tetano
Editor, Senior Moderator
BMJ Open Respir Res
. 2026 Feb 24;13(1):e003728.
doi: 10.1136/bmjresp-2025-003728.
Predicting low oxygen in patients with acute COVID-19 infection isolating at home: a clinical prediction model
Robert Wu[SUP] 1 2 [/SUP], Alex Mariakakis[SUP] 3 4 [/SUP], Eyal de Lara[SUP] 3 [/SUP], Joseph Munn[SUP] 5 [/SUP], Daniyal Liaqat[SUP] 3 6 [/SUP], Salaar Liaqat[SUP] 3 [/SUP], Junlin Chen[SUP] 3 [/SUP], Teresa To[SUP] 7 8 [/SUP], Philip W Lam[SUP] 9 5 [/SUP], Andrew Simor[SUP] 9 5 [/SUP], Adrienne K Chan[SUP] 9 5 [/SUP], Nisha Andany[SUP] 9 5 [/SUP], Sameer Masood[SUP] 2 [/SUP], Nick Daneman[SUP] 9 5 [/SUP], Tiffany Chan[SUP] 5 10 [/SUP], Christopher Graham[SUP] 10 [/SUP], Vikram Comondore[SUP] 11 [/SUP], Alice Y Tu[SUP] 2 [/SUP], Andrea Gershon[SUP] 9 5 [/SUP]
Affiliations
Rationale: Identifying patients who could be safely managed at home versus those needing hospitalisation was a particular concern during early COVID-19. Respiratory viruses remain a concern, including new COVID-19 variants, influenza and respiratory syncytial virus. We developed COVIDFree@Home, a mobile application and clinician dashboard for remote monitoring, to determine if remotely collected measures could predict low oxygen saturation in home-isolating patients with COVID-19.
Methods: We conducted a prospective cohort study of patients newly diagnosed with COVID-19 from three Toronto hospitals between 2020 and 2022. Participants used the COVIDFree@Home app daily to enter symptoms, temperature, heart rate and oxygen saturation at home, which clinicians monitored via an online dashboard. We analysed baseline characteristics and remote monitoring variables to identify predictors of oxygen saturation ≤92%. A random forest classifier was trained to predict low oxygen saturation in the following 2 days. A secondary objective was to identify factors predicting hospitalisation.
Results: Of 431 participants, 376 (87.2%) entered at least one measure. Of the 376, 49 (13%) experienced low oxygen saturation, and 19 (5.1%) were hospitalised. Baseline factors associated with low oxygen saturation included older age, obesity, pre-existing pulmonary disease and Alpha/Beta variant. The classifier predicted future low oxygen saturation with an area under the curve of 0.68 (sensitivity 57%, specificity 72%, positive predictive value 3%, negative predictive value 99%). Key predictive factors included cough, lower baseline oxygen saturation, severe fatigue, higher temperature and higher heart rate. Factors associated with hospitalisation included dyspnoea, fever, Alpha/Beta variant and comorbidities of hypertension, mental illness and diabetes. Patients with a runny nose or sore throat were less likely to be hospitalised.
Conclusions: During the COVID-19 pandemic, remote monitoring along with knowledge of baseline characteristics could predict low oxygen saturation in the next 2 days in people with COVID-19. This approach may help identify individuals needing medical attention during future pandemics, though further model improvement is necessary.
Trial registration number: NCT04453774.
Keywords: COVID-19; Machine Learning; Telemedicine.
. 2026 Feb 24;13(1):e003728.
doi: 10.1136/bmjresp-2025-003728.
Predicting low oxygen in patients with acute COVID-19 infection isolating at home: a clinical prediction model
Robert Wu[SUP] 1 2 [/SUP], Alex Mariakakis[SUP] 3 4 [/SUP], Eyal de Lara[SUP] 3 [/SUP], Joseph Munn[SUP] 5 [/SUP], Daniyal Liaqat[SUP] 3 6 [/SUP], Salaar Liaqat[SUP] 3 [/SUP], Junlin Chen[SUP] 3 [/SUP], Teresa To[SUP] 7 8 [/SUP], Philip W Lam[SUP] 9 5 [/SUP], Andrew Simor[SUP] 9 5 [/SUP], Adrienne K Chan[SUP] 9 5 [/SUP], Nisha Andany[SUP] 9 5 [/SUP], Sameer Masood[SUP] 2 [/SUP], Nick Daneman[SUP] 9 5 [/SUP], Tiffany Chan[SUP] 5 10 [/SUP], Christopher Graham[SUP] 10 [/SUP], Vikram Comondore[SUP] 11 [/SUP], Alice Y Tu[SUP] 2 [/SUP], Andrea Gershon[SUP] 9 5 [/SUP]
Affiliations
- PMID: 41735015
- PMCID: PMC12933795
- DOI: 10.1136/bmjresp-2025-003728
Rationale: Identifying patients who could be safely managed at home versus those needing hospitalisation was a particular concern during early COVID-19. Respiratory viruses remain a concern, including new COVID-19 variants, influenza and respiratory syncytial virus. We developed COVIDFree@Home, a mobile application and clinician dashboard for remote monitoring, to determine if remotely collected measures could predict low oxygen saturation in home-isolating patients with COVID-19.
Methods: We conducted a prospective cohort study of patients newly diagnosed with COVID-19 from three Toronto hospitals between 2020 and 2022. Participants used the COVIDFree@Home app daily to enter symptoms, temperature, heart rate and oxygen saturation at home, which clinicians monitored via an online dashboard. We analysed baseline characteristics and remote monitoring variables to identify predictors of oxygen saturation ≤92%. A random forest classifier was trained to predict low oxygen saturation in the following 2 days. A secondary objective was to identify factors predicting hospitalisation.
Results: Of 431 participants, 376 (87.2%) entered at least one measure. Of the 376, 49 (13%) experienced low oxygen saturation, and 19 (5.1%) were hospitalised. Baseline factors associated with low oxygen saturation included older age, obesity, pre-existing pulmonary disease and Alpha/Beta variant. The classifier predicted future low oxygen saturation with an area under the curve of 0.68 (sensitivity 57%, specificity 72%, positive predictive value 3%, negative predictive value 99%). Key predictive factors included cough, lower baseline oxygen saturation, severe fatigue, higher temperature and higher heart rate. Factors associated with hospitalisation included dyspnoea, fever, Alpha/Beta variant and comorbidities of hypertension, mental illness and diabetes. Patients with a runny nose or sore throat were less likely to be hospitalised.
Conclusions: During the COVID-19 pandemic, remote monitoring along with knowledge of baseline characteristics could predict low oxygen saturation in the next 2 days in people with COVID-19. This approach may help identify individuals needing medical attention during future pandemics, though further model improvement is necessary.
Trial registration number: NCT04453774.
Keywords: COVID-19; Machine Learning; Telemedicine.