Abstract
Non-invasive oxygen saturation ($SpO_2$) is a central vital sign used to shape the management of COVID-19 patients. Yet, there have been no reports quantitatively describing $SpO_2$ dynamics and patterns in COVID-19 patients using continuous $SpO_2$ recordings. We performed a retrospective observational analysis of the clinical information and 27,000 hours of continuous $SpO_2$ high-resolution (1 Hz) recordings of 367 critical and non-critical COVID-19 patients hospitalized at the Rambam Health Care Campus, Haifa, Israel. An absolute $SpO_2$ threshold of 93% most efficiently discriminated between critical and non-critical patients, regardless of oxygen support. Oximetry-derived digital biomarkers (OBMs) computed per 1-hour monitoring window showed significant differences between groups, notably the cumulative time below 93% $SpO_2$ (CT93). Patients with CT93 above 60% during the first hour of monitoring were more likely to require oxygen support. Mechanical ventilation exhibited a strong effect on $SpO_2$ dynamics by significantly reducing the frequency and depth of desaturations. OBMs related to periodicity and hypoxic burden were markedly affected up to several hours before the initiation of mechanical ventilation. In summary, OBMs, traditionally used in the field of sleep medicine research, are informative for continuous assessment of disease severity and response to respiratory support of hospitalized COVID-19 patients. In conclusion, OBMs may improve risk stratification and therapy management of critical care patients with respiratory impairment.
Introduction
In most clinical settings, oxygen saturation ($SpO_2$) is treated as a static number or a discrete measurement taken every few hours. However, COVID-19 is a dynamic respiratory disease where a patient's status can shift rapidly. Standard metrics like the ROX score often rely on these intermittent "snapshots," potentially missing the rich information hidden in the continuous "wiggles" of a patient's oxygen signal. This study sought to quantitatively describe $SpO_2$ dynamics to determine if continuous monitoring could provide earlier warning signs of deterioration and better differentiate between critical and non-critical patients than current clinical guidelines.
Methods and Datasets
The researchers conducted a retrospective observational analysis at the Rambam Health Care Campus in Israel. They extracted high-resolution (1 Hz) $SpO_2$ data directly from bedside monitors, resulting in a dataset of 27,000 hours of continuous recordings. To analyze this data, they used an Oximetry-derived Digital Biomarker (OBM) toolbox, extracting 175 different features categorized into general statistics, complexity, periodicity, desaturations, and hypoxic burden.
The study utilized data from 367 adult COVID-19 patients, divided into 162 critical and 205 non-critical cases. The dataset included 15,000 hours on room air, 8,000 hours on non-invasive oxygen support, and 4,000 hours on mechanical ventilation.
Key Insights and Results
The 93% Threshold and Predictive Power
While the WHO defines severe COVID-19 using a 90% $SpO_2$ threshold, this data suggests that 93% is the most efficient threshold for discriminating between critical and non-critical patients. A patient’s first hour of continuous monitoring proved highly predictive: those with a high CT93 (>60%) in their first hour had a hazard ratio of 3.07 for needing escalated oxygen support compared to those with low CT93.
Signal Texture and Early Warning
Mechanical ventilation doesn't just raise oxygen levels; it fundamentally alters the signal's dynamics by reducing its complexity and the frequency of desaturation events. Certain OBMs related to periodicity and hypoxic burden began to shift several hours before a patient was actually put on a ventilator, suggesting these markers could serve as an early-warning system for clinicians.
As illustrated in the signal traces, critical patients showed deeper and longer desaturations, whereas non-critical patients had more frequent but shallower drops. Mechanical ventilation reduced the frequency of desaturations by a fold change of 1.85 and significantly lowered signal complexity.
Limitations and Future Work
The authors acknowledge that this is a single-center retrospective study, which may limit generalizability. Furthermore, the study did not explore the specific effects of different ventilator settings on the $SpO_2$ signal. Future work could involve integrating these OBMs into wearable devices for home monitoring, creating an automated alert system that identifies when a patient needs hospitalization before they reach a state of crisis.
Summary updated