Identifying Neuro-inflammatory Biomarkers of Generalized Anxiety Disorder from Lymphocyte Subsets based on machine learning approaches, 2025, Lu et al

Identifying Neuro-inflammatory Biomarkers of Generalized Anxiety Disorder from Lymphocyte Subsets based on machine learning approaches

Jingjing Lu, Weiwei Liang, Lijun Cui, Shaoqi Mou, Xuedan Pei, Xinhua Shen, Zhongxia Shen, Ping Shen

Introduction
Activation of the inflammatory response system is involved in the pathogenesis of generalized anxiety disorder (GAD). The purpose of this study was to identify and characterize inflammatory biomarkers in the diagnosis of GAD based on machine learning algorithms.

Methods
The evaluation of peripheral immune parameters and lymphocyte subsets was performed on patients with GAD. Multivariable linear regression was used to explore the association between lymphocyte subsets and the of severity GAD. Receiver operator characteristic (ROC) analysis was used to determine the predictive value of these immunological parameters for GAD. Machine learning technology was applied to classify the collected data from patients in the GAD and healthy control groups.

Results
Of the 340 patients enrolled, 171 were GAD patients and 169 were non-GAD patients as healthy control. The levels of neutrophil (NEU), monocytes (MON) and systemic immune-inflammation index (SII) were significantly elevated in GAD patients (P<0.01), and the count and proportion of immune cells, including CD3+CD4+ T cells, CD3+CD8+ T cells, CD19+ B cells and CD3-CD16+CD56+ NK cells (P<0.001) have undergone large changes.

The classification analysis conducted by machine learning using a weighted ensemble-L2 algorithm demonstrated an accuracy of 95.00±2.04% in assessing the predictive value of these lymphocyte subsets in GAD. In addition, the feature importance analysis score is 0.255, which was much more predictive of GAD severity than for other lymphocyte subsets.

Conclusion
In the presented work, we show the level of lymphocyte subsets altered in GAD. Lymphocyte subsets, specifically CD3+CD4+ T %, can serve as neuroinflammatory biomarkers for GAD diagnostics. Furthermore, the application of machine learning offers a highly efficient approach for investigating neuroinflammatory biomarkers and predicting GAD. Our research has provided novel insights into the involvement of cellular immunity in GAD and highlighted the potential predictive value and therapeutic targets of lymphocyte subsets in this disorder.

Link (Neuropsychobiology) [Paywall]

 

I can't see the full text. 95% accuracy seems high, but since they don't mention a test set, I assume this is just overfitting on the training data.

 

They do mention a test set:

 

Interesting. If it's not buried too deeply, does it say what specific measurements were used in the 95% model?

 

@Nightsong don't even worry about it actually (unless you're interested yourself), I'm not super interested.

All these that they found large differences seem like basic tests that have probably been tested a million times already in GAD.

One example where there was no difference in some of these:

Enhanced Th17 phenotype in individuals with generalized anxiety disorder, 2010, Vieira et al