Adaptive DBS Should Focus on Readouts

Identifying neural correlates of depression is an area of active research for neuromodulation - the idea being that if you can find this measure then you don’t have to talk to the patient 1.

Several terms are thrown around, seemingly at random:

They’re all problematic, but some moreso.

Biomarkers vs Signatures

The term “biomarkers” has a long history driven mainly in the medical world by genetics. A dominant idea there is that the mere presence of a gene should translate to a measurable change in the presence of disease 2.

In a nutshell, the -marker part has binary baggage. Which is problematic for diseases that are increasingly acknowledged as spectrum 3.

“Signature” I tend to like, but it’s geared more towards scientists than it is to the general patient/public population. It may even conflate problematically with a literal signature - so it might be best to refer to this only in methods sections that need to talk through the vector space features are embedded in.

Biometric

Biometric has problematic baggage in its integration into security and privacy-breaching applications. Eg, fingerprinting as a biometric.

This is unfortunate, because “metric” is actually a great root for what we’re trying to do in the DBS space with physiologically derived measurements designed to reflect disease state.

Readouts

I like readouts as a term and am using that in my own work. To me, readout make the causal direction clear - there really isn’t one. These measures read the activity in the brain related to the disease.

Importantly, reading isn’t perfect. There’s noise always involved. Unlike biomarkers and biometrics, which overemphasize the biology in diseases that are only partially rooted there, readout seems to generalize better.

Summary

Stop using “biomarker” unless you want to bring in binarization. Stop using “biometric” if your focus is ethical use of physiologically-derived measures. Readout may be the best term as it better reflects the limitation of physiologic measurements while also enabling us to focus more on ethical applications.

  1. When I was a young graduate student this made me feel just a little uneasy, but otherwise seemed noble. Why bother the patient if you don’t need to? But after seeing how our healthcare system twists “objective” numbers to undermine patient values, I’ve come to despise this framing. Nevertheless, it’s a major driving force in the field today. ↩︎

  2. Genes are way more complicated than this, and we should adopt a more dynamical/energy landscape understanding of genes to see progress (imho). My own conception of this is that genes parametrize the potential landscape - potential in the energy sense, not in the possibility sense - which then affects the trajectories a patient navigates depending, inextricably, on context and drivers. A post for another day. ↩︎

  3. …or, better yet, multidimensional. ↩︎