Distribution of skin cancers by number of Mohs micrographic surgery layers. For all 1131 cases, the mean number of Mohs micrographic surgery layers was 2.06.
Age distribution of patients who underwent Mohs micrographic surgery.
Odds of extensive subclinical spread based on univariate analysis of preoperative tumor size. Odds ratios were directly correlated with increasing size, and any tumor greater than 10 mm was significantly more likely to exhibit extensive subclinical spread compared with tumors measuring less than 5 mm.
Odds of extensive subclinical spread of nonmelanoma skin cancer based on univariate analysis by location. Odds ratios are relative to the cheek. Asterisk indicates statistical significance, with a 2-tailed P<.05.
Odds of extensive subclinical spread of nonmelanoma skin cancer based on univariate analysis by tumor classification. Odds ratios were calculated compared with nodular basal cell carcinoma (BCC). SCC indicates squamous cell carcinoma.
Odds ratios based on univariate analysis for important predictors of extensive subclinical spread in nonmelanoma skin cancer. Preoperative size and patient age were also significant predictors, as in Table 5. BCC indicates basal cell carcinoma; SCC, squamous cell carcinoma.
Batra RS, Kelley LC. Predictors of Extensive Subclinical Spread in Nonmelanoma Skin Cancer Treated With Mohs Micrographic Surgery. Arch Dermatol. 2002;138(8):1043-1051. doi:10.1001/archderm.138.8.1043
In nonmelanoma skin cancer, the clinically visible portion may represent a small fraction of microscopic tumor spread. Previous studies have examined individual risk factors for subclinical spread based on patient and tumor characteristics. However, these risk factors have not been prioritized or studied in combination.
To identify the most predictive risk factors for extensive subclinical tumor spread.
Retrospective analysis of 1131 Mohs micrographic surgical cases. Variables analyzed included patient age, sex, and immune status and lesion size, location, histologic subtype, and recurrence. Logistic regression was applied to identify important combinations of tumor characteristics and to quantify relative odds of spread.
Academic referral center.
Consecutive sample of all referred patients treated by a single Mohs micrographic surgeon in a 3-year period.
Main Outcome Measure
Number of Mohs micrographic surgical layers required to clear a tumor, with 3 or more layers defined as extensive subclinical spread.
The highest-risk tumors, with odds ratios greater than 6.0, were basosquamous and morpheaform basal cell carcinoma (BCC) on the nose, morpheaform BCC on the cheek, and those with a preoperative size greater than 25 mm. Other important risk factors were recurrent and nodular BCC on the nose; location on the eyelid, temple, or ear helix; neck tumors and recurrent BCC in men; and tumor size greater than 10 mm. Patients younger than 35 years were at lower risk. Increasing age and immunocompromise were not significant predictors.
Identification of lesions likely to exhibit extensive subclinical spread can help guide management to ensure complete tumor eradication and thereby reduce the risk of recurrence and its associated morbidity and cost.
APPROXIMATELY 40% to 50% of Americans who live to age 65 years will have skin cancer at least once.1 Nonmelanoma skin cancers, such as basal cell carcinoma (BCC) and squamous cell carcinoma (SCC), comprise the majority, with BCC accounting for more than 75% of all cases.2 Almost all skin cancers are curable if diagnosed and treated promptly; however, if not treated correctly, they can cause extensive destruction of normal tissue and even death.3 The clinically visible tumor surface may only represent one fifth of its local microscopic invasion.4 Therefore, identification of high-risk lesions with extensive subclinical spread can help determine optimal treatment and can significantly reduce morbidity and cost.
The degree to which cancers subtly extend into surrounding tissue is one of the main reasons for recurrence. Unnoticeable extension means that visual estimates of tumor perimeters may be insufficient and that inadequate removal may result.2,5 Any residual tumor left after treatment greatly increases the likelihood of clinical recurrence.6- 8
Because malignant extensions often are not clinically detectable due to the microscopic nature of tumor spread, complete tumor excision relies on microscopic tissue margin control. Of the techniques in practice, Mohs micrographic surgery (MMS) is the most accurate in determining subclinical spread of skin cancer.9,10 In contrast to standard vertical sections, which often sample less than 0.1% of the true surgical margin,11 in MMS, 100% of the surgical margin is examined, and any residual tumor is successively excised.
The number of MMS layers required to clear a tumor is indicative of the degree of subclinical spread. Because the first layer usually takes an average of 3-mm margins around the clinically apparent tumor, it is minimally dependent on preoperative size. A tumor requiring more than 1 or 2 layers for eradication evidences extensive subclinical spread that might surprise the clinician relying solely on clinical appearance. These tumors are much less likely to be adequately treated without microscopic margin control.9,10
Because of its accuracy in determining tumor-free margins, patients referred for MMS typically present with tumors that have a relatively higher statistical likelihood of local recurrence. Indications for MMS have been formulated based on studies12- 14 that tended to focus on a subset of risk factors for subclinical spread. Tumors were compared on the basis of one variable alone, for example, histologic type or anatomic location. Although certain tumor characteristics were found to be "high risk," this was on the basis of statistically significant differences from another characteristic without a general comparison between all groups of characteristics. As a result, there was no way to order or prioritize those differences that were most important. The separate studies that identified individual predictors of subclinical spread of nonmelanoma skin cancers did not quantify the magnitude of risk, which makes interstudy comparison and generalizability difficult.
The purpose of this study is to identify risk factors that are most predictive of extensive subclinical spread of tumor. These findings should expand on previously formulated indications for MMS by identifying the highest-risk lesions for inadequate treatment and likely recurrence. This information should be helpful in recommending appropriate referral and in preoperative planning regarding tumor excision and reconstruction.
Retrospective medical chart analysis was performed for 1131 cases of malignant skin tumors referred for MMS to the Department of Dermatology at the Beth Israel Deaconess Medical Center between July 8, 1996, and July 19, 1999, and excised by the same Mohs surgeon (L.C.K.). All patients who underwent MMS during these 3 years were included in the study except 6 patients whose medical charts were not available. In patients with multiple clinically distinct tumors, each lesion was counted as a separate primary tumor.15 Mohs micrographic surgery was performed on an outpatient basis under local anesthesia using the standard technique.12,13,16 An average of 3-mm margins were taken for each layer.
The number of layers required to clear the tumor was detailed on the operative record, and 3 or more layers was considered the criterion for extensive subclinical spread of tumor. Variables examined included patient age, sex, and immune status and lesion preoperative size, general and specific anatomic location, histologic type, and recurrence status.
For the purposes of analysis, patients were classified into age groups defined as younger than 35, 35 to 44, 45 to 54, 55 to 64, 65 to 74, 75 to 84, and 85 years and older. A patient was considered immunocompromised if medical history included human immunodeficiency virus, acquired immunodeficiency syndrome, lymphoma, leukemia, or organ transplantation. Preoperative size was based on measurements of the largest diameter of each tumor taken using a millimeter ruler. Postoperative defects of MMS were also measured but were not used in the final analysis because previous studies17 had demonstrated wide variability of tissue stretch in different anatomic locations. Preoperative size was analyzed according to the following categories: less than 5, 5 to 9, 10 to 14, 15 to 19, 20 to 24, and 25 mm and greater. Although tumor size greater than 20 mm is a standard indication for MMS, the additional classifications quantified the relative size contribution in combination with other MMS criteria.
Tumors were classified according to anatomic location into the general categories of nose, ear, eyelid, lip, forehead, cheek, chin, eyebrow, temple, neck, trunk, extremity, and scalp. The categories of nose, ear, eyelid, and lip were further subclassified based on specific location. Tumors on the nose were recorded as ala, bridge, dorsum, sidewall, or tip. Tumors on the ear were classified as antihelix, concha, lobule, helix, postauricular, or tragus. Eyelid tumors were specified as lateral canthus, medial canthus, lower eyelid, or upper eyelid. Tumors of the lip were classified as cutaneous or vermilion.
Tumor classification included nodular BCC, morpheaform BCC, basosquamous BCC, recurrent BCC, SCC, SCC in situ, recurrent SCC, adenoid cystic carcinoma, cellular dermatofibroma, desmoplastic trichoepithelioma, dermatofibrosarcoma protuberans, eccrine carcinoma, keratoacanthoma-like SCC, lentigo maligna, lymphadenoma, Merkel cell carcinoma, and nevus sebaceous. For analysis, the category of nodular BCC included superficial BCC, morpheaform BCC included infiltrative BCC and micronodular BCC, and SCC included keratoacanthoma-like SCC. These groupings were determined based on similar histologic appearance on frozen section and previous studies18 demonstrating comparable growth patterns. Tumors with mixed patterns of histologic subtype were classified according to the dominant pattern based on accepted histopathologic criteria. Recurrence status was based on patient history or multiple treatments required during the 3 years of follow-up.
Microsoft Access (Microsoft Corp, Redmond, Wash) was used for data compilation and management. Stata Version 6.0 (Stata Corp, College Station, Tex) was used for statistical testing.
The goal of statistical analysis was to construct a model of the probability of extensive subclinical spread based on explanatory characteristics of tumor. Data were initially tabulated and mean number of MMS layers were compared using 2-sample t tests. Potential explanatory variables were tested using the Pearson χ2 statistic. One-way analysis of variance was used for categorical variables. In all cases, P<.05 was considered statistically significant.
This preliminary analysis guided the selection of variables to be included in the logistic regression.19 Sample sizes for histologic types other than BCC and SCC were insufficient for meaningful statistical analysis, and these 17 cases were excluded. Nineteen cases of lentigo maligna were also excluded because the tumors had been excised using a modified MMS technique with immunohistochemical staining. The logistic model was therefore built based on data derived from 1095 cases of nonmelanoma skin cancers treated with standard MMS.
Univariate analysis identified explanatory variables for the outcome of 3 or more MMS layers. Variables were checked for interaction and confounding, and significant interactions were modeled using interaction variables. The odds of reaching 3 or more MMS layers after controlling for all other variables in the model was calculated for each risk factor. Odds for other histologic subtypes were calculated relative to nodular BCC, odds ratios (ORs) for location were relative to cheek, size was compared with less than 5 mm, and age was compared with the reference group of younger than 35 years. References were chosen based on low probability of subclinical spread.
Alternate multiple logistic regression models were compared using the likelihood ratio test, the Pearson χ2 statistic for goodness of fit, and the percent correct classification table. The pseudo R2, or coefficient of determination, was calculated for a rough estimate of overall fit.20
This process allowed calculation of ORs in the univariate analysis that identified important risk factors and allowed comparison of risk within individual categories. Logistic regression also allowed analysis of the interaction of risk factors and the comparison and quantification of odds of subclinical spread in high-risk combinations. An overall model was built that quantified the ORs for the most significant risk factors contributing to extensive subclinical spread of tumor.
In total, 1131 cases of skin cancers excised using MMS were analyzed; 244 of these tumors (21.6%) required 3 or more MMS layers to be cleared, which was considered indicative of extensive subclinical spread of tumor. The overall mean ± SD number of MMS layers for the data set was 2.06 ± 1.05. The distribution of MMS layers is shown in Figure 1.
The mean ± SD age of patients in the data set was 64.1 ± 15.3 years vs 65.8 ± 14.5 years among patients with extensive subclinical spread. This difference was not significant (P = .13). A summary of patient ages is shown in Figure 2. There were no statistically significant differences in mean MMS layers based on age. However, the group younger than 35 years had a significantly reduced OR of requiring 3 or more MMS layers (OR, 0.12; P = .03).
There were slightly more men than women (55.3% men and 44.7% women overall). A higher percentage of men (23.4%) progressed to 3 or more MMS layers than did women (19.4%), but this difference was not statistically significant (P = .10). Analyzing the distribution of anatomic locations by sex revealed interesting trends (Table 1). Men were significantly more likely to have tumors on the ear, temple, and scalp. Women had more tumors on the nose, the chin, an extremity, and the cutaneous lip.
A similar analysis of histologic subtypes by sex found that women referred for MMS were more likely to have morpheaform BCCs and basosquamous BCCs than were men, although the proportion of both histologic subtypes was relatively low overall (Table 2). Men had significantly higher numbers of BCCs and SCCs.
There were 39 immunocompromised patients (3.4%). The mean number of MMS layers for this group was 2.05, which did not differ significantly from that for other MMS patients (P = .93). Immunocompromised patients were not more likely to require 3 or more MMS layers (P = .58).
Preoperative size was a significant predictor of extensive subclinical spread in a univariate analysis. Increasing ORs from 1.8 to 3.7 were directly correlated with increasing preoperative size greater than 10 mm (Figure 3).
Univariate analysis on the basis of location revealed that the nose, ear, eyelid, temple, and neck had significantly higher odds of extensive subclinical spread than the cheek (Table 3
and Figure 4). Sublocations found to have elevated odds of subclinical spread relative to the cheek included the nasal ala, nasal bridge, nasal tip, ear helix, and lower eyelid.
The odds of subclinical spread for other tumor classifications compared with nodular BCCs in a univariate analysis revealed that morpheaform BCCs were 2.3 times (P<.001), recurrent BCCs were 3.2 times (P<.001), and recurrent SCCs were 4.2 times (P = .01) as likely to exhibit extensive subclinical spread (Figure 5).
Interactions between predictor variables were assessed, and of the many possible combinations, only those of histologic type, anatomic location, and sex were significant predictors of extensive subclinical spread. Anatomic location and tumor type had significant high-risk associations (Table 4). All subtypes of BCC were high risk on the nose, whereas nodular BCC on the ear and morpheaform and recurrent BCC on the cheek exhibited extensive spread.
Analysis of interactions between location and histologic type also identified significant combinations that were low risk. On the ear, SCC had an OR of 0.1, that is, it was 90% less likely to progress to 3 or more MMS layers (P = .048). On the cheek, nodular BCC was the lowest-risk histologic subtype, with an OR of 0.5 (P = .01).
Male sex significantly increased the likelihood of extensive subclinical spread in certain combinations. Recurrent BCC in men had an OR of 5.3 compared with women requiring 3 or more MMS layers (P = .001). Tumors on the neck in men were also more than 3 times as likely (OR, 3.6) as those in women to demonstrate wide microscopic extension (P = .02).
Factors that were significant predictors of extensive subclinical spread in univariate logistic regression models were included in a multivariate analysis. In the final model, the most significant predictors of extensive subclinical spread were anatomic location on the nose of any type of BCC; morpheaform BCC on the cheek; recurrent BCC in men; location on the neck in men; location on the ear helix, eyelid, or temple; and increasing preoperative size. Being younger than 35 years was a significant protective factor for extensive subclinical spread. The ORs are summarized in Table 5 and Figure 6.
Prediction of extensive subclinical spread can help identify tumors that otherwise might be inadequately treated and might result in a high rate of recurrence, with associated tissue destruction, morbidity, and cost. Our findings suggest that the most important predictors of extensive subclinical spread of nonmelanoma skin cancers are basosquamous, morpheaform, nodular, and recurrent BCC subtypes on the nose; morpheaform BCC on the cheek; any tumor on the eyelid, temple, or ear helix; any tumor on the neck in men; recurrent BCC in men; and preoperative size greater than 10 mm. Patients younger than 35 years were significantly less likely to exhibit extensive subclinical tumor spread. Increasing age and immunocompromise were not significant explanatory variables.
The predictors of subclinical spread identified in this study generally correspond with those in previous studies21- 25 that examined individual categories of risk factors. Analysis of interactions between risk factors also identified several novel combinations that were significant predictors of extensive subclinical spread. The model allowed quantification and comparison of risk associated with various tumor characteristics, which has not to our knowledge appeared in the literature.
Ninety-five percent of nonmelanoma skin cancers occur in patients aged 40 to 79 years13,26; however, the relative contribution of increasing age to the risk of subclinical spread of tumor has not been quantified. Our analysis found that although the mean number of MMS layers required tended to increase with age, this trend was not statistically significant. In addition, there was an 88% decreased risk of extensive subclinical spread in our sample among patients younger than 35 years controlling for all other variables. Our results are consistent with those of an earlier study27 that found significantly smaller lesion and defect sizes in young patients and no appreciable differences in tumor location, histologic type, or clinical morphologic features between MMS patients younger than 30 years and those older than 56 years. Although our data indicate that younger patients may exhibit less microscopic tumor spread than other patients referred for MMS, this should not supersede other well-established indications for MMS given the heightened cosmetic concern and the importance of clear margins in this patient population. Further study with larger patient samples may help clarify tumor behavior unique to younger people.
Few previous studies28 have compiled comprehensive data on sex differences in nonmelanoma skin cancer patients referred for MMS. Sex-based analysis in our data set shed light on epidemiologic and referral patterns for MMS. Men were more likely to have tumors on areas left exposed to sun by short or receding hair, such as the ear, temple, and scalp. Women were undergoing MMS in areas often treated by other modalities, such as the extremities, which may result from greater concern for scar size. Sex was significantly associated only with extensive subclinical spread in lesions on the neck in men and recurrent BCC in men; however, this was not an independent risk factor.
Previous studies29 have shown an increased incidence of cutaneous neoplasms among immunosuppressed individuals compared with the general population and a shift in distribution to 2:1 of SCC to BCC. In our sample, immune status was not predictive of local subclinical spread. Our results suggest that although the incidence of skin cancer may be increased among the immunosuppressed population, the risk of extensive microscopic tumor spread was not significantly higher compared with others with tumors indicated for MMS. This may be because of increased surveillance in this population. Moreover, because increased risk of metastasis of nonmelanoma skin cancer has been associated with immunocompromise,30 our results indicate that this risk may be attributable to other modes of spread (eg, hematogenous or lymphangitic) rather than to local tissue infiltration.
Previous studies31 have identified size greater than 20 mm as an indication for MMS based on higher likelihood of recurrence. Preoperative size greater than 10 mm proved to be a significant predictor of marked microscopic spread in our data set, with odds ranging from 2 to 4 times that of tumors measuring less than 5 mm. These findings are consistent with those of other studies32,33 that have shown a gradation of risk of recurrence with increasing tumor size.
Several previous studies28,32,33 have identified high-risk anatomic locations for subclinical spread of nonmelanoma skin cancer that were generally consistent with the significant sites of nose, ear, eyelid, eyebrow, and temple in our analysis. A variety of explanations for the greater potential for spread at these locations has been proposed, including the existence of embryonic fusion planes,25,28 higher density of nerves or the presence of perichondrium and periosteum in close proximity to dermis,34 and high density of midfacial sebaceous glands, which provide "pockets" for tumor islands.35 The highest-risk sites in our study were found to overlay cartilage or bone, which may suggest a greater role for silent perichondral or periosteal spread.
On the other hand, our identification of the neck in men as a high-risk site differs from previous studies,33,36 which found generally low recurrence rates in the neck. Although our sample size was small, none of 9 women compared with 7 of 14 men with neck tumors required 3 or more MMS layers. Because poikiloderma is more common in men, such patients may have tumors with less distinct clinical margins than those in women. Further study may identify anatomic distinctions that help explain the elevated sex risk.
Examination of histologic subtype in relation to subclinical spread in the univariate analysis corresponded with results of previously published studies of high-risk histologic types.18,37- 39 Our study also found recurrent BCC and SCC to evidence 3 to 4 times the risk of extensive subclinical spread compared with primary nodular BCC. Previous studies40 have illustrated that recurrent nonmelanoma skin cancers often display more invasive, sclerotic behavior. Previous treatment may decrease local host defenses, cause multiple foci of unconnected tumor, or cause entrapment of tumor cells in scar tissue that are subsequently released and cause recurrence.41,42 In multivariate analysis, recurrent BCC in men was especially likely to exhibit extensive spread, with odds more than 4 times as great as in women. This may be explained by heightened cosmetic sensitivity in women, who may notice recurrences earlier.
Analysis of the association between anatomic location and histologic subtype yielded several surprising results. Squamous cell carcinoma on the ear, usually considered a high-risk tumor for metastasis,43- 45 was a very low risk for extensive subclinical spread. Because dermatologists already have a high suspicion for SCC on the ear, patients with such tumors were possibly referred relatively earlier for biopsy and ultimately MMS. In addition, SCC in this location may have more distinct clinical margins with less microscopic invasion.
Similarly, although the cheek is usually considered a low-risk anatomic location, it was one of the highest-risk locations for morpheaform BCCs. Although basosquamous carcinoma overall showed similar risk of subclinical spread to nodular BCC, on the nose the risk was drastically increased, and basosquamous carcinoma on the nose was actually the highest-risk lesion in the study. Identification of such combinations may help guide further study to elucidate the tumor biologic characteristics in particular anatomic locations.
Other studies have used various formulas to assess subclinical spread, such as the difference between the largest postoperative and the largest preoperative tumor dimension4 or half this value.39 We focused instead on the number of MMS layers because layers were uniform, were less dependent on preoperative size, and could be compared between sites more consistently than postoperative size.17 Because our study focused on cases from a single surgeon, there was consistency in technique. For tumors of larger size or aggressive histologic type, where classic teaching recommends larger layers, this method actually underestimates subclinical spread such that any risk factors still identified are even less likely to be spurious.
Several other studies4,46- 48 have described statistical indicators of the extent of subclinical tumor growth. Studies are usually confined to either BCCs or SCCs and focus on the risks associated with age,27 sex,28 immune status,29 anatomic location,28,34 or histologic subtype18 alone. A recent study37 of postoperative MMS defects examined multiple variables separately but was limited to the periocular region. We chose a logistic regression model because it would allow risk factors to be assessed simultaneously and, by quantifying risk, would allow comparison and generalizability.
One limitation of our study was the lack of data on previous treatment modalities for the recurrent tumors. Recurrent skin cancers after radiation therapy have been shown to exhibit greater tumor extension than tumors that recur after other types of treatment.41 It would have been interesting to learn the microscopic tumor proliferation associated with various previous treatments. In addition, our study lacked information on the duration of the lesion before MMS. Tumor age has been shown to affect subclinical extension only indirectly through an increase in tumor diameter48 and was not independently predictive of recurrence.49 However, inclusion of these data in our model may have yielded more information.
Another limitation was that our study focused on a biased population: patients already identified as having higher-risk tumors indicated for MMS. The relatively high proportion of skin cancers with extensive subclinical spread in our sample yielded statistical power to identify the most predictive factors. Although these results are generalizable for nonmelanoma skin cancer in MMS patients, further research may allow the extension of these predictors to guide management in the general population.
A strength of our study was the large sample size of 1131 MMS cases and, specifically, 1095 nonmelanoma skin cancers. From an epidemiologic standpoint, our data were consistent with other referral-based MMS practices at tertiary care centers. The distribution of sex, age, anatomic locations, and histologic types for patients with tumors was comparable to other MMS studies2,27,50- 52 in university settings. More than 3% of cases in the sample were histologic types other than BCC or SCC, which likely reflects the tertiary care setting of this study. These other histologic types, such as lentigo maligna or dermatofibrosarcoma protuberans, are subject to much more variability in terms of MMS technique.53,54 We therefore chose to focus on 1095 cases of nonmelanoma skin cancer treated according to standard MMS.
By analyzing and controlling for the interaction of risk factors, our final model for prediction of subclinical spread allows identification of high-risk tumors with greater specificity. A risk scale based on these predictors has been developed and validated.55 Our results indicate that certain histologic subtypes are especially problematic in specific locations, such as basosquamous carcinoma and nodular BCC on the nose, whereas other low-risk locations such as the neck or cheek are much more likely to have extensive subclinical spread for men or for morpheaform BCC, respectively. By quantifying this risk, a patient referred for MMS can be adequately prepared given the odds of a more extensive surgery. Moreover, the surgeon may be more likely to plan his or her schedule to accommodate a lesion likely to require multiple stages, to be especially vigilant in histopathologic examination of tumors identified as high risk, and to plan for appropriate post-MMS reconstruction or repair. Most important, identification of lesions likely to require more extensive MMS may help ensure complete tumor eradication.
Accepted for publication October 10, 2001.
Corresponding author and reprints: Larisa C. Kelley, MD, Department of Dermatology, Beth Israel Deaconess Medical Center, 330 Brookline Ave, Boston, MA 02215.