Article Text

Download PDFPDF

1117 SLE phenotypes formed from machine learning and their associations with cognitive impairment
  1. Michelle Barraclough1,
  2. Lauren Erdman1,
  3. Andrea Knight1,
  4. Juan Pablo Diaz-Martinez1,
  5. Kathleen Bingham1,
  6. Jiandong Su2,
  7. Mahta Kakvan2,
  8. Maria Carmela Tartaglia1,
  9. Lesley Ruttan1,
  10. Joan Wither1,
  11. May Y Choi3,
  12. Marvin J Fritzler3,
  13. Dennisse Bonilla2,
  14. Dorcas Beaton1,
  15. Ben Parker4,
  16. Robin Green1,
  17. Patti Katz5,
  18. Ian N Bruce4 and
  19. Zahi Touma1
  1. 1University of Toronto
  2. 2University Health Network
  3. 3University of Calgary
  4. 4University of Manchester
  5. 5University of California San Francisco


Background Cognitive impairment (CI) in SLE is highly prevalent. Several factors are associated with CI: depression, pain, fatigue, medications, as well as more specific SLE factors such as disease damage, and autoantibodies. We aimed to phenotype CI in SLE using machine learning techniques to enable personalised targeted treatments.

Methods SLE patients aged 18-65 years attending a single completed the ACR Neuropsychological Battery (ACR-NB) cognitive assessment. Z-scores on all 19 tests of ACR-NB. ACR-NB tests were reduced using principal component analysis (PCA) to generate a factor score (CI Factor Score).

Demographic, clinical data, and patient reported outcomes including, SF-36, LupusQoL, the PDQ-20 (perceived cognitive deficits), Beck Depression Inventory-II, Beck Anxiety Inventory, and the fatigue severity scale (FSS) were analysed using similarity network fusion (SNF) to identify patient subtypes. Differences between the SNF identified subtypes were evaluated using Kruskal-Wallis tests and chi-square tests.

Results Of 301 patients, 89% were women, mean age and disease duration at study visit 40.9 ± 12.1 years. The CI Factor score accounted for 28.8% of the variance and was associated predominantly with executive function and verbal memory. The SNF defined three subtypes (1, 2 and 3 with 60, 112, and 129 patients respectively) with distinct patterns in health-related quality of life (HRQoL), depression, anxiety, fatigue, fibromyalgia, medication usage, and damage. The CI Factor Score was significantly different between the subtypes. Examining specific cognitive domains revealed the most significant differences in the language processing and executive function tests. Subtype 3 performed worst on the majority of cognitive domains). Further exploration revealed statistical differences with depression, anxiety, fatigue, and fibromyalgia between the subtypes (figure 1). Differences were also found relating to organ involvement within the last ten years and damage within specific organs. No differences were found for SLE disease activity. Subtype 3 had higher levels of all conditions and disease damage, Subtype 2 had lower levels and Subtype 1 mixed levels.

Abstract 1117 Figure 1

Variables with significant differences between the three phenotyped subtypes. Box and whisker plots: blue=subtype 1, red =subtype and purple=subtype 3. Bar charts: red=number of participants with variable and blue=number without

Conclusion The subtype with the greatest psychiatric and disease burden and reduced HRQoL performed worse on cognitive testing, specifically in domains of language processing and executive function. This subtype also had more musculoskeletal (MSK) and cardiovascular involvement. MSK involvement affects pain levels, which can impact cognition. Cardiovascular damage may be linked to cerebral small vessel disease, which is known to affect cognitive function SLE patients. Overall, these results aid with phenotyping CI in SLE and provide a baseline for our future longitudinal results.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See:

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.