Geisinger and Eisai to test real-world validity of AI-powered Passive Digital Marker (PDM) in detecting early cognitive impairment and dementia

by

Research col­lab­o­ra­tion will test nov­el algorithm’s effec­tive­ness on Geisinger data (press release):

Geisinger and Eisai Inc. today announced a col­lab­o­ra­tive effort to study the poten­tial effec­tive­ness of an arti­fi­cial intel­li­gence (AI) tool in the detec­tion of cog­ni­tive impair­ment that could iden­ti­fy demen­tias, includ­ing Alzheimer’s dis­ease (AD). If effec­tive, the AI tool could poten­tial­ly be devel­oped to sup­port the ear­ly detec­tion and stag­ing of cog­ni­tive impair­ment and demen­tia, lead­ing to appro­pri­ate addi­tion­al test­ing for the clin­i­cal, bio­log­i­cal diag­no­sis and treat­ment of demen­tias such as AD.

The research col­lab­o­ra­tion will study the use of an algo­rithm trained on a set of de-iden­ti­fied patient data to iden­ti­fy indi­vid­u­als like­ly to have cog­ni­tive impair­ment. The algo­rithm, known as a Pas­sive Dig­i­tal Mark­er (PDM), was devel­oped and test­ed by researchers at Pur­due Uni­ver­si­ty and Indi­ana Uni­ver­si­ty … “AI tech­nol­o­gy has the poten­tial to trans­form med­i­cine,” said Yass­er El-Man­za­lawy, Ph.D., prin­ci­pal inves­ti­ga­tor and assis­tant pro­fes­sor of Trans­la­tion­al Data Sci­ence and Infor­mat­ics at Geisinger. “AI-based tools can effi­cient­ly scan mas­sive amounts of health­care data and iden­ti­fy hid­den pat­terns. These pat­terns can be used to detect dis­eases, like can­cer and demen­tia, at an ear­ly stage…”

“As an imple­men­ta­tion sci­en­tist, it is always excit­ing to have oth­er sci­en­tists eval­u­ate the repro­ducibil­i­ty of the per­for­mance of our pas­sive dig­i­tal mark­er in very dif­fer­ent pop­u­la­tions,” said Malaz Bous­tani, M.D., Richard M. Fair­banks Pro­fes­sor of Aging Research at Indi­ana Uni­ver­si­ty. “Repro­ducibil­i­ty is the cor­ner­stone of sci­en­tif­ic progress.”

About the Passive Digital Marker (PDM):

Pre­dict­ing demen­tia with rou­tine care EMR data (Arti­fi­cial Intel­li­gence in Med­i­cine). From the Abstract:

Our aim is to devel­op a machine learn­ing (ML) mod­el that can pre­dict demen­tia in a gen­er­al patient pop­u­la­tion from mul­ti­ple health care insti­tu­tions one year and three years pri­or to the onset of the dis­ease with­out any addi­tion­al mon­i­tor­ing or screen­ing. The pur­pose of the mod­el is to auto­mate the cost-effec­tive, non-inva­sive, dig­i­tal pre-screen­ing of patients at risk for dementia.

Towards this pur­pose, rou­tine care data, which is wide­ly avail­able through Elec­tron­ic Med­ical Record (EMR) sys­tems is used as a data source. These data embody a rich knowl­edge and make relat­ed med­ical appli­ca­tions easy to deploy at scale in a cost-effec­tive man­ner. Specif­i­cal­ly, the mod­el is trained by using struc­tured and unstruc­tured data from three EMR data sets: diag­no­sis, pre­scrip­tions, and med­ical notes…

The results show that the com­bined mod­el is gen­er­al­iz­able across mul­ti­ple insti­tu­tions and is able to pre­dict demen­tia with­in one year of its onset with an accu­ra­cy of near­ly 80% despite the fact that it was trained using rou­tine care data. More­over, the analy­sis of the mod­els iden­ti­fied impor­tant pre­dic­tors for demen­tia. Some of these pre­dic­tors (e.g., age and hyper­ten­sive dis­or­ders) are already con­firmed by the lit­er­a­ture while oth­ers, espe­cial­ly the ones derived from the unstruc­tured med­ical notes, require fur­ther clin­i­cal analysis.

News in Context: