Palliation bias is being overlooked in neonatal hypothermia trials

I thank Edwards and Azzopardi for their review1 of the trials on therapeutic hypothermia after perinatal asphyxia.2,3,4 Unfortunately, the evaluation of data quality fails to acknowledge a major shortcoming common to all three reported trials—“palliation bias”. A clear palliation bias in the control arm completely invalidates the use of death as an outcome measure, be it singly or in the composite form including disability. The efficacy study reported by Eicher et al2 is the only one that gives some account of the contribution that withdrawal of care makes to the death statistic, and it serves well to show that this is highly important. The criteria for withdrawal of care are not standardised and can be assumed to vary with differences in philosophy across units and between caregivers. More importantly, two factors must be recognised that will erroneously inflate the death rate in the control arm: (a) a commitment to treatment for 72 h in the cooling arm, which discourages early withdrawal in the ventilator‐dependent infant, and (b) parents who agree to participate in the trial are primed by a consent process that underlines the prospect of benefit in the presence of treatment. It is also incorrect to assume that the addition of disability to form the composite outcome will buffer this problem by taking into account the disability rate in the “extra survivors”.