Education: The one-armed bandit

Readers of psychiatric research should beware the all too common, one-armed bandit.

A number of effects happen to the groups over the course of a trial (placebo group dashed line, active treatment solid line): 

Natural improvement: In chronic conditions sick patients tend to get better (and then worsen again). Clinical trials purposefully enroll patients with severe symptoms so it’s fair to assume some improvement may simply be due to the natural course of the disease.

The study effect: Sometimes patients and raters to behave differently when they know they are participating in a trial. The regular visits of a psychiatric clinical trial sometimes ensures a better level of healthcare than patients would receive outside the trial.

The placebo effect: Sometimes patients and raters report an improvement in symptoms in a blinded trial simply because they believe they may be receiving a novel treatment. 

(the above three effects are often lumped together and mislabelled “placebo response”)

The treatment effect: Finally the elusive treatment effect, only patients in the active treatment arm receive this effect from the pharmacological properties of the treatment they are taking.

The raison d’être of the RCT is to isolate and report the observed treatment effect. In order to do this a control arm is necessary to separate the treatment effect from the others (by simple subtraction).

With a single arm study (without a control arm) it is impossible to determine what proportion of the observed change was due to treatment alone. Worse still investigators often assign all observed change to the treatment effect thus falling for the one-armed bandit fallacy.

Little can be said about efficacy, effectiveness, safety or just about anything from an uncontrolled, one-armed design in psychiatry. Journals should report results from these designs judiciously.

In our opinion patients, physicians and psychiatric research are not  well served by this design.

Suggested reading:
Hawthorne effect