Comprehensive Summary
This paper by Saygili et al. studies the usage of machine learning models to help with prognosis of Adrenocortical carcinoma (ACC). ACC is a rare and aggressive type of tumor growth that is difficult to treat and prognosticate. There exists a dataset named S-GRAS which uses a number of measures to quantify the prognosis of ACC. The researchers use this dataset to their machine learning model and to validate it. The best version of their ML model was then published as a web-based tool for patients to assess their ACC risk. Three models performed the best at predicting disease progression and overall mortality and had a sensitivity and validity of 65% and 81% in the validation cohort. Their best ML model, Quadratic discriminant analysis, was particularly apt at predicting 5-year overall mortality with an AUC of 0.85. Their best models, Quadratic Discriminant Analysis, Light Gradient Boosting Machine, and Ada Boost Classifier, were implemented into their website and patients can currently use this site to access their own ACC predictions.
Outcomes and Implications
ACC is known to be very difficult to track and may become aggressive very unexpectedly. Thus, it is extremely important to develop tools for patients to track their progression as consistent direct medical testing is more difficult to do regularly. These ML models are currently available for patient use and can be employed on their own discretion. Currently, medical professions treating ACC find it challenging to manage frequent advanced calculations to assess risk and the usage of machine learning models for this task would greatly help streamline the patient care process for ACC.