Development and validation of a nomogram to predict the risk of post-stroke complex regional pain syndrome

Abstract

Objective: This study aims to assess risk factors and build a nomogram model to facilitate the early recognition of post-stroke complex regional pain syndrome (CRPS).

Methods: A total of 587 stroke patients admitted to Dongguan Hospital of Guangzhou University of Traditional Chinese Medicine from September 2021 to October 2024 were initially included in this study. After exclusions, 376 patients were selected. Among these, there were 90 patients with post-stroke CRPS, while the non-stroke CRPS group consisted of 286 patients. Feature selection and optimization to generate the predictive model and nomogram were performed using LASSO regression and multivariable logistic regression analysis. We also utilized calibration plots, receiver operating characteristic (ROC) curves, decision curves (DCA), and clinical impact curves (CIC) for model validation.

Results: LASSO regression analysis and multivariate logistic regression identified gender, age, NIHSS score, cervical spondylosis, sleep disorders, fasting blood glucose (FBG), and albumin (ALB) as significant predictors. The nomogram model showcased reliable predictive effectiveness, achieving an area under the curve (AUC) of 0.858 (95% CI, 0.801-0.915). Both DCA and CIC demonstrated that the nomogram model holds substantial clinical utility.

Conclusion: This study has developed a novel predictive model for post-stroke CRPS, providing a valuable tool to facilitate the early detection of high-risk patients in a clinical environment.

Keywords: LASSO; nomogram; post-stroke complex regional pain syndrome; prediction model; stroke.

FIGURE 1

Flowchart.

FIGURE 2

(A) The profile plot shows the LASSO coefficients for 31 features plotted against the log(lambda) sequence. Seven variables were chosen based on the optimal lambda. (B) A vertical dashed line is drawn to represent the value determined through 10-fold cross-validation. The optimal adjusted parameter log(λ) is displayed on the horizontal axis, while the deviance (binomial deviance) is shown on the vertical axis.

FIGURE 3

Nomogram for predicting CRPS in post-stroke patients.

FIGURE 4

The ROC curve for the nomogram predicting post-stroke CRPS (A). In the training set (A), the AUC was 0.858 (95% CI: 0.801–0.915). When a cutoff point was set at a risk probability of 0.180, the specificity and sensitivity of the predicted results were 77.3% and 80.7%, respectively. In the validation set presented in (B), the AUC was 0.740 (95% CI: 0.636–0.844), with a threshold of 0.099, a specificity of 0.633, and a sensitivity of 0.818.

FIGURE 5

Calibration curve for the post-stroke CRPS nomogram (A,B). The diagonal dashed line signifies perfect prediction by the optimal model. The proximity of the red solid line to this dashed line indicates the model’s predictive performance; the closer they are, the more accurate the predictions.

FIGURE 6

Decision curve analysis (DCA) for the post-stroke CRPS nomogram prediction (A,B). The y-axis measures the net benefit obtained from using the model. The red line in the figure represents the clinical diagnostic model for post-stroke CRPS. In contrast, the black horizontal line (None line) and the gray diagonal line (All line) represent the extreme cases of “no intervention” and “all intervention,” respectively.

FIGURE 7

Clinical impact curve (CIC) analysis for the post-stroke CRPS nomogram prediction (A,B). The y-axis represents the number of individuals at risk. The red line indicates the predicted number of events occurring for post-stroke CRPS, whereas the red dashed line depicts the actual occurrences.