PracticalDermatology

Artificial or augmented intelligence (AI) and machine learning are changing the way medicine is practiced across disciplines and specialties, and dermatology is no exception.

The latest study shows that machine learning can help identify early melanoma survivors who face a higher risk of cancer recurrence and help doctors determine which patients would likely benefit from increased surveillance or aggressive immunotherapy even at the initial stages of their disease.

“We have performed the largest and most comprehensive study to date assessing the ability of machine-learning algorithms to predict early-stage melanoma recurrence using clinicopathologic features extracted from electronic health records data in a real-world clinical setting,” say researchers who were led by Yevgeniy R. Semenov, MD, an assistant professor of dermatology at Massachusetts General Hospital in Boston.

Most patients with early-stage melanoma are treated with surgery, but today, patients with more advanced cancers often receive immune checkpoint inhibitors in addition to surgery. Being able to better predict who is most likely to benefit from these drugs can cut back on unnecessary risk among those who aren’t as likely to respond or those who will do well with surgery alone. Adjuvant immunotherapy is not without unpredictable and significant risks so being able to accurately identify who should undertake those risks would be of high clinical value.

What the Study Showed

Dr. Semenov and colleagues assessed the effectiveness of algorithms based on machine learning that used data from patient electronic health records to predict melanoma recurrence. They performed two types of prediction by using nine machine-learning algorithms: (1) melanoma recurrence classification and (2) time-to-event melanoma recurrence risk prediction. This new method was validated in a study published in npj Precision Oncology.

For the study, they reviewed 1,720 early-stage melanomas from the General Brigham health care system and the Dana-Farber Cancer Institute and extracted 36 clinical and pathologic features of these cancers to predict patients’ recurrence risk with machine learning algorithms. This study also included social determinants of health such as patient median income and insurance type.

Tumor thickness and mitotic rate, or rate of cancer cell division, were identified as the most predictive features, the new study showed. Other predictive factors included the American Joint Committee on Cancer (AJCC) stage, median income, insurance type, and age at diagnosis.

Looking Ahead

It’s exciting to see more high-quality studies assessing and validating algorithms that can lead to better outcomes for our patients. The predictive capabilities of these models can benefit from the incorporation of additional features such as gene expression profiling, digital histopathology images, and novel tumor biomarkers. The AI field is exciting, and I expect that with more investigation, we will know better how to interpret and apply these tools in the coming years. Predicting early-stage melanoma recurrence rate is likely just the tip of the iceberg.