fetching data ... 
Background: Systemic lupus erythematosus (SLE) is a chronic autoimmune disease that affects multiple organs and systems. However, in addition to physical damage, patients with SLE have an increased risk of mental health disorders, such as depression and anxiety.[1] Unfortunately, the diagnosis of these mental illnesses is delayed or missed in regular clinical practice.[2] Given the increased prevalence and susceptibility to mental disorders among SLE patients, it is important to recognize people suffering from anxiety and depression as early as possible in order to provide them with psychological support.
Objectives: To develop a predictive regression model to calculate the expected value of the Short Form Health Survey (SF-36) mental component using information about patient, drug usage, disease duration and activity level.
Methods: A cohort of patients with systemic lupus erythematosus (n=120), collected at the Rheumatology Hospital of St. Petersburg, was used to train and test various machine learning models. This cohort included both juvenile onset (n=66) and adult onset (n=54) SLE patients, with a female prevalence of 92.5%, a mean age of 33.5 ± 8.7 years, and a disease duration of 5.9 ± 6 years. The study utilized only anonymized data. The parameters chosen for the model included the patient’s age, age of SLE debut, disease duration, number of SLE exacerbations, BMI, SLEDAY-2K, and the current dose of prednisolone taken by the patient. The dataset consisted of 70% of patients (n=84) for training the models and 30% of patients (n=36) for testing. The creation of the model involved using the scikit-learn package in the Python programming language. To identify the machine learning algorithm with the best performance for handling our data, Random Forest, XGBoost, K-Nearest Neighbors (KNN), and Linear Regression models were implemented. Each algorithm underwent 10-fold cross-validation to assess the performance and generalization ability of a model. The performance of the different models was evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Each of our model gives as output the predicted value of SF-36 mental component that represented by continuous value from 0 to 100. While higher scores on the SF-36 mental component signify better mental health and well-being, lower scores may indicate difficulties or limitations in these aspects.
Results: In evaluating the metrics for developed models, the Linear Regression model yielded a MAE of 5.91 and RMSE of 7.07, the KNN model showed a MAE of 5.98 and RMSE of 6.99, the XGBoost model demonstrated a MAE of 5.79 and RMSE of 7.11, while the Random Forest model produced a MAE of 6.51 and RMSE of 7.98. The KNN model has the lowest RMSE (6.99), indicating better accuracy in predicting SF-36 mental component values compared to other models. Meanwhile, the XGBoost model exhibited the lowest MAE, suggesting better overall accuracy in predicting the absolute errors of SF-36. Linear Regression and XGBoost models have relatively close performance metrics, with the latter slightly outperforming in terms of MAE. Random Forest has the highest MAE and RMSE among the models, suggesting that it might not perform as well as the other models on this particular dataset.
Conclusion: The results underscore the potential of these models to aid in early identification of mental health issues in SLE patients, with the KNN and XGBoost models showing promise for clinical application. Further research and validation are warranted to enhance the robustness of these predictive models in improving mental health outcomes for individuals with SLE.
REFERENCES: [1] Moustafa AT, Moazzami M, Engel L, Bangert E, Hassanein M, Marzouk S, et al. Prevalence and metric of depression and anxiety in systemic lupus erythematosus: A systematic review and meta-analysis. Semin Arthritis Rheum. 2020;50(1):84–94.
[2] Liao, J., Kang, J., Li, F. et al. A cross-sectional study on the association of anxiety and depression with the disease activity of systemic lupus erythematosus. BMC Psychiatry 22, 591 (2022).
Acknowledgements: NIL.
Disclosure of Interests: None declared.