V-044 | Automated detection of cognitive symptom severity and mild cognitive impairment in Parkinson’s disease

Cognition, Behavior, and Memory
Author: Franco Javier Ferrante | Email: francoferrante95@gmail.com

---

Franco J. Ferrante^1,2,3, Daniel Escobar Grisales⁴; María Fernanda López^5,6; Pamela Lopes da Cunha^1,2; Lucas Federico Sterpin¹; Jet M.J. Vonk⁷; Pedro Chana Cuevas^8,9; Claudio Estienne¹⁰; Eugenia Hesse^1,11; Lucía Amoruso^1,13,14; Juan Rafael Orozco Arroyave^4,12 & Adolfo M. García^1,15,16,*

1 Centro de Neurociencias Cognitivas, Universidad de San Andrés, Vito Dumas 284 (B1644BID), Victoria, Buenos Aires, Argentina.

2 Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Godoy Cruz 2290, Piso 9 (C1425FQB), Ciudad Autónoma de Buenos Aires, Argentina.

3 Facultad de Ingeniería, Universidad de Buenos Aires, Colón 850 (C1063), CABA, Argentina.

4 GITA Lab, Faculty of Engineering, University of Antioquia Universidad de Antioquia, Street 67 No. 53 – 108 Of. 18-310, Medellín, 050010, Antioquia, Colombia.

5 Hospital Nacional Posadas, Avenida Presidente Arturo U. Illia s/n y Marconi Morón 386, (B1684) El Palomar, Provincia de Buenos Aires.

6 Universidad de Flores, Av. Rivadavia 5741 (C1406GLA), Caballito, Buenos Aires, Argentina.

7 Memory and Aging Center, Department of Neurology, University of California San Francisco, CA (94158), USA

8 Centro de Trastornos del Movimiento (CETRAM), Av. José Joaquín Prieto 7271 (9120501), Lo Espejo, Región Metropolitana, Chile.

9 Facultad de Ciencias Médicas, Universidad de Santiago de Chile, Avenida Libertador Bernardo O’Higgins n°3363 (9170022), Estación Central, Santiago, Chile

10 Instituto de Ingeniería Biomédica, Universidad de Buenos Aires, Av. Paseo Colón 850, 4to. piso, (C1063ACV), Buenos Aires, Argentina.

11 Departamento de Matemática y Ciencias, Universidad de San Andrés, Vito Dumas 284 (B1644BID), Victoria, Provincia­ de Buenos Aires, Argentina.

12 Pattern Recognition Lab, Martensstr. 3, Erlangen, 91058, Germany.

13 Basque Center on Cognition, Brain, and Language (BCBL), San Sebastian, Spain

14 Ikerbasque, Basque Foundation for Science, Bilbao, Spain

15 Global Brain Health Institute, University of California San Francisco, 1651 4th St (94143), San Francisco, CA, USA; and Trinity College Dublin, 42A Pearse St (D02 R123), Dublin, Ireland.

16 Departamento de Lingüística y Literatura, Facultad de Humanidades, Universidad de Santiago de Chile, Av. Bernardo O’Higgins 3363 (9160000), Estación Central, Santiago, Chile.

Establishing cognitive status is vital in Parkinson’s disease (PD) assessments, as neuropsychological deficits undermine daily functionality. Here we validate a scalable digital language analysis framework on a well-powered cohort.
384 PD patients completed two semantic and one phonemic fluency tasks. We extracted six variables from each word: semantic variability (SV), granularity, length, frequency, neighborhood and concreteness. In Experiment-1, these variables were used in a random forest regression to predict Mattis dementia rating scale scores as indices of cognitive severity. In Experiment-2 we used these variables to compare [via generalized linear model (GLM)] and classify (via ridge regression) between 49 patients with and 50 patients without mild cognitive impairment (MCI).
In Experiment-1, we found a strong correlation between real and predicted Mattis scores (R = 0.52, 95% CI [0.50-0.54], p < .001), driven by SV and granularity. In Experiment-2, patients with MCI produced significantly less varied and granular responses than those without MCI. Word properties achieved robust patient classification (mean [95% CI] AUC = 0.84 [0.84-0.85], sensitivity = 0.74 [0.72-0.76], specificity = 0.77 [0.75-0.80]), again led by SV and granularity.
Our results show that digital word property analysis predicts cognitive symptom severity and distinguishes between cognitive phenotypes of PD, enabling scalable neuropsychological screenings.