The effect of COVID-19 epidemic on vital signs in hospitalized patients: a pre-post heat-map study from a large teaching hospital

The Lombardy SARS-CoV-2 outbreak in February 2020 represented the beginning of COVID-19 epidemic in Italy. Hospitals were flooded by thousands of patients with bilateral pneumonia and severe respiratory, and vital sign derangements compared to the standard hospital population. We propose a new visual analysis technique using heat maps to describe the impact of COVID-19 epidemic on vital sign anomalies in hospitalized patients.

We conducted an electronic health record study, including all confirmed COVID-19 patients hospitalized from February 21st, 2020 to April 21st, 2020 as cases, and all non-COVID-19 patients hospitalized in the same wards from January 1st, 2018 to December 31st, 2018. All data on temperature, peripheral oxygen saturation, respiratory rate, arterial blood pressure, and heart rate were retrieved. Derangement of vital signs was defined according to predefined thresholds. 470 COVID-19 patients and 9241 controls were included. Cases were older than controls, with a median age of 79 vs 76 years in non survivors (p = < 0.002). Gender was not associated with mortality.

Overall mortality in COVID-19 hospitalized patients was 18%, ranging from 1.4% in patients below 65 years to about 30% in patients over 65 years. Heat maps analysis demonstrated that COVID-19 patients had an increased frequency in episodes of compromised respiratory rate, acute desaturation, and fever. COVID-19 epidemic profoundly affected the incidence of severe derangements in vital signs in a large academic hospital. We validated heat maps as a method to analyze the clinical stability of hospitalized patients. This method may help to improve resource allocation according to patient characteristics.
Materials and methodsConsecutive acute ischemic stroke patients who received reperfusion therapies at IRCCS Humanitas Research Hospital were included. The Italian NLP Bidirectional Encoder Representations from Transformers (BERT) model was trained with AL to automatically extract clinical variables from electronic health text. Simulated active learning performances were evaluated on a set of labels representing patients’ comorbidities, comparing Bayesian Uncertainty Sampling by Disagreement (BALD) and random text selection. Prognostic models predicting patients’ functional outcomes using Gradient Boosting were trained on manually labelled and semi-automatically extracted data and their performance was compared.ResultsThe active learning process initially showed null performance until around 20% of texts were labelled, possibly due to root layers freezing in the BERT model, yet overall, active learning improves model learning efficiency across most comorbidities. Prognostic modelling showed no significant difference in performance between models trained on manually labelled versus semi-automatically extracted data, indicating effective prediction capabilities in both settings.ConclusionsWe developed an efficient language model to automate the extraction of clinical data from Italian unstructured health texts in a cohort of ischemic stroke patients. In a preliminary analysis, we demonstrated its potential applicability for enhancing prediction model accuracy.