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

Machine learning for healthcare often trains models on de-identified datasetswith randomly-shifted calendar dates, ignoring the fact that data weregenerated under hospital operation practices that change over time. Thesechanging practices induce definitive changes in observed data which confoundevaluations which do not account for dates and limit the generalisability ofdate-agnostic models. In this work, we establish the magnitude of this problemon MIMIC, a public hospital dataset, and showcase a simple solution. We augmentMIMIC with the year in which care was provided and show that a model trainedusing standard feature representations will significantly degrade in qualityover time. We find a deterioration of 0.3 AUC when evaluating mortalityprediction on data from 10 years later. We find a similar deterioration of 0.15AUC for length-of-stay. In contrast, we demonstrate that clinically-orientedaggregates of raw features significantly mitigate future deterioration. Oursuggested aggregated representations, when retrained yearly, have predictionquality comparable to year-agnostic models.

Rethinking clinical prediction: Why machine learning must consider year of care and feature aggregation