Cox regression with interaction between site of recurrence and time to recurrence. The y-axis represents the hazard rate, which here is an estimate of the relative risk of death of disease for the variable in question.
Survival rate curves for time to recurrence (at 1 and 12 mo) for patients who initially received single-modality treatment (typically surgery), comparing local with regional recurrence.
Survival rate curves for time to recurrence (at 1 and 12 mo) for patients who initially received combined-modality treatment (typically surgery with adjuvant radiotherapy), comparing local with regional recurrence.
Kernohan MD, Clark JR, Gao K, Ebrahimi A, Milross CG. Predicting the Prognosis of Oral Squamous Cell Carcinoma After First Recurrence. Arch Otolaryngol Head Neck Surg. 2010;136(12):1235-1239. doi:10.1001/archoto.2010.214
To describe the clinicopathologic features of oral squamous cell carcinoma in patients who develop locoregional recurrence of disease, to identify factors that predict prognosis in the subset of patients treated with salvage surgery, and to determine the adjusted effect of time to recurrence.
A head and neck cancer institute in Sydney, New South Wales, Australia.
A total of 77 patients who underwent salvage surgery for oral squamous cell carcinoma that had been treated initially by surgery, radiotherapy, or surgery with postoperative radiotherapy.
Main Outcome Measures
Univariable and multivariable analysis of clinical and pathologic risk factors.
Median time to recurrence from initial treatment was 7.5 months (range, 0.9-143.9 mo), with 86% of recurrences occurring within the first 24 months. Surgical salvage was attempted in 77 patients who had experienced recurrence at the primary site (n = 39), ipsilateral neck (n = 27), and contralateral neck (n = 11). Time to recurrence, initial treatment modality, and site of failure were independent prognostic variables.
The relationship of these prognostic variables displays a dynamic interaction. Initial combined-modality treatment and shorter time to recurrence were associated with worse outcome, while the effect of site of recurrence (local vs regional) was dependent on an interaction with the time to recurrence. The result of this interaction was that local recurrence was worse for those who experienced it early (eg, <6 mo after the initial treatment) and nodal recurrence was worse for those who experienced it late (eg, ≥6 mo after the intial treatment).
Treating local or regional recurrence of oral squamous cell carcinoma (OSCC) involves balancing the risk of morbidity against the likely benefit to the patient in terms of his or her chance of survival or locoregional control. Clinical and pathologic risk factors are used to assess the risk of recurrence of OSCC. Of the known prognostic factors, only tumor size (T stage) and nodal status (N stage) are included in the current TNM staging system1 commonly used to stratify the risk of recurrence and survival at the time of initial presentation. After initial treatment, when a patient presents with a local or regional recurrence, the decision to treat further is based on the risk of morbidity vs the likely benefit to that patient in terms of his or her chance of survival or locoregional control.
Currently, there is a paucity of published data2- 5 to guide health care professionals regarding which clinicopathologic factors can be used to predict survival once the patient has developed recurrent disease. Data from our institution6 suggest that 80% of locoregional recurrences happen within the first 2 years after initial therapy and that time to recurrence (TTR) is an important predictor of ultimate successful salvage.
The primary aim of this study was to identify the clinical or pathologic factors that predict the prognosis of patients with recurrent OSCC who undergo salvage surgery. Based on the belief that such prognostic information would be helpful with treatment planning and patient counseling, we sought to determine, in particular, whether TTR was an important predictor when adjusted for other variables.
The Sydney Head and Neck Cancer Institute has prospectively collected data on more than 9000 patients. At the time of this analysis, 533 had been treated for OSCC diagnosed between April 28, 1988, and February 15, 2006. Patients were included in this study if they had a documented first recurrence of OSCC, had no previous treatment for the recurrence at another institution, and had been treated with curative intent. Patients were excluded if they had distant metastases at the time of recurrence (n = 14). The site most proximal to the initial primary tumor was recorded as the recurrence site; hence, concurrent locoregional recurrence was considered to be local recurrence. Clinical staging at initial presentation was carried out according to American Joint Committee on Cancer1 criteria.
A total of 117 patients fulfilled these criteria. The male to female ratio was 2:1 and the median age was 68 years (range, 29-90 y). Surgical salvage for recurrence undertaken with curative intent was performed in 77 of 117 patients (66%): 39 for recurrence at the primary site, 27 for that at the ipsilateral neck, and 11 for that at the contralateral neck. Surgery was performed by 3 surgeons in 97% of these cases. To determine the prognostic significance of clinicopathologic variables in predicting successful surgical salvage as measured by disease-specific survival rate from first recurrence, data were extracted from the prospective database and analyzed in 4 categories.
The initial American Joint Committee on Cancer TNM clinical T stage, clinical N stage, and pathologic N stage were used. Histopathologic variables included data from histopathology reports noting resection margin status (clear of tumor or involved), perineural invasion at the periphery of the resection (present or absent), lymphovascular invasion (present or absent), and extracapsular lymph node spread (present or absent).
The treatment applied at initial presentation was recorded as single-modality (surgery only or radiotherapy only) or combined-modality treatment (surgery with adjuvant radiotherapy with or without concurrent chemotherapy). Recurrence was considered local if it occurred in the oral cavity and regional if it occurred in the cervical lymph nodes. The TTR was recorded as the time from initial treatment until local or regional recurrence was confirmed pathologically. The TTR was recorded as a continuous variable; however, the data were dichotomized at 6 and 12 months for univariate comparisons.
Disease-specific survival rate was calculated from the time of recurrence to death of disease using the Kaplan-Meier method.7 Univariate comparisons were performed using the log-rank test.8 Variables with P <.05 on univariate analysis were included in the initial multivariable model. Multivariable analysis was performed using a stepwise backward Cox regression,9 and owing to the existence of relatively few events, variables with P >.05 were excluded. Continuous variables were appropriately transformed to maintain linearity, and potential interactions were examined.
The proportional hazards model assumes that there is a linear relationship between continuous variables and the effect on survival rate. Therefore, TTR has been transformed to maintain this assumption. When the effect on survival rate or a certain variable (eg, site of recurrence) changes depending on the value of another variable (eg, time to recurrence), there is an interaction between these variables. This difference can be accounted for by including an interaction term in the model to better explain the effect on survival. All analyses were performed using Stata statistical software, version 9 (StataCorp LP, College Station, Texas).
Salvage surgery was performed in 77 patients. The initial TNM classification for those patients is summarized in Table 1. Surgery as the initial single-modality treatment (excision of primary tumor with or without a neck dissection) was performed in 41 of the 77 patients. Radiotherapy as the single-modality treatment was delivered to only 5 patients. Combined-modality treatment was delivered to 31 patients (surgery with radiotherapy in 29 and chemotherapy in 2).
Neck dissection was performed as part of the initial management in 40 patients (28 unilateral and 12 bilateral), and 37 patients had no neck dissection as part of their initial treatment, of whom 7 had adjuvant radiotherapy. Of the 30 patients who had received observation only for regional recurrence, 13 developed ipsilateral neck disease and 1 developed contralateral neck disease (recurrence at the primary site occurred in 16 of these 30 patients). The most common oral cavity subsite that displayed recurrence was the tongue, followed by the floor of the mouth, the retromolar region, and the buccal mucosa. Median TTR from initial treatment was 7.5 months for the entire group (range, 0.9-143.9 mo), with 86% of recurrences happening within the first 2 years.
Patients with recurrences that were considered potentially curable and operable underwent salvage surgery and radiotherapy if those recurrences had not been previously irradiated. A few recurrences were reirradiated. Patients with recurrences that were considered incurable were treated with palliative chemotherapy or referred to palliative care. The overall disease-specific survival for the surgical salvage group was 50% at 5 years.
All variables analyzed are listed in Table 2. Variables significant at P < .05 that were considered for multivariable analysis were initial treatment modality (single or combined), site of recurrence (local or regional), TTR (≤12 mo vs >12 mo), clinical T stage at diagnosis, and lymphovascular invasion. Lymphovascular invasion was not selected because only 11 patients exhibited this feature.
Variables studied in the initial multivariable model were T stage (T1 and T2 vs T3 and T4), TTR (<12 months vs ≥12 months), site of recurrence (local vs regional), and initial treatment modality (single vs combined). By using the backward stepwise method, T stage (P = .83) and TTR (P = .25) were removed from the model. Initial treatment modality (combined) and regional recurrence showed significances of P = .01 and P = .04, respectively. Because our initial hypothesis centered on the effect of TTR when adjusted for potential confounders, TTR was reintroduced to the model as a continuous variable with a log transformation. Interaction terms were included between TTR and the remaining variables. The final multivariable model is shown in Table 3.
After adjusting for the effect of initial treatment (combined-modality hazard ratio = 2.3, 95% confidence interval, 1.1-4.6, P = .02) and site of recurrence (P = .08), there is strong evidence that shorter TTR is associated with reduced survival rate (P = .007). Hazard ratios associated with TTR and site of recurrence are dynamic owing to a significant interaction between TTR and site of recurrence (P = .009). Therefore, in those patients whose recurrence happens early (eg, <6 mo after initial treatment), worse outcomes are associated with those whose recurrences are located in the primary site compared with those whose recurrences are located in the neck. In contrast, for those patients whose recurrences happen late (eg, ≥6 mo after initial treatment), worse outcomes are associated with regional recurrence. The TTR at which the hazard ratios are equal for local and regional recurrence is approximately 5 months (log 1.5), as shown in Figure 1.
Predicted survival curves calculated for patients based on TTR, site of recurrence, and initial treatment modality are shown in Figure 2 for those who received single-modality treatment and Figure 3 for those who received combined-modality treatment. Table 3 and Table 4 describe the predicted 2-year disease-specific survival rates for patients whose recurrences happened at 1, 3, 6, and 12 months after initial single-modality and combined-modality treatment, respectively.
We reviewed our prospective database to determine factors that influence the likelihood of successful surgical salvage of first recurrence of oral cavity cancer. An initial treatment that involved more than 1 modality (typically surgery and radiotherapy) and shorter TTR were associated with worse outcome. In addition, the effect of local or regional recurrence was not uniform but interacted with the TTR. The result of this interaction was that local recurrence was worse for those to whom recurrences happened early, and nodal recurrence was worse for those to whom recurrences happened late.
The published outcomes3,5,10- 12 after salvage for oral cavity recurrence vary widely. Comparison is further complicated by the fact that many reports combine the oropharynx and the oral cavity as a single group despite marked differences in prognosis and salvage options. Disease-specific survival after treatment of OSCC by salvage surgery in the current study was 50% at 5 years. In contrast, the meta-analysis by Goodwin12 of 32 studies of recurrent squamous cell carcinoma of the upper aerodigestive tract reported a 5-year overall survival of 43% for patients with oral cancers treated by surgery. The meta-analysis included 1633 patients with just more than half of recurrences located in the oral cavity or oropharynx and nearly all patients (99%) having undergone radiotherapy prior to disease recurrence. The author comments on the difficulties in combining data and reports on 109 patients with recurrent cancer treated at their institution, where 2-year survival after surgical salvage was 44%.
Clinicopathologic factors that influence disease control and survival at initial treatment are well described2- 5; however, few data exist, to our knowledge, to support their usefulness in the setting of recurrence. Deciding which patients should undergo salvage surgery can be challenging and is often based on personal experience without the support of reliable evidence. Many studies3,5 represent combinations of heterogeneous groups, making practical application difficult.
In the current analysis, initial combined-modality treatment was associated with a 1.3-fold increased risk of death of disease compared with patients treated with surgery only. Presumably, the poor outcome reflects a combination of more advanced disease at initial presentation, resistant tumor biology, and limited salvage options. These findings suggest that patients whose disease fails maximal combination therapy have a low likelihood of successful salvage; we recommend that such patients be counseled accordingly. In certain instances salvage may be possible, such as when the initial surgery and/or radiotherapy was suboptimal. Examples include patients with early neck recurrence who had not previously undergone neck dissection, those with local recurrence who had initial positive tumor margins (and no other adverse prognostic features), those with delayed or interrupted initial radiotherapy, those whose disease recurred outside the radiotherapy treatment field, and those whose initial radiotherapy dose was inadequate.
Many authors10- 14 agree that the frequency of recurrence within the first 2 years is much higher than that in subsequent years. More than 80% of recurrences happened within the first 2 years in our study, which supports the current strategy for more intense and frequent surveillance during the first few years after initial treatment. It is logical that biologically aggressive tumors will recur quickly, which is supported by studies5,10,13 that conclude that early recurrence predicts poor outcome. Stell,13 in his report of a cohort of 515 patients who experienced recurrences after radiotherapy for head and neck cancer, suggested that TTR was the most significant factor in survival.
Our results suggest that the relationship between TTR and prognosis is complicated because the effect of TTR also depends on the recurrence site. Early local recurrence is associated with a very poor prognosis, whereas late local recurrence gives the patient a much higher chance of successful salvage. In contrast, neck recurrence is less dependent on the TTR (as indicated by the flatter hazard gradient in Figure 1). It is possible that the slightly improved outcomes observed with early neck recurrence may reflect initial understaging resulting in observation rather than neck dissection.
The data used in this study do not benefit from the additional information now available from more recent investigative tools, such as positron-emission tomography/computed tomography and human papilloma virus status testing. In addition, this study does not include analysis of the effect of smoking status on recurrence risk, although we currently record this prospectively.
These data can be interpreted with the following explanatory scenarios: patient A develops local recurrence at 12 months after initial treatment for an OSCC. The predicted 2-year survival for salvage therapy is 81% if the initial treatment involved a single modality and 62% if the initial treatment involved a combination of surgery and radiotherapy. Therefore, local recurrence at 12 months may still have an optimistic outcome and salvage surgery should be considered. Conversely, local recurrence at 3 months predicts a very poor outcome (18%; 2-y survival) for patients who had combined-modality treatment initially. Patient B develops regional recurrence at 12 months after initial treatment for an OSCC. The predicted 2-year survival for salvage therapy would be much less optimistic at 52% if initial treatment involved a single modality and 23% if initial treatment involved surgery and radiotherapy.
Therefore, neck failure despite maximal initial therapy is unlikely to be cured, and palliative care is often the more appropriate option. This may not apply to early neck failures for which neck dissection was not performed.
Treatment planning for patients with recurrent OSCC is always a challenge. This retrospective study provides new information regarding survival prediction for these patients and demonstrates the interaction of clinically relevant prognostic factors that reflect variation in disease biology and behavior. These data and unique analyses have identified the independent prognostic variables as TTR, initial treatment modality (single or combined), and site of failure (local or regional). The relationship of these variables is not fixed; the variables have a dynamic interaction.
Correspondence: Michael D. Kernohan, FDSRCS, FRCS, MSc, Sydney Head and Neck Cancer Institute, Royal Prince Alfred Hospital, PO Box M142, Missenden Road, Camperdown, NSW 2050, Australia.
Submitted for Publication: April 1, 2010; final revision received August 6, 2010; accepted September 25, 2010.
Author Contributions: Drs Kernohan, Clark, and Milross had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Kernohan, Clark, Gao, and Milross. Acquisition of data: Kernohan, Clark, and Gao. Analysis and interpretation of data: Kernohan, Clark, Gao, Ebrahimi, and Milross. Drafting of the manuscript: Kernohan, Clark, and Gao. Critical revision of the manuscript for important intellectual content: Kernohan, Clark, Ebrahimi, and Milross. Statistical analysis: Clark, Gao, and Ebrahimi. Administrative, technical, and material support: Kernohan and Gao. Study supervision: Clark and Milross.
Previous Presentation: This study was presented at the annual meeting of the American Head and Neck Society; April 29, 2010; Las Vegas, Nevada.
Financial Disclosure: None reported.