Effect of a scoring system and protocol for sedation on duration of patients' need for ventilator support in a surgical intensive care unit

We would like to highlight two potential problems with the analysis
of the data presented.

First, from Figure 1, the data do not seem to follow a normal
distribution. When a small number of cases have extreme values, mean
scores may be biased. It is common for measures of duration to have
extreme scores, because they have a minimum of zero, and can be very
short, but have (almost) no limitation on how long they can be. In this
case, a very small number of individuals can have a large effect on the
results of a study. It is possible that the mean scores, particularly in
the pre-intervention case, may be biased by the small number of cases of
longer duration, which can be seen in Figure 1. In these circumstances a
transformation (such as a logarithmic transformation) can be used to
normalise the data prior to analysis, or alternatively a non-parametric
approach may be used.

Second, the data may show some signs of seasonality, that is, a
naturally occurring variation over the year. The two periods are not
directly comparable - the pre-intervention period includes the winter
month of November, which appears to be a peak. The post-intervention
period ends before that time. The lack of direct comparability of the two
time periods might suggest that the differences found may be, in part, an
artefact of the time periods used, rather than wholly due to the
intervention. A modelling approach that can include a seasonal component,
such as ARIMA, may be more appropriate in these circumstances.