Prognostic factors for recurrence and survival in anal cancer

Generating hypotheses from the mature outcomes of the first United Kingdom Coordinating Committee on Cancer Research Anal Cancer Trial (ACT I)

Authors


Abstract

BACKGROUND:

Only 2 prospective studies have previously reported prognostic factors for anal cancer, European Organization for Research and Treatment of Cancer trial 22861 (EORTC 22861) and Radiation Therapy Oncology Group trial 98-11 (RTOG 98-11). Both of those trials reported that clinically positive lymph nodes and male sex predicted poorer overall survival (OS). The EORTC 22861 trial indicated that the same factors were prognostic for locoregional control. In the current report, the authors investigated potential prognostic factors from the first United Kingdom Coordinating Committee on Cancer Research Anal Cancer Trial (ACT I), in which patients were randomized to receive either radiotherapy alone or chemoradiation (CRT) with concurrent 5-fluorouracil/mitomycin C.

METHODS:

In the ACT I trial, associations between several baseline characteristics and 3 endpoints were investigated: locoregional failure (LRF), anal cancer death (ACD), and OS. The analyses were restricted to 292 patients who received CRT, which subsequently became standard treatment. A score was derived using multivariable Cox regression to identify the set of factors that, together, had the best prognostic performance. This score was then validated with a large, independent prospective trial (the ACT II trial).

RESULTS:

Palpable, clinically positive lymph nodes were associated with LRF (P = .012), a greater risk of ACD (P = .031), and decreased OS (P = .006) in multivariable analyses. Men had worse outcomes than women for LRF (P = .036), ACD (P = .039), and OS (P = .008). On average, a lower hemoglobin level had an adverse effect on ACD (P = .008), and a higher white blood cell count had an adverse effect on OS (P = .001). However, external validation of the score was poor for LRF (area under the curve [AUC] = 54%) but was better for ACD (AUC = 67%) and OS (AUC = 63%).

CONCLUSIONS:

The results from this analysis of the ACT I trial supported evidence for palpable lymph nodes and male sex as prognostic factors for LRF and OS, and lower hemoglobin levels and a higher white blood cell count were identified as prognostic factors for ACD and OS, respectively. Cancer 2013. © 2012 American Cancer Society.

INTRODUCTION

Epidermoid cancer of the anus (ECA) is uncommon and may demonstrate an indolent natural history with a low rate of distant metastases.1-4 The first United Kingdom Coordinating Committee on Cancer Research (UKCCCR) Anal Cancer Trial (ACT I) compared chemoradiation (CRT) (5-fluorouracil [5-FU] and mitomycin C [MMC] plus and radiotherapy [RT]) with RT alone in patients with ECA and confirmed the benefit of CRT.4, 5 Contemporary trials support this advantage.6, 7 Recent trials using cisplatin have failed to improve on this schedule.8-11

The outcome of patients with ECA varies, but several trials have suggested that CRT results in a locoregional recurrence-free survival rate of approximately 60% to 80%4, 6-8, 10 and an overall survival (OS) rate of 65% to 78%5, 7, 11 at 3 to 5 years. Prognostic factors represent easily measurable objective characteristics obtained from long-term observations within a population that can offer information on possible outcome, irrespective of treatment. Knowledge of validated clinical and molecular prognostic variables in anal cancer can be relevant to the individual patient, can refine and allow individualization of different treatment strategies, and can direct rational follow-up.

To our knowledge, European Organization for Research and Treatment of Cancer trial 22861 (EORTC 22861)6 and the Radiotherapy Treatment and Oncology Group trial 98-11 (RTOG 98-11)12 are the only 2 randomized studies to report analyses of prognostic factors. Results from EORTC 22861 indicated that skin ulceration, lymph node involvement, and male sex were independent variables associated with locoregional failure (LRF) and OS in multivariate analysis.6 CRT reduced the LRF rate but did not improve OS.6

The RTOG 98-11 analysis could not establish prognostic factors for LRF but supported 2 of the poor prognostic factors previously reported by the EORTC (clinically involved lymph nodes and men) and also established a tumor size >5 cm in greatest dimension as an independent variable predicting both poorer disease-free survival (DFS) and OS.12 These factors have yet to be validated in an additional prospective data set.

The objective of the current article was to define potentially new prognostic factors from analysis of the ACT I trial that can be validated on the large ACT II prospective trial (n = 940). Our findings would add to evidence of prognostic factors established by the EORTC 228616 and RTOG 98-1112 studies.

MATERIALS AND METHODS

We conducted an analysis of patients with ECA in the UKCCCR ACT I trial. Patients were randomized to receive either RT alone (n = 285) or CRT (n = 292) combined with 5-FU during the first and final weeks of radiotherapy and MMC as a single intravenous bolus injection on day 1 of the first cycle of chemotherapy between 1987 and 1994. Disease was staged according to the 1985 International Union Against Cancer (UICC) classification.13 Full details of the trial design, procedures, and CRT schedules have been previously published.4

The primary endpoint of the ACT I trial was local failure, either from disease or from complications of treatment, as indicated by the need for major surgical intervention. Five-year OS was a secondary endpoint. Death notifications in the United Kingdom were received from the Office for National Statistics.

In the current analysis, we considered 3 endpoints: 1) LRF at 2 years (because 85% of LRF recorded events had occurred by then), which was defined as patients with pelvic, inguinal, or local relapse (unlike the original report, this did not include patients who underwent colostomy in the absence of recurrent disease); 2) anal cancer death (ACD) at 5 years (because, by then, 86% of ACD recorded events had occurred; patients who died of any other cause were censored at date of death); and 3) OS at 5 years. We did not analyze colostomy, because an evaluation of colostomy has not been performed in the same manner for the 2 trials—the ACT II evaluation is more detailed (capturing the date of pretreatment colostomy and the reason for colostomy formation). For each analysis, we investigated the group of patients who had been at risk for at least 2 or 5 years or who had experienced an event by that time. For example, for 2-year LRF, we only considered patients who had been followed for at least 2 years or those who developed LRF within the first 2 years.

Statistical Methods

We investigated several baseline characteristics: sex, age, tumor site, T-classification, lymph node status, hemoglobin, and platelet and white blood cell (WBC) counts. All analyses on the ACT I trial were restricted to 292 patients who received CRT, because this is now standard treatment.

The prognostic performance of each factor on its own was determined by examining the true-positive rate (TPR) (or sensitivity; ie, the proportion of patients who had the event of interest [eg, LRF at 2 years] who had a certain characteristic or whose blood measurement exceeded a specified cutoff) and the false-positive rate (FPR) (or a specificity of 1; ie, the proportion of patients without the event of interest who had a certain characteristic or whose blood measurement exceeded a specified cutoff). A factor with good prognostic performance had a high TPR and a low FPR. The likelihood ratio was calculated as TPR ÷ FPR; thus, the higher the likelihood ratio, the stronger the prognostic factor. The prognostic performance of each baseline variable was examined on its own according to the endpoints time to LRF, ACD, and OS. All blood values were used as continuous variables. All variables were simultaneously included in a Cox proportional hazards model, and backwards elimination was used to produce the final model, in which the selected variables were associated independently with outcomes (each had a P value < .05 after adjustment for the other factors in the model). A prognostic score was then generated from the linear predictor of the variables from the final Cox model (using the hazard ratio [HR] estimate for each factor).

The score was validated externally using an independent, large, randomized study: the ACT II trial.14, 15 That study was based on 940 patients who were recruited between 2001 and 2008 and received CRT (radiotherapy to 50.4 Gray [Gy]; 5-FU 1000 mg/m2 daily as a continuous, 24-hour infusion over 4 days in weeks 1 and 5 with either MMC 12 mg m2 as an intravenous bolus on day 1 of chemotherapy or cisplatin 60 mg/m2 by as an intravenous infusion on days 1 and 29).10 That trial is currently in long-term follow-up. The prognostic score was obtained for each ACT II patient, and the TPR and likelihood ratio were calculated at fixed FPR values. At the time of the current analyses, the following numbers of events were reported in the ACT II trial: 130 (2-year LRF), 137 (5-year ACD), and 186 (5-year OS).

Prognostic test performance also was examined using receiver operating curves and area under the curve (AUC) analysis. In such analyses, the closer the AUC is to 1, the better the prognostic model. Stata version 12.1 statistical software package (Stata Corporation, College Station, Tex) was used to analyze the data.

RESULTS

Patient Characteristics

Of the 292 patients who received CRT in the ACT I trial, 9 patients who had metastases at the time of entry were excluded, leaving 283 patients for the ACD and OS analyses. For the time to LRF, an additional 8 patients who died or relapsed within 6 weeks of the end of initial treatment who, thus, were considered refractory to treatment were excluded (n = 275).

Table 1 provides data on the prognostic performance of each baseline characteristic separately. The best factor for 5-year ACD is palpable, clinically involved inguinal lymph nodes (PCIINs), which had a TPR of 36%, an FPR of 12%, and a likelihood ratio of 2.94 (among patients who died of anal cancer within 5 years, the proportion of those with palpable lymph nodes was 2.94 times greater than among those who did not have the event [ACD]). None of the other factors on their own were striking; that is, the TPR was not materially higher than the FPR.

Table 1. Prognostic Performance of Individual Baseline Characteristics for Locoregional Failure (2 Years), Anal Cancer Death (5 Years), and Overall Survival (5 Years) Among Patients in the First Anal Cancer Trial (ACT I) who Received Chemoradiationa
 2-Year Locoregional Status5-Year Anal Cancer Death Status5-Year Death Status
Characteristic: Positive-Marker LevelTPR % (68 Events)FPR % (177 Nonevents)LRTPR % (72 Events)FPR % (161 Nonevents)LRTPR % (113 Events)FPR % (161 Nonevents)LR
  • Abbreviations: FPR, false positive rate; LR, likelihood ratio; TPR, true-positive rate; WBC, white blood cells.

  • a

    The TPR indicates the percentage of patients with an event, with the given characteristic; the FPR is the percentage of patients without an event, with the given characteristic; and the LR is calculated as TPR ÷ FPR, indicating the strength of the prognostic factor.

Age, y         
 ≥5565720.8974681.0981681.19
 ≥6535450.7943391.1252391.36
 ≥7022240.9324211.1231211.47
 ≥75690.65670.811071.42
Sex         
 Men53361.4654361.5055361.52
 Women47640.7446640.7245640.71
SiteN = 66N = 173 N = 70N = 157 N = 111N = 157 
 Anal margin24231.0517260.6624260.93
 Anal canal76770.9983741.1276741.02
Tumor classificationN = 64N = 170 N = 68N = 153 N = 109N = 153 
 T2-T495841.1391851.0789851.05
 T3-T467491.3668471.4463471.35
 T422102.1924122.0017121.48
Lymph node statusN = 65N = 170 N = 69N = 154 N = 110N = 154 
 Palpable31152.0136122.9434122.73
 Impalpable69850.8264880.7366880.76
Hemoglobin, g/dLN = 50N = 143 N = 49N = 129 N = 84N = 129 
 ≥12.570770.9169780.9070780.91
 ≥13.540500.7933520.6339520.76
 ≥14.528281.0018300.6125300.83
WBC count ×109/LN = 49N = 143 N = 48N = 129 N = 83N = 129 
 ≥694821.1592821.1292821.11
 ≥863531.1965501.2865501.29
 ≥1029261.1033241.3934241.40
Platelets, ×109/LN = 50N = 140 N = 49N = 126 N = 83N = 126 
 ≥25068661.0280641.2477641.20
 ≥30044441.0155401.3952401.31
 ≥35026241.1031211.4830211.46

Table 2 provides findings from the Cox regression models based on all baseline factors (for the 3 endpoints). Men and PCIINs were independent risk factors for all 3 endpoints (LRF, ACD, and OS).

Table 2. Multivariable Analyses of Locoregional Failure, Anal Cancer Death, and Overall Survival Among Patients who Received Chemoradiation in the First Anal Cancer Trial (ACT I)a
 Adjusted HR (95% CI)
FactorLocoregional Failure, N = 265Anal Cancer Death, N = 222Overall Survival, N = 220
  • Abbreviations: CI, confidence interval; HR, hazard ratio.

  • a

    Each HR was adjusted for all other factors in the same model. The table lists the factors that were selected from backward elimination Cox regression modeling.

Sex   
 Women1.001.001.00
 Men1.60 (1.03-2.49)1.80 (1.03-3.16)1.56 (1.12-2.17)
 P.036.039.008
Lymph node status   
 Impalpable1.001.001.00
 Palpable1.87 (1.15- 3.05)1.83 (1.06-3.18)1.74 (1.17-2.58)
 P.012.031.006
Age, per 5-y increase1.25 (1.15-1.35)
 P  < .001
White blood cells, per unit increase in ×109/L1.11 (1.04-1.18)
 P  .001
Hemoglobin, per unit increase in g/dL0.81 (0.70-0.95)
 P .008 

Sex

LRF, ACD, and OS were better for women than for men. Univariate effects revealed a clear disadvantage for men (Figs. 1A, 2A, 3A). In multivariable analysis, compared with women, men had an increased risk of LRF by 60% (HR, 1.60; 95% CI, 1.03-2.49; P = .036), an 80% increased risk of ACD (HR, 1.80; 95% CI, 1.03-3.16; P = .039), and a 56% greater risk of any death (HR, 1.56; 95% CI, 1.12-2.17; P = .008).

Figure 1.

The time to locoregional failure is illustrated among patients who received chemoradiation according to (A) sex (no. of events: 41 of 157 women [26%] vs 43 of 118 men [36%]; P = .017) and (B) lymph node status (no. of events: 57 of 207 patients with impalpable lymph nodes [28%] vs 23 of 58 patients with palpable lymph nodes [40%]; P = .006).

Figure 2.

The time to anal cancer death is illustrated among patients who received chemoradiation according to (A) sex (no. of events: 41 of 161 women [25%] vs 47 of 122 men [39%]; P = .005) and (B) lymph node status (no. of events:, 57 of 213 patients with impalpable lymph nodes [27%] vs 27 of 60 patients with palpable lymph nodes [45%]; P < .001).

Figure 3.

The time to death from any cause is illustrated among patients who received chemoradiation according to (A) sex (no. of events: 92 of 161 women [57%] vs 100 of 122 men [82%]; P < .001) and (B) lymph node status (no. of events:, 140 of 213 patients with impalpable lymph nodes [82%] vs 45 of 60 patients with palpable lymph nodes [75%]; P = .001).

Clinical lymph node status

PCIINs were associated with a poorer outcome compared with impalpable lymph nodes. Univariate effects indicated an improvement for impalpable lymph nodes (Figs. 1B, 2B, 3B). The patients who had PCIINs had a 87% increased risk of LRF (HR, 1.87; 95% CI, 1.15-3.05; P = .012), a 83% greater risk of ACD (HR, 1.83; 95% CI, 1.06-3.18; P = .031), and worse OS (HR, 1.74; 95% CI, 1.17-2.58; P = .006). The proportional hazards assumption was not violated for any of the 3 endpoints.

Hemoglobin also was associated independently with ACD. On average, a single-unit (g/dL) increase in hemoglobin was associated with a 19% reduction in the risk of ACD (P = .008) after adjusting for sex and lymph node status. For example, comparing a baseline hemoglobin level of 11 g/dL versus 10 g/dL or 16 g/dL versus 15 g/dL, on average, will suggest a 19% reduction in the risk of ACD.

In addition, the WBC count was identified as an independent risk factor for OS. The adjusted HR indicated an average 11% increase in the risk of any death per 1 × 109/L increase in the WBC count (HR, 1.11; 95% CI, 1.04-1.18; P = .001). This was irrespective of whether the baseline WBC count was in the normal range. Unsurprisingly, increased age is also was prognostic for any death (HR, 1.25; 95% CI, 1.15-1.35; P < .001).

Validation of the Prognostic Score in the Independent Data Set: ACT II

The prognostic performance using the above factors selected by Cox regression modeling is provided in Table 3 (also see Box 1). The likelihood ratio for values of the prognostic score for 2-year LRF does not suggest a particularly good performance, because the TPRs are close to the FPRs, and all likelihood ratio values are <2, which is far below the recommended level of 10 for a strong prognosis16 (AUC, 54%; 95% CI, 48%-60%). The prognostic performance is better, but only moderate, for 5-year ACD (AUC: 67%; 95% CI, 61%-73%) and OS (AUC: 63%; 95% CI, 58%-68%).

Table 3. Independent Validation of Scores Derived From the First Anal Cancer Trial (ACT I) Applied to the ACT II Trial Among Patients who Received Chemoradiationa
FPR, %bTPR, %bLRcPrognostic Score Cut-Off
  • Abbreviations: FPR, false positive rate; LR, likelihood ratio; TPR, true-positive rate.

  • a

    The prognostic score is based on the combination of selected factors measured at baseline (locoregional failure: sex and lymph node status; anal cancer death: sex, lymph node status, and hemoglobin level; overall survival: sex, lymph node status, age, and white blood cell count).

  • b

    The FPR indicates the proportion of patients who did not have the outcome of interest (eg, those who did not have locoregional failure at 2 years) whose prognostic score exceeded the specified cut-off; the TPR indicates the proportion of patients who did have the outcome of interest (eg, those who had locoregional failure at 2 years) whose prognostic score exceeded the specified cut-off; the LR is calculated as TPR ÷ FPR.

  • c

    The prognostic score for locoregional failure is based only on categorical variables (sex and lymph node status); and, because the score has only 4 values, there is not always an FPR of 5%, 10%, 15%, 20%, 25%, or 50%.

Two-year locoregional failure (sex and lymph node status)b
 11201.88≥1.09
 32371.16≥0.62
 56601.07≥0.47
Five-year anal cancer death (sex, lymph node status, and hemoglobin level)b   
 5285.62≥−1.75
 10343.29≥−1.90
 15392.60≥−2.04
 20432.16≥−2.10
 25461.85≥−2.19
 50731.45≥−2.52
Five-year overall survival (sex, lymph node status, age, and white blood cell count)   
 5132.64≥4.76
 10232.34≥4.56
 15362.40≥4.35
 20412.04≥4.25
 25461.83≥4.14
 50641.27≥3.77

Box 1: prognostic score

In determining the prognostic score for locoregional failure (LRF) we first assessed the baseline factors that are independently associated with the outcome, using Cox regression. In the multivariable model, only sex and nodal status were associated with LRF. The hazard ratio (HR) for each factor is given in Table 2. A linear predictor (LP) was then generated from the model containing just sex and nodal status:

LP = loge HRsex x sex + loge HRnodes x nodes

Where loge refers to natural logarithms; HRsex is the adjusted HR for the effect of sex on time to LRF, comparing males vs. females, and sex is coded as 0 (females) and 1 (males); HRnodes is the adjusted HR for the effect of nodal status on time to LRF, comparing those with palpable nodes (1) vs. impalpable nodes (0).

In the case of this model the LP can take just one of four different values:

Female and impalpable nodes: LP = loge (1.60) x 0 + loge (1.87) x 0 = 0

Female and palpable nodes: LP = loge (1.60) x 0 + loge (1.87) x 1 = 0.63

Male and impalpable nodes: LP = loge (1.60) x 1 + loge (1.87) x 0 = 0.47

Male and palpable nodes: LP = loge (1.60) x 1 + loge (1.87) x 1 = 1.10

The higher the linear predictor the greater the risk is of failing locoregionally.

This prognostic score based upon the ACT I data was then externally validated using the ACT II dataset

DISCUSSION

The current analyses support previous evidence that PCIINs and men with ECA are prognostic factors for LRF and OS in addition to ACD. For the ACT I study, data on skin ulceration were not available. Presenting hemoglobin level and an increased initial WBC count may be further useful prognostic factors for ACD and OS, respectively. Age is prognostic for OS only. Tumor site (canal vs margin), clinical T-classification, and platelets were not prognostic for LRF, ACD, or OS.

Lymph node involvement appears to be an important prognostic factor for LRF and OS.7, 16, 17 In multivariate analyses of retrospective, nonrandomized data sets, tumor size and lymph node status are the strongest prognostic factors for a poor outcome after CRT.18-20 One retrospective study suggested that tumor site was important.21 In addition, several factors in retrospective analyses indicated a higher risk of developing metastatic disease, ie, increasing T-classification, N-stage, and basaloid histology.18 The American National Cancer Data Base has identified men, age ≥65 years, T-classification, N-stage, distant metastases, or poorly differentiated histology as prognostic factors for survival from almost 20,000 patients who were treated from 1985 to 2000.3

To avoid the recognized limitations of retrospective analyses, we used the ACT II data to externally validate the 3 models developed in ACT I, and, hence, identified the covariates as potentially prognostic and applicable to future patients with anal cancer. ACT II is a recent data set, which is now mature (median follow-up, 5.1 years) and reflects current practice in the UK.

To our knowledge, only 4 phase 3 interventional anal cancer studies have been published to date.4, 6-8 The EORTC 22861 trial (n = 110)6 reported that their prognostic factor analysis indicated that palpable lymph nodes (LRF, P = .003; OS, P = .0003), men (LRF, P = .05; OS, P = .01), and skin ulceration (LRF, P = .003; OS, P = .005) were prognostic, but tumor size was not. T-classification, tumor length (<5 cm vs >5 cm), and involved part of the circumference (< 1/3 versus 1/3 to 2/3 versus > 2/3 of the circumference) did not exhibit any prognostic value for LRF or OS. The RTOG 8704 study (n = 310) has yet to report a prognostic factor evaluation. The RTOG 98-11 study (644 patients) confirmed 2 of the previously reported EORTC prognostic factors (palpable lymph nodes and men) and established the greatest tumor dimension as an independent variable predicting DFS and OS.12 These factors have not been validated by an additional prospective data set.

Data from the current analysis support findings from 2 previously reported prognostic factors (lymph node status and sex) as independent pretreatment variables that predict LRF and OS. Whereas RTOG 98-11 examined DFS, EORTC 22861 and ACT I assessed the prognostic factors for LRF. OS may be a less useful endpoint in anal cancer than ACD. After 12 years of follow-up in ACT I,5 there was a large treatment difference in locoregional control favoring CRT, but no statistically significant treatment effect on OS was reported, although a 14% reduction in the risk of death was suggested for CRT. This finding may reflect the possibility of surgical salvage and more competing risks in the elderly.

One novel finding was the impact of hemoglobin on ACD, which was documented in previous retrospective studies.22, 23 Anemia may exacerbate tumor hypoxia, leading to radioresistance and promoting aggressive phenotypes, tumor progression, and distant metastases. The effect of hemoglobin is usually on LRF rather than survival. However, neither the EORTC 22861 trial6 nor the RTOG 98-11 trial12 examined the pretreatment hemoglobin level as a prognostic variable. The use of hemoglobin as a risk-stratification factor may be useful in future trials and may provide a method with which to identify patients who might benefit from future methods of hypoxic modification of RT.24

Inflammation plays a role in the progression of cancer. Markers like the absolute WBC count and the neutrophil-to-lymphocyte ratio appear to have prognostic value.25 The current findings are in keeping with a retrospective analysis in squamous cell cancers of the cervix, in which patients who had a high pretreatment WBC count (≥10,000/μL) had a significantly higher treatment failure rate (P < .0001) and shorter OS (P < .0001) than patients who had lower levels, suggesting that the WBC count is an independent prognostic factor.26

There are limitations to this study. The prognostic model for 2-year LRF was not a good predictor with the ACT II data set (AUC, 54%) but was better for 5-year ACD (AUC, 67%). The finding that ACT I data validate poorly with ACT II data may reflect the finding that the 2 trials were conducted 10 years apart, indicating how much treatment has changed in terms of imaging (staging with computed tomography, magnetic resonance imaging, and positron emission tomography has become more common and may have led to lead-time bias). Other differences include the avoidance of a gap in ACT II and more modern techniques of RT and surgical salvage. The lack of impact of T-classification probably reflects the finding that the ACT I trial used 1985 UICC staging based on anatomic extent, whereas patients in the ACT II trial were staged according the 1990 UICC staging system27 based on tumor size. Also, the LRF rate was much better in the ACT II trial.

In the future, clinical, epigenetic, genetic, and germ-line variables could be used to develop a model that predicts response to therapy and prognosticates outcome. The first step in developing such models may be the establishment of reliable prognostic factors and their interaction.

Approximately 9% of newly diagnosed patients fit into the worst prognostic category (men/PCIIN). The current analysis does not provide data to suggest that these groups should be treated differently. Newer tools, such as positron emission tomography, could be incorporated into initial staging and into the early assessment of response in such a poor prognostic group.

In conclusion, the current analysis of ACT I complements 2 other studies, adding further evidence that palpable inguinal lymph node status and sex are independently prognostic for OS, LRF, and ACD. In addition, after adjusting for sex and lymph node status, the presenting hemoglobin level is a further prognostic factor for ACD. An increase in the initial WBC count may be prognostic for worse OS. Neither tumor size nor skin ulceration could be evaluated in our analysis of prognostic factors. The current study provides an appropriate strategy with which to alter existing risk-stratification systems.

We recommend hemoglobin as a novel additional candidate biomarker for prospective trials to assess risk-adapted therapies. Absolute and cutoff values of hemoglobin could easily and reliably be used between different laboratories to classify high-risk versus low-risk groups, either to direct novel therapeutic strategies or to increase the RT dose.

Clinically involved inguinal lymphadenopathy is also a strong prognostic marker. We do not yet know whether computed tomography-defined lymphadenopathy provides a similar prognostic value. Hence, stratification of both early stage and more advanced groups may be possible in future trials according to these factors and may be used to categorize different risk groups.

Acknowledgements

We thank Professor Allan Hackshaw, Deputy Director, Cancer Research UK and University College London Cancer Trials Center Cancer Institute, University College London for his helpful advice.

FUNDING SOURCES

This work was supported by the UK Coordinating Committee on Cancer Research.

CONFLICT OF INTEREST DISCLOSURES

The authors made no disclosures.

Ancillary