Selection bias introduced in a randomized clinical trial due to employee instigated deletion of newborn resuscitation videos.

Abstract

Evaluating if employee instigated deletion of newborn resuscitation videos resulted in selection bias compromising the validity of the NeoBeat trial. The NeoBeat randomized clinical trial was conducted at Stavanger University Hospital, Norway, to evaluate if continuously displayed ECG heart rate starting from birth resulted in a higher proportion of treated newborns receiving ventilation within 60 seconds. Videos of resuscitations were used to evaluate time to ventilation. Employees could request deletion of videos by e-mail. Halfway through the trial, protests over video recordings led to 12-month pause in data collection. An anonymous solution for employee video deletion request was implemented, resulting in a surge of video deletions. To investigate possible selection bias, proxies for severity of birth asphyxia before and after introduction of the anonymous opt-out solution were compared. In total, 7343/10,362 (71%) newborns of gestational age ≥34 weeks were enrolled before and 2566/4417 (58%) after implementation of the anonymous opt-out solution. Time of birth and video were available for 171 and 80 newborns, respectively. Implementation of an anonymous deletion request solution resulted in loss of 0 and 14 cases before and after, leaving 171 and 66 cases for comparison. There were no differences in participant characteristics, but significantly higher Apgar scores, shorter duration of ventilation, and fewer admissions after implementation of the anonymous opt-out solution. The introduction of anonymous employee deletion of videos resulted in systematic differences among included newborns, strongly suggesting a selection bias. The trial was terminated halfway without power to conclude on outcomes.