Sampling plans for use of rapid adenosine triphosphate (ATP) monitoring must overcome variability or suffer statistical invalidity

Several authors have commented on the dangers of overstating the usefulness of these commercial ATP devices, the risks of alternative sources of ATP, the lack of correlation with specific pathogens of concern, the amount of ATP present within any particular cells or bacterial species, and the measurement variability that undermines statistical measures applied to the research.4–7 In this regard, and of specific concern in terms of method in Visrodia et al.,1 is the way that ATP measurements and samples were obtained—for example, samples from the brush and flush sampling were divided into only 2 parts, with one part apparently used for a single ATP test and the other part tested for protein residues. Reliance by Visrodia et al.1 upon the sample means of groups of singular ATP readings is undermined by the knowledge of variability where the standard deviation can be as high as 40% of the data mean for the individual brand of device used.8 The authors themselves note the risk of singular testing in the body of the discussion: “to sample more than one… and to use more than 1 rapid indicator,” but we wonder how the statistical assumptions hold valid without multiple (replicate) samples taken for the ATP testing. [...]the scale of RLU is completely relative and cannot be used interoperatively between differently branded devices.2,3 Second, the variability for each of the brands is so high that without a sampling approach that accounts for multiple samples at any one point, the ability of the scientists involved to meaningfully apply statistical methods renders the article subject to first principle flaws.9 Reporting the RLU readings on a log scale is not the same as taking multiple samples, identifying the median value, and then log plotting the data.