AI sleep apnoea screening looks excellent in pooled data and unpredictable in any one clinic

The edition in brief

Today's pulmonology desk is about the difference between a pooled estimate and what a test will do in your clinic. A meta-analysis of 60 studies of artificial-intelligence screening tools for obstructive sleep apnoea, 47 contributing to the pooled analysis, reports sensitivities of 0.94, 0.87 and 0.83 at apnoea-hypopnoea index thresholds of 5, 15 and 30 events per hour, with specificities of 0.77, 0.81 and 0.91. The summary areas under the curve run from 0.907 to 0.943. The prediction intervals tell a different story: specificity at a threshold of 5 has a 95% prediction interval from 0.30 to 0.96. Certainty of evidence was rated low or very low, and the authors call for prospective validation before routine use. A systematic review of 49 studies and 2864 preterm infants found that synchronising nasal intermittent positive pressure ventilation reduced extubation failure compared with nasal CPAP (risk ratio 0.38, 95% CI 0.17 to 0.85), with consistent reductions in breathing effort and patient-ventilator asynchrony. Neurally adjusted ventilatory assist achieved the best synchronisation, with trigger delay around 35 ms. In MARIPOSA, median time to symptomatic progression was 43.6 months with amivantamab plus lazertinib against 29.3 months with osimertinib (hazard ratio 0.69, 95% CI 0.57 to 0.83). A supplement to the intravitreal dexamethasone implant application has been approved, contents unspecified.