Nomogram to Predict the Benefit of IntensiveTreatment for LocoregionallyAdvancedHeadandNeck CancerLoren K. Mell1, Hanjie Shen1, Phuc Felix Nguyen-Tan2, David I. Rosenthal3, Kaveh Zakeri1,Lucas K.Vitzthum1, Steven J. Frank3, Peter B. Schiff4, AndyM.Trotti III5, James A. Bonner6,Christopher U. Jones7, Sue S. Yom8,Wade L. Thorstad9, Stuart J.Wong10,George Shenouda11, John A. Ridge12, Qiang E. Zhang13, and Quynh-Thu Le14
Abstract
Purpose: Previous studies indicate that the benefit of ther-apy depends on patients' risk for cancer recurrence relative tononcancer mortality (w ratio). We sought to test the hypoth-esis that patients with head and neck cancer (HNC) with ahigher w ratio selectively benefit from intensive therapy.
Patients and Methods: We analyzed 2,688 patients withstage III–IVB HNC undergoing primary radiotherapy (RT)with or without systemic therapy on three phase III trials(RTOG 9003, RTOG 0129, and RTOG 0522). We used gen-eralized competing event regression to stratify patients accord-ing to w ratio and compared the effectiveness of intensivetherapy as a function of predicted w ratio (i.e., w score).Intensive therapywas defined as treatment on an experimentalarm with altered fractionation and/or multiagent concurrentsystemic therapy. A nomogram was developed to predict
patients' w score on the basis of tumor, demographic, andhealth factors. Analysis was by intention to treat.
Results: Decreasing age, improved performance status,higher body mass index, node-positive status, P16-negativestatus, and oral cavity primary predicted a higher w ratio.Patients with w score �0.80 were more likely to benefit fromintensive treatment [5-year overall survival (OS), 70.0% vs.56.6%;HRof 0.73, 95%confidence interval (CI): 0.57–0.94; P¼ 0.016] than those with w score <0.80 (5-year OS, 46.7% vs.45.3%; HR of 1.02, 95%CI: 0.92-1.14; P¼ 0.69; P¼ 0.019 forinteraction). In contrast, the effectiveness of intensive therapydid not depend on risk of progression.
Conclusions: Patients with HNC with a higher w scoreselectively benefit from intensive treatment. A nomogramwasdeveloped to help select patients for intensive therapy.
IntroductionAlthough the effectiveness of intensive therapy [e.g., concurrent
chemotherapy or altered fractionation (AFX)] for locoregionallyadvanced head and neck cancer has been established, there isconsiderable controversy surrounding which subsets of patientsare most likely to benefit from this approach. In particular, theeffectiveness of intensive therapy in patients who are older, orwho have comorbidities, or who have relatively favorable riskdisease [e.g., human papillomavirus (HPV)-associated disease,nonsmokers] is unclear.
Traditionally, risk stratificationmodels used in cancer outcomesresearch have focused on the effects of treatments and risk factorson endpoints such as overall survival (OS) or progression-freesurvival (PFS). A problem is that these endpoints do not differ-entiate effects on primary events, such as disease recurrence orcancer mortality, from competing events, such as death fromcomorbid illness. As a result, suchmodels are suboptimal, becausethey pool patients at high risk for cancer events with patients athigh risk for competing events, even though these groups havedifferent expected benefit from intensive therapy (1–9). Thus,staging systems and nomograms that predict for OS and PFS arelikely to be suboptimal for selecting patients with head and neckcancer (HNC) for intensive therapeutic regimens.
Previous studies indicate that in patients with competing risks,the hazard for cancer relative to competing mortality events
1Department of Radiation Medicine and Applied Sciences, University ofCalifornia San Diego, La Jolla, California. 2Department of Radiation Oncology,Centre Hospitalier de l'Universit�e de Montreal, Montreal, Quebec, Canada.3Department of Radiation Oncology, The University of Texas MD AndersonCancer Center, Houston, Texas. 4Department of Radiation Oncology, New YorkUniversity School of Medicine, New York, New York. 5Department of RadiationOncology, Moffitt Cancer Center, Tampa, Florida. 6Department of RadiationOncology, Hazelrig-Salter Radiation Oncology Center, The University ofAlabama at Birmingham, Birmingham, Alabama. 7Radiological Associates ofSacramento, Sacramento, California. 8Department of Radiation Oncology,University of California, San Francisco, San Francisco, California. 9Departmentof Radiation Oncology, Washington University School of Medicine, St. Louis,Missouri. 10Division of Hematology Oncology, Department of Medicine, MedicalCollege of Wisconsin, Milwaukee, Wisconsin. 11Department of RadiationOncology, McGill University Health Centre, Montreal, Quebec, Canada.12Department of Surgical Oncology, Fox Chase Cancer Center, Philadelphia,Pennsylvania. 13NRG Oncology Statistics and Data Management Center,Philadelphia, Pennsylvania. 14Department of Radiation Oncology, StanfordUniversity School of Medicine, Stanford, California.
Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).
Corresponding Author: Loren K. Mell, University of California, San Diego, 3855Health Sciences Drive, MC0843, La Jolla, CA 92093. Phone: 858-246-0471;Fax: 858-822-5568; E-mail: [email protected]
(i.e., w ratio) is a key determinant of treatment benefit (7, 9–11).In particular, older patients with higher w ratios may be goodcandidates for more intensive therapy; conversely, youngerpatients with lower w ratios may not be. Further work is neededto define factors that critically affect w ratios and correlate themwith treatment effects. Correspondingly, newer methods havebeen developed to quantify effects on the w ratio, with consid-erable improvement in risk stratification compared with standardmodels (10–13). However, it is not knownwhether the benefit ofmore intensive treatment varies according to w ratio and, inparticular, whether this is a more effective method to predictwhich patients aremost likely to benefit from intensive treatment.The goal of this study was to develop a model to identify patientswith locally advanced HNC with a higher w ratio and to test thehypothesis that such patients selectively benefit from treatmentintensification.
Patients and MethodsPopulation, sampling methods, and treatment
We studied 2,688 patients with locoregionally advanced (stageIII–IVB) HNC treated on three clinical trials: RTOG 9003(NCT00771641), RTOG 0129 (NCT00047008), and RTOG0522 (NCT00265941). Details of these protocols have beenpublished previously (14–17). Written informed consent wasobtained for all patients. The study was conducted in accordancewith recognized ethical guidelines and was approved by theinstitutional review boards at all participating institutions.
Briefly, patients onRTOG9003were randomized to oneof fourarms: hyperfractionated radiotherapy (HFX: 81.6 Gy in 68 frac-tions twice a day over 7 weeks), delayed concomitant boostradiotherapy (DCB: 72 Gy in 42 fractions over 6 weeks), splitcourse radiotherapy (SC: 67.2 Gy in 42 fractions over 6 weeks), orstandard fractionation (SFX: 70 Gy in 35 fractions over 7 weeks).For the purpose of this analysis, HFX and DCB were consideredAFX, whereas SC and SFX were not. Patients on RTOG 9003 didnot receive chemotherapy. Patients on RTOG 0129were random-ized to either AFX or SFX and received chemotherapy (two cyclesof cisplatin 100 mg/m2 weeks 1 and 4 of chemoradiotherapy forpatients receiving AFX, and three cycles at the same dose weeks 1,4, and 7 for patients receiving SFX). Patients on RTOG 0522 were
randomized to either cetuximab (400 mg/m2 loading dose,followed by 250mg/m2weekly) or no cetuximab, and all patientsreceived AFX (six fractions per week) along with concurrentcisplatin (two cycles of cisplatin 100 mg/m2 weeks 1 and 4 ofchemoradiotherapy). All human investigations were performedafter approval by a local human investigations committee and inaccordance with an assurance filed with and approved by theDepartment of Health & Human Services.
OutcomesProgression-free survival time was defined as the time from
randomization to thefirst recurrence of disease, or death fromanycause, or censoring. Overall survival time was defined as the timefrom randomization to death from any cause, or censoring. Timeto recurrence and time to cancer-specificmortality were defined asthe time from randomization to first recurrence (or cancer-relatedmortality), with competing mortality events treated as censored.Time to competing mortality for recurrence was defined as timefrom randomization to death from any cause, in the absence of arecurrence event, with recurrence events treated as censored.Correspondingly, time to competing mortality for cancer mor-tality was defined as time from randomization to death from anycause, in the absence of a cancer mortality event, with cancermortality events treated as censored.
Statistical analysisThe study followed TRIPOD guidelines (18). The statistical
approach involved twomain steps: (i) development of amodel toseparate patients by w ratio and (ii) validation of the model as amethod to predict treatment effects (i.e., variation in treatmenteffects as a function ofw).Overall survival was used as the primaryoutcome assessment for model validation, because this endpointwas not used in model development and represents an outcomeof clear clinical benefit to patients.
Kaplan–Meier functions were used to plot PFS and OS andcumulative incidence functions were used to plot competingevents with respect to time. The basehaz function in R (version3.4.2) was used to estimate cumulative hazards. Forest plots wereused to analyze treatment effects within risk strata, according tointention to treat. Proportional hazards assumptions were testedusing the Grambsch–Therneau method (cox.zph function in R).
We trained risk scores for recurrence, competing mortality,and PFS using data from the control arms from the threestudies (Supplementary Fig. S1), based on the linear predictorfrom a multivariable Cox proportional hazards regres-sion (19). For RTOG 9003, the SC and SFX arms were collec-tively considered the control group. For the multivariablemodels, we selected the following candidate variables forinclusion, based on their availability in all three trials andpotential association with disease recurrence (15–17, 20, 21)and/or competing mortality (2–4, 11, 19, 22, 23): age (per 10years; continuous), female sex, black/African–American race(vs. other), white/Caucasian race (vs. other), body mass index(BMI; �20 kg/m2 vs. >20), ECOG performance status (0 vs.1–2), marital status (married vs. other/unknown), anemia(yes/no), education history (any college vs. other/unknown)—as a proxy for socioeconomic status (SES), primary site (oral cavityvs. oropharynx vs. hypopharynx vs. larynx), T stage (0–2 vs. 3 vs.4), andNstage (0 vs. 1–2a vs. 2b–2c vs. 3). Anemiawasdefined formales as a baseline hemoglobin �13.5 g/dL and for females asa baseline hemoglobin �12.5 g/dL. For patients with known
Translational Relevance
The effectiveness of intensive therapy for patients with headand neck cancer (HNC) who are older, or who have comor-bidities, or who have favorable risk human papilloma virus–associated disease is unclear. Traditional risk stratificationmodels pool patients at high risk for cancer events withpatients at high risk for competing events, even though thesegroups have different expected benefit from intensive therapy.Studies indicate that the hazard for cancer recurrence relative tocompeting mortality (w ratio) is a key determinant of treat-ment benefit, with newer regression methods developed toquantify effects on this ratio. This is the first study to examinethe effectiveness of intensive therapy for HNC as a function ofw ratio. We found that patients with a predicted w > 0.80 hadimproved overall survivalwith intensive therapy, using pooleddata from three randomized controlled trials.
smoking and tumor P16 status, we included pack-years (�10 vs.>10) and P16 (positive vs. negative) as covariates. P16 wasanalyzed as a prognostic factor for both oropharyngeal andnonoropharyngeal sites, based on several studies that have founddifferences in outcomes by P16 or HPV status in both oropha-ryngeal (20, 21) and nonoropharyngeal HNC (24–26).
All variables were normalized by subtracting the sample meanand dividing by the sample standard deviation. The mean BMIvaluewas imputed for 194patientswithmissingdata, using singleimputation. Risk scores based on the linear predictor were gen-erated taking the inner product of the coefficient vector with theindividual patient's data vector, as described previously (11). Riskstrata were defined according to quantiles of the risk scoredistribution. We compared results with a standard model devel-oped to stratify patients with oropharyngeal cancer (20, 21). Notethat this model also stratifies patients with nonoropharyngealcancer (Supplementary Fig. S2).
For modeling effects of covariates, we used generalized com-peting event (GCE) regression based on a proportional relativehazards model (11, 27, 28). A detailed description of the GCEmodeling approach is provided below. In brief, the ratio of thecause-specific hazard for recurrence (l1) versus the cause-specifichazard for competingmortality (l2) is represented aswþ, whereasthe ratio of the cause-specific hazard for recurrence (l1) to thehazard for anyprogression-free event (l1þl2) is represented asw.We use the terms w and wþ ratio to refer to observed values,whereas w and wþ score refers to values of w and wþ, respectively,predicted by the GCE model.
For GCE regression, the same variables were used as in the Coxproportional hazards models after normalization. Separateregression models were built for both cause-specific events(i.e., disease recurrence) and competing mortality (i.e., death inthe absence of disease recurrence, with the cause-specific eventtreated as censored). Treatment-related deaths were classified ascompeting mortality events. To test the sensitivity of our conclu-sions to model specification, reduce overfitting, and facilitateclinical implementation, we generated a parsimonious GCEmodel using backward stepwise regression to exclude variablesfrom the regression if P > 0.20 and by consolidating N stage (0 vs.1–3). In thismodel, only age, performance status, BMI, oral cavitysite, N stage, and P16 status had P < 0.20 and thuswere retained inthe final nomogram, which was trained on the subset of controlswith known P16 status (N ¼ 602).
GCE risk scores were generated by taking the inner product ofthe (normalized) individual patient's data vector with thedifference of the coefficient vector for cause-specific events andcompeting mortality. For 95% confidence intervals (CIs) ofestimates, we employed the gcerisk package in R (27). Risk stratawere defined according to quantiles of the GCE risk scoredistribution. Tests of treatment effects and interactions includ-ed random effects for study and age (29). All P values are two-sided.
GCE modelFor mutually exclusive events of type k, we posit the following
proportional relative hazards model:
vþk tjXð Þ ¼ vþ
k0ðtÞ exp bþ0
k GCEX� �
ðAÞ
where
vþk0ðtÞ ¼ lk0ðtÞ=Sj„k lj0ðtÞ ðBÞ
Here, lk0(t) is the baseline hazard for an event of type k,Sj„klj0(t)is the baseline cause-specific hazard for the set of events compet-ing with event type k, X is a vector of covariates, and bþk GCE is thevector of effects (coefficients) on the covariates. From this model,it can be shown that
bþk GCE ¼ ðbk � bj„kÞ ðCÞwhere bk and bj„k represent effects on the baseline hazard forevent type k and competing events, respectively, from the Cox
proportional hazard model. We use bþk GCE ¼ bk-bj„k as the esti-
mator for bþk GCE and vþk0ðtÞ ¼ Lk0ðtÞ=Lj„k 0ðtÞ, where Lk0ðtÞ and
Lj„k 0ðtÞ represent the Nelson–Aalen estimators18 for the cumu-lative hazard for event type k and the set of competing events attime t, respectively.We estimate the predicted value of vþ
k ðtjddÞ foran individual with given data vector d as:
vþk tjddð Þ ¼ vþ
k0 tð Þexp bþk GCE � dd
� �ðDÞ
Note then that expðbþk GCEÞ is the estimate of the wþ ratio, whichquantifies how the relative hazards for primary and competingevents change in response to changes in covariates.
We define the omega value as the ratio of the hazard for anevent of type k to the hazard for all events:
vk tð Þ ¼ lkðtÞ=lðtÞ ðEÞ
and estimate the predicted omega value as:
vk tjddð Þ ¼ vþk tjddð Þ= 1þ vþ
k tjddð Þ� � ðFÞ
Note that while vk ranges from 0 to 1 inclusive, vþk ranges from 0
to¥. For k¼2, a valueof vþ1 ¼ 1 means thehazard for event type1
equals the hazard for event type 2, and therefore v1 ¼ v2 ¼ 0.5. Forthe purpose of this study, we defined wþ as the ratio of the hazardfor disease recurrence to the hazard for competing mortality in theabsence of recurrence, and w as the ratio of the hazard for diseaserecurrence to the hazard for recurrence or death from any cause. Allvaluesofw areunscaledunlessotherwise specified. Scaled estimateswere obtained by factoring out the baseline wþ values.
Sample size estimatesWe used the power calculator described by Pintilie (30) to
estimate sample sizes for hypothetical randomized trials with aprimary endpoint of PFS, assuming balanced randomization,accrual time of 3 years, follow-up time of 2 years, two-sided a¼0.05, and b¼0.20.We considered two events, cancer recurrence(k¼ 1) and competing mortality (k¼ 2), and assumed an HR forcancer recurrence (q1) of 0.5 and an HR for competing mortality(q2) of 1. Under varying v1 values, we allowed the HR for anyevent (q) to vary according to the equation:
� ¼ v1� �1 þ 1� v1ð Þ ¼ vþ1 � �1 þ 1
� �= vþ
1 þ 1� � ðGÞ
Final GCE risk scoreR functions to define the GCE risk score and scaled predicted w
For this calculation, AGE is in years, BMI is in kg/cm2, ECOG12is 1 if ECOGperformance status is >0 and 0 otherwise, OC is 1 fororal cavity tumors and 0 otherwise, N0 is 1 if there is no nodalinvolvement and 0 otherwise, and P16 is 1 for P16-positivetumors and 0 otherwise. The factor 2.6 is the mean baseline wþ
estimate from the control sample.
ResultsSample characteristics are provided in Supplementary Table S1.
Comparisons of model estimates for the entire control group andthe subset with known smoking history and P16 status appear
in Table 1. Factors predicting a higher ratio were decreasing age,improved performance status, higher BMI, node-positive status,P16-negative status, and oral cavity primary. It is interesting tocompare and contrast effect estimates from Cox versus GCEmodels. Although patients with poorer OS or PFS are typicallyidentified as candidates for more intensive treatment, the GCEmodel indicates that patients with advanced age, poorer perfor-mance status, hypopharynx site, and advanced T category, forexample, have a reduced hazard for cancer events relative tocompeting mortality, implying that such patients are relatively lesslikely to benefit from treatment intensification. Moreover, somefactors, such as N3 category andmarital, education, and smokingstatus, are attenuated in theGCEmodel due tooffsetting effects onrecurrence and competing mortality.
Compared with standardmodels, GCEmodels improved strat-ification according tow ratio within each risk group, with increas-ing w from low risk to high risk according to both model pre-dictions and observations (Supplementary Table S2). Agreementbetweenpredictedw (i.e.,w score) andobservedw ratioswashigh,indicating excellent model fit and validity. The observed 3-yearw and wþ ratios for the whole cohort were 0.719 and 2.56,
Table 1. Comparison of Cox versus GCE models in all controls (left columns) and complete cases with known smoking and P16 status (right columns)
All controls (N ¼ 1352)Subset with known smoking and
respectively. The observed 3-year w and wþ ratios for the subsetwith known p16 and smoking status were 0.738 and 2.87,respectively.
As shown in Fig. 1, OS differed markedly across risk groupsdefined by standardmodels (Fig. 1A andC),whereasGCEmodelsshow little correspondence betweenOS and risk level when risk isdefined by w score (Fig. 1B and D). This suggests that, paradox-ically, patients with a better predicted survival (and higher wscore) could bemore likely to benefit from intensive treatment (byvirtue of being much less likely to die from noncancer causes).This is further shown in Fig. 2, which plots the cumulativeincidences of cancer recurrence and competing mortality withinrisk groups. Note that with standard risk stratificationmodels, theprobability of both cancer andnoncancermortality is increased inthe highest risk strata relative to the GCE model, whereas theconverse is true of the lower risk strata, further supporting GCEmodel validity. This is because while standard models aredesigned to separate groups according to PFS and OS, GCEmodels are designed to optimize the ratio of competing eventsin order to favor a particular event of interest.
Patients with the highest w score (�0.80)—representing thehighest quintile—were more likely to benefit from intensivetreatment (5-year OS, 70.0% vs. 56.6%; HR of 0.73, 95% CI:0.57–0.94;Wald P¼ 0.016) than thosewithw score <0.80 (5-yearOS, 46.7% vs. 45.3%; HR of 1.02, 95% CI: 0.92–1.14; Wald P ¼0.69; P ¼ 0.019 for interaction). For patients with known P16status, the GCE nomogram similarly identified a statisticallysignificant benefit from treatment intensification in patients withw score�0.80 (HR of 0.67; 95% CI: 0.47–0.95, Wald P¼ 0.027);in contrast, we did not find statistically significant treatmenteffects in the high-risk subgroups defined by standard modelsoverall (Fig. 3), or in any of the trials separately. Treatmentintensification was also associated with statistically significantimprovement in OS in patients with w score �0.80 in the RTOG9003 trial separately (Supplementary Fig. S3). These resultsappeared robust over a range of potential cut-off points near thew score of 0.80 (Supplementary Fig. S4A and S4B). Calibrationplots showed excellent discriminatory ability, with betterfitting athigher predicted w values (Supplementary Fig. S4C). A nomo-gram for calculating an individual's w score appears in Fig. 4.
Model estimates and performance were similar when patientswith missing BMI data were omitted from the analysis. We foundevidenceof efficiencygainswith theGCEmodel relative to standardmodelsundervaryingdefinitionsof"high risk" (Table2), due to thehigher ratio of primary to competing events. However, this analysisdoes not account for efficiency loss that could result from a lowerevent rate. Although the incidence of competing mortality waslower in the high-risk group using the GCE model, the lowerincidence of cancer recurrence offsets some of the efficiency gains,indicating correlation between primary and competing events. Assuch, GCE models could be less efficient than models designed topredict recurrence, but this conclusion was sensitive to the lack ofP16 status for themajority of the cohort. It is noteworthy, however,that sample sizeestimateswere similarwith thevariousapproaches,despite a marked reduction in the overall event rate in the "high-risk" group defined by the GCE model.
DiscussionIn this study we found that intensive treatment differentially
benefits patients with a higher relative recurrence risk (w score
�0.80). Previous studies involving HNC and other disease siteshave found that w scores could be used to identify patients with agreater likelihood to benefit from intensive therapy (3, 10–13).This study is the first to examine treatment effects within riskgroups defined by this factor. We found evidence to support thehypothesis that relative recurrence risk is an important predictorof treatment effectiveness in the HNC population.
Advantages of this study were its large sample size and largenumber of known predictors of both cancer-related and compet-ing events. Randomization also mitigated the impact of selectionbias, which presents a problem studying treatment effects in otherdata sources. A limitation of this study, however, is the hetero-geneity in the treatment and populations across the three trials."Intensive treatment" was defined relative to the baseline (con-trol) group and thus included AFX (with or without concurrentchemotherapy), or chemoradiotherapy with concurrent targetedtherapy, depending on the trial. We were also unable to controldirectly for some predictors, such as comorbidity and SES, thatlikely would have helped optimize themodel relative to standardapproaches. For example, income was prognostic for both cancerrecurrence and competing mortality but had to be omittedbecause it was not collected for all trials. Although previousstudies have found increased survival for patients undergoingtreatment at high-volume centers (31–33), radiotherapy qualityin RTOG trials is considered to be high. The incidence of non-cancer mortality in this cohort was also lower than has beenobserved in prior studies (2), indicating the exclusion of manypatients at risk for competing events.
Variation in the definition of "intensive treatment" is a poten-tial limitation of our study; however, the intent was to comparethe effectiveness of intensity with respect to the control arm, andour results were unaffected whether we applied fixed or random-effects models. Data from large randomized trials or meta-analyses involving homogeneous treatments (in particular, withan established survival benefit) will be important for furthermodel validation. Note that both RTOG 0129 and RTOG 0522failed to reject the null hypothesis; in the absence of effectivetherapy, it is not possible to identify subpopulations that wouldbenefit. Future studies involvingmore trials thatmet their primaryendpoint would be helpful to determine how treatment effectsand toxicity vary with the w ratio. However, we did observe asurvival advantage with AFX in the high-risk group from RTOG9003, lending support to the hypothesis that w score is a usefulpredictive marker.
Interpreting the effects of particular covariates in this studyshould be undertaken with caution, since CIs were fairly wide(leading to some differences in interpretation across samples).Lack of consistency and incomplete collection of key prognosticvariables hampers efforts to compare risk models, requiring us toretrain multivariable models in new samples; however, GCEmodels have previously been validated in population-basedstudies (10, 11). It should also be noted that age cutoffs �50(and >70) have been previously associatedwith a selective benefit(or lack thereof) of treatment intensification in HNC, includingthe RTOG 0522 trial included in this analysis (16, 34, 35). How-ever, age as a sole criterion for treatment selection is generally notfavored (36), because other health factors can influence theappropriate intensity of therapy. In this study, age �50 years wasnot predictive of a treatment benefit in the whole cohort.
GCE regression is a modeling approach with clear differencesrelative to standard risk stratification methods. It contrasts with
Nomogram for Intensive Therapy for Head/Neck Cancer
Overall survival by risk strata. A, Cox model in the whole cohort. B, Generalized competing event (GCE) model in the whole cohort. C, Fakhry and colleagues (21)nomogram in patients with known smoking history and P16 status. D, GCE nomogram in patients with known P16 status.
other nomograms (21, 37, 38) in that instead of predictingpatients' risk for event-free survival, which is preferable for prog-nostication, the GCE model seeks to predict the ratio of cancerevents to competing events, which is considered preferable as apredictive model. Further studies are required to establish its
advantages over standard methods, especially in the postopera-tive setting and the larger population not participating in trials,whowe expect would have differing risk for competing events. Animportant limitation is that the cutoff of 0.80 for the w score,although robust, was not chosen a priori; our results should thus
Figure 2.
Competing event incidences by risk score. A, Cox model in the whole cohort. B,Generalized competing event (GCE) model in the whole cohort. C, Fakhry andcolleagues (21) nomogram in patients with known smoking history and P16 status. D, GCEmodel in patients with known P16 status.
Nomogram for Intensive Therapy for Head/Neck Cancer
Interaction between experimental therapy andw score. A,Whole cohort. Left: w score <0.80; right: w score�0.80. B, Patients with known P16 status. Left: wscore <0.80; right: w score�0.80.
Nomogram to predict patients' relative hazard for recurrence based on GCE regression model. ECOG, Eastern Cooperative Oncology Group.
Table 2. Comparison of sample size estimates within variously defined high-risk groups
Cancer recurrence[3-year cumulativeincidence (%)]
Competing mortality[3-year cumulativeincidence (%)] HRa N
Whole cohortHighest tertileCox model for OS 48.4 25.3 0.672 442Cox model for recurrence 50.7 21.9 0.651 364GCE modela 36.6 6.4 0.574 307
Highest quintileCox model for OS 51.9 26.7 0.670 409Cox model for recurrence 54.6 24.3 0.654 346GCE modela 36.6 5.5 0.565 293
Subset with known P16 and smoking statusHighest tertileFakhry model for OS 52.3 20.7 0.642 331Cox model for recurrence 53.4 18.5 0.629 298GCE model 39.7 8.9 0.592 315
Highest quintileFakhry model for OS 54.4 24.6 0.655 351Cox model for recurrence 58.9 19.6 0.625 264GCE model 34.6 11.0 0.600 301
Abbreviations: GCE, generalized competing event; OS, overall survival.aProjected HR for recurrence or death from any cause from Eq. (G), based on observed w ratios.
www.aacrjournals.org Clin Cancer Res; 2019 OF9
Nomogram for Intensive Therapy for Head/Neck Cancer
be considered hypothesis generating and should be validated infuture studies. Ascertaining optimal cutoffs to define "high-risk"groups remains an area of investigation, especially with modelscontrolling for comorbidity and other geriatric/frailty assess-ments. Perhaps most interestingly, our findings suggest that ahigher absolute risk for recurrence/progression does not neces-sarily confer a higher likelihood to benefit from intensive therapy(or greater power to detect treatment effects). This is becausepatients with a low risk for both recurrence and competingmortality may benefit as much from aggressive treatmentapproaches as patients with high risk for both events.
In summary, here we propose a method to predict an under-reported but meaningful quantity for individual patients (i.e.,relative recurrence risk, or w ratio), with a clinically relevantinterpretation (i.e., a value >50% means the individual's hazardfor cancer recurrence exceeds thehazard for competingmortality).Our findings indicate that patients with a higher relative recur-rence risk, indicated by a w score �0.80, selectively benefit fromintensive therapy. This approach is being implemented prospec-tively in the NRG-HN004 trial, alongwith a nomogram to informclinical practice and trial design (comogram.org). Furtherresearch, however, is needed to optimize GCE models and toascertain which patients derive the greatest benefit from intensivetherapy.
Disclosure of Potential Conflicts of InterestD.I. Rosenthal is an employee of Merck. S.J. Frank is an employee of
Boston Scientific and Varian Medical; reports receiving commercial researchgrants from Hitachi, Eli Lilly, and Elekta; and holds ownership interest(including patents) in C4 Imaging. J.A. Bonner is an employee of Bristol-Myers Squibb, Eli Lilly, Merck Serono, and Cel-Sci. S.S. Yom is an employeeof Galera; reports receiving commercial research grants from Bristol-MyersSquibb, Merck, Genentech, and BioMimetix; and reports receiving other
remuneration from Springer and UpToDate. No potential conflicts of inter-est were disclosed by the other authors.
Authors' ContributionsConception and design: L.K. Mell, H. Shen, Q.-T. LeDevelopment of methodology: L.K. Mell, H. Shen, K. Zakeri, S.J. WongAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): L.K. Mell, P.F. Nguyen-Tan, D.I. Rosenthal,A.M. Trotti III, J.A. Bonner, C.U. Jones, S.S. Yom, W.L. Thorstad, S.J. Wong,G. Shenouda, J.A. Ridge, Q.E. ZhangAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): L.K. Mell, H. Shen, D.I. Rosenthal, K. Zakeri,L.K. Vitzthum, S.J. Frank, P.B. Schiff, A.M. Trotti III, J.A. Bonner, S.J. Wong,G. Shenouda, J.A. Ridge, Q.E. ZhangWriting, review, and/or revision of the manuscript: L.K. Mell, H. Shen,P.F. Nguyen-Tan, D.I. Rosenthal, K. Zakeri, L.K. Vitzthum, S.J. Frank,P.B. Schiff, A.M. Trotti III, J.A. Bonner, C.U. Jones, S.S. Yom, W.L. Thorstad,S.J. Wong, G. Shenouda, J.A. Ridge, Q.E. Zhang, Q.-T. LeAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): L.K. Mell, L.K. Vitzthum, J.A. Bonner, S.J. WongStudy supervision: L.K. Mell
AcknowledgmentsThis project was supported by grants U10CA180868 (NRG Oncology
Operations), U10CA180822 (NRG Oncology SDMC) from the NationalCancer Institute (NCI; NRG Oncology/RTOG 9003, NCT00771641, https://clinicaltrials.gov/ct2/show/NCT00771641, NRG Oncology/RTOG 0129,NCT00047008, https://clinicaltrials.gov/ct2/show/NCT00047008, and NRGOncology/RTOG 0522, NCT00265941, https://clinicaltrials.gov/ct2/show/NCT00265941), and Eli Lilly.
The costs of publicationof this articlewere defrayed inpart by the payment ofpage charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Received June 4, 2019; revised July 9, 2019; accepted August 13, 2019;published first August 16, 2019.
References1. Argiris A, Brockstein BE, Haraf DJ, Stenson KM, Mittal BB, Kies MS, et al.
Competing causes of death and second primary tumors in patients withlocoregionally advanced head and neck cancer treated with chemora-diotherapy. Clin Cancer Res 2004;10:1956–62.
2. Mell LK, Dignam JJ, Salama JK, Cohen EE, Polite BN, Dandekar V, et al.Predictors of competingmortality in advanced head and neck cancer. J ClinOncol 2010;28:15–20.
3. Rose BS, Jeong JH, Nath SK, Lu SM, Mell LK. Population-based study ofcompeting mortality in head and neck cancer. J Clin Oncol 2011;29:3503–9.
4. KwonM, Roh JL, Song J, Lee SW, Kim SB, Choi SH, et al. Noncancer healthevents as a leading cause of competing mortality in advanced head andneck cancer. Ann Oncol 2014;25:1208–14.
5. Mell LK, Weichselbaum RR. More on cetuximab in head and neck cancer.N Engl J Med 2007;357:2201–2.
6. Dignam JJ, Kocherginsky MN. Choice and interpretation of statistical testsused when competing risks are present. J Clin Oncol 2008;26:4027–34.
7. Mell LK, Jeong JH. Pitfalls of using composite primary end points in thepresence of competing risks. J Clin Oncol 2010;28:4297–9.
8. Mell LK, Zakeri K, Rose BS. On lumping, splitting, and the nosology ofclinical trial populations and end points. J Clin Oncol 2014;32:1089–90.
9. Mell LK, Carmona R, Gulaya S, Lu T, Wu J, Saenz CC, et al. Cause-specificeffects of radiotherapy and lymphadenectomy in stage I-II endometrialcancer: a population-based study. J Natl Cancer Inst 2013;105:1656–66.
10. Carmona R, Gulaya S, Murphy JD, Rose BS, Wu J, Noticewala S, et al.Validated competing event model for the stage I-II endometrial cancerpopulation. Int J Radiat Oncol Biol Phys 2014;89:888–98.
11. Carmona R, Zakeri K, Green G, Hwang L, Gulaya S, Xu B, et al. Improvedmethod to stratify elderly patients with cancer at risk for competing events.J Clin Oncol 2016;34:1270–7.
12. Zakeri K, Rose BS, D'Amico AV, Jeong JH, Mell LK. Competing events andcosts of clinical trials: analysis of a randomized trial in prostate cancer.Radiother Oncol 2015;115:114–9.
13. Zakeri K, Rose BS, Gulaya S, D'Amico AV, Mell LK. Competing event riskstratification may improve the design and efficiency of clinical trials:secondary analysis of SWOG 8794. Contemp Clin Trials 2013;34:74–9.
14. Fu KK, Pajak TF, Trotti A, Jones CU, Spencer SA, Phillips TL, et al. ARadiation Therapy Oncology Group (RTOG) phase III randomized studyto compare hyperfractionation and two variants of accelerated fraction-ation to standard fractionation radiotherapy for head and neck squamouscell carcinomas: first report of RTOG 9003. Int J Radiat Oncol Biol Phys2000;48:7–16.
15. Nguyen-Tan PF, Zhang Q, Ang KK, Weber RS, Rosenthal DI, Soulieres D,et al. Randomized phase III trial to test accelerated versus standardfractionation in combination with concurrent cisplatin for head and neckcarcinomas in the Radiation Therapy Oncology Group 0129 trial: long-term report of efficacy and toxicity. J Clin Oncol 2014;32:3858–66.
16. Ang KK, Zhang Q, Rosenthal DI, Nguyen-Tan PF, Sherman EJ, Weber RS,et al. Randomized phase III trial of concurrent accelerated radiation pluscisplatin with or without cetuximab for stage III to IV head and neckcarcinoma: RTOG 0522. J Clin Oncol 2014;32:2940–50.
17. Beitler JJ, ZhangQ, Fu KK, Trotti A, Spencer SA, Jones CU, et al. Final resultsof local-regional control and late toxicity of RTOG 9003: a randomizedtrial of altered fractionation radiation for locally advanced head and neckcancer. Int J Radiat Oncol Biol Phys 2014;89:13–20.
Mell et al.
Clin Cancer Res; 2019 Clinical Cancer ResearchOF10
18. Collins GS, Reitsma JB, Altman DG,Moons KG. Transparent reporting of amultivariable prediction model for individual prognosis or diagnosis(TRIPOD): the TRIPOD statement. Br J Cancer 2015;112:251–9.
19. Cox DR. Regression models and life tables. J R Stat Soc Series B StatMethodol 1972;B34:187–220.
20. Ang KK, Harris J, Wheeler R, Weber RS, Rosenthal DI, Nguyen-Tan PF, et al.Human papillomavirus and survival of patients with oropharyngeal can-cer. N Engl J Med 2010;363:24–35.
21. Fakhry C, Zhang Q, Nguyen-Tan PF, Rosenthal DI, Weber RS, Lambert L,et al. Development and validation of nomograms predictive of overall andprogression-free survival in patients with oropharyngeal cancer. J ClinOncol 2017;35:4057–65.
22. Zakeri K, MacEwan I, Vazirnia A, Cohen EE, Spiotto MT, Haraf DJ, et al.Race and competing mortality in advanced head and neck cancer.Oral Oncol 2014;50:40–4.
23. Park A, Alabaster A, Shen H, Mell LK, Katzel JA. Undertreatment of womenwith locoregionally advanced head and neck cancer. Cancer 2019;125:3033–9.
24. Chung CH, Zhang Q, Kong CS, Harris J, Fertig EJ, Harari PM, et al. p16protein expression and human papillomavirus status as prognostic bio-markers of nonoropharyngeal head and neck squamous cell carcinoma.J Clin Oncol 2014;32:3930–8.
25. Bryant AK, Sojourner EJ, Vitzthum LK, Zakeri K, Shen H, Nguyen C, et al.Prognostic role of p16 in nonoropharyngeal head and neck cancer. J NatlCancer Inst 2018;110:1393–9.
26. Tian S, Switchenko JM, Jhaveri J, Cassidy RJ, Ferris MJ, Press RH, et al.Survival outcomes by high-risk human papillomavirus status in nonor-opharyngeal head and neck squamous cell carcinomas: a propensity-scored analysis of the National Cancer Data Base. Cancer 2019;125:2782–2793.
27. Shen H, Carmona R, Mell LK. Generalized competing event model: gceriskR package. R (CRAN). Available from: https://cran.r-project.org/.
28. LunnM,McNeil D. ApplyingCox regression to competing risks. Biometrics1995;51:524–32.
29. Michiels S, Baujat B, Mah�e C, Sargent DJ, Pignon JP. Random effectssurvival models gave a better understanding of heterogeneity in individualpatient data meta-analyses. J Clin Epidemiol 2005;58:238–45.
30. Pintilie M. Competing risks: a practical perspective. Chichester, England:John Wiley & Sons; 2006. p.115–26.
31. Wuthrick EJ, ZhangQ,MachtayM, Rosenthal DI, Nguyen-Tan PF, Fortin A,et al. Institutional clinical trial accrual volume and survival of patients withhead and neck cancer. J Clin Oncol 2015;33:156–64.
32. Boero IJ, Paravati AJ, Xu B, Cohen EE, Mell LK, Le QT, et al. Importance ofradiation oncologist experience among patients with head-and-neck can-cer treated with intensity-modulated radiation therapy. J ClinOncol 2016;34:684–90.
33. David JM, Ho AS, Luu M, Yoshida EJ, Kim S, Mita AC, et al. Treatment athigh-volume facilities and academic centers is independently associatedwith improved survival in patients with locally advanced head and neckcancer. Cancer 2017;123:3933–42.
34. Bourhis J, Overgaard J, Audry H, Ang KK, Saunders M, Bernier Jet al. Meta-Analysis of Radiotherapy in Carcinomas of Head and neck (MARCH)Collaborative Group. Hyperfractionated or accelerated radiotherapy inhead and neck cancer: a meta-analysis. Lancet 2006;368:843–54.
35. Pignon JP, le Ma�tre A, Maillard E, Bourhis J; MACH-NC CollaborativeGroup. Meta-analysis of chemotherapy in head and neck cancer (MACH-NC): an update on 93 randomised trials and 17,346 patients.Radiother Oncol 2009;92:4–14.
36. Wildiers H, Mauer M, Pallis A, Hurria A, Mohile SG, Luciani A, et al. Endpoints and trial design in geriatric oncology research: a joint Europeanorganisation for research and treatment of cancer–Alliance for ClinicalTrials in Oncology–International Society Of Geriatric Oncology positionarticle. J Clin Oncol 2013;31:3711–8.
37. Wang SJ, Patel SG, Shah JP, Goldstein DP, Irish JC, Carvalho AL, et al. Anoral cavity carcinoma nomogram to predict benefit of adjuvant radiother-apy. JAMA Otolaryngol Head Neck Surg 2013;139:554–9.
38. Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms inoncology: more than meets the eye. Lancet Oncol 2015;16:e173–80.
www.aacrjournals.org Clin Cancer Res; 2019 OF11
Nomogram for Intensive Therapy for Head/Neck Cancer
Published OnlineFirst August 16, 2019.Clin Cancer Res Loren K. Mell, Hanjie Shen, Phuc Felix Nguyen-Tân, et al. Locoregionally Advanced Head and Neck CancerNomogram to Predict the Benefit of Intensive Treatment for
Updated version
10.1158/1078-0432.CCR-19-1832doi:
Access the most recent version of this article at:
Material
Supplementary
http://clincancerres.aacrjournals.org/content/suppl/2021/02/25/1078-0432.CCR-19-1832.DC1Access the most recent supplemental material at:
E-mail alerts related to this article or journal.Sign up to receive free email-alerts
To order reprints of this article or to subscribe to the journal, contact the AACR Publications
Permissions
Rightslink site. (CCC)Click on "Request Permissions" which will take you to the Copyright Clearance Center's
.http://clincancerres.aacrjournals.org/content/early/2019/10/18/1078-0432.CCR-19-1832To request permission to re-use all or part of this article, use this link
