On 2015 Dec 06, Mark Bolland commented:

This comment provides background information to an earlier comment.

We were interested in why different meta-analyses of vitamin D supplements came to different conclusions, and noticed that different meta-analyses used different data from this trial. We identified several errors/inconsistencies in the text and emailed the lead author requesting clarification about the data in March 2014. He responded that the data in the Tables were correct. We replied that there were inconsistencies within the Tables as well and asked for clarification, but he did not respond.

Therefore in April 2014, we contacted the editor of Osteoporosis International advising the editor of the inconsistencies and requested that they be corrected. We felt this is particularly important as it is a highly cited, influential trial reporting benefits of vitamin D supplements on falls, and the inconsistencies occur in the treatment group numbers and the primary and secondary outcomes, so have a major bearing of the interpretation of the trial results. Our primary concern was/is to use the correct data for the trial in meta-analyses. In May 2014, the editor passed on an extract of the lead author’s response and indicated that an erratum would be published. However, we felt the lead author’s response was inadequate because it left a number of uncorrected inconsistencies/errors in the text. We pointed this out to the editor. We are not sure what action the editor took, but no erratum was published. We followed up with 2 further emails to the editor over the next year, and in April 2015, the editor advised that it seemed unlikely that an erratum would be forthcoming.

Therefore, we requested the permission to summarize the issues in a very brief letter. The editor agreed and a short letter summarizing the six errors/inconsistencies in the article was published (Osteoporos Int 2015;26:2713 Bolland MJ, 2015) with a response from the author (Osteoporos Int. 2015;26:2715-6 Pfeifer M, 2015). Unfortunately, in his response the author chose to correct only 2 of the errors identified and introduced a further inconsistency.

In a further letter to the editor, we therefore highlighted the remaining errors/inconsistencies and the consequence of using different possible data combinations from this trial for meta-analyses. The editor indicated that the issues we raised were important but probably irresolvable and offered to publish the letter in Archives of Osteoporosis. We indicated that our preference was that the data were corrected in the original publication rather than our letter being published. We think it quite straightforward to make the simple necessary corrections to two tables and a couple of sentences of text. We feel it is essential that the corrections occur as the errors are in the most important data from the trial: the treatment group numbers and the primary outcome data for falls. The editor considered the issue further and then published our letter (Arch Osteoporos. 2015;10:43 Bolland MJ, 2015) along with a response from the author (Arch Osteoporos. 2015;10:42 Pfeifer M, 2015).

In this second response, the author has still not corrected the identified errors, so, because of the inconsistencies/errors in the data, it is not possible to be sure how many women were in each randomized treatment group, how many women had a fall during the trial, how many total falls occurred in each treatment group, or what is the breakdown of participants by numbers of falls (no falls, 1 fall, 2 falls, etc).

Therefore, we think that the study should be excluded from meta-analyses and systematic reviews and its results viewed with caution until consistent data are provided by the authors with some explanation as to how the errors/inconsistencies occurred.

Mark Bolland, Andrew Grey. University of Auckland

On 2015 Dec 06, Mark Bolland commented:

There are several errors/inconsistencies in the data related to treatment group numbers and numbers of falls and fractures in this paper. These include:

• 1. The text and Table 1 reports 121 participants in each treatment group, but Table 3 reports 120 for Ca and 122 for CaD.
• 2. Table 3 reports that 75 participants fell with Ca and 49 with CaD, but the breakdown of fallers in Table 3 sums to 71 for Ca and 53 for CaD.
• 3. The text reports the total number of falls as 171 with Ca and 76 with CaD, but Table 3 reports 169 with Ca and 106 with CaD.
• 4. The breakdown of total falls in Table 3 sums to at least 111 with CaD, which is greater than the total falls for CaD reported in the text (76) and Table 3 (106).
• 5. The text reports the mean number of falls per group as 1.41 with Ca and 0.63 with CaD which are incompatible with the reported total number of falls in the text and Table 3.
• 6. The text reports 13 participants with fracture with Ca whereas Table 3 reports 12.

These are discussed in two letters to the editor (Osteoporos Int 2015;26:2713 Bolland MJ, 2015 and Arch Osteoporos. 2015;10:43 Bolland MJ, 2015) with responses by the lead author to each letter (Osteoporos Int. 2015;26:2715-6 Pfeifer M, 2015 and Arch Osteoporos. 2015;10:42 Pfeifer M, 2015).

Given the unwillingness/inability of the author to provide consistent data from the trial, we think it should be excluded from systematic reviews or meta-analyses until consistent data are provided by the authors, with some explanation as to how the errors/inconsistencies occurred.

We have described the full sequence of events in a separate comment.

Mark Bolland, Andrew Grey. University of Auckland

On 2015 Dec 06, Mark Bolland commented:

There are several errors/inconsistencies in the data related to treatment group numbers and numbers of falls and fractures in this paper. These include:

• 1. The text and Table 1 reports 121 participants in each treatment group, but Table 3 reports 120 for Ca and 122 for CaD.
• 2. Table 3 reports that 75 participants fell with Ca and 49 with CaD, but the breakdown of fallers in Table 3 sums to 71 for Ca and 53 for CaD.
• 3. The text reports the total number of falls as 171 with Ca and 76 with CaD, but Table 3 reports 169 with Ca and 106 with CaD.
• 4. The breakdown of total falls in Table 3 sums to at least 111 with CaD, which is greater than the total falls for CaD reported in the text (76) and Table 3 (106).
• 5. The text reports the mean number of falls per group as 1.41 with Ca and 0.63 with CaD which are incompatible with the reported total number of falls in the text and Table 3.
• 6. The text reports 13 participants with fracture with Ca whereas Table 3 reports 12.

These are discussed in two letters to the editor (Osteoporos Int 2015;26:2713 Bolland MJ, 2015 and Arch Osteoporos. 2015;10:43 Bolland MJ, 2015) with responses by the lead author to each letter (Osteoporos Int. 2015;26:2715-6 Pfeifer M, 2015 and Arch Osteoporos. 2015;10:42 Pfeifer M, 2015).

Given the unwillingness/inability of the author to provide consistent data from the trial, we think it should be excluded from systematic reviews or meta-analyses until consistent data are provided by the authors, with some explanation as to how the errors/inconsistencies occurred.

We have described the full sequence of events in a separate comment.

Mark Bolland, Andrew Grey. University of Auckland

On 2015 Dec 06, Mark Bolland commented:

This comment provides background information to an earlier comment.

We were interested in why different meta-analyses of vitamin D supplements came to different conclusions, and noticed that different meta-analyses used different data from this trial. We identified several errors/inconsistencies in the text and emailed the lead author requesting clarification about the data in March 2014. He responded that the data in the Tables were correct. We replied that there were inconsistencies within the Tables as well and asked for clarification, but he did not respond.

Therefore in April 2014, we contacted the editor of Osteoporosis International advising the editor of the inconsistencies and requested that they be corrected. We felt this is particularly important as it is a highly cited, influential trial reporting benefits of vitamin D supplements on falls, and the inconsistencies occur in the treatment group numbers and the primary and secondary outcomes, so have a major bearing of the interpretation of the trial results. Our primary concern was/is to use the correct data for the trial in meta-analyses. In May 2014, the editor passed on an extract of the lead author’s response and indicated that an erratum would be published. However, we felt the lead author’s response was inadequate because it left a number of uncorrected inconsistencies/errors in the text. We pointed this out to the editor. We are not sure what action the editor took, but no erratum was published. We followed up with 2 further emails to the editor over the next year, and in April 2015, the editor advised that it seemed unlikely that an erratum would be forthcoming.

Therefore, we requested the permission to summarize the issues in a very brief letter. The editor agreed and a short letter summarizing the six errors/inconsistencies in the article was published (Osteoporos Int 2015;26:2713 Bolland MJ, 2015) with a response from the author (Osteoporos Int. 2015;26:2715-6 Pfeifer M, 2015). Unfortunately, in his response the author chose to correct only 2 of the errors identified and introduced a further inconsistency.

In a further letter to the editor, we therefore highlighted the remaining errors/inconsistencies and the consequence of using different possible data combinations from this trial for meta-analyses. The editor indicated that the issues we raised were important but probably irresolvable and offered to publish the letter in Archives of Osteoporosis. We indicated that our preference was that the data were corrected in the original publication rather than our letter being published. We think it quite straightforward to make the simple necessary corrections to two tables and a couple of sentences of text. We feel it is essential that the corrections occur as the errors are in the most important data from the trial: the treatment group numbers and the primary outcome data for falls. The editor considered the issue further and then published our letter (Arch Osteoporos. 2015;10:43 Bolland MJ, 2015) along with a response from the author (Arch Osteoporos. 2015;10:42 Pfeifer M, 2015).

In this second response, the author has still not corrected the identified errors, so, because of the inconsistencies/errors in the data, it is not possible to be sure how many women were in each randomized treatment group, how many women had a fall during the trial, how many total falls occurred in each treatment group, or what is the breakdown of participants by numbers of falls (no falls, 1 fall, 2 falls, etc).

Therefore, we think that the study should be excluded from meta-analyses and systematic reviews and its results viewed with caution until consistent data are provided by the authors with some explanation as to how the errors/inconsistencies occurred.

Mark Bolland, Andrew Grey. University of Auckland