Study shows the power of electroencephalography and machine learning to help predict response to psychotherapy (or lack thereof) in patients with PTSD

A fas­ci­nat­ing study just pub­lished in Nature Men­tal Health aimed at help­ing pre­dict the out­comes of psy­chother­a­py for patients with Post-Trau­mat­ic Stress Dis­or­der (PTSD) using Machine Learn­ing (ML) and elec­troen­cephalog­ra­phy (EEG) data.

PTSD is a men­tal health con­di­tion trig­gered by expe­ri­enc­ing or wit­ness­ing a trau­mat­ic event; two evi­dence-based treatments–Prolonged Expo­sure (PE) and Cog­ni­tive Pro­cess­ing Ther­a­py (CPT)–are com­mon­ly used to help patients, with var­ied results.

In this study, the researchers used a super­vised machine learn­ing approach and high-den­si­ty rest­ing-state EEG (rsEEG) record­ings to pre­dict indi­vid­ual psy­chother­a­py out­comes. They iden­ti­fied a pre­treat­ment EEG con­nec­tiv­i­ty sig­na­ture in the eyes-open theta fre­quen­cy range that was pre­dic­tive of patients’ respons­es to both PE and CPT. 

“Not only could EEG ML pre­dict treat­ment, but mod­els trained on one ther­a­py could pre­dict the oth­er. Not only could EEG ML pre­dict respon­ders, but it could also iden­ti­fy non-respon­ders.…peo­ple for whom nei­ther ther­a­py works.” — Dr. Amit Etkin, Founder and CEO at Alto Neu­ro­science and Adjunct Pro­fes­sor at Stan­ford University

These find­ings are con­sis­tent with pre­vi­ous fMRI-based stud­ies on func­tion­al con­nec­tiv­i­ty abnor­mal­i­ties and treat­ment-asso­ci­at­ed changes in PTSD. The use of EEG in this study offers a more afford­able and clin­i­cal­ly scal­able neu­roimag­ing tool com­pared to fMRI, mak­ing it more acces­si­ble for clin­i­cal applications.

The study shows how bio­mark­ers can poten­tial­ly help match treat­ment-to-indi­vid­ual (or at least to pro­file of individuals):

• Pre­dic­tion of treat­ment out­comes: By pre­dict­ing indi­vid­ual respons­es to two major types of psy­chother­a­py for PTSD patients, bio­mark­ers can help clin­i­cians bet­ter select treat­ments to improve ther­a­py outcomes.
• Iden­ti­fi­ca­tion of treat­ment-resis­tant patients: Bio­mark­ers can also help iden­ti­fy patients who may not respond well to exist­ing psy­chother­a­py approaches.

In sum­ma­ry, this study used machine learn­ing mod­els and EEG con­nec­tiv­i­ty data to pre­dict psy­chother­a­py out­comes in PTSD patients, find­ing a sig­na­ture that was sep­a­rate­ly pre­dic­tive of the two major types of psy­chother­a­py cur­rent­ly in prac­tice: Pro­longed Expo­sure (PE) and Cog­ni­tive Pro­cess­ing Ther­a­py (CPT). In doing so it con­tributes to a bet­ter under­stand­ing of the neu­ro­bi­ol­o­gy of PTSD, pro­mot­ing fur­ther research on the use of cost-effec­tive neu­roimag­ing tools, and poten­tial­ly improv­ing treat­ment out­comes for patients who are resis­tant to cur­rent treat­ments. Future research include the need for more com­pre­hen­sive analy­ses with larg­er sam­ple sizes and test­ing the method in more diverse populations.

The Study:

Machine learn­ing-based iden­ti­fi­ca­tion of a psy­chother­a­py-pre­dic­tive elec­troen­cephalo­graph­ic sig­na­ture in PTSD (Nature Men­tal Health). From the Abstract:

Although psy­chother­a­py is at present the most effec­tive treat­ment for post­trau­mat­ic stress dis­or­der (PTSD), its effi­ca­cy is still lim­it­ed for many patients, due main­ly to the sub­stan­tial clin­i­cal and neu­ro­bi­o­log­i­cal het­ero­gene­ity in the dis­ease … This study inves­ti­gates whether indi­vid­ual patient-lev­el rest­ing-state EEG con­nec­tiv­i­ty can pre­dict psy­chother­a­py out­comes in PTSD. We devel­oped a treat­ment-pre­dic­tive EEG sig­na­ture using machine learn­ing applied to high-den­si­ty rest­ing-state EEG col­lect­ed from mil­i­tary vet­er­ans with PTSD. The pre­dic­tive sig­na­ture was dom­i­nat­ed by theta fre­quen­cy EEG con­nec­tiv­i­ty dif­fer­ences and was able to gen­er­al­ize across two types of psychotherapy—prolonged expo­sure and cog­ni­tive pro­cess­ing ther­a­py. Our results also advance a bio­log­i­cal def­i­n­i­tion of a PTSD patient sub­group who is resis­tant to psy­chother­a­py, which is cur­rent­ly the most evi­dence-based treat­ment for the con­di­tion. The find­ings sup­port a path towards clin­i­cal­ly trans­lat­able and scal­able bio­mark­ers that could be used to tai­lor inter­ven­tions for each indi­vid­ual or dri­ve the devel­op­ment of nov­el treatments.

The Study in Context:

• Machine-learn­ing study finds EEG brain sig­na­tures that pre­dict response to anti­de­pres­sant treatments
• Alto Neu­ro­science rais­es $60M (equi­ty + cred­it) to help fix the “tri­al and error” approach to psy­chi­atric medication