Non-linear dimension reduction defines subgroups of patients with acute COVID-19 in secondary care

Add to your list(s) Download to your calendar using vCal

• David Greenwood
• Wednesday 26 May 2021, 12:00-13:00
• online via zoom.

If you have a question about this talk, please contact Dr. Marc Haber.

The COVID -19 pandemic has led to over 3 million deaths. Clinical outcomes are heterogeneous and there is intense interest in defining patient subgroups that may be used to determine prognosis and develop tailored treatment options. We applied uniform manifold approximation and projection (UMAP), an algorithm for non-linear dimension reduction, to three independent cohorts of patients with acute COVID -19 to discover novel sub-populations on the day of admission to secondary care. UMAP was combined with Gaussian mixture model (GMM) clustering analysis to define clusters within a multi-site cohort (n= 6099). The clusters identified a range of differential clinical features of patients when observed in temporally independent cohorts from a single hospital (n=996, n=1011). Deconvolution of clinical features within each cluster identified unexpected inter-relationships between clinical variables that may potentially be targeted for therapeutic management. Integration of large, diverse clinical data sets using UMAP dimension reduction and clustering can therefore identify novel patient subgroups in an unsupervised way. This could represent a powerful approach for understanding pathogenesis and guiding therapeutic management in a range of disorders.

This talk is part of the CCB seminars series.

This talk is included in these lists:

• online via zoom

Note that ex-directory lists are not shown.