Big data and black-box medical algorithms
From the article:
"Machine-learning algorithms have been predicted to come into widespread use in the areas of prognosis, radiology, and pathology within the next few years, and diagnosis within the next decade, substantially increasing the power and ubiquity of existing clinical decision support software (1). Recent examples include a deep neural network that is able to identify skin cancer based solely on images of skin lesions, performing as well as board-certified dermatologists (2). A different algorithm identifies trauma patients in need of intervention to reduce the risk of hemorrhage, increasing the chance of prompt intervention without the need for consistent expert monitoring (3).
Consumers already have access to some machine-learning algorithms, such as smartphone apps that aim to identify developmental disorders in young children (4). Further afield from basic medical practice, algorithms can guide the allocation of scarce resources across health systems (5). How should algorithms like these be validated, regulated, and integrated into medical practice to ensure that they perform well in different populations at different times?"
Read more here!