Prediction of low pulse oxygen saturation in COVID-19 using remote monitoring post hospital discharge

Background: Monitoring systems have been developed during the COVID-19 pandemic enabling clinicians to remotely monitor physiological measures including pulse oxygen saturation (SpO₂), heart rate (HR), and breathlessness in patients after discharge from hospital. These data may be leveraged to understand how symptoms vary over time in COVID-19 patients. There is also potential to use remote monitoring systems to predict clinical deterioration allowing early identification of patients in need of intervention.

Methods: A remote monitoring system was used to monitor 209 patients diagnosed with COVID-19 in the period following hospital discharge. This system consisted of a patient-facing app paired with a Bluetooth-enabled pulse oximeter (measuring SpO₂ and HR) linked to a secure portal where data were available for clinical review. Breathlessness score was entered manually to the app. Clinical teams were alerted automatically when SpO₂ < 94 %. In this study, data recorded during the initial ten days of monitoring were retrospectively examined, and a random forest model was developed to predict SpO₂ < 94 % on a given day using SpO₂ and HR data from the two previous days and day of discharge.

Results: Over the 10-day monitoring period, mean SpO₂ and HR increased significantly, while breathlessness decreased. The coefficient of variation in SpO₂, HR and breathlessness also decreased over the monitoring period. The model predicted SpO₂ alerts (SpO₂ < 94 %) with a mean cross-validated. sensitivity of 66 ± 18.57 %, specificity of 88.31 ± 10.97 % and area under the receiver operating characteristic of 0.80 ± 0.11. Patient age and sex were not significantly associated with the occurrence of asymptomatic SpO₂ alerts.

Conclusion: Results indicate that SpO₂ alerts (SpO₂ < 94 %) on a given day can be predicted using SpO₂ and heart rate data captured on the two preceding days via remote monitoring. The methods presented may help early identification of patients with COVID-19 at risk of clinical deterioration using remote monitoring.