On 2016 Jan 14, Brandon George commented:

When preparing the data and analysis code for public release, we noticed minor errors in the values reported in the paper. The errors and corrected values are detailed below, which do not substantially alter the conclusions. We apologize for the discrepancies and note that the data from the study and code for the appropriate analyses is available at ICPSR (https://www.openicpsr.org/repoEntity/show/53793?versionId=53793V1).

-In Table 2, the Mean(SD) for peak power in the HIIT group should be 957.7(209.7), and the p-value for the baseline difference between intervention groups should be 0.3875. This was due to a programming error.

-In Table 3, the p-value for the test of an overall change over time in QUICKI (complete case analysis) should be 0.3670, not 0.0254. This was due to a transcription error. Although the nominal significance of QUICKI changed, the overall pre-post change of insulin sensitivity (Si) was still significant and as Si is considered a more sensitivity measure of glucose tolerance and is the basis of our conclusion that MIT or HIIT intervention may improve glucose tolerance.

-In Table 3, the ITT analysis was done using a slightly different definition of subject compliance than was used in the complete case analysis, where subjects’ proportion of completed workouts was used as a covariate. The change in the p-values of the ITT analyses using the same covariate is small and there was no change in nominal significance. The exact values can be calculated using the available code and data at ICPSR.

On 2016 Jan 14, Brandon George commented:

When preparing the data and analysis code for public release, we noticed minor errors in the values reported in the paper. The errors and corrected values are detailed below, which do not substantially alter the conclusions. We apologize for the discrepancies and note that the data from the study and code for the appropriate analyses is available at ICPSR (https://www.openicpsr.org/repoEntity/show/53793?versionId=53793V1).

-In Table 2, the Mean(SD) for peak power in the HIIT group should be 957.7(209.7), and the p-value for the baseline difference between intervention groups should be 0.3875. This was due to a programming error.

-In Table 3, the p-value for the test of an overall change over time in QUICKI (complete case analysis) should be 0.3670, not 0.0254. This was due to a transcription error. Although the nominal significance of QUICKI changed, the overall pre-post change of insulin sensitivity (Si) was still significant and as Si is considered a more sensitivity measure of glucose tolerance and is the basis of our conclusion that MIT or HIIT intervention may improve glucose tolerance.

-In Table 3, the ITT analysis was done using a slightly different definition of subject compliance than was used in the complete case analysis, where subjects’ proportion of completed workouts was used as a covariate. The change in the p-values of the ITT analyses using the same covariate is small and there was no change in nominal significance. The exact values can be calculated using the available code and data at ICPSR.