An exploratory investigation of some statistical summaries of oximeter oxygen saturation data from preterm babies

Aim. To explore the potential usefulness of the mean, standard deviation (SD), and coefficient of variation (CV = SD/mean) of oximeter oxygen saturations in the clinical care of preterm babies. Methods. This was an exploratory investigation involving 31 preterm babies at 36 weeks postmenstrual age. All babies were healthy, but two were considered to be clinically unstable and required greater attention. Each baby's oxygen saturations were recorded using an oximeter and summarized by the mean, SD, and CV. The potential usefulness of each measure was assessed by its ability to distinguish the two unstable babies from the others. This was achieved using box plots and hierarchical clustering together with the Calinski-Harabasz (CH) index to quantify clustering performance (higher CH index indicates stronger clustering outcome). Results. The box plots flagged both unstable babies as outliers and none of the other babies. Successful clustering of the stable and unstable babies was achieved using the CV (CH = 72.8) and SD (CH = 63.3) but not with the mean. Conclusion. Taking the box plots and clustering results together, it seems plausible that variability might be more effective than mean level for detecting instability in oxygen saturation in preterm babies and that the combination of variability and level through the CV might be even better.