Search

The interpretation of lung auscultation is highly subjective and relies on non-specific nomenclature. Computer-aided analysis has the potential to better standardize and automate evaluation. We used 35.9 hours of auscultation audio from 572 pediatric outpatients to develop DeepBreath : a deep learning model identifying the audible signatures of acute respiratory illness in children. It comprises a convolutional neural network followed by a logistic regression classifier, aggregating estimates on recordings from eight thoracic sites into a single prediction at the patient-level. Patients were either healthy controls (29%) or had one of three acute respiratory illnesses (71%) including pneumonia, wheezing disorders (bronchitis/asthma), and bronchiolitis). To ensure objective estimates on model generalisability, DeepBreath is trained on patients from two countries (Switzerland, Brazil), and results are reported on an internal 5-fold cross-validation as well as externally validated (extval) on three other countries (Senegal, Cameroon, Morocco). DeepBreath differentiated healthy and pathological breathing with an Area Under the Receiver-Operator Characteristic (AUROC) of 0.93 (standard deviation [SD] ± 0.01 on internal validation). Similarly promising results were obtained for pneumonia (AUROC 0.75 ± 0.10), wheezing disorders (AUROC 0.91 ± 0.03), and bronchiolitis (AUROC 0.94 ± 0.02). Extval AUROCs were 0.89, 0.74, 0.74 and 0.87 respectively. All either matched or were significant improvements on a clinical baseline model using age and respiratory rate. Temporal attention showed clear alignment between model prediction and independently annotated respiratory cycles, providing evidence that DeepBreath extracts physiologically meaningful representations. DeepBreath provides a framework for interpretable deep learning to identify the objective audio signatures of respiratory pathology. [ABSTRACT FROM AUTHOR]

Background: Blended learning, defined as the combination of traditional face‐to‐face instructor‐led learning and e‐learning course, has never been validated as a teaching method for the effective use of manual defibrillators in cardiopulmonary resuscitation. Aim: To evaluate whether paediatric emergency and critical care providers exposed to a blended learning session performed better and recalled more defibrillator skills than those exposed to face‐to‐face learning only. Study design: A two‐period prospective, stratified, single‐centre, simulation‐based, randomized, controlled trial. Methods: Registered nurses and postgraduate residents from either a paediatric emergency department or an intensive care unit were randomly assigned to a blended learning or face‐to‐face learning sessions on the recommended use of a manual defibrillator. Participants' adherence to recommendations was assessed by testing defibrillator skills in three consecutive paediatric cardiopulmonary scenarios performed on the day of the training and once again 2 months later. The primary endpoint was the number of errors observed during defibrillation, cardioversion, and transcutaneous pacing at the time of the initial intervention. Results: Fifty participants were randomized to receive the intervention and 51 to the control group. When pooling all three procedures, the median total errors per participant was lower (2 [IQR: 1‐4]) in providers exposed to blended learning than in those exposed to face‐to‐face learning only (3 [IQR: 2‐5]; P =.06). The median of total errors per procedure was also lower. However, both training methods appeared insufficient to maintain appropriate skill retention over time as a repetition of procedures 2 months later without any refresher learning session yielded more errors in both groups. Conclusions: Learners exposed to blended learning showed a reduced number in the total amount of errors compared with those exposed to face‐to‐face learning alone, with waning of skills over time. Relevance to clinical practice: Proficiently teaching the use of a manual defibrillator can be performed through blended learning. [ABSTRACT FROM AUTHOR]

Antigen-based rapid diagnostic tests (RDTs) are used in children despite the lack of data. We evaluated the diagnostic performance of the Panbio-COVID-19 Ag Rapid Test Device (P-RDT) in children. Symptomatic and asymptomatic participants 0 to 16 years old had two nasopharyngeal swabs (NPS) for both reverse transcription-PCR (RT-PCR) and P-RDT. A total of 822 participants completed the study, of which 533 (64.9%) were symptomatic. Among the 119 (14.5%) RT-PCR-positive patients, the P-RDT sensitivity was 0.66 (95% confidence interval [CI] 0.57 to 0.74). Mean viral load (VL) was higher among P-RDT-positive patients than negative ones ( P  < 0.001). Sensitivity was 0.91 in specimens with VL of >1.0E6 IU/ml (95% CI 0.83 to 0.99) and decreased to 0.75 (95% CI 0.66 to 0.83) for specimens >1.0E3 IU/ml. Among symptomatic participants, the P-RDT displayed a sensitivity of 0.73 (95% CI 0.64 to 0.82), which peaked at 1.00 at 2 days post-onset of symptoms (DPOS) (95% CI 1.00 to 1.00), then decreased to 0.56 (95% CI 0.23 to 0.88) at 5 DPOS. There was a trend toward lower P-RDT sensitivity in symptomatic children <12 years (0.62 [95% CI 0.45 to 0.78]) versus ≥12 years (0.80 [95% CI 0.69 to 0.91]; P  = 0.09). In asymptomatic participants, the P-RDT displayed a sensitivity of 0.43 (95% CI 0.26 to 0.61). Specificity was 1.00 in symptomatic and asymptomatic children (95% CI 0.99 to 1.00). The overall 73% and 43% sensitivities of P-RDT in symptomatic and asymptomatic children, respectively, was below the 80% cutoff recommended by the World Health Organization. We observed a correlation between VL and P-RDT sensitivity, as well as variation of sensitivity according to DPOS, a major determinant of VL. These data highlight the limitations of RDTs in children, with the potential exception in early symptomatic children ≥12yrs.