Gene expression-based biomarkers associated with disease recurrence in patients with cutaneous squamous cell carcinoma may help in identifying a population subset considered to have high-risk disease.
Chrysalyne Schmults, MD, MSCE
Gene expression-based biomarkers associated with disease recurrence in patients with cutaneous squamous cell carcinoma (cSCC) may help in identifying a population subset considered to have high-risk disease.
In the United States, cSCC is diagnosed more frequently than any other cancer except basal cell carcinoma, with an estimated 700,000 new cases diagnosed each year.1,2 The widespread incidence and relatively low mortality rate of cSCC has led to its exclusion from national cancer registries such as The Surveillance, Epidemiology and End Results (SEER) Program.
Because the precise incidence of cSCC is not known, data regarding associated metastases and deaths remain tentative. The standard of care in cSCC is surgical removal of the primary lesion, which is curative in most cases involving early-stage disease. The outlook, however, is not always positive for patients with cSCC: approximately 3% of patients are at risk for nodal metastasis, and as many as 8700 individuals in the United States (or approximately 1% of those affected) die each year as a result of cSCC.1,3,4Researchers are currently attempting to refine existing staging systems to better distinguish between patients with low-risk disease and those with high-risk disease.3-7 Current staging systems rely on clinical features, not genetic signatures.
Chrysalyne D. Schmults MD, MSCE, director of the Mohs and Dermatologic Surgery Center at Brigham & Women’s Hospital, Boston, and an associate professor at Harvard Medical School, notes that current staging systems do not adequately identify cSCC recurrences and metastases, due to low sensitivity levels. “[These screening systems] are prone to misidentify patients as high-risk who will not go on to experience secondary events, meaning these systems have a low positive predictive value,” she said. “There is an unmet clinical need for an objective predictor of cSCC recurrence and metastasis.”
Schmults presented a poster at the 2018 American Society of Clinical Oncology (ASCO) Annual Meeting detailing the development of a gene expression signature associated with cSCC.7 “Identifying the subset of patients at risk of recurrence is critical for development of clinical trials in cSCC, which has no FDA-approved treatments and very few phase II trials,” she said in an interview. “Therefore, we set out to develop a gene expression-based biomarker associated with disease recurrence and metastasis in cSCC.”
Schmults emphasized that the relatively low morbidity and mortality statistics of cSCC belie a distressing clinical trend. “Only about 15% of deaths in this disease occur in patients who have internal metastases. That means that 85% of cSCC deaths occur in people with uncontrolled local disease, or local and nodal disease,” she observed. “That makes these deaths underrecognized and underappreciated...these patients don’t make their way to a cancer center to see a medical oncologist because so many of them never got to that stage where internal metastases would make it clear that they needed therapy.”
She went on to add that “after patients fail surgery and radiation several times, it is common for them to become ill with this large tumor burden. Then they often die from the disease.” Schmults and colleagues identified 73 candidate genes for analysis. They developed a multicenter protocol, ultimately collecting primary cSCC tumors and their accompanying clinical data from 14 US medical centers. They analyzed the tumors for messenger RNA expression of the genes potentially associated with cSCC metastasis. Patients included in the study were diagnosed later than 2006 and received at least 3 years of follow-up care if their cSCC had not recurred.7
Investigators accrued 541 samples. Of these, 305 cases included gene expression data. The investigators further refined the development set to 221 cases. Within the development set, there were 25 recurrences including 18 local and 13 metastases.7 To achieve predictive modeling, the study team used significantly varied genes and multiple machine-learning methods. Researchers also performed k-fold cross validation and bootstrapping and evaluated performance metrics.7
They recorded various demographic factors among the development cohort. Of the 221 patients, the median age was 74 among all patients, including those who did not experience a recurrence. Among the 25 patients who did experience recurrence, the median age was 69; however, according to the Pearson correlation test, this P value was not statistically significant. Males composed 74% of the cohort; of those who experienced recurrence, 84% were men, although the threshold for statistical significance was not met.7
A total of 6 patient attributes did have statistically significant P values. These include being immunocompromised and having a larger tumor diameter; median 2.9 cm among recurrent cases versus median 1.4 cm in the total cohort (P <.0001, for both). Three attributes had P values of <.001: Values for poorly differentiated or undifferentiated status, Clark Level IV/V, and the presence of perineural invasion were significant (P <.0001). Invasion into subcutaneous fat occurred in 10% of the total cohort and 12% of patients with recurrence (P =.015).7
The investigators also assessed probe performance. They considered gene expression to be detectable for a sample if the cycling time value was less than 40. Probe performance analysis showed that 69 of the target genes were expressed in 75% to 100% of the samples in recurrent and/or nonrecurrent cases.7
Six genes demonstrated consistent expression across all tested samples. The researchers used these genes as controls to normalize expression values of the remaining genes. Eighteen genes expressed differently between recurrent and nonrecurrent cases. Evaluation of the genes with multiple predictive modeling methods compared with existing staging methods included American Joint Committee on Cancer (AJCC) Version 7, AJCC Version 8, and Brigham and Women’s Hospital staging methods.7
Schmults’s team identified an optimal model for local recurrence activity that was 75% sensitive, 92% specific, had a 50% positive predictive value (PPV), and a 96% negative predictive value (NPV) for recurrence. The PPV of 50% compared favorably with AJCC staging at approximately 24% and BWH staging at approximately 18%, while maintaining a high NPV.7The investigators concluded that high-risk cSCC patients can be identified when machine-learning is applied to gene expression data. Characterizing their prognostic test as “robust,” they concluded that applying the test in clinical settings can guide postsurgical treatment planning for highrisk cSCC patients, such as nodal staging or adjuvant radiation. The test may also help identify patients who would benefit from enrollment in clinical trials examining systemic therapies for cSCC. The data suggest that a test for predicting recurrence and metastases outcomes in cSCC patients is possible, Schmults said, noting that further sample collection and model development are underway.