Post-stroke sequelae risk score (PSSRS): a 16-year retrospective study developing an internally validated multi-algorithm model for post-stroke sequelae
4 affiliations
- 1Department of Cerebrovascular Surgery, Neurosurgery Center, The Neurosurgery Institute of Guangdong Province, Zhujiang Hospital, The National Key Clinical Specialty, Southern Medical University
- 2School of Biomedical Engineering, Southern Medical University
- 3School of Nursing, Southern Medical University
- 4School of Mechanical Engineering, Dongguan University of Technology
Abstract
Introduction: Post-stroke sequelae (PSS) encompass persistent functional impairments after stroke and may benefit from early risk stratification. We developed and internally validated a time-to-event prediction model based on routinely available clinical and biochemical variables.
Method: This retrospective single-center cohort included 2,095 patients hospitalized with acute stroke at Zhujiang Hospital between 2008 and 2024. Thirty-seven candidate variables were assessed. The cohort was split 7:3 into training ( n = 1,467) and internal test ( n = 628) sets before preprocessing, feature selection, tuning, and model comparison. Univariate Cox, multivariable Cox, least absolute shrinkage and selection operator (LASSO), and random survival forest (RSF) analyses identified eight intersection variables. A total of 101 survival-model pipelines were compared using training-set resampling; the selected model was then evaluated in the internal test set.
Result: The StepCox (forward) + RSF pipeline showed the highest prespecified training-resampling performance. Its Harrell C-index was 0.87 in the training set and 0.85 in the internal test set. Test-set time-dependent AUROCs at 1, 3, and 5 years were 0.800 (95% CI, 0.760–0.840), 0.830 (0.790–0.870), and 0.820 (0.780–0.860), respectively; corresponding AUPRCs were 0.748, 0.772, and 0.761, and Brier scores were 0.140, 0.170, and 0.180. Calibration slopes were 0.96, 0.93, and 0.90. LDH, CK-MB, and DBIL were among the leading contributors in the model-interpretation analyses.
Conclusion: PSSRS provides an internally validated framework for estimating individualized risk of post-stroke sequelae. External temporal and multicenter validation is required before clinical implementation.
Related
- Exploratory clinical predictors and internally validated models of in-hospital mortality after PCI in obese patients with acute myocardial infarction and documented cardiogenic shock: a single-center retrospective study.
- Needle-Free Bedside and Anthropometric Measures Show Weak Discrimination of an Adverse Blood-Biomarker Profile in Middle-Aged and Older Mexican Adults: An Internally Validated Machine-Learning Study
- Identification of candidate inverse miRNA–mRNA pairs and an internally validated tissue-based diagnostic model for benign and malignant breast lesions
- Clinicopathological predictors and an internally validated exploratory nomogram for 2-year recurrence after hepatic resection of hepatocellular carcinoma ≤5 cm.
- The vitality infant algorithm: A validated tool to classify food allergy outcomes in research studies
BibTeX
@article{dong2026poststroke,
title = {{Post-stroke sequelae risk score (PSSRS): a 16-year retrospective study developing an internally validated multi-algorithm model for post-stroke sequelae}},
author = {Yiqun Dong and Rongji Lu and Yuying Zhang and Yuteng Huang and Ziyi Wang and Jiman Sun and Siyi Li and Yuxin Jia and Ying Tian and Haodong Hu and Xifeng Li and Yanchao Liu and Chuanzhi Duan},
journal = {Frontiers in Neurology},
year = {2026},
month = oct,
doi = {10.3389/fneur.2026.1833445},
url = {https://www.frontiersin.org/articles/10.3389/fneur.2026.1833445/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Post-stroke sequelae risk score (PSSRS): a 16-year retrospective study developing an internally validated multi-algorithm model for post-stroke sequelae
AU - Yiqun Dong
AU - Rongji Lu
AU - Yuying Zhang
AU - Yuteng Huang
AU - Ziyi Wang
AU - Jiman Sun
AU - Siyi Li
AU - Yuxin Jia
AU - Ying Tian
AU - Haodong Hu
AU - Xifeng Li
AU - Yanchao Liu
AU - Chuanzhi Duan
PY - 2026
DA - 2026/10/15
JO - Frontiers in Neurology
DO - 10.3389/fneur.2026.1833445
UR - https://www.frontiersin.org/articles/10.3389/fneur.2026.1833445/full
AB - Introduction: Post-stroke sequelae (PSS) encompass persistent functional impairments after stroke and may benefit from early risk stratification. We developed and internally validated a time-to-event prediction model based on routinely available clinical and biochemical variables. Method: This retrospective single-center cohort included 2,095 patients hospitalized with acute stroke at Zhujiang Hospital between 2008 and 2024. Thirty-seven candidate variables were assessed. The cohort was split 7:3 into training ( n = 1,467) and internal test ( n = 628) sets before preprocessing, feature selection, tuning, and model comparison. Univariate Cox, multivariable Cox, least absolute shrinkage and selection operator (LASSO), and random survival forest (RSF) analyses identified eight intersection variables. A total of 101 survival-model pipelines were compared using training-set resampling; the selected model was then evaluated in the internal test set. Result: The StepCox (forward) + RSF pipeline showed the highest prespecified training-resampling performance. Its Harrell C-index was 0.87 in the training set and 0.85 in the internal test set. Test-set time-dependent AUROCs at 1, 3, and 5 years were 0.800 (95% CI, 0.760–0.840), 0.830 (0.790–0.870), and 0.820 (0.780–0.860), respectively; corresponding AUPRCs were 0.748, 0.772, and 0.761, and Brier scores were 0.140, 0.170, and 0.180. Calibration slopes were 0.96, 0.93, and 0.90. LDH, CK-MB, and DBIL were among the leading contributors in the model-interpretation analyses. Conclusion: PSSRS provides an internally validated framework for estimating individualized risk of post-stroke sequelae. External temporal and multicenter validation is required before clinical implementation.
SN - 1664-2295
ER -