Can we at least learn to fail faster?

For clinicians and (more important) patients, the current trial‐and‐error process of finding an effective depression treatment is frustrating and discouraging. Our ability to accurately match individual patients with specific medications is embarrassingly poor1. And, given the delayed symptomatic response to most depression treatments, the cycle time for each trial‐and‐error is as long as two months. It is therefore not surprising that many patients starting depression treatment become discouraged and never return.

As R. Perlis2 clearly describes, more accurate prediction or personalized treatment selection is not yet in sight. It may not even be just over the horizon. Much of the research that claims to support personalization of treatment is really more relevant to general prediction of depression outcome or general prediction of treatment response than to selection of specific treatments for individuals1. I refer to this mis‐application of evidence as “trying to answer a four‐group question with a two‐group research design”.

Stated statistically, personalized or precision treatment selection depends on interaction effects rather than main effects. If we hope to detect interactions rather than just main effects, research to support precision medicine for depression will certainly require much larger samples than we are accustomed to. More important, selection of and testing for promising interactions or moderators will likely require a clearer understanding of treatment mechanisms and more precise measures of outcome.

While accurate prediction of treatment success may be off in the distance, we are probably closer to faster detection of depression treatment failure. And “failing faster” would be a significant improvement on the current state. Even though depression treatment guidelines often advise waiting six weeks or more to assess the effectiveness of antidepressant medication, evidence from placebo‐controlled trials consistently demonstrates separation between active medication and placebo as early as seven days3. Even more promising, direct assessment of the neuropsychological “building blocks” of depression may allow even more rapid discrimination of treatment success or failure – identifying treatments unlikely to work earlier than traditional clinical measures.

For example, C. Harmer and colleagues at Oxford have shown that biased processing of emotional information (measured by a computerized task resembling a video game) can change within hours of a first dose of antidepressant medication4. We may soon welcome the day when we tell patients: “Download this app, take this pill tonight, and send me your test results in the morning. We can decide tomorrow if this medication is worth continuing”. That scenario would be a dramatic improvement over our current advice to “take this medication for a month, and we can decide then if it was worth the wait”.

The National Institute of Mental Health's Research Domain Criteria (RDoC) scheme5 helps to reveal the connection between these two goals (precision prediction of treatment success and rapid detection of treatment failure). Under the RDoC scheme, we hope to resolve the heterogeneous category of depression into more crisply defined components or building blocks. Any individual case of depression would represent some admixture of more fundamental elements such as decreased sensitivity to reward, impaired executive function, and over‐valuation of negative emotional stimuli.

Following this scheme, performance‐based assessment of those RDoC components could facilitate advances in both directions: faster detection of treatment failure and more accurate prediction of treatment success. Stated statistically, discovery of mediators (processes that explain or account for the success of any specific treatment) will inform the discovery of moderators (pre‐treatment characteristics identifying individuals for whom that treatment will be successful). Ultimately, this “experimental medicine” approach would also facilitate the development of more specific (and more effective) new treatments.

I expect that advances in precision medicine for depression will likely come sooner from neuropsychology than from genomics.

Gregory E. Simon
Group Health Research Institute, Seattle, WA, USA