The 2000 College of American Pathologists (CAP) guidelines recommend that a characterization of carcinomas of the upper aerodigestive tract, including tongue cancer, should include depth of invasion (DI) and the presence of lymphovascular invasion (LVI) or perineural invasion (PNI).
This study included patients who were diagnosed with cancer of the oral tongue, who underwent tumor resection, and who were reported to either the Metropolitan Atlanta and Rural Georgia Surveillance, Epidemiology, and End Results (SEER) registry or the Los Angeles SEER registry. The authors assessed the completeness of pathology reporting with respect to the documentation of PNI or LVI and DI. Generalized estimating equations were used to examine factors that influenced reporting while taking into consideration clustering of observations within the hospitals. Univariate and multivariate survival analyses were conducted to examine the impact of tumor invasion characteristics on mortality while controlling for other prognostic factors.
DI reporting increased from 13% between 1997 and 1999 to 23% between 2000 and 2004 after the CAP issued its recommendations; whereas mode of invasion (the presence of LVI and/or PNI) reporting for the same period increased from 13% to 38%. The observed increase in reporting was most pronounced in the first 2 years (2000 and 2001) and appeared to decline again afterward. Tumor invasion >3 mm in depth and the presence of PNI were among the strongest predictors of survival in multivariate analyses.
Cancer of the tongue constitutes approximately 33% of all oral-pharyngeal cancers, with greater than 10,000 new cases and nearly 1900 deaths estimated in the United States in 2008 alone.1 Although stage at diagnosis serves as an important predictor of survival in patients with tongue cancer, it is not the only prognostic factor. A certain proportion of early stage tumors exhibit aggressive behavior and have an unexpectedly poor prognosis. For this reason, it is important to rely on additional tumor characteristics that would identify more aggressive cancers at the time of diagnosis.
Literature indicates that the presence of lymphovascular and perineural tumor invasion as well as increased depth of subepithelial invasion need to be viewed as important prognostic factors in addition to traditional diagnostic characteristics, such as tumor size, lymph node involvement, and the presence of distant metastases.2 On the basis of these observations, the protocol issued by the College of American Pathologists (CAP) currently recommends that a characterization of tongue carcinomas should evaluate the mode of invasion (LVI, PNI, or none) as well as the extent of subepithelial connective tissue and muscle invasion measured from the basement membrane in millimeters. This protocol initially was published in 19983 and, in 2000, was adopted as the official CAP recommendation.
Despite the apparent importance of the CAP recommendations, to our knowledge, the frequency and consistency of reported mode and depth of tumor invasion at a population level have not been examined. In addition, previous studies evaluating the impact of tumor invasion on cancer survival usually were based on single-institution patient series and rarely used multivariate analyses.
The objectives of this study were to evaluate of the effect of CAP recommendations on the frequency and completeness of reported information related to the depth and mode of tumor invasion and to examine the relation of depth and mode of invasion to patient survival while controlling for other patient-related, provider-related, tumor-related, and treatment-related characteristics. The current communication is the first report from this project, which was supported by the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) Program and was awarded as the Rapid Response Surveillance Study grant to 2 sites: the Metropolitan Atlanta and Rural Georgia (MARGA) SEER registry and the Los Angeles (LA) SEER registry. According to the SEER data from 2000 through 2005, the annual tongue cancer incidence in both Atlanta and Los Angeles was 2.6 per 100,000, whereas the corresponding rate for rural Georgia was 3.1 per 100,000.
MATERIALS AND METHODS
The data for this project were pooled from 2 separately funded study sites that used very similar research protocols and identical data collection instruments. The initial population for 1 study included 229 patients who were diagnosed with tongue cancer between January 1, 1998 and December 31, 2004 while residing in 1 of the 15 counties covered by the MARGA registry. The other study sought to examine pathology reports from 631 patients with tongue cancer reported to the LA SEER registry with years of diagnosis 1997 and 2002-2003. Figure 1 shows that pathology reports were available for 81.2% of the MARGA patients and for 77% of the LA SEER cases. Among the retrieved pathology reports, 4.4% either did not provide evidence of tongue cancer or indicated dates of diagnosis that were outside of the study interval. Only patients that underwent tumor resection were evaluated; nonsurgical pathology reports were not considered, because those are not expected to report depth and mode of invasion of the primary tumor site. Patients with cancers located in the base of the tongue and those with in situ disease also were excluded (Fig. 1).
Pathology reports for all eligible patients were obtained from hospitals and pathology laboratories. Trained data abstractors reviewed the pathology reports and extracted relevant data, which included patient demographic characteristics, clinical information, macroscopic pathology findings, and the results of microscopic evaluation of the specimens.
The data collection instrument was developed by the LA SEER registry in close collaboration with the Atlanta site to facilitate future data pooling. The data were entered by 1 of the coauthors (G.S. in Atlanta or L.B. in Los Angeles) directly from pathology reports. Then, for quality-control purposes, the resulting data were validated independently by another coauthor (M.G. in Atlanta or L.L. in Los Angeles) who cross-checked the electronic data entries against the pathology reports and made necessary corrections if an error was identified. Errors were identified and corrected in <5% of data entry fields in Atlanta. No data entry errors were identified in the Los Angeles study site.
The respective study protocols received institutional review board approvals from Emory University and the Georgia Department of Human Resources (Atlanta site) and from the University of Southern California (Los Angeles site).
We assessed the completeness of pathology reporting with respect to the presence of perineural invasion (PNI) or lymphovascular invasion (LVI) and depth of invasion (DI) as specifically identified in the CAP guidelines. Additional information obtained from pathology reports included other tumor characteristics, such as stage, grade, and surgical margin status, as well as patient demographic characteristics, such as race, age, sex, and address. Patient addresses were geocoded to the level of census tract and then assigned an area-based measure of socioeconomic status (SES) based on the percentage of individuals in the census tract living below the poverty level.4 High area-based SES was defined as including <10% of the population in the census tract living below the poverty level, medium area-based SES was defined as including from 10.1% to 19.9% of the population in the census tract living below the poverty level, and low area-based SES was defined as ≥20% of the population in the census tract living below the poverty level.
All hospitals and laboratories that established tongue cancer diagnoses and reported cases to their respective regional SEER registries were categorized according to the average number of cases per year contributed to the study by each facility. When appropriate, the data extracted from pathology reports were supplemented by additional information from SEER. SEER data also were used to obtain each patient's vital status and date of death or, for those who remained alive, the date of last contact.
Univariate data analyses compared the frequencies of reporting DI and LVI/PNI across different patient characteristics. Kaplan-Meier curves were constructed to examine patient survival according to the depth and mode of tumor invasion. Multivariate analyses were performed using generalized estimating equations (GEE) for binary data with a logit-link function or Cox regression models, as appropriate.
In the GEE analyses, the association between each binary outcome variable (inclusion of depth or mode of invasion in the pathology report) and each independent variable (age, sex, race, area SES, and hospital patient volume) was analyzed among patients who were diagnosed after the CAP guidelines became available (2000 or later). The GEE modeling approach allowed us to obtain multivariate-adjusted odds ratios (ORs) while accounting for potential clustering of observations (because of common reporting practices) within each hospital.
Cox regression models were used to examine the association between patient survival and various patient-related and tumor-related characteristics with the primary focus on depth and mode of tumor invasion as independent variables of interest. In addition to socioeconomic and demographic patient characteristics, covariates in the survival models also included the presence of regional or distant metastases, tumor grade, tumor size, and surgical margin status.
The associations between outcome and predictor variables were expressed as adjusted ORs in the GEE models or as adjusted hazards ratios (HRs) in the Cox survival models and are reported along with the corresponding 95% confidence intervals (CIs). Proportional hazards assumptions were tested for the Cox survival models by examining log-minus-log plots for each variable. The Cox regression models were used only if the proportional hazards assumptions were not violated (ie, when the log-minus log curves did not cross).5 All models were examined for interactions and collinearity among covariates. The analyses were performed using SPSS 11.5 for Windows (LEAD Technologies, Inc., Chicago IL) and/or SAS 8.02 for Windows (SAS Institute, Inc., Cary, NC) statistical software packages.
Frequency of Reporting Depth and Mode of Invasion
Among 339 patients who met the study inclusion criteria, only 68 patients (20%) had pathology reports that included subepithelial DI; and, in 103 patients (30%), there was a documented mode of invasion. Table 1 shows that patients who had tumor DI reported did not differ from those without documented DI with respect to their age (P = .17), sex (P = .81), race (P = .14), or reporting hospital case volume (P = .27), but they were significantly more likely (P < .01) to reside in more affluent neighborhoods (defined as census tract with less than 10% of population living below the poverty level). A similar comparison of patients with and without reported evidence of LVI or PNI demonstrated no discernable differences with respect to race (P = .97), sex (P = .44), or area SES (P = .12); however, pathology reports with documented mode of invasion (LVI and/or PNI) were significantly more common among younger patients (P < .01) and were less likely to come from hospitals that reported fewer than 1 case per year (P < .01). A greater proportion of pathology reports from Atlanta included depth and mode of invasion information compared with the LA study site (both P values <.01).
Table 1. Sociodemographic Characteristics of Patients Who Had Oral Tongue Cancer With and Without Information on Depth and Mode of Invasion: Pooled Los Angeles and Metropolitan Atlanta/Rural Georgia Surveillance, Epidemiology, and End Results Data (1997-2004)
LA indicates Los Angeles; SES, socioeconomic status.
P values were calculated using the Pearson chi-square test.
High area-based SES was defined as <10% of the census tract population living below the poverty level. Low SES was defined as ≥20% of the population living below the poverty level; in 1 patient, the measure of SES was missing.
Based on the number of patients who contributed to the total study sample.
DI reporting increased from 13% between 1997 and 1999 to 23% between 2000 and 2004 after the CAP issued its recommendations; whereas reporting on the mode of invasion (the presence of LVI and/or PNI) for the same period increased from 13% to 38%. Figure 2 shows that, in almost all years with the exception of 2000, the frequency of DI reporting was lower that the frequency of LVI/PNI reporting. Compared with pre-CAP, years there was a substantial increase in the proportion of pathology reports documenting DI and LVI and/or PNI during 2000 and 2001. In 2000, mode and depth of invasion were reported in 32% and 63% of cases, respectively; whereas, in 2001, the corresponding percentages were 85% for mode of invasion and 50% for DI. There also was evidence that the initial improvement in reporting was followed by a decline, which nevertheless did not reach the pre-2000 levels. The results limited to Atlanta SEER were similar to those observed for the entire dataset (Fig. 2).
Table 2 provides the results from the GEE analyses evaluating the association between various patient characteristics and reporting of depth and mode of invasion for tongue cancers that were newly diagnosed after January 1, 2000. Age, race, and sex were not associated significantly with either DI or LVI/PNI reporting. Compared with patients who resided in relatively affluent areas (<10% of the population living in poverty), patients who resided in the neighborhoods with >20% of the population living below the poverty level were significantly less likely to have reported DI (OR, 0.33; 95% CI, 0.13-0.86), whereas the corresponding OR for LVI/PNI also decreased (0.57), but the 95% CI included unity. Higher volume hospitals and laboratories (ie, those that reported at least 3 cases of tongue cancer per year during the study period) were more likely to report DI than lower volume hospitals (OR, 0.47), but the difference was not statistically significant (95% CI, 0.12-1.77). With respect to mode of invasion, pathology reports from facilities that reported less than 1 case per year were significantly less likely to include LVI/PNI information (OR, 0.01; 95% CI, 0.001-0.14) relative to pathology reports from high-volume hospitals and laboratories (Table 2).
Table 2. Multivariate Analyses Evaluating the Association Between Patient Characteristics and Reporting of the Depth and Mode of Invasion (Presence of Lymphovascular or Perineural Invasion) Among Patients Newly Diagnosed With Cancer of the Oral Tongue Based on College of American Pathologist Recommendations According to Pooled Los Angeles and Metropolitan Atlanta/Rural Georgia Surveillance, Epidemiology, and End Results Data (2000-2004)*
Depth of Invasion
Mode of Invasion
95% Confidence Limits
95% Confidence Limits
SES indicates socioeconomic status.
Multivariate analyses were performed using generalized estimating equation models that accounted for clustering of observations within hospitals.
High area-based SES was defined as <10% of the census tract population living below the poverty level. Low SES was defined as >20% of the census tract population living below the poverty level.
Based on the number of patients who contributed to the total study sample.
The mean and median follow-up for the entire study group were 45 months and 40 months, respectively. Overall, the 5-year survival rate in the study group was 57.9%, and the median duration of survival was 67 months.
Kaplan-Meier analyses of patient survival according to reported depth and mode of invasion are shown in Figures 3 and 4, respectively. The analyses were conducted on a subset of patients for whom these variables were reported (n = 68 for DI and n = 103 for LVI/PNI). The differences in survival among patients who had different DI were examined using cutoff values of 3 mm and 10 mm. The difference was most pronounced when patients who had >3 mm subepithelial invasion were compared with those who had tumor invasion ≤3 mm. This difference in survival was far less obvious when patients who had DI of 3.1 mm to 10 mm were compared with patients who had DI >10 mm (Fig. 3). The log-rank test across all 3 DI categories was significant (P = .002). Figure 4 shows that there was a clear and significant decrease in survival among patients who had tumors with PNI compared with patients who had tumors with no evidence of invasion. The effect of LVI was somewhat less pronounced, but the log-rank test indicated a statistically significant difference across the 3 categories (P = .001).
Table 3 provides the results of multivariate Cox regression analyses, which evaluated the association between patient survival and both mode and depth of tumor invasion. An examination of the log-minus-log plots demonstrated that the proportional hazards assumption was violated when DI was examined as a continuous variable or as polynomial categorical variable. By contrast, log-minus-log plots for the DI variable dichotomized at ≤3 mm versus >3 mm did not cross, indicating that the proportional hazards requirements were met.
Table 3. Multivariate Cox Regression Analyses Evaluating the Association Between Various Patient Characteristics and Survival Among Patients With Tongue Cancer (n=339): Pooled Los Angeles and Metropolitan Atlanta/Rural Georgia Surveillance, Epidemiology, and End Results Data (1997-2004)*
95% Confidence Limits
SES indicates socioeconomic status.
Adjusted for study site.
High area-based SES was defined as <10% of the census tract population living below the poverty level. Low SES was defined as ≥20% of the census tract population living below the poverty level.
Based on the number of patients who contributed to the total study sample.
For the DI variable, patients with DI ≤3 mm were used as the reference group and were compared with 2 categories: patients who had DI >3 mm and patients who did not have DI reported. For the LVI/PNI variable, patients whose pathology reports specifically indicated no LVI/PNI, were used as the reference group and were compared with 3 categories: patients who had LVI only, patients who had PNI (with or without LVI), and patients who did not have LVI/PNI status reported.
After controlling for study site, patient demographic/socioeconomic characteristics (age, sex, race, and neighborhood SES), facility case volume, tumor size, grade, tumor location, presence of regional or distant metastases, surgical margin status, and radiation treatment, DI >3 mm, compared with ≤3 mm, was associated with a strong and statistically significant elevated HR (indicating a decrease in survival) of 12.40 (95% CI, 1.64-93.90), whereas the HR for patients who did not have DI reported was 9.13 (95% CI, 1.25-66.43). The corresponding results demonstrated a statistically significant association with mortality for PNI (HR, 2.43; 95% CI, 1.08-5.50) but not for LVI (HR, 1.75; 95% CI, 0.68-4.50). Additional factors that demonstrated a statistically significant, independent association with poor survival included age ≥65 years (HR, 2.20; 95% CI, 1.55-3.12) compared with age <65 years, and regional lymph node involvement (HR, 1.54; 95% CI, 1.02-2.31) and distant metastases (HR, 3.25; 95% CI, 1.52-6.97) compared with localized disease. There also was evidence of a gradual decrease in survival with increasing tumor size: Compared with tumors that measured ≤10 mm in greatest dimension (reference category), lesions that measured from 11 mm to 20 mm, from 21 mm to 30 mm, and >30 mm were associated with HRs of 1.10 (95% CI, 0.64-1.87), 1.23 (95% CI, 0.71-2.13), and 1.85 (95% CI, 1.06-3.22), respectively. The corresponding results for race, sex, area SES, facility volume, surgical margin status, tumor grade, and radiation treatment did not indicate a statistically significant departure from the null (Table 3).
It is important to ensure that tumor resection (or incisional biopsy) specimens are of sufficient size and depth to include part of the advancing front of the tumor, as noted in a recent review of pitfalls and procedures in the histopathologic diagnosis of oral cancers.6 In addition, there appears to be a consensus that an appropriate evaluation of tumor invasion should include an assessment of tumor cells ahead of the main invasive front to determine the presence of lymphovascular emboli and neural spread.6-8
Our search of the literature failed to identify any previous studies evaluating the frequency of reporting DI and LVI/PNI in a general population-based sample of patients with oral tongue cancer. On the basis of our data, DI was reported in only approximately 20% of all eligible patients who were diagnosed with tongue cancer between 1997 and 2004, whereas mode of invasion for the same period was reported in <33% of cases. There was evidence that those proportions increased substantially after the CAP issued its recommendations in 2000; however, the increase in CAP protocol adherence was most pronounced in the first 2 years (2000 and 2001) and appeared to decline again in the subsequent years. Although data from the 2 registries represent geographically and demographically diverse study sites, it is important to keep in mind that the findings from our study may not apply to the entire US population.
There was evidence that patients with tongue cancer who resided in poverty areas were less likely to have depth and mode of tumor invasion documented in their pathology reports, although the association was only significant for DI and not for LVI/PNI. Mode of tumor invasion (LVI or PNI) was reported more commonly for cases diagnosed at hospitals or laboratory facilities that reported more than 3 cases per year; whereas reporting of DI did not seem to differ by facility case volume. These findings need to be viewed with caution because of the limitations of our data. For example, it might be important to examine the relation between completeness of pathology reporting and the level of training or specialization of individual pathologists, a factor that was not available in our study.
The data from the current study confirmed that DI might serve as an important prognostic factor for patients with tongue cancer. By using the cutoff level of >3 mm versus ≤3 mm, the association was strong (HR, 13), statistically significant (P < .001), and independent of patient demographics, other disease-related features, and treatment received. We also observed that the presence of PNI, but not LVI, was associated with a significant increase in mortality among patients with tongue cancer. The lack of statistical significance in the LVI analyses most likely is because our data were restricted to patients with oral cancer who underwent surgical resection, which was the most appropriate study group for the DI analyses. When survival was examined for all patients with tongue cancer (including base of tongue cancer), the result for LVI was significant (HR, 2.79; 95% CI, 1.21-6.48).
Our literature search identified several studies that examined subepithelial DI and/or the presence of PNI and LVI and their association with mortality and/or disease progression among patients with tongue cancer. Although the evidence is not entirely consistent across studies, on balance supports, it the importance of measuring DI and assessing LVI and PNI as part of the histologic workup for patients with newly diagnosed tongue cancer.9-21 It is important to emphasize that the relation between DI and survival in our data are not consistent with a simple biologic gradient. We produced no evidence that a gradual increase in DI beyond 3 mm is associated with a further increase in mortality. It is possible that the observed results reflect a threshold effect in which a DI of 3 mm is an indicator of a transition from superficial to deep invasion.6 Similar to our analyses, the majority of previously published studies examined the effect of DI on survival by using the cutoff levels of >3 mm or ≥4 mm.11, 15, 16, 18-20, 22 The biologic rationale for using this particular cutoff level needs to be understood, but the consistency of results across studies is noteworthy. We also confirmed previous reports indicating the prognostic value of tumor size11, 23, 24 and the lack of an association between survival and tumor grade.12, 25-27
The literature regarding histologic features and prognosis of tongue cancer is rather voluminous and goes back at least 20 years; however, it is important to point out several differences between our report and other studies published to date. To our knowledge, all previous studies were based on data from single institutions, whereas our analyses were population-based. Many studies examined survival in relation to DI or LVI/PNI alone,10, 14-17, 20, 23, 28 some examined both depth and mode of invasion,9, 11-13, 18, 19, 21, 22 and relatively few conducted multivariate analyses (usually relating DI and LVI/PNI to the presence of neck metastases).9, 11, 21 To our knowledge, only 1 other study evaluated the role of depth and mode of invasion simultaneously in a multivariate survival model; however, the results for DI in that study were not reported, perhaps because the authors chose to use tumor thickness as the variable of interest.9
Unlike hospital-based patient series, our sample, although it was relatively large (n = 339), included a high percentage of patients with no information on depth or mode of invasion. This lack of data on a large proportion of patients has to be considered a limitation; however, our primary objective was to evaluate the quality and availability of the histologic information and the potential importance of this information in an unselected sample that included all patients with surgical tongue cancer who were identified in a given geographic area over a given period.
The survival in our study group was better than the average survival in all patients with tongue cancer, because we only considered the patients who underwent surgical resection. These patients are expected to have a better prognosis than patients who have inoperable tumors. Our analyses of the data for all oral cancer cases reported to the MARGA and LA SEER registries from 1997 through 2004 demonstrated a 5-year survival rate of just over 50%; whereas the 5-year survival rate in our study group was nearly 58%. Nevertheless, higher overall survival in our study population is not expected to bias the within-group comparisons.
Another limitation of our survival analyses is the use of all-cause mortality, which fails to distinguish between deaths from tongue cancer and deaths caused by comorbidities. Conversely, it has been demonstrated that population-based studies that rely on cancer-specific mortality suffer from misclassification bias because of inaccuracies of the causes of death documented on death certificates.29-31
In summary, our population-based study results support previous reports that mode and depth of invasion are critical prognostic factors among patients with tongue cancer. It is worth pointing out that patients who had unknown DI (presumably representing a combination of patients with poor and good survival) had a statistically significant 9-fold increase in mortality compared with patients whose DI was documented as ≤3 mm. Thus, it appears that, for patients who have unknown histologic tumor characteristics, it may be difficult to determine prognosis even after evaluating traditional predictors, such as age, disease stage, and tumor size. Our finding that CAP recommendations led to an improvement in pathology reporting followed by an apparent decline underscores the need for continuous monitoring of adherence to the CAP protocol.