Key Points: Survival Outcomes with Concurrent Chemoradiation for Locally Advanced Elderly Head and Neck Cancer Patients #hncs16 @ASTRO_org

Key Points: Survival Outcomes with Concurrent Chemoradiation for Locally Advanced Elderly Head and Neck Cancer Patients #hncs16 @ASTRO_org

User Photo

4 years
Want to watch this again later?
Sign in to add this video to a playlist. Login
0 0
The NCDB is a joint project of the Commission on Cancer of the American College of

Surgeons and the American Cancer Society. It is a hospital-based registry that represents 70% of

all cancer cases in the US, drawing data from more than 1,500 commission-accredited cancer

programs. The NCDB contains detailed information on disease stage, risk-factors specific to

HNSCC cancer, and receipt of treatment including radiation dose, treatment site, and

chemotherapy delivered during the first course of treatment. The data used in the study are

derived from a de-identified 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. The NCDB has

established criteria to ensure the data submitted meet specific quality benchmarks. The following

NCDB analysis was performed with the approval of our local institutional review board.

We initially queried patients with oropharynx, larynx, and hypopharynx cancer (all

histologies) diagnosed between 1998 and 2011; nasopharynx, nasal cavity, and oral cavity cases

were not included (Fig. 1). Patients included in the primary query received RT +/- chemotherapy,

had known follow up, and had complete TNM staging; patients with metastatic disease at

presentation and those undergoing upfront surgery were excluded. The cohort was next limited to

squamous cell carcinoma histologic codes (International Classification of Disease for Oncology

[third edition] histology code 8052, 8070-8078), non-palliative cases, and only those ?71 years

of age at diagnosis. The age cutoff of ?71 years was chosen to facilitate direct comparisons to

MACH-NC.10 The resulting cohort were then limited to American Joint Committee on Cancer

(AJCC) stage III and IV (T1-2, N(+) or T3-4, N0-3) and known comorbidity score; T1-2 N0

were removed as RT alone is considered standard of care. In the final analysis, all patients

considered to have received CRT had known chemotherapy start date 14 days before or after the


Page 5 of 31 Cancer

RT start date; chemotherapy outside the 14 day window was not considered concurrent and

excluded from the final analysis.

Patient Demographics and Treatment Variables

Potentially relevant patient and treatment characteristics were included. Age was categorized as

71-76 years and ? 77 years. Race was categorized as White, African-American, and all others.

Insurance status was defined by the NCDB and included not insured, private insurance/managed

care, Medicaid, Medicare, other government, and unknown. Metropolitan, urban, and rural

residence was coded based on published files by the US Department of Agriculture Economic

Research Service. Median household income in the patient zip code was assessed as quartiles

relative to the US population. Patient comorbidities were categorized as 0, 1, or ?2 according to

Charlson-Deyo (CD) comorbidity scores.11 Institution type was classified as community cancer

program, comprehensive community cancer program, and academic/research program including

National Cancer Institute (NCI)-designated comprehensive cancer centers. Clinical T and N-
category were based on the AJCC staging guidelines.12 Each patient stage was based on the

edition corresponding to their year of diagnosis (AJCC 5th, 6th, or 7th edition). Between editions,

no changes were made to the N-category. Patients with T4a or T4b were combined as T4 in the

analysis given this was a change between AJCC editions. The primary endpoint was OS.

Statistical Analysis


Cancer Page 6 of 31

All statistical analyses were performed using SPSS V22.0 (SPSS Inc., Chicago, IL). Pearson chi-
square tests were used to assess associations between categorical variables and treatment


OS interval was calculated from the date of diagnosis to the date of death. OS was first

examined using the Kaplan Meier method. Univariate survival analysis (UVA) was performed

with the log-rank test and unadjusted Cox proportional hazards models to estimate hazard ratios

(HR); HR>1 corresponded to worse OS. Patient and clinical variables were selected a priori.

Variables included: age, receipt of concurrent chemotherapy, gender, race, insurance, residence,

distance from facility, CD-comorbidity score, facility type, year of diagnosis, tumor site, T-
category, and N-category. Multivariate Cox regression analysis was performed using OS as

outcomes with a significance level of p<0.05. The proportional hazards assumption was assessed

using a test of Schoenfeld residuals for covariates in all final models and returned no significant

results.13 Subgroup analyses including the same variables used in the Cox regression model for

the entire cohort were performed for tumor site (oropharynx, larynx, hypopharynx) and for

patients documented as receiving a complete course of RT (n=1,898). A complete course of RT

was defined as 66-81.6 Gy in 1.2-2.0 Gy per fraction based on recommendations from the

National Comprehensive Cancer Network (NCCN) guidelines and results from the Radiation

Therapy Oncology Group (RTOG) 9003 trial.14,15

Multivariate logistic regression models were used to assess the association between

receipt of CRT and patient/treatment characteristics including age, race, facility type and

duration of RT treatment (days between start and end of RT); the median of 51 days was

chosen as the cutoff for RT duration.


Page 7 of 31 Cancer

To further account for confounding variables, propensity score matching (PSM) was

performed for patients treated with CRT or RT alone accounting for the same variables used in

the multivariate model: age, receipt of concurrent chemotherapy, gender, race, insurance,

residence, distance from facility, CD-comorbidity score, facility type, year of diagnosis,

tumor site, T-category, and N-category. The propensity score was calculated using logistic

regression to estimate the probability of receiving CRT vs. RT. One-to-one propensity matching

without replacement was performed using caliper match algorithm described by Coca-
Perraillon,16 with the caliper width set to 0.05 times the standard deviation of the logit of the

propensity score.17 Survival outcomes were assessed using a log-rank test and the hazard ratio

was determined by univariate Cox regression.

Recursive Partitioning Analysis (RPA) was used to quantify the effect of CRT compared

to RT alone in all significantly different tumor subgroups. RPA was implemented in the Matlab

computing environment (Mathworks Inc, Natick, MA) using the methods described by Ciampi et

al,18 which proceeded briefly as follows. In RPA, the entire cohort of patients is broken into

subsets based on OS. The splitting of a group of patients (or a cluster in RPA) by age, stage,

comorbidity score into subclusters is a fully automated process and not determined by the

statistician and/or clinician. T-category (T1, T2, T3, T4), N-category (N0, N1, N2, N3), and

comorbidity score (0, 1, ?2) were analyzed as ordinal variables; age was continuous. At

each node, the algorithm recursively iterates through every possible binary partition of these

patients according to age, T-category, N-category, and comorbidity score. The partition that

results in the most statistically significant difference in OS is used to divide the cluster into two

subclusters, and the process is repeated. In this way, the algorithm continues to recursively

partition the total patient population into smaller and smaller subclusters until there are no


Cancer Page 8 of 31

longer any statistically significant partitions possible. Once the entire cohort of patients has been

partitioned, the effect of chemotherapy is tested in each subcluster using multivariate Cox

regression analysis (adjusted for the same factors described previously).
Up Next Autoplay