A risk prediction model for post-stroke depression in Chinese stroke survivors based on clinical and socio-psychological features

Rui Liu, Yingying Yue, Haitang Jiang, Jian Lu, Aiqin Wu, Deqin Geng, Jun Wang, Shenghua Li, Hua Tang, Xuesong Lu, Kezhong Zhang, Tian Liu, Yonggui Yuan, Qiao WangOncotarget 2017 · DOI

TL;DR

This paper develops a robust risk prediction model for post-stroke depression (PSD) in Chinese stroke survivors, integrating clinical and socio-psychological factors. The study highlights the significant role of socio-psychological elements in predicting PSD, offering a practical tool for early detection and intervention.

What Problem Does This Paper Address?

Post-stroke depression (PSD) is a common and debilitating complication following a stroke, significantly impacting patient recovery and quality of life. While numerous factors have been associated with PSD, a comprehensive and clinically applicable model for early risk prediction has been lacking. Existing models often focus on a limited set of factors, neglecting the crucial interplay of socio-psychological elements that can influence a patient's vulnerability to depression after a stroke. This paper aims to bridge this gap by developing a predictive model that incorporates a wide range of clinical and socio-psychological features to facilitate early identification of high-risk individuals.

Methods

The researchers employed a two-step approach to develop their risk prediction model.

  1. Multivariable Backward Logistic Regression: This standard statistical technique was used on a training dataset (70% of the participants) to identify significant predictors of PSD. The model was built by iteratively removing non-significant variables, aiming to find the most parsimonious yet effective set of predictors.
  2. Decision Tree Model: To provide a more intuitive and clinically applicable tool, the logistic regression model was converted into a decision tree. This method visually represents a series of "if-then" rules, making it easier for clinicians to assess risk based on a sequence of assessments. The decision tree was pruned to optimize its complexity and classification accuracy.

Both models were evaluated using Receiver Operating Characteristic (ROC) curves to assess their discriminatory power, and accuracy metrics were also reported.

Datasets

The study recruited 562 stroke patients in China. Data was collected on demographic information, medical history, vascular risk factors, functional status post-stroke, and various socio-psychological factors. The participants were assessed at one month post-stroke. The dataset was randomly split into a training set (n=393) for model development and a testing set (n=169) for validation. No new benchmark datasets were introduced; the study utilized existing scales and patient data.

Key Insights and Ideas

  • Socio-psychological factors are paramount: Beyond clinical history (like brain infarction), personality traits (neuroticism), emotional processing (alexithymia), life events, and social support emerged as critical predictors of PSD. This underscores the importance of a holistic approach to stroke recovery, acknowledging the patient's internal and external environment.
  • A practical decision tree for clinicians: The conversion of the logistic model into a decision tree provides a user-friendly tool. This "if-then" rule-based system simplifies the risk assessment process, making it more accessible for routine clinical use compared to complex statistical models.
  • Early detection is feasible: The model's performance, particularly the decision tree's accuracy, suggests that PSD risk can be effectively identified within one month post-stroke, allowing for timely intervention.

Key Results

The study identified several significant predictors for PSD:

  • Clinical Factor: History of brain cerebral infarction (OR, 3.84)
  • Socio-psychological Factors:
    • Eysenck Personality Questionnaire - Neuroticism/Stability (EPQ_N) (OR, 1.18)
    • Life Event Scale (LES) (OR, 0.99)
    • Social Support Rating Scale (SSRS) (OR, 0.91)
    • 20 items Toronto Alexithymia Scale (TAS) (OR, 1.06)

The logistic regression model achieved an Area Under the Curve (AUC) of 0.85 for the ROC, with an accuracy of 0.76.

The decision tree model also achieved an AUC of 0.85 and demonstrated a higher accuracy of 0.86, with a sensitivity of 0.70 and specificity of 0.83. The tree model generated 11 distinct rules for risk prediction.

Limitations and Future Work

The authors acknowledge several limitations:

  • Generalizability: The findings may not be generalizable to all stroke features, such as biochemical indices and lesion location, which were not fully explored.
  • Timing of Assessment: Predictors were assessed at one month post-stroke. It's suggested that assessing at different time points might offer a clearer picture of risk factors and improve model effectiveness.
  • Cut-off Values: The decision rules and cut-off points are not absolute and may require optimization in future studies to enhance accuracy.

Future research should aim to:

  • Integrate a broader range of risk factors, including biochemical and lesion-specific data.
  • Conduct longitudinal studies to track PSD development over different timeframes.
  • Further optimize the decision tree model and its cut-off values for improved clinical utility and accuracy.

Summary updated

Abstract

This study provided an effective risk model for PSD and indicated that the socio-psychological factors were important risk factors of PSD.
BibTeX
@inproceedings{liu2017risk,
  author = {Rui Liu and Yingying Yue and Haitang Jiang and Jian Lu and Aiqin Wu and Deqin Geng and Jun Wang and Jianxin Lü and Shenghua Li and Hua Tang and Xuesong Lu and Kezhong Zhang and Tian Liu and Yonggui Yuan and Qiao Wang},
  title = {{A risk prediction model for post-stroke depression in Chinese stroke survivors based on clinical and socio-psychological features}},
  booktitle = {Oncotarget},
  year = {2017},
  doi = {10.18632/oncotarget.16907},
}
Figure