Friday, December 11, 2015
The objective of this workshop is to present problems of growing relevance in healthcare and discuss how advanced machine learning techniques can be used to address them through interactions between clinicians and ML-researchers.
Clinical practice follows a conceptually simple process: a patient shows symptoms of some malignancy and becomes subjected to the medical system. In a perfectly observed case, a clinician could read off the internal state of a patient (diagnosis) and understand a list of causes for these symptoms followed by actionable suggestions for how to optimally treat the patient to reach a more beneficial state. In practice, however, both the patients as well as the diseases and treatment-related actions are only available through disconnected and imperfect observational data without full access to the system connecting them causally. Thus, one of the chief concerns of the medical sciences is to use empirical observations of various measurements, symptoms, tests, treatments and outcomes to build empirical models of diseases and patient responses and act on them, aided by life-science which often provides causal knowledge of sub-processes on a molecular level that can help inform clinicians.
Recent years have seen an unprecedented rise in availability and size of collections of such clinical data, aided by the advent of the digital age. These rich data sources present opportunities to apply and develop machine learning methods to solve problems faced by clinicians and to usher in new forms of medical practice that are infeasible without modern machine learning methods.
Of particular interest to this year's workshop is statistical modeling. The role of modeling in healthcare is two-fold. First, it provides clinicians with a tool to aid exploration of hypotheses in a data-driven way. Second, it furnishes evidence-based clinically actionable predictions. Examples of machine learning tasks include disease progression models, where patients and diseases are characterized by states that evolve over time, or dose-response models, where the treatment details involving complex and often combinatorial therapies can be inferred in a data driven way to optimally treat individual patients. Such methods face many statistical challenges such as accounting for confounding effects like socioeconomic backgrounds or genetic alterations in subpopulations. Models take the role of in-silico testbeds that make use of not only large collections of patient records but also detailed patient specific data to probe medical hypothesis in a manner precise for a single patient.
In this workshop we want to bring together clinicians with problems associated with empirical data and machine learning researchers working on healthcare solutions. The goal is to have a discussion to understand clinical needs and the technical challenges presented by the needs including interpretable techniques which can adapt to noisy, dynamic environments and the biases inherent in the data due to being generated by the current standard of care.
Theofanis Karaletsos, Rajesh Ranganath, Suchi Saria, David Sontag
We acknowledge generous contributions from our sponsors Google DeepMind, Memorial Sloan Kettering Cancer Center, the National Science Foundation, and NVIDIA (who donated a Titan X for the best contribution award)