Early Detection of Basal Cell Carcinoma of Skin From Medical History.

Abstract

Background and objectives: Basal cell carcinoma (BCC) is the most common form of skin cancer, originating from basal cells in the skin's outer layer. It frequently arises from prolonged exposure to ultraviolet (UV) radiation from the sun or tanning beds. Although BCC rarely metastasizes, it can cause significant local tissue damage if left untreated. Early detection is essential to prevent extensive damage and potential disfigurement. The United States Preventive Services Task Force (USPSTF) currently remains uncertain about the benefits and potential harms of routine skin cancer screenings in asymptomatic individuals. This paper evaluates the accuracy of predicting BCC using patients' medical histories to address this uncertainty and support early detection efforts.

Methods: We analyzed the medical histories of 405,608 patients, including 7733 with BCC. We categorized 25,154 diagnoses into 16 body systems based on the hierarchy in the Systematized Nomenclature of Medicine (SNOMED) ontology. For each body system, we identified the most severe condition present. Logistic Least Absolute Shrinkage and Selection Operator (LASSO) regression was then employed to predict BCC, using demographic information, body systems, and pairwise and triple combinations of body systems, as well as missing value indicators. The dataset was split into 90% for training and 10% for validation. Model performance was evaluated using McFadden's R2, Percentage Deviance Explained (PDE), and cross-validated with the area under the receiver operating characteristic curve (AUC).

Results: Diagnoses related to the Integument system showed an 8-fold higher likelihood of being associated with BCC compared to diagnoses related to other systems. Older (age from 60 to 69) white individuals were more likely to receive a BCC diagnosis. After training the model, it achieved a McFadden's R2 of 0.286, an AUC of 0.912, and a PDE of 28.390%, reflecting a high level of explained variance and prediction accuracy.

Conclusions: This study underscores the potential of LASSO Regression models to enhance early identification of BCC. Extant medical history of patients, available in electronic health records, can accurately predict the risk of BCC. Integrating such predictive models into clinical practice could significantly improve early detection and intervention.