![]() One of the main difficulties in developing robust RL algorithms for healthcare is the highly confidential nature of clinical data. In particular, recommendations made by RL algorithms may not be safe if the training data omit variables that influence clinical decision making, or if the effective sample size is small 12. Nonetheless, some authors have highlighted the lack of reproducibility and potential for patient harm inherent in these methods 11. There is thus vast potential for RL algorithms to optimise personalised treatment regimens, as shown by early research on antiretroviral therapy in HIV 7, 8, radiotherapy planning in lung cancer 9, and management of sepsis 10. Clinicians often rely, at least in part, on clinical judgement to prescribe sequences of treatments, because the clinical evidence base is incomplete and available evidence may not represent the diversity of real-life clinical states. ![]() This generally requires modifying the duration, dosage, or type of treatment over time and is challenging due to patient heterogeneity in responses to treatments, potential relapses, and side-effects. Reinforcement learning for health care: promises and challengesĬlinicians treating individuals with chronic disorders ( e.g., human immunodeficiency virus (HIV) infection) or with potentially life-threatening conditions ( e.g., sepsis) often prescribe a series of treatments to maximise the chances of favourable outcomes. To address this challenge, this paper introduces the Health Gym project–a collection of highly realistic synthetic medical datasets that can be freely accessed to facilitate the development of machine learning (ML) algorithms, with a specific focus on RL. Health-related data is, however, not as easily accessible due to privacy concerns around the disclosure of private information. The success of RL was greatly facilitated by the availability of standard benchmark problems: tasks with publicly available datasets which allowed the research community to develop, test, and compare RL algorithms ( e.g., OpenAI Gym 4, DeepMind Lab 5, and D4RL 6). Recent studies that combine RL with neural networks have achieved super-human performances in tasks from video games 2 to complex board games 3. Reinforcement learning 1 (RL) is an area of artificial intelligence (AI) which learns a behavioural policy–a mapping from states to actions–which maximises a cumulative reward in an evolving environment. Furthermore, the risk of sensitive information disclosure associated with the public distribution of the synthetic datasets is estimated to be very low. The distributions of variables, and correlations between variables and trends in variables over time in the synthetic datasets mirror those in the real datasets. The datasets were created using a novel generative adversarial network (GAN). ![]() ![]() The three synthetic datasets described in this paper present patient cohorts with acute hypotension and sepsis in the intensive care unit, and people with human immunodeficiency virus (HIV) receiving antiretroviral therapy. Here we introduce the Health Gym - a growing collection of highly realistic synthetic medical datasets that can be freely accessed to prototype, evaluate, and compare machine learning algorithms, with a specific focus on reinforcement learning. This has hampered the development of reproducible and generalisable machine learning applications in health care. Clinical data are usually not openly available due to their confidential nature. In recent years, the machine learning research community has benefited tremendously from the availability of openly accessible benchmark datasets.
0 Comments
Leave a Reply. |