Predictive model for the therapeutic effect of bilateral subthalamic nucleus deep brain stimulation on the freezing of gait in Parkinson's disease

Abstract

Background: Freezing of gait (FOG) is a major disabling symptom that affects the quality of life of patients with Parkinson's disease (PD). To date, notions regarding the effects of deep brain stimulation of the subthalamic neucleus (STN-DBS) on FOG remain controversial. Therefore, we developed a prediction model based on the influence of bilateral deep brain stimulation (DBS) of the subthalamic nucleus (STN) on FOG in patients with PD.

Methods: We collected data from 104 PD participants with FOG who underwent STN-DBS at Xuanwu Hospital between September 2017 and June 2022. The patients were divided into a training set (70%; n = 68) and a validation set (30%; n = 36). The selected characteristics in the LASSO regression were used in multivariate logistic regression to build the prediction model. The receiver operating characteristic (ROC) curves were constructed for the training and validation sets to verify the model's efficiency.

Results: Independent variables in the prediction model included Unified Parkinson's Disease Rating Scale II (UPDRS II), UPDRS IV, leg rigidity, Montreal Cognitive Assessment (MoCA) score, and Mini-Mental State Examination (MMSE) score. The prediction model formula is as follows: Logit(y) = -1.0043 + 0.159 × UPDRS II + 0.030 × UPDRS IV - 1.726 × leg rigidity + 0.121 × MoCA + 0.036 × MMSE. To validate the model, we analyzed the ROC curves of the training and validation sets. The area under the ROC curve (AUC) of internal validation was 0.869 (95% confidence interval [CI]: 0.771-0.967) and the AUC of external validation was 0.845 (95% CI: 0.6526-1). The calibration plots showed good calibration.

Conclusion: The model we developed can effectively assist clinicians in assessing the efficacy of deep brain stimulation of the bilateral subthalamic nucleus for freezing of gait in Parkinson's disease patients. This approach can support the formulation of personalized treatment plans and has the potential to improve patient outcomes.

Keywords: Parkinson’s disease; deep brain stimulation; freezing of gait; prediction model; subthalamic nucleus.

Figure 1

Variable selection using the LASSO binary logistic regression model. (A) LASSO coefficients of 30 variables. The optimal penalization coefficient (lambda) is identified in the LASSO model where optimal λ results in five features with non-zero coefficients. (B) The partial likelihood deviance (binomial deviance) curve is plotted against log(lambda). The left vertical line is dotted at the minimum criteria of λ, and the right vertical line is dotted at one standard error of the minimum λ. LASSO, least absolute shrinkage and selection operator.

Figure 2

(A) Nomogram including MDS-UPDRS II, MDS-UPDRS IV, Legs rigidity, MoCA, and MMSE. (B) One patient is randomly selected, and the probability of postoperative FOG symptom improvement in this patient is predicted based on the five characteristic indicators of the nomogram.

Figure 3

(A) The x-axis represents the predicted probability of FOG symptom improvement 1 year postoperatively, while the y-axis represents the actual probability. The diagonal line represents perfect prediction by an ideal model. The higher the ideal curve, the actual curve, and the calibration curve overlap, the better the model accuracy. (B) The x-axis represents the threshold probability, and the y-axis represents the net benefit. The solid line parallel to the x-axis represents patients who showed no postoperative improvement, the gray oblique solid line represents those who showed all improvement postoperatively, and the red curve represents the prediction model (nomogram) developed in the present study.

Figure 4

The area under the ROC curve is used to determine the predictive ability of the prediction model. The x-axis represents the false positive rate predicted by the model; while the y-axis represents the true positive rate predicted by the model. The red curve represents the performance of the nomogram. (A) The performance of the model in the training set. (B) The performance of the model in the validation set.