Related Experiment Video
Updated: Mar 25, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Validating a Pragmatic Approach to Cognitive Screening in Stroke Prevention Clinics Using the Montreal Cognitive
Richard H Swartz1, Megan L Cayley2, Krista L Lanctôt2
1From the Departments of Medicine (Neurology) (R.H.S., B.J.M.) and Psychiatry (K.L.L., N.H.), and Dalla Lana School of Public Health (K.E.T.), University of Toronto, Toronto, ON, Canada; Departments of Medicine (Neurology) (R.H.S., M.L.C., B.J.M.) and Psychiatry (K.L.L., N.H.), and Hurvitz Brain Sciences Research Program (R.H.S., K.L.L., B.J.M, N.H.), Sunnybrook Health Sciences Centre, Toronto, ON, Canada; Heart and Stroke Foundation Canadian Partnership for Stroke Recovery, Toronto, ON, Canada (R.H.S., K.L.L.); University of Toronto Stroke Program, Toronto, ON, Canada (R.H.S., M.L.C.); Department of Clinical Neurosciences, Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada (E.E.S.); Department of Medicine (Neurology), Hamilton General Hospital, Hamilton Health Sciences, McMaster University, Hamilton, ON, Canada (D.J.S.); and St. Michael's Hospital, Applied Health Research Centre of the Li Ka Shing Knowledge Institute, Toronto, ON Canada (A.C., K.E.T.). rick.swartz@sunnybrook.ca.
Background And Purpose:
The Montreal Cognitive Assessment (MoCA) is used commonly to identify cognitive impairment (CI), but there are multiple published cut points for normal and abnormal. We seek to validate a pragmatic approach to screening for moderate-severe CI, by classifying patients into high-, intermediate-, and low-risk categories.
Methods:
A total of 390 participants attending an academic Stroke Prevention Clinic completed the MoCA and more detailed neuropsychological testing. Between April 23, 2012 and April 30, 2014, all consecutive new referrals to the regional Stroke Prevention Clinic who were English-speaking, not severely aphasic, and could see and write well enough to complete neuropsychological testing were assessed for inclusion, and consenting patients were enrolled. CI was defined as ≥2 SDs below normal for age and education on at least 2 cognitive subtests. A single cut point for CI was compared with 2 cut points (high sensitivity and high specificity) generated using receiver operator characteristic and area under the curve analyses. The intermediate-risk group contained those scoring between the 2 cut points.
Results:
Thirty-four percent of participants had a symptomatic or silent stroke, 34% were seen for possible or probable transient ischemic attack, and 32% were diagnosed with other vascular or nonvascular conditions. Using a single cut point, sensitivity and specificity were optimal with MoCA ≤22, (sensitivity=60.4%, specificity=89.9%, area under the curve=0.801, positive predictive value=48.5%, negative predictive value=93.5%, positive likelihood ratio=6, and negative likelihood ratio=0.4). Using 2 cut points, sensitivity was optimal with MoCA ≥28 (sensitivity=96.2%, negative predictive value =97.6%, and negative likelihood ratio=1.27), and specificity was optimal with MoCA ≤22 (specificity=89.9%, positive predictive value=48.5%, and positive likelihood ratio=6).
Conclusions:
Stratifying participants into 3 categories facilitates the identification of a homogenous group at low risk for CI, as well as 2 other groups with intermediate and higher risk. This approach could facilitate clinical care pathways and patient selection for research.
