English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Most medical treatment decisions are sequential in nature. Hence, there issubstantial hope that reinforcement learning may make it possible to formulateprecise data-driven treatment plans. However, a key challenge for mostapplications in this field is the sparse nature of primarily mortality-basedreward functions, leading to decreased stability of offline estimates. In thiswork, we introduce a deep Q-learning approach able to obtain more reliablecritical care policies. This method integrates relevant but noisy intermediatebiomarker signals into the reward specification, without compromising theoptimization of the main outcome of interest (e.g. patient survival). Weachieve this by first pruning the action set based on all available rewards,and second training a final model based on the sparse main reward but with arestricted action set. By disentangling accurate and approximated rewardsthrough action pruning, potential distortions of the main objective areminimized, all while enabling the extraction of valuable information fromintermediate signals that can guide the learning process. We evaluate ourmethod in both off-policy and offline settings using simulated environments andreal health records of patients in intensive care units. Our empirical resultsindicate that pruning significantly reduces the size of the action space whilestaying mostly consistent with the actions taken by physicians, outperformingthe current state-of-the-art offline reinforcement learning method conservativeQ-learning. Our work is a step towards developing reliable policies byeffectively harnessing the wealth of available information in data-intensivecritical care environments.

Pruning the Way to Reliable Policies: A Multi-Objective Deep Q-Learning Approach to Critical Care