Reading Our Minds: New book issues strong call to action to modernize psychiatry

by

The Rise of Big Data Psy­chi­a­try (The Wall Street Journal):

As a physi­cian, I need to fig­ure out three things when a new patient walks into my office: what their life is typ­i­cal­ly like, what has changed that made them seek treat­ment and what I can do to help them. It’s a com­plex prob­lem, and most fields of med­i­cine approach it by tak­ing mea­sure­ments. If I were a car­di­ol­o­gist eval­u­at­ing a patient’s chest pain, for instance, I would speak with the patient, but then I would lis­ten to their heart and mea­sure their pulse and blood pres­sure. I might order an elec­tro­car­dio­gram or a car­diac stress test, tools that weren’t avail­able a cen­tu­ry ago.

Because I’m a psy­chi­a­trist, how­ev­er, I eval­u­ate patients in pre­cise­ly the same way that my pre­de­ces­sors did in 1920: I ask them to tell me what’s wrong, and while they’re talk­ing I care­ful­ly observe their speech and behav­ior. But psy­chi­a­try has remained large­ly immune to mea­sure­ment. At no point in the exam­i­na­tion do I gath­er numer­i­cal data about the patient’s life or behav­ior, even though tools for tak­ing such mea­sure­ments already exist. In fact, you like­ly are car­ry­ing one around in your pock­et right now. Keep read­ing essay HERE, adapt­ed from the new book Read­ing Our Minds: The Rise of Big Data Psy­chi­a­try by psy­chi­a­trist Daniel Barron.

Relevant Study:

A machine learn­ing approach pre­dicts future risk to sui­ci­dal ideation from social media data (NPJ Dig­i­tal Med­i­cine). From the Abstract:

  • Machine learn­ing analy­sis of social media data rep­re­sents a promis­ing way to cap­ture lon­gi­tu­di­nal envi­ron­men­tal influ­ences con­tribut­ing to indi­vid­ual risk for sui­ci­dal thoughts and behav­iors. Our objec­tive was to gen­er­ate an algo­rithm termed “Sui­cide Arti­fi­cial Intel­li­gence Pre­dic­tion Heuris­tic (SAIPH)” capa­ble of pre­dict­ing future risk to sui­ci­dal thought by ana­lyz­ing pub­licly avail­able Twit­ter data. We trained a series of neur­al net­works on Twit­ter data queried against sui­cide asso­ci­at­ed psy­cho­log­i­cal con­structs includ­ing bur­den, stress, lone­li­ness, hope­less­ness, insom­nia, depres­sion, and anx­i­ety. Using 512,526 tweets from N?=?283 sui­ci­dal ideation (SI) cas­es and 3,518,494 tweets from 2655 con­trols, we then trained a ran­dom for­est mod­el using neur­al net­work out­puts to pre­dict bina­ry SI sta­tus. The mod­el pre­dict­ed N?=?830 SI events derived from an inde­pen­dent set of 277 sui­ci­dal ideators rel­a­tive to N?=?3159 con­trol events in all non-SI indi­vid­u­als with an AUC of 0.88 (95% CI 0.86–0.90). Using an alter­na­tive approach, our mod­el gen­er­ates tem­po­ral pre­dic­tion of risk such that peak occur­rences above an indi­vid­ual spe­cif­ic thresh­old denote a ~7 fold increased risk for SI with­in the fol­low­ing 10 days … We val­i­dat­ed our mod­el using region­al­ly obtained Twit­ter data and observed sig­nif­i­cant asso­ci­a­tions of algo­rithm SI scores with coun­ty-wide sui­cide death rates across 16 days in August and in Octo­ber, 2019, most sig­nif­i­cant­ly in younger indi­vid­u­als. Algo­rith­mic approach­es like SAIPH have the poten­tial to iden­ti­fy indi­vid­ual future SI risk and could be eas­i­ly adapt­ed as clin­i­cal deci­sion tools aid­ing sui­cide screen­ing and risk mon­i­tor­ing using avail­able technologies.

News in Context: