Development and Validation of Machine Learning-Based Models for Predicting Postoperative Depression Risk in Patients With Ovarian Cancer.

Saved in:

Bibliographic Details
Title:	Development and Validation of Machine Learning-Based Models for Predicting Postoperative Depression Risk in Patients With Ovarian Cancer.
Authors:	Zhao, Jitong^1,2, Pei, Kaige^1,2, Liu, Junhan^1,2, Bian, Ce^1,2, Ling, Chen^1,2 lingchen@scu.edu.cn
Source:	Actas Espanolas de Psiquiatria. 2026, Vol. 54 Issue 2, p480-499. 20p.
Subjects:	Prediction models, Mental depression, Random forest algorithms, Machine learning, Feature selection, Risk assessment, Ovarian cancer
Abstract:	Objective: To develop machine learning-based prediction models for postoperative depression risk in patients with ovarian cancer and to evaluate their predictive performance and clinical application value. Methods: Clinical data from 850 postoperative patients with ovarian cancer were retrospectively analysed. Postoperative depression risk was defined as positive when Patient Health Questionnaire-9 (PHQ-9) score was ≥10. Feature selection was performed using least absolute shrinkage and selection operator (LASSO) regression and Boruta algorithm, with the intersection of both methods determining the final predictive variables. Data were randomly divided into training and validation sets at a 7:3 ratio. Five prediction models were constructed: logistic regression, random forest, support vector machine, extreme gradient boosting (XGBoost), and neural network. Model performance was evaluated through area under the receiver operating characteristic curve (AUC), Brier score, calibration curves, and decision curve analysis. SHapley Additive exPlanations (SHAP) method was employed to interpret the feature contributions of the optimal model, and a nomogram was constructed to facilitate clinical application. Results: Among 850 patients, 268 (31.5%) were positive for postoperative depression risk. Feature selection identified 13 predictive variables: age, operation time, length of hospital stay, pain score, white blood cell count, albumin, C-reactive protein, CA125, education level, history of depression/anxiety, postoperative insomnia, fatigue, and opioid analgesic use. Among the five models, random forest demonstrated superior performance with an AUC of 0.776 in the validation set, a Brier score of 0.182, sensitivity of 0.771, and an F1 score of 0.792, along with satisfactory calibration and clinical net benefit. SHAP analysis revealed that pain score, postoperative insomnia, albumin level, and opioid use contributed substantially to model predictions. A nomogram based on logistic regression model was constructed for intuitive individual risk assessment. Conclusion: The machine learning-based prediction models for postoperative depression risk in patients with ovarian cancer demonstrated satisfactory discriminative ability and clinical utility, with random forest model showing optimal performance. A clinical nomogram was additionally constructed to enable individualised and visual risk quantification suitable for bedside application. Together, these tools facilitate early identification of high-risk patients and provide evidence for clinical intervention. [ABSTRACT FROM AUTHOR]
	Copyright of Actas Espanolas de Psiquiatria is the property of Maria Lopez-Ibor and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	MedicLatina

Description
Abstract:	Objective: To develop machine learning-based prediction models for postoperative depression risk in patients with ovarian cancer and to evaluate their predictive performance and clinical application value. Methods: Clinical data from 850 postoperative patients with ovarian cancer were retrospectively analysed. Postoperative depression risk was defined as positive when Patient Health Questionnaire-9 (PHQ-9) score was ≥10. Feature selection was performed using least absolute shrinkage and selection operator (LASSO) regression and Boruta algorithm, with the intersection of both methods determining the final predictive variables. Data were randomly divided into training and validation sets at a 7:3 ratio. Five prediction models were constructed: logistic regression, random forest, support vector machine, extreme gradient boosting (XGBoost), and neural network. Model performance was evaluated through area under the receiver operating characteristic curve (AUC), Brier score, calibration curves, and decision curve analysis. SHapley Additive exPlanations (SHAP) method was employed to interpret the feature contributions of the optimal model, and a nomogram was constructed to facilitate clinical application. Results: Among 850 patients, 268 (31.5%) were positive for postoperative depression risk. Feature selection identified 13 predictive variables: age, operation time, length of hospital stay, pain score, white blood cell count, albumin, C-reactive protein, CA125, education level, history of depression/anxiety, postoperative insomnia, fatigue, and opioid analgesic use. Among the five models, random forest demonstrated superior performance with an AUC of 0.776 in the validation set, a Brier score of 0.182, sensitivity of 0.771, and an F1 score of 0.792, along with satisfactory calibration and clinical net benefit. SHAP analysis revealed that pain score, postoperative insomnia, albumin level, and opioid use contributed substantially to model predictions. A nomogram based on logistic regression model was constructed for intuitive individual risk assessment. Conclusion: The machine learning-based prediction models for postoperative depression risk in patients with ovarian cancer demonstrated satisfactory discriminative ability and clinical utility, with random forest model showing optimal performance. A clinical nomogram was additionally constructed to enable individualised and visual risk quantification suitable for bedside application. Together, these tools facilitate early identification of high-risk patients and provide evidence for clinical intervention. [ABSTRACT FROM AUTHOR]
ISSN:	11399287
DOI:	10.62641/aep.v54i2.2177