It would be helpful if Dr Woelk clarified the rather confusing
material presented in his paper on behalf of the Remotiv/Imipramine Study
Group1. First, table 2 claims to present an analysis of “change from
baseline score” for the Hamilton depression rating scale. Since the
baseline itself is included as a covariate, whether raw outcomes, or
difference from baseline is used has no effect on the “difference of least
square means” and its associated confidence interval reported in the
table2. It does, however, affect the least square means themselves. Are
these means of the outcomes, or means of the difference from baseline? The
former is implied in the abstract but the latter is implied in the heading
to the table. If the latter is the case, the overall (non-significant)
difference is in favour of Imipramine rather than St John’s wort and the
statement that “lower scores indicate greater improvement” is false.
Furthermore there appears to be a considerable increase in variance
during the trial. The standard error of the difference between groups at
baseline can be calculated from table 1 to be 0.35. The corresponding
standard error for the outcome variable can be estimated from the
confidence limits in table 2 to be about 0.58. There is thus an increase
in variance by a factor of more than 2.7, despite the fact that the model
for analysis includes the baseline as an explanatory covariate. The
disparity in variances between raw outcome scores and baselines must be
even greater. A ratio of more than 3.5 to 1 would be implied by a modest
correlation of 0.5. Are these results correct?
If the BMJ required authors to provide an email address on their
submissions, it would be much easier for others to check such details.
Declaration of interest: the author is a consultant to the
pharmaceutical industry.
References
1. Woelk H. Comparison of St John's wort and imipramine for treating
depression: randomised controlled trial. BMJ 2000;321:536-539.
2. Senn SJ. Statistical Issues in Drug Development. Chichester: John
Wiley, 1997.
Rapid Response:
Depressed and confused
It would be helpful if Dr Woelk clarified the rather confusing
material presented in his paper on behalf of the Remotiv/Imipramine Study
Group1. First, table 2 claims to present an analysis of “change from
baseline score” for the Hamilton depression rating scale. Since the
baseline itself is included as a covariate, whether raw outcomes, or
difference from baseline is used has no effect on the “difference of least
square means” and its associated confidence interval reported in the
table2. It does, however, affect the least square means themselves. Are
these means of the outcomes, or means of the difference from baseline? The
former is implied in the abstract but the latter is implied in the heading
to the table. If the latter is the case, the overall (non-significant)
difference is in favour of Imipramine rather than St John’s wort and the
statement that “lower scores indicate greater improvement” is false.
Furthermore there appears to be a considerable increase in variance
during the trial. The standard error of the difference between groups at
baseline can be calculated from table 1 to be 0.35. The corresponding
standard error for the outcome variable can be estimated from the
confidence limits in table 2 to be about 0.58. There is thus an increase
in variance by a factor of more than 2.7, despite the fact that the model
for analysis includes the baseline as an explanatory covariate. The
disparity in variances between raw outcome scores and baselines must be
even greater. A ratio of more than 3.5 to 1 would be implied by a modest
correlation of 0.5. Are these results correct?
If the BMJ required authors to provide an email address on their
submissions, it would be much easier for others to check such details.
Declaration of interest: the author is a consultant to the
pharmaceutical industry.
References
1. Woelk H. Comparison of St John's wort and imipramine for treating
depression: randomised controlled trial. BMJ 2000;321:536-539.
2. Senn SJ. Statistical Issues in Drug Development. Chichester: John
Wiley, 1997.
Competing interests: No competing interests