adjuvant RT alone, and 81 (29.3%) with no adjuvant treatment surveillance. Patients receiving definitive RT were generally older and presented with more advanced disease. Patients undergoing surgery were more likely to have a tonsil primary site. Only 19 patients underwent nontransoral surgery; therefore, all surgical procedures are reported together.

Median follow-up for surviving patients was 32.8 months (interquartile range, 19.5-53.1 months). Table 2 details the 3-year overall survival and the univariable and adjusted variables associated with overall survival. There were several clinical and socioeconomic variables associated with adverse overall survival on univariable analysis, including older age (oldest quartile vs the youngest: HR, 2.47; 95% CI, 1.44-4.23), female sex (HR, 1.97; 95% CI, 1.45-2.70), and higher comorbidity score (≥2 vs 0: HR, 1.68; 95% CI, 1.23-2.30). Patients with tonsil site (tonsil vs base of tongue site: HR, 0.71; 95% CI, 0.52-0.97), no tobacco exposure (HR, 0.33; 95% CI, 0.15-0.74), and higher socioeconomic status (highest income quartile vs the lowest: HR, 0.54; 95% CI, 0.37-0.80) experienced improved survival.

However, on multivariable analysis, only older age (oldest quartile vs the youngest: HR, 1.81; 95% CI, 1.17-2.80) and female sex (HR, 1.69; 95% CI, 1.19-2.39) were significantly associated with inferior overall survival. Patients with higher socioeconomic status (highest income quartile vs the lowest: HR, 0.68; 95% CI, 0.42-1.11) experienced improved survival.

Enteral feeding was evaluated using 3 metrics: any enteral feeding within the first year of diagnosis, gastrostomy use in the past 6 months from the end of the last treatment, and gastrostomy use in the past 12 months from the end of the last treatment. A total of 518 patients (58.6%) experienced enteral feeding within the first year of diagnosis. On multivariable analysis, tobacco exposure (odds ratio [OR], 1.50; 95% CI, 1.07-2.12) and advanced stage (OR, 2.51; 95% CI, 1.77-3.57) were associated with an increased risk of any gastrostomy dependence. Patients with tonsil site (tonsil vs base of tongue site: OR, 0.65; 95% CI, 0.48-0.88) were less likely to require gastrostomy (Table 3). In addition, PS was associated with an absolute reduction of 18.3% (95% CI, 15.6%-20.8%) and a significantly decreased risk of gastrostomy use (OR, 0.57; 95% CI, 0.42-0.77). In a secondary model in which chemotherapy was included as an independent variable, the use of PS was no longer associated with a reduction in gastrostomy (OR, 1.20; 95% CI, 0.95-1.51).

By 6 months after the end of treatment (Table 4), only 114 of 756 patients (15.1%) from the entire cohort were still using gastrostomy feedings. Independent variables associated with medium-term dependence were restricted to tobacco exposure. The absolute decrease in gastrostomy use after PS compared with definitive RT was modest (absolute difference, 5.8%; OR, 0.67; 95% CI, 0.41-1.08).

After 1 year from the end of treatment, 63 of 658 patients (9.6%) were gastrostomy dependent. The absolute difference between definitive RT and PS was 3.6% (10.7% RT vs 7.1% PS) and nonsignificant on univariable and multivariable analyses. There were no variables substantially associated with long-term dependence in this analysis.

Fifty-three patients (6.0%) underwent esophageal dilation (for presumed stricture) within the first year. Only advanced age (fourth vs first quartile: OR, 3.72; 95% CI, 1.55-9.00) was associated with stricture on multivariable analysis. Treatment type (PS vs RT: OR, 1.02; 95% CI, 0.54-1.94) was not associated with the risk of dilation.

In total, 74 patients (8.4% of the cohort) experienced a bone toxic effect within the first year of completing treatment. There were no independent variables associated with bone toxic effects, including primary treatment modality (PS vs RT: OR, 0.92; 95% CI, 0.54-1.56).

The total adjusted payer (commercial plan paid) costs for the first year were $107 949 for definitive RT and $103 897 for PS (adjusted difference, $5969; 95% CI, −$3836 to $17 066) (Table 5). The total adjusted out-of-pocket (patient paid) costs for the first year were $4861 for definitive RT and $4686 for PS (adjusted difference, $562; 95% CI, −$158 to $1443). Most patient-borne costs were attributable to outpatient out-of-pocket costs, although approximately $700 was spent on prescriptions during the first year in both cohorts. In the adjusted comparisons of plan-paid and patient-paid costs, inpatient costs were significantly higher for surgical patients and outpatient costs significantly higher for patients undergoing RT.

Given that oropharyngeal cancer is afflicting a younger and generally healthier population that is medically capable of receiving most modalities of treatment, efficacy and toxic effect comparisons between RT and PS are particularly important. In this study, we found that the 2 local treatments yielded comparable absolute survival outcomes; a modest absolute survival advantage with surgery was observed, although the wide CI around the estimate prevents any conclusions. Additional research with a larger sample size and more granular details on patient stage, HPV status, and comorbidity is needed to provide a more definitive answer regarding comparative treatment effectiveness. The difference in these treatments was the significant increase in enteral feedings among patients receiving primary RT at some point during the first year of therapy. However, this absolute difference was clinically and statistically insignificant by 6 months after the end of treatment. The risks of other well-known complications of head and neck cancer therapy (esophageal stricture and osteonecrosis) were low and nearly equivalent between the 2 therapies. Moreover, the payer- and patient-borne costs of therapy were similar.

Retrospective comparative analyses are compromised by small sample sizes and selection bias but generally confirm equivalent survival with different adverse effect profiles. For example, Ling et al⁸ performed a single-institution retrospective analysis that incorporated patient-reported quality-of-life outcomes. In this analysis, more than 40% of patients received transoral robotic surgery alone, and patients receiving PS experienced superior salivary outcomes. However, the results for RT and surgery plus adjuvant RT were similar. No other comparisons were statistically significant. Yeh and colleagues⁹ performed a largely qualitative systemic review of CRT and PS, again finding similar oncologic outcomes with more RT-related adverse effects in the population that received definitive RT. Although adjuvant RT adverse effects, such as mucositis, were poorly described, there was a distinct increase in gastrostomy use in the primary RT population. However, this article⁹ specified that patients undergoing transoral robotic surgery had lower T and N stage, which may explain much of this difference.

Our analysis provides additional support for a decrease in short-term gastrostomy use with initial surgery. Because this was a claims-based analysis with limited patient-specific information, this increase in gastrostomy use with definitive RT may be explained by unmeasured confounding factors, such as performance status and lower-volume disease, or a general bias toward elective gastrostomy in patients treated with definitive RT. In any case, the result provides important information to discuss with patients who are debating between the 2 therapies. From a mechanistic perspective, once chemotherapy use was considered as an effect modifier, the primary local therapy became nonsignificant, indicating that concurrent CRT, in the definitive or adjuvant setting, may be the primary driver of gastrostomy use. Because avoidance of a gastrostomy is a frequent consideration in deciding between the 2 therapies, initial surgical resection and subsequent avoidance of chemotherapy may be favored.

We also found several other variables associated with short- and long-term gastrostomy use, including tobacco exposure, base of tongue primary site, and stage. Of interest, tobacco exposure was associated with gastrostomy use at any point and at 6 months, a finding not typically seen in dysphagia studies. This result may be associated with the improved performance status of nonsmokers and patients with HPV-positive disease,¹⁰ or there may be a biological explanation to the finding, such as reduced oxygenation to key swallowing structures.

Finally, this analysis found no meaningful difference in cost between the 2 primary treatment modalities. One retrospective study¹¹ suggested that transoral surgery was cost saving vs RT, but this single-institutional analysis did not clarify the source of the cost and did not have the patient numbers to adjust for baseline patient characteristics. In contrast, we found no meaningful difference in first-year payer- or patient-borne costs after multivariable adjustment between PS or definitive RT. The increased outpatient cost with RT was counterbalanced by the increased hospitalization costs from initial surgery. The first-year out-of-pocket cost for these patients was more than $4000, which is a high sum for patients experiencing a potentially debilitating disease and treatment, highlighting the importance of considering financial factors in discussing treatment with patients.¹²

The treatment landscape of favorable HPV-positive OPSCC is changing rapidly, with multiple prospective studies^13,14 already suggesting that RT dose de-escalation is oncologically sound. Because there are multiple paradigms of treatment deintensification, including primary transoral robotic surgery and reduced-intensity adjuvant therapy (including none at all), it is unclear how the comparative effectiveness of these 2 therapies will evolve. The short-term and long-term adverse effects of both approaches will certainly improve over time, and the prospective collection of patient-reported outcomes is important to facilitate meaningful comparisons between treatments.

This analysis has methodologic limitations inherent in retrospective and claims-based studies. Perhaps most important is that HPV status was not available for each patient, which is a significant limitation given the association of HPV status with survival. One recent, large database analysis¹⁵ suggested that survival outcomes are comparable between modalities for HPV-positive patients but improved with surgery for HPV-negative patients; we were not able to further evaluate this question. Similarly, the available stage information may not have captured the true extent of the disease, especially in the population receiving definitive RT, which has typically more advanced disease. Because claims data cannot be used to accurately assign an RT dose, one cannot be entirely confident that all patients receiving RT received curative-intent treatment, which would bias the survival results against the RT cohort. Many important data were not patient specific, such as indexes of socioeconomic status, and the patient comorbidity score was based on patient claims and may not accurately reflect true performance status or medical condition; the claims likely did not reflect any functional deficit from the disease. These latter limitations are particularly important given that swallowing function and the need for gastrostomy are associated with pretreatment performance status, dysphagia, and weight loss, which were not captured in the claims or registry databases. Moreover, although this analysis used cancer registry–based disease data, the data were not particularly granular. Finally, because surgical patients are typically chosen because of their smaller-volume disease and presumed superior recovery and rehabilitation potential, unmeasured confounders are almost certainly going to favor the surgical cohort, affecting survival and toxic effect results. That most outcomes were equivalent suggests that these treatment modalities are similar in most long-term functional domains.

The analysis of this unique data set suggests that there are comparable oncologic, toxic effects, and cost outcomes after definitive RT or PS for patients with OPSCC. It appears that surgical therapy may reduce the need for enteral support during and shortly after treatment, and this benefit is probably attributable to sparing patients concurrent CRT. However, an open and patient-specific question is whether sequelae from surgery, such as hemorrhage or fistula, are warranted to minimize gastrostomy insertion, especially when the difference may be largely short term. Randomized clinical trials of RT vs transoral surgery in patients with localized OPSCC may help to resolve this comparison in a subset of patients.¹⁶ Pending the results of these trials, the present data suggest that these competing local therapies are equally viable options for this disease, with the final decision hinging on patient-specific medical and disease characteristics and, ultimately, individual preference.

Accepted for Publication: July 2, 2018.

Corresponding Author: David J. Sher, MD, MPH, Department of Radiation Oncology, University of Texas Southwestern Medical Center, 5810 Forest Park Dr, Dallas, TX 75390 (david.sher@utsouthwestern.edu).

Published Online: September 20, 2018. doi:10.1001/jamaoto.2018.1929

Author Contributions: Dr Sher had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Sher, DeVries.

Acquisition, analysis, or interpretation of data: Sher, Agiro, Zhou, Day.

Drafting of the manuscript: Sher, Day.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Sher, Agiro, Zhou, Day.

Obtained funding: Sher, DeVries.

Administrative, technical, or material support: Sher, Agiro, DeVries.

Supervision: Sher, Day, DeVries.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.

Funding/Support: This study was supported by research grant ROI2015-915 from the Radiation Oncology Institute.

Role of the Funder/Sponsor: The funding source 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.

Meeting Presentation: This study was presented at the American Society for Radiation Oncology Annual Meeting; September 24, 2017; San Diego, California.