January 29th, 2025: In two academic papers, a team of researchers led by Newcastle University demonstrates how Artificial Intelligence can determine the course and severity of aggressive skin cancers, such as Merkel cell carcinoma (MCC). This enhanced clinical decision-making by generating personalised predictions of treatment-specific outcomes for patients and their doctors.

The international team combined machine learning with clinical expertise to develop the web-based system called “DeepMerkel” which offers the power to predict MCC treatment-specific outcomes based on personal and tumor-specific features.

The team suggests that this system could be applied to other aggressive skin cancers for precision prognostication, the enhancement of informed clinical decision-making, and improved patient choice.  

MCC

MCC is a rare but highly aggressive skin cancer. It can be difficult to treat - typically affecting older adults with weakened immune systems who present with advanced disease associated with poor survival.

Dr. Tom Andrew, a Plastic Surgeon and CRUK funded PhD student at Newcastle University and first author said; “DeepMerkel is allowing us to predict the course and severity of a Merkel cell carcinoma enabling us to personalise treatment so that patients are getting the optimal management.

“Using AI allowed us to understand subtle new patterns and trends in the data which meant on an individual level, we can provide more accurate predictions for each patient.

“This is important as in the 20 years up to 2020, the number of people diagnosed with this cancer has doubled and while it is still rare it is an aggressive skin cancer which is increasingly affecting older people.”

The research was conducted with Penny Lovat, Professor of Dermato-oncology, Newcastle University, and Dr Aidan Rose, Senior Clinical Lecturer, Newcastle University and Consultant Plastic Surgeon at Newcastle Hospitals NHS Foundation Trust.

Dr Rose said; “Being able to accurately predict patient outcomes is critical when guiding clinical decision making. This is particularly important when treating aggressive forms of skin cancer in a complex patient group which typically results in difficult, and sometimes life-changing, choices being made regarding treatment options. The developments we have made using AI allow us to provide personalised survival predictions and inform a patient’s medical team of the optimal treatment.”

In two complementary publications in Nature Digital Medicine and the Journal of the American Academy of Dermatology, the team describes how using advanced statistical and machine learning methods they developed the web-based prognostic tool for MCC.

Method

In Nature Digital Medicine, the team describes how they employed explainability analysis and the data to reveal new insights into mortality risk factors for the highly aggressive cancer, MCC. They then combined deep learning feature selection with a modified XGBoost framework, to develop a web-based prognostic tool for MCC which they termed DeepMerkel.

Analysing the data from nearly 11,000 patients in 2 countries, the researchers describe in the Journal of the American Academy of Dermatology how DeepMerkel was able to accurately identify high-risk patients at an earlier stage of cancer. This allows medics to make more informed decisions about when to use radical treatment options and intensive disease monitoring.

Patients first

The team hopes that DeepMerkel will provide better information for patients to make decisions with their medical teams about the best treatment for them as individuals.

Dr Andrew added: “With further investment, the exciting next step for our team is to further develop DeepMerkel so that the system can present options to help advise clinicians on the best treatment pathway open to them.”

The next step is to integrate the DeepMerkel website into routine clinical practice and broaden the scope of its use into other tumour types.

References:

• Andrew TA, Alrawi M, Plummer R, Reynolds NJ, Sondak V, Brownell I, Lovat PE, Rose A and Shalhout S (2024). A Hybrid Machine Learning Approach for the Personalized Prognostication of Aggressive Skin Cancers. Nature Digital Medicine Ref NPJDIGITALMED-11157R1
• Andrew TW, Erdmann S, Alrawi M, Plummer R, Shalhout SZ, Sondak V, Brownnell I, Lovat PE and Rose A (2024.) A multivariable disease-specific model enhances prognostication beyond current Merkel cell carcinoma staging: An international cohort study of 10,958 patients - ScienceDirect. Journal of the American Academy of Dermatology. :  Ms. No. JAAD-D-24-01601R3