Pulse oximeter perfusion index for assessment of brachial plexus block: a holy grail or a design fail?

Editor—I read with interest the recently published article by Abdelnasser and colleagues¹ regarding prediction of successful supraclavicular brachial plexus block using the pulse oximeter perfusion index (PI). The article represents an addition to other articles evaluating the usefulness of the pulse oximeter perfusion index in objective assessment of brachial plexus blocks.^2–4

I have two concerns that I need to address regarding the study by Abdelnasser and colleagues.¹

First, the authors used the PI ratio in addition to absolute PI values, because of the known high variability of the absolute PI values. Surprisingly, a cut-off absolute PI value of >3.3 at 10 min after injection showed a sensitivity of 100%, specificity of 100%, positive predictive value of 100%, negative predictive value of 100%, and an area under the receiver operating characteristic curve of 1. Such an outstanding predictive ability is very rarely seen in biological studies. When applying the cut-off PI value of >3.3 on the unblocked limb of the same cohort of patients at 10 min after injection, >25% of the patients will be diagnosed falsely to have a successful block, because the 75th percentile of PI is 3.5 (Table 2 of the article by Abdelnasser and colleagues).¹ This can be explained by the presence of bias, because blinding of the assessors was not mentioned to be part of the study design. Another explanation might be that the baseline PI values were not comparable between the successful group and the failed group. Using the guidelines provided by Arnold and colleagues⁵ to evaluate box plots of the baseline PI values of the successful and the failed groups may support this explanation (Fig. 1 of the article by Abdelnasser and colleagues).¹

Second, the authors mentioned that they compared the PI between blocked and non-blocked limbs using anova for repeated measures. Comparison between blocked and non-blocked limbs can be done with Student’s unpaired t-test (for normally distributed data) or the Mann–Whitney U-test (for non-normally distributed data); however, the data presented are spatially paired, and utilization of Student’s paired t-test (for normally distributed data) or the Wilcoxon signed-rank test (for non-normally distributed data) would be more appropriate and powerful. On the other hand, comparison of changes in PI over time in each limb should be done by anova for repeated measures (for normally distributed data) or Friedman’s test (for non-normally distributed data).

Declaration of interest

None declared.

References