Journal of affective disorders · 2026
Predicting lithium-associated thyroid dysfunction through dimension reduction to establish a practical machine learning model: A retrospective multicenter study.
СтатьяМногоцентровое исследованиеНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Jheng Shu-Yu, Chan Fan-Ying, Chang Fang-Yung, Cheng Jen-Kai, Chen Hsiang-Yin
Аннотация
BACKGROUND: Thyroid dysfunction is a prevalent side effect among patients using lithium and links to refractory mood disorders. Existing predictive models are limited by high dimensionality; therefore, developing a simplified model that maintains performance is essential for clinical use.
OBJECTIVE: This study aimed to develop a machine learning model to predict lithium-associated thyroid dysfunction and reduce model dimensionality to enhance clinical applicability.
METHOD: This multicenter retrospective study included patients who received lithium carbonate between January 2010 and December 2021. Models were developed using XGBoost, Support Vector Machine, and Logistic Regression, with SHAP-based feature selection strategy applied to construct a simplified model. Performance and SHAP interpretability were compared between the original and simplified models at both global and individual levels.
RESULT: This study included 1595 patients and 113 developed thyroid adverse events after lithium treatment. XGBoost demonstrated the best performance, achieving an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.773 and Area Under the Precision-Recall Curve (AUPRC) of 0.444. Simplified model constructed by 7 features achieved AUROC and AUPRC values of 0.802 and 0.460, respectively. DeLong test and bootstrap analysis showed no significant differences between the simplified and original models, and SHAP analysis revealed similar feature importance.
CONCLUSION: This study developed a simplified, clinically applicable XGBoost model for predicting lithium-associated thyroid dysfunction using 7 clinical features. SHAP analysis successfully helped the clinician identify feature contributions and take interventions to prevent thyroid dysfunctions. Future studies on different populations are warranted to extrapolate the models for clinical use.
OBJECTIVE: This study aimed to develop a machine learning model to predict lithium-associated thyroid dysfunction and reduce model dimensionality to enhance clinical applicability.
METHOD: This multicenter retrospective study included patients who received lithium carbonate between January 2010 and December 2021. Models were developed using XGBoost, Support Vector Machine, and Logistic Regression, with SHAP-based feature selection strategy applied to construct a simplified model. Performance and SHAP interpretability were compared between the original and simplified models at both global and individual levels.
RESULT: This study included 1595 patients and 113 developed thyroid adverse events after lithium treatment. XGBoost demonstrated the best performance, achieving an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.773 and Area Under the Precision-Recall Curve (AUPRC) of 0.444. Simplified model constructed by 7 features achieved AUROC and AUPRC values of 0.802 and 0.460, respectively. DeLong test and bootstrap analysis showed no significant differences between the simplified and original models, and SHAP analysis revealed similar feature importance.
CONCLUSION: This study developed a simplified, clinically applicable XGBoost model for predicting lithium-associated thyroid dysfunction using 7 clinical features. SHAP analysis successfully helped the clinician identify feature contributions and take interventions to prevent thyroid dysfunctions. Future studies on different populations are warranted to extrapolate the models for clinical use.