COVID–19–related pneumonia requires different modalities of Intensive Care Unit (ICU) interventions at different times to facilitate breathing, depending on severity progression. The ability for clinical staff to predict how patients admitted to hospital will require more or less ICU treatment on a daily basis is critical to ICU management. For real datasets that are sparse and incomplete and where the most important state transitions (dismissal, death) are rare, a standard Hidden Markov Model (HMM) approach is insufficient, as it is prone to overfitting. In this paper we propose a more sophisticated ensemble-based approach that involves training multiple HMMs, each specialized in a subset of the state transitions, and then selecting the more plausible predictions either by selecting or combining the models. We have validated the approach on a live dataset of about 1,000 patients from a partner hospital. Our results show that rare events, as well as the transitions to the most severe treatments outperform state of the art approaches.

An HMM–ensemble approach to predict severity progression of ICU treatment for hospitalized Covid–19 patients / Mandreoli, Federica; Motta, Federico; Missier, Paolo. - (2021).

An HMM–ensemble approach to predict severity progression of ICU treatment for hospitalized Covid–19 patients

Federica Mandreoli;Federico Motta
;
2021

Abstract

COVID–19–related pneumonia requires different modalities of Intensive Care Unit (ICU) interventions at different times to facilitate breathing, depending on severity progression. The ability for clinical staff to predict how patients admitted to hospital will require more or less ICU treatment on a daily basis is critical to ICU management. For real datasets that are sparse and incomplete and where the most important state transitions (dismissal, death) are rare, a standard Hidden Markov Model (HMM) approach is insufficient, as it is prone to overfitting. In this paper we propose a more sophisticated ensemble-based approach that involves training multiple HMMs, each specialized in a subset of the state transitions, and then selecting the more plausible predictions either by selecting or combining the models. We have validated the approach on a live dataset of about 1,000 patients from a partner hospital. Our results show that rare events, as well as the transitions to the most severe treatments outperform state of the art approaches.
Mandreoli, Federica; Motta, Federico; Missier, Paolo
File in questo prodotto:
File Dimensione Formato  
2021_12_15_oral_presentation.pdf

accesso aperto

Descrizione: Special Session: Machine Learning in Health
Tipologia: Post-print dell'autore (bozza post referaggio)
Dimensione 721.1 kB
Formato Adobe PDF
721.1 kB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

Caricamento pubblicazioni consigliate

Licenza Creative Commons
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11380/1284098
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact