Isn't this just playing with the measurements until you get the outcome you want?

Full paper here if you can link to it http://journals.cambridge.org/download.php?file=%2F13134_5E7D3226392237DCEE8B5EE570DB2E65_journals__PSM_PSM40_02_S0033291709006102a.pdf&cover=Y&code=2a8a1b32514f0cfbb38b329bcd41b762

Well, I don't think so. HAMD17 should in no way be seen as a "gold standard" in measuring depression severity. I guess that's partially why the Montgomery-Åberg depression rating scale (MADRS) was created.

Psychometrically speaking, the HAMD17 isn't a homogenous scale and fares quite badly in item response analyses (Rasch-analyses). Now, the HAMD6 was created because it corresponds better with the clinical severity of depression and fulfill item response criteria so that it can be used as proper severity rating scale . And the HAMD6 wasn't created a posteriori to better the outcome of antidepressant trials, so it isn't "playing with the measurements".

I raise the question because this is one element completely overlooked in the whole "antidepressants doesn't work"-discussion.

There's still the question of what it is measuring and how this is validated - i.e. is it a valid measure of severity of depression or is it just a measure that picks up the actions of antidepressants in patients?

Also, isn't the 'antidepressants don't work' conclusion somewhat independent of the measures used since this seems to have been decided on by patients who either reject them because they are not deriving clinical benefit or the large numbers that have been classified as non-responders?

The HAMD6 was created when very experienced psychiatrists rated patients depression severities on a scale from 0-10. HAMD was co-administered, and the 6 items on the HAMD6 were sufficient to describe clinical severity and fulfilled item response requirements as a severity scale. So no antidepressants or other interventions were used to created the HAMD6, and therefore it wasn't created to just pick up actions of antidepressants (but isn't that why we do antidepressant trials?).

And I don't agree that the measure isn't important in the debate over antidepressants. The somewhat magical 3 point difference requirement from the NICE and failure to find large drug/placebo differences is very much a scale-sensible subject!

The only real test of antidepressant activity is whether a drug relieves depression or not - Regrettably the 3 point requirement has allowed just noticeable differences that make no real difference to the patient to be accepted as successes - unsurprisingly these drugs have failed when they have reached the clinic

Hence, isn't the most parsimonious explanation of the failure to find large drug/placebo differences that the drugs don't actually relieve depression in the great majority of patients?

Well, a possible problem with that view is explained in the Bech 2010 article I linked to. Measuring effect size based on heterogenous scales that measure both effect and side effects is problematic and the interpretation of these findings is somewhat problematic and not per se evidence for the drugs not working. And, even though response rates aren't 100% to antidepressants, the clinical experience is that the drugs work.

I would be interesting if the meta-analyses could be re-done on HAMD6 data or to se whole meta-analyses on MADRS data.

The evidence presented by drug companies like Eli Lilly when seeking approval from the FDA for drugs like Prozac showed there were at best only marginal effects (e.g. Leo J, Lacasse JR (2008) The Media and the Chemical Imbalance Theory of Depression Soc 45:35–45)

Since then, evidence for the effectiveness of the antidepressants has been exaggerated by publication bias in the scientific literature (Pigott HE, Leventhal AM, Alter GS, Boren JJ (2010) Efficacy and Effectiveness of Antidepressants: Current Status of Research Psychother Psychosom 79:267–279) and by misinformation presented in the popular media (Leo J, Lacasse JR (2008) The Media and the Chemical Imbalance Theory of Depression Soc 45:35–45)

The reality is that response rates to currently prescribed antidepressants may be as low as 40-50% (e.g. Trivedi MH et al (2006) Evaluation of outcomes with citalopram for depression using measurement-based care in STAR*D: implications for clinical practice. Am J Psychiatry 163, 28-40 Colman I, et al (2011) Predictors of long-term prognosis of depression. 183: 1969–76).

It must also be recognised that clinical experience is not a precise measure and can be influenced by a number of external factors - including the size of drug company advertising budgets (e.g. Lacasse JR, Leo J (2005) Serotonin and depression: A disconnect between the advertisements and the scientific literature. PLoS Med 2(12): e392).

My colleauges very recent study indicates 100% sensitivity (or correct detection), and 91% specificity (or correct exclusion) for the Hebrew version of the brief HAM-D

Article Psychometric properties of responses by clinicians and older...

One of the problems with the Hamilton depression scales is that they are clinician (usually psychiatrist) rated. Clinician rated scales offer a perspective, but are problematic in introducing systematic error into the process. Ideally two or three perspectives should be used from patient, clinician, researcher, significant other. Where only one perspective is available i would always favour self rating using a scale with adequate psychometric properties.

The development of briefer scales is based on a misunderstanding of the role of reliability in measurement. They are inevitably less reliable and as a result make the findings less replicable.