Clustering of antipsychotic-naïve patients with schizophrenia based on functional connectivity from resting-state electroencephalography

Karen S. Ambrosen, Fanny Fredriksson, Simon Anhøj, Nikolaj Bak, Edwin van Dellen, Livia Dominicus, Cecilie K. Lemvigh, Mikkel Sørensen, Mette Ødegaard Nielsen, Kirsten Borup Bojesen, Birgitte Fagerlund, Birte Glenthøj, Bob Oranje, Lars Kai Hansen, Bjørn H. EbdrupEuropean 2023 · DOI

TL;DR

This study uses resting-state electroencephalography (rsEEG) and machine learning to identify distinct subgroups within antipsychotic-naïve schizophrenia patients based on their Default Mode Network (DMN) functional connectivity. These data-driven subgroups show clinically relevant differences in psychopathology and cognitive profiles, suggesting that DMN connectivity can help stratify patients and reveal underlying biological heterogeneity in early-stage schizophrenia.

What Problem Does This Paper Address?

Schizophrenia is a complex disorder often characterized by disruptions in large-scale brain networks, particularly the Default Mode Network (DMN). While DMN abnormalities are consistently observed, their precise clinical implications and how they might define distinct patient subgroups remain unclear. Previous research has yielded conflicting results regarding the direction and magnitude of DMN connectivity changes, and there's a lack of consensus on how to best analyze EEG data for functional connectivity. This study aims to address this by applying unsupervised machine learning to rsEEG functional connectivity within the DMN to identify potential subgroups of antipsychotic-naïve, first-episode schizophrenia patients and investigate if these subgroups have distinct clinical and cognitive characteristics.

Methods

The researchers employed a multi-step data-driven approach:

  1. rsEEG Data Acquisition: Resting-state EEG data was collected from 37 antipsychotic-naïve, first-episode schizophrenia patients and 97 matched healthy controls.
  2. Source Localization: EEG signals were source-localized to identify activity in key DMN regions (Medial Prefrontal Cortex, Precuneus Cortex, and Lateral Parietal Cortex).
  3. Functional Connectivity Estimation: The Phase Lag Index (PLI) was used to measure undirected functional connectivity between these DMN regions. PLI is chosen for its ability to mitigate volume conduction artifacts common in EEG. Connectivity was analyzed separately for different frequency bands (delta, theta, alpha, beta).
  4. Dimensionality Reduction (PCA): Principal Component Analysis (PCA) was applied to reduce the dimensionality of the connectivity data within each frequency band, creating a more manageable set of features.
  5. Unsupervised Clustering (GMM): A Gaussian Mixture Model (GMM) was used to cluster the schizophrenia patients based on their DMN connectivity in the PCA space. This unsupervised approach allows for the discovery of natural groupings without pre-defined labels. The optimal number of clusters was determined using cross-validation.
  6. Subgroup Characterization: The identified clusters were then analyzed for differences in demographic, psychopathological (using PANSS), and cognitive measures.
  7. Predictive Modeling (SVM): Support Vector Machines (SVM) were used to test if psychopathological or cognitive profiles could predict subgroup membership, further validating the clinical relevance of the identified clusters.

Datasets

The study utilized data from two previously established multimodal cohorts:

  • The Pan European Collaboration on Antipsychotic Naïve Schizophrenia (PECANS)
  • The Pan European Collaboration on Antipsychotic Naïve Schizophrenia II (PECANSII)

These cohorts provided data from 37 antipsychotic-naïve, first-episode schizophrenia patients and 97 matched healthy controls. No new benchmarks were introduced; the study leveraged existing, well-characterized patient and control groups.

Key Insights and Ideas

  • DMN Connectivity as a Stratification Tool: The study demonstrates that functional connectivity within the DMN, even at the early stages of schizophrenia, can be used to identify distinct patient subgroups. This moves beyond a "one-size-fits-all" approach to understanding schizophrenia.
  • Frequency-Specific Subgroups: Distinct subgroups were identified in the theta (4-8 Hz) and beta (12-30 Hz) frequency bands, suggesting that different oscillatory dynamics within the DMN may underpin different pathophysiological profiles.
  • Clinical Relevance of Subgroups: The identified subgroups, while not differing significantly on basic demographics, showed meaningful differences in psychopathology and cognitive function, indicating their clinical relevance.
  • Negative Symptoms as a Key Predictor: For the theta subgroups, negative symptoms emerged as a primary driver for subgroup prediction, highlighting their importance in understanding DMN dysfunction in this population.
  • Cognitive Profiles Differentiate Beta Subgroups: For the beta subgroups, cognitive measures, particularly mental flexibility, verbal IQ, and reaction time, were predictive of subgroup membership, suggesting a link between DMN connectivity and cognitive processing in these distinct groups.

Key Results

  • Theta Subgroups: Two theta subgroups were identified. While they didn't differ in connectivity, one subgroup tended to have poorer functioning and more negative symptoms. Psychopathology, particularly negative symptoms, predicted theta subgroup membership with 69.4% accuracy (p=0.003).
  • Beta Subgroups: Two beta subgroups were identified, showing differences in DMN connectivity compared to healthy controls. One subgroup had increased connectivity, while the other showed decreased connectivity in specific DMN connections. These beta subgroups were predicted by cognitive measures with 63.6% accuracy (p=0.034), driven by mental flexibility, verbal IQ, and reaction time.
  • No Delta or Alpha Subgroups: The delta and alpha frequency bands did not yield distinct patient clusters using this methodology.

Limitations and Future Work

  • Sample Size: The study acknowledges that the relatively small sample size, particularly for subgrouping, may limit the ability to detect subtle differences and associations. Replication in larger, independent samples is recommended.
  • Methodological Variability: The authors note the ongoing lack of consensus on optimal EEG functional connectivity analysis methods, which can impede cross-study comparisons.
  • Ground Truth for Clustering: As with all cluster analysis, the identified subgroups are based on data-driven patterns and may not perfectly align with clinically defined categories. The authors addressed this by validating subgroups with clinical and cognitive measures.
  • Future Directions: The study suggests that data-driven methods like clustering, applied to functional connectivity, are valuable for recognizing pathophysiological patterns in early-stage schizophrenia. Future work could explore the predictive power of these subgroups for treatment response and long-term outcomes. Further investigation into the specific mechanisms linking theta and beta band connectivity to clinical symptoms is also warranted.

Summary updated

Abstract

Schizophrenia is associated with aberrations in the Default Mode Network (DMN), but the clinical implications remain unclear. We applied data-driven, unsupervised machine learning based on resting-state electroencephalography (rsEEG) functional connectivity within the DMN to cluster antipsychotic-naïve patients with first-episode schizophrenia. The identified clusters were investigated with respect to psychopathological profile and cognitive deficits. Thirty-seven antipsychotic-naïve, first-episode patients with schizophrenia (mean age 24.4 (5.4); 59.5% males) and 97 matched healthy controls (mean age 24.0 (5.1); 52.6% males) underwent assessments of rsEEG, psychopathology, and cognition. Source-localized, frequency-dependent functional connectivity was estimated using Phase Lag Index (PLI). The DMN-PLI was factorized for each frequency band using principal component analysis. Clusters of patients were identified using a Gaussian mixture model and neurocognitive and psychopathological profiles of identified clusters were explored. We identified two clusters of patients based on the theta band (4-8 Hz), and two clusters based on the beta band (12-30 Hz). Baseline psychopathology could predict theta clusters with an accuracy of 69.4% (p = 0.003), primarily driven by negative symptoms. Five a priori selected cognitive functions conjointly predicted the beta clusters with an accuracy of 63.6% (p = 0.034). The two beta clusters displayed higher and lower DMN connectivity, respectively, compared to healthy controls. In conclusion, the functional connectivity within the DMN provides a novel, data-driven means to stratify patients into clinically relevant clusters. The results support the notion of biological subgroups in schizophrenia and endorse the application of data-driven methods to recognize pathophysiological patterns at earliest stage of this syndrome.
BibTeX
@inproceedings{ambrosen2023clustering,
  author = {Karen S. Ambrosen and Fanny Fredriksson and Simon Anhøj and Nikolaj Bak and Edwin van Dellen and Livia Dominicus and Cecilie K. Lemvigh and Mikkel Sørensen and Mette Ødegaard Nielsen and Kirsten Borup Bojesen and Birgitte Fagerlund and Birte Glenthøj and Bob Oranje and Lars Kai Hansen and Bjørn H. Ebdrup},
  title = {{Clustering of antipsychotic-naïve patients with schizophrenia based on functional connectivity from resting-state electroencephalography}},
  booktitle = {European},
  year = {2023},
  doi = {10.1007/s00406-023-01550-9},
}
Figure