Feasibility,tolerance and effectiveness of enteral feeding in critically ill patients in prone position: More can be less with inappropriate analysis

Editor,

We read with interest the study by Savio and colleagues assessing enteral feeding in the prone position and discussed it at our regional journal club. ¹ We thank the authors for their timely contribution at the dawn of the COVID-19 pandemic, during which prone positioning has become a routine intervention. Whilst the paper has doubtless proved useful in informing decisions about nutrition, ² we wish to draw attention to a common statistical error; treating interventions as groups, leading to incorrect analysis.

Firstly, Savio et al. state that the independent sample t-, Mann-Whitney U-, and chi-square tests were used. All of these assume independence of observations or subjects (i.e. different participants in each group).^3,4 However, Table 1 (as mentioned in reference 1) describes a single cohort (n = 47), who in this study were exposed to two interventions: supine and prone positioning. However, analyses were performed on the assumption that there were two independent unpaired groups, effectively (and incorrectly) doubling the sample size. Paired tests of the single cohort for time-indexed summary statistics would have been more appropriate. A more detailed analysis with linear mixed models would have been a further improvement by assessing sequences of supine and prone feeding over time, for example with position (i.e. prone/supine) as a fixed effect, and the patients as a random effect, as the simplest model. The repeated measures over time could be further modelled with period as a fixed effect or the subjects being assigned with random coefficients for slope and intercept.

Secondly, we would like to draw attention to a potentially important error: Savio et al. state that “interruption in feeding was not statistically significant between prone hours (12.3%) and supine hours (24.1%) (p = 0.344).” However, Tables 2 and 3 (as mentioned in reference 1) show that, amongst all patients, there were 260.75 h of interruption out of 2831.6 h in the prone position (9.2%), and 878.75/15,506 in the supine position (5.67%). Not only do these results differ from those in the text, but the effect is opposite. Coincidentally, we note from Table 3 (as mentioned in reference 1) that 12.3% and 24.1% correspond to the percentage of nasogastric aspirates with a volume of >250 ml in the prone and supine position, respectively.

Finally, the data presentation in this manuscript is unusual and somewhat at odds with the statistical tests used. Both the Student t- and the Mann-Whitney U- tests involve comparison of central tendency and spread, whilst the chi-squared test compares frequencies.^3,4 However, most results are presented as percentages, either of prescribed doses of nutrition, or of total cumulative time in the prone or supine position. The only result presented as mean ± standard deviation was gastric residual volume (5.3 ± 3.9 ml prone and 15.1 ± 18.5 ml supine). However the large standard deviations indicate skewed data and therefore this should have been presented as median and interquartile range, and analysed with the Wilcoxon matched pairs test. Whilst percentages provide a useful summary, their use as the sole method of data presentation makes checking the statistical analyses impossible, which would have been useful with respect to the discrepancies identified above.

Based on the delivery of nutrition and interruption in feeding, ¹ we agree that this study demonstrates the clinical feasibility of enteral feeding in the prone position; our recent experience caring for patients with COVID-19 is consistent with this. However, the statistical inconsistencies in this paper undermine its credibility. Given that this work has already influenced practice, ² we invite the authors to take the opportunity to revise and update their analyses as appropriate.