Chest physiotherapy and outcomes in ICU

Sir: Thank you for the opportunity to reply to correspondents Beraldo and Timenetsky, who raise concerns regarding our use of statistics and suggest that we should have used ventilator-free days, as suggested by Schoenfeld and Bernard, as an alternative to the standard Kaplan–Meier analyses [1]. We considered the use of ventilator-free days as a new approach to analysing data. We had a number of reservations. Firstly, it is a new technique which has not been evaluated other than by those who suggested it. Secondly, the concept of ventilator-free days is not as tangible as the median time to become ventilator-free, which is generally clearer; however our main concern, which we raised at the initial presentation of the data at the European Society of Intensive Care meeting in Barcelona, was that it was an arbitrary calculation. Like Kaplan–Meier it effectively censures death, but the choice of 28 days is arbitrary. For example, at the meeting we demonstrated that the null hypothesis was rejected variably depending on the arbitrary cut-off point, the significance reducing progressively if the cut-off point increased from 10 to 35 days. The majority of our patients were discharged from intensive care by the second week; thus, a cut-off point of 28 days made no sense since the end point of the study had long passed. Mortality was separately analysed. Consequently, we opted to use conventional methodology. A second concern was the possible influence of the non-significantly higher APACHE II ICU day-1 risk of death among the physiotherapy group, which might have biased the results adversely. We propose that the groups had similar risks of death and actual mortality rates, suggesting equipoise rather than bias. It is well known that acuity-score-based probability of death estimates have considerable intrinsic error particularly with small populations of heterogeneous patients, such as in our study. Perhaps the correspondents are suggesting that such was the impact of physiotherapy that it improved outcome among the “sicker” physiotherapy group and matched the non-physiotherapy patients. This would have been an over-interpretation, and unlikely given that few critical care interventions on small study groups of heterogeneous patients, such as ours, have been demonstrated to improve outcome. Furthermore, the correspondents should bear in mind that the median statistic based on the Kaplan–Meier curve looks at the first 50% of patients in each group, whereas the statistic for APACHE II risk of death was based on all the patients in each group. Secondly, APACHE II reflects acuity on day 1. These patients were recruited after 48 h, and for this reason SOFA scores were included to indicate acuity at the time of recruitment; again, these were non-significantly different. The correspondents suggest that the greater number of neurosurgical patients in the physiotherapy group might have introduced bias against the physiotherapy group. We disagree. Neurosurgical patients invariably had a tracheotomy and had the advantage of meeting an easier ventilator-free end point, namely spontaneous ventilation on a CPAP circuit through a tracheotomy determined a priori. By contrast, the control group had fewer tracheotomies and had to achieve the more difficult ventilator-free end point of breathing spontaneously on a face mask.