Search

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site’s data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare. Federated learning, a method for training artificial intelligence algorithms that protects data privacy, was used to predict future oxygen requirements of symptomatic patients with COVID-19 using data from 20 different institutes across the globe.

ObjectivesTo assess the effectiveness of prone positioning to reduce the risk of death or respiratory failure in non-critically ill patients hospitalized with COVID-19DesignPragmatic randomized clinical trial of prone positioning of patients hospitalized with COVID-19 across 15 hospitals in Canada and the United States from May 2020 until May 2021.SettingsPatients were eligible is they had a laboratory-confirmed or a clinically highly suspected diagnosis of COVID-19, required supplemental oxygen (up to 50% fraction of inspired oxygen [FiO2]), and were able to independently prone with verbal instruction. (NCT04383613).Main Outcome MeasuresThe primary outcome was a composite of in-hospital death, mechanical ventilation, or worsening respiratory failure defined as requiring at least 60% FiO2 for at least 24 hours. Secondary outcomes included the change in the ratio of oxygen saturation to FiO2 (S/F ratio).ResultsA total of 248 patients were included. The trial was stopped early on the basis of futility for the pre-specified primary outcome. The median time from hospital admission until randomization was 1 day, the median age of patients was 56 years (interquartile range [IQR] 45,65), 36% were female, and 90% of patients were receiving oxygen via nasal prongs at the time of randomization. The median time spent prone in the first 72 hours was 6 hours total (IQR 1.5,12.8) for the prone arm compared to 0 hours (0,2) in the control arm. The risk of the primary outcome was similar between the prone group (18 [14.3%] events) and the standard care group (17 [13.9%] events), odds ratio 0.92 (95% CI 0.44 to 1.92). The change in the S/F ratio after 72 hours was similar for patients randomized to prone compared to standard of care.ConclusionAmong hypoxic but not critically patients with COVID-19 in hospital, a multifaceted intervention to increase prone positioning did not improve outcomes. Adherence to prone positioning was poor, despite multiple efforts. Subsequent trials of prone positioning should aim to develop strategies to improve adherence to awake prone positioning.What is already known on this topicProne positioning is considered standard of care for mechanically ventilated patients who have severe acute respiratory distress syndrome. Recent data suggest prone positioning is beneficial for patients with COVID-19 who are requiring high flow oxygen. It is unknown of prone positioning is beneficial for patients not on high flow oxygen.What this study addsProne positioning is generally not well tolerated and innovative approaches are needed to improve adherence. Clinical and physiologic outcomes were not improved with prone positioning among hypoxic but not critically ill patients hospitalized with COVID-19.

BACKGROUND: Transthoracic echocardiography is routinely performed in patients with stroke or transient ischemic attack (TIA) to help plan secondary stroke management, but recent data evaluating its usefulness in this context are lacking. We sought to evaluate the value of echocardiography for identifying clinically actionable findings for secondary stroke prevention. METHODS: We conducted a multicentre cohort study of patients admitted to hospital with stroke or TIA between 2010 and 2015 at 2 academic hospitals in Toronto, Ontario, Canada. Clinically actionable echocardiographic findings for secondary stroke prevention included cardiac thrombus, patent foramen ovale, atrial myxoma or valvular vegetation. We identified patient characteristics associated with clinically actionable findings using logistic regression. RESULTS: Of the 1862 patients with stroke or TIA we identified, 1272 (68%) had at least 1 echocardiogram. Nearly all echocardiograms were transthoracic; 1097 (86%) were normal, 1 (0.08%) had an atrial myxoma, 2 (0.2%) had a valvular vegetation, 11 (0.9%) had a cardiac thrombus and 66 (5.2%) had a PFO. Patent foramen ovale was less likely among patients older than 60 years (adjusted odds ratio [OR] 0.34, 95% confidence interval [CI] 0.20–0.57), with prior stroke or TIA (adjusted OR 0.31, 95% CI 0.09–0.76) or with dyslipidemia (adjusted OR 0.39, 95% CI 0.15–0.84). Among the 130 patients with cryptogenic stroke who had an echocardiogram (n = 110), a PFO was detected in 19 (17%) on transthoracic echocardiogram. INTERPRETATION: Most patients with stroke or TIA had a normal echocardiogram, with few having clinically actionable findings for secondary stroke prevention. Clinically actionable findings, specifically PFO, were more common in patients with cryptogenic stroke.

BackgroundCOronaVirus Disease 2019 (COVID-19) can be challenging to diagnose, because symptoms are non-specific, clinical presentations are heterogeneous, and false negative tests can occur. Our objective was to assess the utility of lymphocyte count to differentiate COVID-19 from influenza or community-acquired pneumonia (CAP).MethodsWe conducted a cohort study of adults hospitalized with COVID-19 or another respiratory infection (i.e., influenza, CAP) at seven hospitals in Ontario, Canada.The first available lymphocyte count during the hospitalization was used. Standard test characteristics for lymphocyte count (×109/L) were calculated (i.e., sensitivity, specificity, area under the receiver operating curve [AUC]). All analyses were conducting using R.ResultsThere were 869 hospitalizations for COVID-19, 669 for influenza, and 3009 for CAP. The mean age across the three groups was 67 and patients with pneumonia were older than those with influenza or COVID19, and approximately 46% were woman. The median lymphocyte count was nearly identical for the three groups of patients: 1.0 ×109/L (interquartile range [IQR]:0.7,2.0) for COVID-19, 0.9 ×109/L (IQR 0.6,1.0) for influenza, and 1.0 ×109/L (IQR 0.6,2.0) for CAP. At a lymphocyte threshold of less than 2.0 ×109/L, the sensitivity was 87% and the specificity was approximately 10%. As the lymphocyte threshold increased, the sensitivity of diagnosing COVID-19 increased while the specificity decreased. The AUC for lymphocyte count was approximately 50%.InterpretationLymphocyte count has poor diagnostic discrimination to differentiate between COVID-19 and other respiratory illnesses. The lymphopenia we consistently observed across the three illnesses in our study may reflect a non-specific sign of illness severity. However, lymphocyte count above 2.0 ×109/L may be useful in ruling out COVID-19 (sensitivity = 87%).

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC)0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.

The COVID-19 pandemic has posed novel infection-control challenges for hospitals around the globe. One infection-control strategy that has been widely used in the context of other outbreaks is patient cohorting. This strategy refers to the placement of all patients exposed to the same laboratory-confirmed infectious agent in one location within the hospital. Little is known about the current utilization of this strategy with non-critically ill COVID-19 patients. An international multicenter, survey study was conducted to identify what strategies are planned or in place for patients with COVID-19 who are not critically ill. The survey was distributed from March 23-29th, 2020 to GIM physicians in Canada, USA, Denmark, Singapore, Hong Kong, and England. Of the 31 hospitals, 29 (94%) indicated that they plan on cohorting all GIM patients with COVID-19 to one location in the hospital. Among these 29 hospitals, 23 (79%) had implemented the plan at the time of the survey. The primary reasons for this decision were to limit the spread of COVID-19 and conserve PPE use. In conclusion, in the face of a novel virus there is near unanimity in the practice of patient cohorting as a potential mitigation strategy.

Clinical trials are underpowered to detect serious but rare adverse events (AEs). We applied a novel data mining approach to detect potential AEs for canagliflozin (CANA), the first marketed SGLT2 inhibitor, prior to public awareness of its potential safety concerns. In a U.S. commercial claims dataset (3/2013-9/2015), two pair-wise cohorts of patients with T2DM initiating CANA or a comparator, i.e., a DPP-4i or a GLP-1RA, were identified and propensity score matched. We used variable ratio matching with up to 4 comparators for each CANA initiator (44,733 CANA vs. 99,458 DPP-4i; and 55,974 CANA vs. 74,727 GLP-1RA). Using a tree-based scan statistic data mining method, we assessed thousands of incident outcomes in CANA vs. comparator initiators, scanning for statistical AE alerts after adjusting for multiple testing. Incident outcomes were defined by hierarchical groupings of clinically related ICD-9 codes. For inpatient and emergency room diagnoses, diabetic ketoacidosis was the only severe AE associated with CANA (p Disclosure M. Fralick: None. M. Kulldorff: None. S. Wang: Research Support; Self; Boehringer Ingelheim International GmbH, Johnson & Johnson, Novartis Pharmaceuticals Corporation. S. Schneeweiss: None. D. Redelmeier: None. E. Patorno: Other Relationship; Self; Boehringer Ingelheim International GmbH. Funding National Institute on Aging (K08AG055670)

Source Citation Etminan M, Sodhi M, Ganjizadeh-Zavareh S, et al. Oral fluoroquinolones and risk of mitral and aortic regurgitation. J Am Coll Cardiol. 2019;74:1444-50. 31514945