Machine-learning study finds EEG brain signatures that predict response to antidepressant treatments

by

–

Brain-wave pat­tern can iden­ti­fy peo­ple like­ly to respond to anti­de­pres­sant, study finds (Stan­ford Med­i­cine press release):

“A new method of inter­pret­ing brain activ­i­ty could poten­tial­ly be used in clin­ics to help deter­mine the best treat­ment options for depres­sion, accord­ing to a study led by researchers at the Stan­ford School of Medicine.

Stan­ford researchers and their col­lab­o­ra­tors used elec­troen­cephalog­ra­phy, a tool for mon­i­tor­ing elec­tri­cal activ­i­ty in the brain, and an algo­rithm to iden­ti­fy a brain-wave sig­na­ture in indi­vid­u­als with depres­sion who will most like­ly respond to ser­tra­line, an anti­de­pres­sant mar­ket­ed as Zoloft …

“This study takes pre­vi­ous research show­ing that we can pre­dict who ben­e­fits from an anti­de­pres­sant and actu­al­ly brings it to the point of prac­ti­cal util­i­ty,” said Amit Etkin, MD, PhD, pro­fes­sor of psy­chi­a­try and behav­ioral sci­ences at Stan­ford. “I will be sur­prised if this isn’t used by clin­i­cians with­in the next five years.”

Instead of func­tion­al mag­net­ic res­o­nance imag­ing, an expen­sive tech­nol­o­gy often used in stud­ies to image brain activ­i­ty, the sci­en­tists turned to elec­troen­cephalog­ra­phy, or EEG, a much less cost­ly technology…

The paper is one of sev­er­al based on data from a fed­er­al­ly fund­ed depres­sion study launched in 2011 — the largest ran­dom­ized, place­bo-con­trolled clin­i­cal tri­al on anti­de­pres­sants ever con­duct­ed with brain imag­ing — which test­ed the use of ser­tra­line in 309 med­ica­tion-free patients. The mul­ti­cen­ter tri­al was called Estab­lish­ing Mod­er­a­tors and Biosig­na­tures of Anti­de­pres­sant Response for Clin­i­cal Care, or EMBARC. Led by Trive­di, it was designed to advance the goal of improv­ing the tri­al-and-error method of treat­ing depres­sion that is still in use today.

“It often takes many steps for a patient with depres­sion to get bet­ter,” Trive­di said. “We went into this think­ing, ‘Wouldn’t it be bet­ter to iden­ti­fy at the begin­ning of treat­ment which treat­ments would be best for which patients?’”

The Study:

An elec­troen­cephalo­graph­ic sig­na­ture pre­dicts anti­de­pres­sant response in major depres­sion (Nature Biotechnology)

  • Abstract: Anti­de­pres­sants are wide­ly pre­scribed, but their effi­ca­cy rel­a­tive to place­bo is mod­est, in part because the clin­i­cal diag­no­sis of major depres­sion encom­pass­es bio­log­i­cal­ly het­ero­ge­neous con­di­tions. Here, we sought to iden­ti­fy a neu­ro­bi­o­log­i­cal sig­na­ture of response to anti­de­pres­sant treat­ment as com­pared to place­bo. We designed a latent-space machine-learn­ing algo­rithm tai­lored for rest­ing-state elec­troen­cephalog­ra­phy (EEG) and applied it to data from the largest imag­ing-cou­pled, place­bo-con­trolled anti­de­pres­sant study (n?=?309). Symp­tom improve­ment was robust­ly pre­dict­ed in a man­ner both spe­cif­ic for the anti­de­pres­sant ser­tra­line (ver­sus place­bo) and gen­er­al­iz­able across dif­fer­ent study sites and EEG equip­ment. This ser­tra­line-pre­dic­tive EEG sig­na­ture gen­er­al­ized to two depres­sion sam­ples, where­in it reflect­ed gen­er­al anti­de­pres­sant med­ica­tion respon­siv­i­ty and relat­ed dif­fer­en­tial­ly to a repet­i­tive tran­scra­nial mag­net­ic stim­u­la­tion treat­ment out­come. Fur­ther­more, we found that the ser­tra­line rest­ing-state EEG sig­na­ture indexed pre­frontal neur­al respon­siv­i­ty, as mea­sured by con­cur­rent tran­scra­nial mag­net­ic stim­u­la­tion and EEG. Our find­ings advance the neu­ro­bi­o­log­i­cal under­stand­ing of anti­de­pres­sant treat­ment through an EEG-tai­lored com­pu­ta­tion­al mod­el and pro­vide a clin­i­cal avenue for per­son­al­ized treat­ment of depression.

The Study in Context: