Figure. Kaplan-Meier curves with accompanying life tables for the outcomes of interest. Cases represent tumors. A, Local recurrence (LR). B, Nodal metastasis (NM). C, Disease-specific death. D, All-cause death.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Schmults CD, Karia PS, Carter JB, Han J, Qureshi AA. Factors Predictive of Recurrence and Death From Cutaneous Squamous Cell Carcinoma: A 10-Year, Single-Institution Cohort Study. JAMA Dermatol. 2013;149(5):541–547. doi:10.1001/jamadermatol.2013.2139
Author Affiliations: Departments of Dermatology, Brigham and Women's Hospital (Drs Schmults, Han, and Qureshi and Mr Karia), and Massachusetts General Hospital (Dr Carter), Harvard Medical School, Boston, Massachusetts.
Importance Although most cases of cutaneous squamous cell carcinoma (CSCC) are easily cured with surgery or ablation, a subset of these tumors recur, metastasize, and cause death. We conducted the largest study of CSCC outcomes since 1968.
Objective To identify risk factors independently associated with poor outcomes in primary CSCC.
Design A 10-year retrospective cohort study.
Setting An academic hospital in Boston.
Participants Nine hundred eighty-five patients with 1832 tumors.
Main Outcomes and Measures Subhazard ratios for local recurrence, nodal metastasis, disease-specific death, and all-cause death adjusted for presence of known prognostic risk factors.
Results The median follow-up was 50 (range, 2-142) months. Local recurrence occurred in 45 patients (4.6%) during the study period; 36 (3.7%) developed nodal metastases; and 21 (2.1%) died of CSCC. In multivariate competing risk analyses, independent predictors for nodal metastasis and disease-specific death were a tumor diameter of at least 2 cm (subhazard ratios, 7.0 [95% CI, 2.2-21.6] and 15.9 [4.8-52.3], respectively), poor differentiation (6.1 [2.5-14.9] and 6.7 [2.7-16.5], respectively), invasion beyond fat (9.3 [2.8-31.1] and 13.0 [4.3-40.0], respectively), and ear or temple location (3.8 [1.1-13.4] and 5.9 [1.3-26.7], respectively). Perineural invasion was also associated with disease-specific death (subhazard ratio, 3.6 [95% CI, 1.1-12.0]), as was anogenital location, but few cases were anogenital. Overall death was associated with poor differentiation (subhazard ratio, 1.3 [95% CI, 1.1-1.6]) and invasion beyond fat (1.7 [1.1-2.8]).
Conclusions and Relevance Cutaneous squamous cell carcinoma carries a low but significant risk of metastasis and death. In this study, patients with CSCC had a 3.7% risk of metastasis and 2.1% risk of disease-specific death. Tumor diameter of at least 2 cm, invasion beyond fat, poor differentiation, perineural invasion, and ear, temple, or anogenital location were risk factors associated with poor outcomes. Accurate risk estimation of outcomes from population-based data and clinical trials proving the utility of disease-staging modalities and adjuvant therapy is needed.
Approximately 250 000 cutaneous squamous cell carcinomas (CSCCs) are diagnosed annually in the United States.1Although most cases are easily cured with surgical excision or ablative modalities, a subset of tumors recur, metastasize, and cause death.2 The risk of nodal metastasis (NM) in cohort and tumor registry studies has ranged from 2.0% to 5.8%.3-9 A single cohort study reported a risk of disease-specific death (DSD) of 1.5%.3
Factors reportedly associated with recurrence, metastasis, and DSD include perineural or lymphovascular invasion, poorly differentiated histological findings, tumor diameter of at least 2 cm, tumor invasion beyond the dermis, location on the lip or the ear, local recurrence (LR), and immunocompromise.3,10-12 Tumors with such features have been referred to as high-risk CSCCs. However, a precise definition of high-risk CSCC is lacking. Risk factors are often concurrent. Studies using multivariate analysis to define risk factors with individual prognostic impact have been underpowered to quantify risks of poor outcomes associated with different risk factors.3,10
This study was undertaken to define risk factors independently associated with poor outcomes on multivariate analysis adjusted for the presence of other concurrent risk factors. Because poor outcomes, particularly DSD, are relatively rare in CSCC, Quiz Ref IDwe used a 10-year retrospective cohort design to ensure a sufficiently large cohort to estimate risks.
The Department of Pathology electronic database at Brigham and Women's Hospital was searched using the keywords squamous and carcinoma to locate all pathology reports with a diagnosis of CSCC from January 1, 2000, through December 31, 2009. Reports identified by this search were then read by one of us (P.S.K.), and cases of noncutaneous SCC, in situ CSCC, and recurrent CSCC were excluded. We reviewed the electronic medical records of included patients to determine whether outcomes of interest occurred, including LR, NM, distant metastases (DM), death due to CSCC (ie, DSD), and all-cause death (ACD). Physician notes from the following providers were considered sources of outcome data: primary care physicians, dermatologists, and physicians involved in treatment of the primary tumor or recurrences (including departments of surgical oncology, medical oncology, head and neck surgery, plastic surgery, and radiation oncology). For the LR, NM, and DSD end points, cases were censored on the date of the last follow-up from these providers or on the date of death as per the Social Security Death Index. If death occurred more than 3 months after the date of the last follow-up, the case was considered lost to follow-up and was censored on the date of the last follow-up rather than on the date of death. For the ACD end point, cases were censored on the date of death if dead and on the date of medical record review by the investigator if alive as confirmed by the Social Security Death Index.
Additional information recorded from the medical record included a history of radiation treatment, burns, or other scarring processes in the area of the tumor and the immune status of the patient (presence of immunosuppression and cause). Tumors were assigned to 1 of the following 7 anatomic regions: head and neck (excluding the ear and temple), ear, arms or hands, legs or feet, trunk, perianal region, and genitalia. Features of the primary tumor, including the anatomic location, differentiation, depth of invasion, infiltrative or desmoplastic growth pattern, and presence of perineural or lymphovascular invasion, were obtained from pathology reports. Slides of tumors with perineural invasion were reviewed in a blinded fashion (by C.D.S. and J.B.C.) to record the diameter of the largest involved nerve, the number of nerves involved, the location of nerve invasion, and the tissue level of the perineural invasion. Unless otherwise stated in the pathology report, tumors were considered to be well differentiated, noninfiltrative, confined to the dermis, and free of lymphovascular invasion. Tumor diameter was obtained from the clinical notes recording the clinical diameter. If the notes did not state diameter, diameter was recorded as the smallest measurement of the gross excisional specimen recorded on the pathology report (eg, if the excised specimen was 5 cm long and 2 cm wide, the tumor was considered to have a diameter of 2 cm). Although this measurement includes marginal tissue, it also incorporates tissue shrinkage in processing (which ranges from 18%-31%).13-15 Whether the excised specimen's diameter or the clinical diameter most closely approximates the true tumor diameter is unknown; therefore, either measurement was considered valid in this study.
Treatment of the primary tumor and clearance of surgical margins were also recorded. Data regarding demographic factors, including the patient's age at diagnosis of the primary tumor, sex, race, and ethnicity were collected. Local recurrence was considered to have occurred if a pathology report documented invasive CSCC in the same anatomic location as a prior CSCC and a medical record note from the treating physician confirmed that this second lesion was considered a recurrence of the index primary tumor. Nodal metastasis was defined as pathologically confirmed CSCC in a draining nodal basin of the primary CSCC with no other potential source (such as a history of a mucosal primary CSCC that could drain to the same nodal basin). In cases of NM with a history of more than 1 primary CSCC in the skin region draining to the affected nodal basin, the primary tumor considered to have resulted in NM was that considered by the treatment team to have been the source of NM. If the treatment team had not determined which lesion was the source of NM, the case was reviewed by a physician (C.D.S.), and the lesion most likely to have led to NM was determined on the basis of temporality. Preference was given to tumors diagnosed within the nodal region within 2 years of the diagnosis of NM because most metastases occur within 2 years.3 Distant metastasis was determined by histological and/or radiological findings and documentation by the treatment team. Disease-specific death was considered to have occurred if the treatment team documented that the patient died of a specific CSCC or of complications that directly arose from CSCC. All-cause death was determined by review of the medical records and the Social Security Death Index. Patients who did not develop an outcome of interest were censored on the date of death or the date of the medical record review if alive.
Statistical analysis and model building were performed on primary tumors. We analyzed baseline demographic variables and CSCC clinical data using descriptive statistics and frequency tabulation. The Cox proportional hazards model was used to determine univariate and multivariate associations of risk factors with development of ACD. We used the competing risk analysis of Fine and Gray16 for the end points of NM, LR, and DSD. Competing risk models were chosen over Cox proportional hazards models for NM, LR, and DSD because the Cox models assume that no other competing events preclude the occurrence of events of interest. Cox proportional hazards models are therefore not ideally suited to this population of elderly CSCC patients because death from non-CSCC causes occurs commonly and the risk of patients dying of these other causes before they have an opportunity to develop disease recurrence, metastasis, or DSD is quite high. Statistical models for competing risks are used when an individual may experience more than 1 type of failure at the time of censoring. These competing causes of failure lead to dependent censoring and can produce biased estimates of hazard ratios or survival functions when using standard methods. In our analyses of particular outcomes (eg, LR), other outcomes (eg, deaths other than DSD) are treated as competing risks. Models from competing risks make estimation of cause-specific hazard functions possible (subhazard ratios [SHRs], which are interpreted similarly to hazard ratios) and cumulative incidence functions.16,17 The proportional hazards assumption (which applies to competing risk and Cox proportional hazards models) was checked via Schönfeld residual plots.
Multivariate models were built through forward stepwise variable addition followed by backward elimination. In this form of model building, modeling begins with the variable with the largest effect estimate on univariate modeling. Other variables added are based on the next variable with the largest effect estimate and are retained in the model if the Wald test comparing the smaller model with the larger model was significant at P ≤ .05 or if the P value of the χ2 test comparing the 2 models was borderline (>.05 to .10) and if addition of the variable changed the SHR by at least 10%. Life tables and Kaplan-Meier curves were generated to illustrate event-free survival for LR and NM, disease-specific survival for DSD, and overall survival for ACD. All statistical tests were performed using a 2-sided 5% type I error rate. Analyses were performed using commercially available software (STATA Statistical Software, Release 12; StataCorp). The Partners Human Research Committee approved this study.
The pathology database search yielded 1980 cases of primary invasive (non–in situ) CSCC. After medical record review, 148 cases were excluded because of insufficient primary tumor information, leaving 1832 tumors in 985 patients in the study cohort. Table 1 summarizes the characteristics of the study population. Most of the patients were white, with an equal sex distribution. The median age was 71 (range, 37-93) years. The median follow-up time was 50 (range, 2-142) months. Most patients (722 [73.3%]) had a single tumor; 209 (21.2%), 2 to 4 tumors; and 54 (5.5%), 5 or more tumors. Immunosupression was present in 143 patients (14.5%) owing to organ transplant (n = 58), rheumatoid arthritis (n = 36), chronic lymphocytic leukemia (n = 23), human immunodeficiency virus (n = 1), and other causes (n = 25). Tumor characteristics are summarized in Table 2. A wide variety of body locations were represented in the study, but only 9 tumors (0.5%) were located in the anogenital region. Most tumors were less than 2 cm in diameter (1557 [85.0%]), well differentiated (1208 [65.9]%), and confined to the dermis (1639 [89.5%]). Radiological imaging was ordered rarely (69 cases). Most imaging studies (n = 51) were performed at the time of disease recurrence. The remaining 18 studies obtained at primary presentation were performed in cases of deeply invasive tumors (beyond the dermis).
The following outcomes of interest occurred during the study period. Quiz Ref IDLocal recurrence occurred in 50 tumors (2.7%) and 45 patients (4.6%); NM, in 36 tumors (2.0%) and 36 patients (3.7%); DSD, in 21 tumors (1.1%) and 21 patients (2.1%); and ACD, in 613 tumors (33.5%) and 284 patients (28.8%). Three patients had multiple LRs from separate primary tumors. Otherwise, events arose from a single primary tumor. Eight tumors resulted in LR, NM, and DSD; 13, LR and NM; 8, LR and DSD; and 4, NM and DSD. Nine tumors resulted in multiple LRs, 3 led to multiple (2) NMs, and 1 tumor resulted in 2 separate DMs.
In 6 of the 36 tumors with NM, the patient had had more than 1 primary CSCC in the skin area draining to the affected nodal basin. In 2 of these patients, the treatment team had determined which primary CSCC was responsible for the NM. In 4 patients, an investigator (C.D.S.) made the determination, as detailed in the Methods section. Distant organ metastasis was rare, occurring in only 4 cases (0.2% of tumors and 0.4% of patients). Occurrence of DM was established by histological analysis in 2 of the 4 tumors and by radiological studies and documentation by the treatment team in all 4. All 21 cases of DSD were documented by the treatment team in the medical record. Most DSDs (17 of 21) were due to uncontrollable locoregional recurrence rather than distant organ disease. Given the relatively elderly patient population, death from non-CSCC causes was common, occurring in 26.7% of the cohort during the study period.
Results of univariate modeling of possible risk factors are summarized in Table 3, showing risk factors associated with the 4 outcomes of interest: LR, NM, DSD, and ACD. Results of multivariate modeling are summarized in Table 4. Five risk factors predicted outcomes. Tumor diameter of 2 cm or larger was associated with an elevated risk of LR (SHR, 5.6 [95% CI, 2.9-10.7]), NM (7.0 [2.2-21.6]), and DSD (15.9 [4.8-52.3]). Poor tumor differentiation predicted LR (SHR, 3.3 [95% CI, 1.8-5.9]), NM (6.1 [2.5-14.9]), DSD (6.7 [2.7-16.5]), and ACD (hazard ratio, 1.3 [1.1-1.6]). Tumor invasion beyond subcutaneous fat was associated with LR (SHR, 7.2 [95% CI, 3.4-15.3]), NM (9.3 [2.8-31.1]), DSD (13.0 [4.3-40.0]), and ACD (hazard ratio, 1.7 [1.1-2.8]). Perineural invasion predicted LR (SHR, 3.2 [95% CI, 1.5-7.0]), and DSD (3.6 [1.1-12.0]). Tumor location on the ear or temple was associated with LR (SHR, 4.8 [95% CI, 2.0-11.6]), NM (3.8 [1.1-13.4]), and DSD (5.9 [1.3-26.7]). Anogenital location (although this occurred in only 9 cases) was also significantly associated with poor outcomes, with NM in 1 of 5 perianal and 2 of 4 penile or vulvar cases and death in 2 of 5 perianal and 1 of 4 penile or vulvar cases. Location on the lip was not significantly associated with any outcome of interest, nor was treatment modality. The Figure shows Kaplan-Meier curves and life tables for LR, NM, DSD, and ACD. The life tables show that 98% of LR, 100% of NM, and 81% of DSD occur within 4 years of CSCC diagnosis.
Although most CSCC is easily cured, a small subset of patients experience poor outcomes. In the present study, 3.7% of CSCC patients developed NM and 2.1% died of CSCC. These findings are similar to the 4% and 1.5% figures reported in the only other cohort study of CSCC outcomes to date.3 The subset of tumors likely to result in these poor outcomes (termed high-risk CSCC) has not been defined consistently. For example, the American Joint Committee on Cancer uses criteria including tumor diameter larger than 2 cm, poor cellular differentiation, depth of invasion more than 2 mm or to the reticular dermis (Clark level IV), perineural invasion, or ear or mucosal lip location to classify high-risk tumors.18 In contrast, the National Comprehensive Cancer Network19 uses an expanded definition that also includes rapidly growing or recurrent tumors and those occurring in immunosuppressed patients. The National Comprehensive Cancer Network proposes a threshold of 4 mm in depth and moderate differentiation to define high risk. Prognostic estimates have not been published for this high-risk tumor subgroup. Clinicians currently have little evidence to guide decisions regarding nodal staging and adjuvant therapy.20,21 Subsequently, management of high-risk CSCC varies widely.22 Prognosis is poor once DM or unresectable locoregional recurrence develops,23-25 so determining which tumors or patients are at high risk for poor outcomes would be advantageous, allowing aggressive adjuvant treatment earlier in the course of the disease. Furthermore, accurate classification of high-risk CSCC will allow for inclusion of such patients in clinical trials regarding nodal staging and adjuvant therapy, which are nearly nonexistent in CSCC at this time.26 Such trials may greatly aid clinical decision making in high-risk CSCC.
To our knowledge, this study represents the first cohort study of CSCC outcomes large enough to calculate risks of poor outcomes with multivariate modeling accounting for the presence of multiple risk factors. Quiz Ref IDThe following 5 risk factors were found to contribute to poor outcomes: a tumor diameter of at least 2 cm, poor differentiation, depth of invasion beyond the subcutaneous fat, perineural invasion, andQuiz Ref IDlocation on the ear, temple, or anogenital region. This study suggests that these 5 risk factors may be among the most important prognostic factors in CSCC.
Although this study is the largest to date, it is underpowered to assess all risk factors with prognostic significance. Quiz Ref IDFor example, immunosuppression has been associated with poor CSCC outcomes in numerous studies, but only a small fraction of the study population was immunosuppressed. Mohs surgery has been associated with better outcomes in high-risk CSCC.12 Because the bulk of cases (69.5%) in this cohort were treated with standard excision, the impact of different treatment modalities could not be fully evaluated. In addition, risks of poor outcomes for specific unusual clinical presentations of CSCC, such as tumors in nonwhite patients (n = 21), could not be adequately assessed. Some factors that may have prognostic significance could not be evaluated, such as millimeters of tumor depth. Pathologists do not routinely report depth in CSCC primarily because most CSCCs are diagnosed via results of shave biopsies that transect the specimen, precluding accurate measurement of depth in millimeters. In addition, CSCCs often have substantial exophytic growth that may not be prognostically significant, and large or infiltrative tumors are often found to be deeply invasive on horizontal tissue sections (during Mohs or margin-controlled excision), from which vertical depth cannot be assessed. Given these difficulties in accurately measuring depth in CSCC, we did not attempt to record it retrospectively. The logistical difficulties in measuring depth in CSCC may make the tissue level of invasion a more reliable prognostic indicator in this disease. This subject deserves future study.
Another limitation is that this cohort was drawn from a single academic medical hospital. Such hospitals often see more patients with difficult tumors or advanced disease. Exclusion of recurrent tumors and focus on primary tumors minimizes such potential patient selection bias, but the population may still be higher risk than the CSCC patient population at large. This difference may inflate point estimates of risk. Accurate point estimates of risk await population-based studies of CSCC outcomes. However, the relative importance of different prognostic factors is unlikely to differ substantially between this cohort and patients with CSCC in general. Thus, the 5 risk factors found to predict poor outcomes in this study should be applicable to populations outside academic hospitals.
In this retrospective cohort study design, follow-up procedures were not uniform. Patients were censored in the study when follow-up by primary care physicians, dermatologists, or other physicians treating the tumor ceased. Local recurrence may have been underreported. However, we should not assume that such misclassification error would differ by risk factor, so bias is unlikely to have occurred on this basis. Nodal metastases, DMs, and DSDs are unlikely to have been underreported. In addition, reporting of risk factors on pathology reports may not have been uniform among pathologists or over time. However, a χ2 sensitivity analysis showed no significant difference between the first and last halves of the study inclusion period in the proportion of cases reported to have each histological risk factor.
In this study, patients with invasive CSCC had a 2.1% to 3.7% risk of metastasis or death. Tumor diameter of at least 2 cm, deep invasion, poor cellular differentiation, perineural invasion, and ear, temple, or anogenital location were risk factors for poor outcomes. These 5 risk factors may be among the most significant drivers of CSCC outcomes, but further studies are needed to replicate our findings. Clinical trials regarding nodal staging and adjuvant therapy may be targeted at patients with 1 or more such risk factors. Optimal management of high-risk CSCC awaits accurate risk estimation of outcomes from population-based data and clinical trials proving the utility of disease-staging modalities and adjuvant therapy.
Correspondence: Chrysalyne D. Schmults, MD, MSCE, Department of Dermatology, Brigham and Women's Hospital, 1153 Centre St, Ste 4349, Jamaica Plain, MA 02130 (firstname.lastname@example.org).
Accepted for Publication: October 15, 2012.
Author Contributions: Dr Schmults and Mr Karia had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the analysis. Study concept and design: Schmults. Acquisition of data: Karia and Carter. Analysis and interpretation of data: Karia, Carter, Han, and Qureshi. Drafting of the manuscript: Qureshi. Critical revision of the manuscript for important intellectual content: All authors. Statistical analysis: Schmults, Karia, Han, and Qureshi. Administrative, technical, or material support: Karia and Carter. Study supervision: Schmults and Qureshi.
Conflict of Interest Disclosures: None reported.
Online-Only Material: Listen to an author interview about this article, and others, at http://archderm.jamanetwork.com/multimedia.aspx#AuthorInterviews
Additional Contributions: George F. Murphy, MD, helped with extracting pathology reports.
Create a personal account or sign in to: