Predicting the onset of freezing of gait in Parkinson’s disease

Freezing of gait (FoG) refers to a sudden inability to initiate or continue gait [1]. As a common symptom with increasing frequency as Parkinson’s disease (PD) progresses, it has a significant impact on the patient quality of life [1]. FoG is hard to study due to its transient occurrence and multifaceted pathophysiology. The locomotor network, especially the subthalamic nucleus (STN), globus pallidus internus (GPi), and substantia nigra pars reticulata (SNr) can all contribute to the movement breakdown [2]. Moreover, cognitive and limbic networks are also involved in this gait abnormality [3]. While effective therapies for FoG are still under investigation, early prediction of FoG may identify future patients for preventive management.

Several methods are adopted to predict FoG onset [4]. Wearable sensors, which can objectively detect the gait disturbance, are accessible to a limited number of patients [5]. Clinical variables and neuro-physiological biomarkers, however, are more common in clinical use. Various associations between FoG and clinical observations, genetic variations, Cerebrospinal fluid (CSF) biomarkers as well as neuro-imaging features have been discovered [6‐14]. Motor factors such as rigidity, postural instability, and bradykinesia are predictive of FoG [6]. Non-motor factors, including specific cognitive deficits, mood disorders, and autonomic dysfunctions, are also associated with FoG occurrence [7‐10]. CSF biomarkers such as β-amyloid 1–42 (Abeta₄₂) and gene mutations including APOE ε4 have been associated with the symptom onset [8]. Striatal dopaminergic denervation, which can be examined using dopamine transporter (DAT) scans and single photon emission computed tomography (SPECT) imaging, is also associated with FoG pathology [15]. Although past studies have identified several FoG risk factors, few have combined these factors to predict FoG onset.

Instead of focusing on one aspect of clinical or imaging assessments, our study analyzed a comprehensive battery of indicators. Motor and non-motor factors, genetic characteristics, CSF biomarkers as well as the neuroimaging parameters were evaluated using longitudinal data of five-year visit from the PPMI cohort. Our goal was to determine the early symptoms and characteristics exhibited in PD patients before FoG occurrence.

In this study, we evaluated various aspects of potential risk factors to predict FoG onset in early developed PD patients. There were 183 patients enrolled in the study among whom 68 patients (37.2%) developed FoG in the follow up duration of five years. Overall, 31.1% of patients (57/183) reported ‘freezing when walking’ in their activities of daily living, while 17.5% of patients (32/183) were defined as FoG by the investigator in clinic (Table S1, Figure S1). The two groups showed good homogeneity in the baseline characteristics except for the differences in age (p = 0.03) and age at symptom onset (p = 0.05). Reasons for the differences observed are unknown but the small sample size may explain at least a part of it. The influence of the differences is likely to be limited as the two variables were excluded from the backward stepwise selection and age was adjusted as a covariate. The prevalence of FoG in the present study (37.2%) was comparable to previous studies, as a meta-analysis found that the weighted prevalence of FoG in early-stage PD patients with a disease duration ≤ 5 years was 37.9% [18]. The lower rate of FoG detected in clinic reflects the transient nature of FoG symptoms, suggesting a combination of questionnaire with regard to daily living can be more sensitive to reflect FoG development in PD patients.

Freezing of gait has been recognized as a neuronal integration failure caused by a multilevel brain network. It is influenced by cognitive, sensory-perceptual, and affective manipulations, and can be induced by several occasions such as turning, fatigue, confined spaces, and stressful situations [19, 20]. This study found that motor factors, along with non-motor factors including cognitive functions, mood and CSF Abeta predisposed PD patients to FoG development. While motor features have been recognized to correlate with FoG [4, 7], it was noticed that motor subtypes also correlated with non-motor features [21]. To further evaluate the difference in TD/PIGD subtype, we analyzed baseline features of different motor phenotypes (Table S2) in PD patients. No difference in Abeta, fatigue and SDMT scores was found significant at baseline between TD and PIGD subtypes. However, there were differences observed in education years, HVLT Retention, number of e4 alleles in APOE genotype and depression apart from the tremor score in groups. Although none of these factors were identified as risk factors in our studies, they may indirectly contribute to FoG through their influence on motor phenotypes. Previous studies have reported the associations between these factors and FoG. Therefor, we can not exclude their potential value in the prediction of FoG onset [10, 22]. As was shown by previous study conducted by Kim et al., PIGD score was a strong predictor of FoG [8]. In our analysis,the predictive power of PIGD score was stronger than that of TD or PIGD subtype. Besides, PIGD score also showed a correlation with fatigue and SDMT score in our analysis, supporting again the coexistence and inter-relationship of motor and non-motor symptoms in the development FoG.

Non-motor features are considered of increasing importance in the development of FoG [19]. In our study, fatigue and cognitive deficit (represented as lower SDMT scores), emerged as independent FoG predictors. In PD patients, fatigue is a major triggering factor of FoG [20]. It is associated with FoG occurrence in clinically observed FoG as well as self-reported FoG, despite late or early onset [23]. However, it is influenced by motor phenotype [21]. Noticeably, this study showed a correlation between PIGD score and fatigue (r = 0.22, p = 0.003, Table S3) at baseline. The control of PIGD symptoms and fatigue may be originiated from the same or adjacent neuronal circuit in FoG development while further investigations are warranted. Accumulating evidence has suggested cognitive effects, specifically, the executive functions, attention, and visuospatial functions to FoG occurrence [24]. In our study, SDMT, which is a commonly used instrument to evaluate cognitive functions especially attention deficit, was identified as an independent risk factor. While studies suggested visual and motor con-founders in SDMT interpretations [25], a recent study using gaze analysis technique excluded the confounding effects, further demonstrating the role of cognitive functions in SDMT performance in PD patients [26]. In this study, we demonstrate the role of SDMT performance in FoG prediction, which implicates therapeutics for cognitive rehabilitation, especially for attention improvement might help delay FoG onset in PD patients.

Low CSF Abeta₄₂ levels are regarded as a biological fluid marker for Alzheimer's disease [27]. In this study, CSF Abeta₄₂ also correlated with FoG in PD patients. The decreased levels of CSF Abeta are associated with cognitive impairments and gait symptoms in PD [27, 28]. Extra-nigral pathologies, represented as the increased neocortical Abeta deposition, can significantly increase the risk of FoG development [29]. However, it remains unclear how the reduced level of CSF Abeta_42, or the deposition of Abeta₄₂ contributes to the motor dysregulation in PD patients [29].

Our study has several limitations. First, it is a retrospective study with a limited number of subjects in the PPMI study. Participants in PPMI were generally well educated, which may not be representative of the other group. Second, we could not distinguish the medication states of patients identified by MDS-UPDRS Part II, thus restricting our study of FoG under different medication conditions. Third, we did not analyze the severity of FoG. The combination of UPDRS II and UPDRS III for FoG identification made it difficult to find a consistent standard to measure the severity. Besides, UPDRS II may be less sensitive to detect FoG compared with the FoG questionnaire (FOG-Q) [30]. The number of freezers may still be underestimated in the analysis. Fourth, we built our model from the complete-case analysis. Data with missing values were deleted and this may introduce selection bias. The difference in results was indeed observed. As in the multivariate logistic regression analysis on 251 patients, variables that were selected in the model were PIGD score, SCOPA-AUT gastrointestinal (GI) score, SDMT score, mean striatum DAT uptake and Abeta. Nevertheless, all the variables in our model were significantly associated with FoG onset in the patient set with missing values, showing good consistency in the results. In this study, we evaluated FoG with both self-reported rating scales and examinations from clinical specialists. We integrated a comprehensive battery of clinical, biochemical, and imaging assessments and emphasized several independent risk factors. Future prospective studies integrating the identified factors under different medication status may further demonstrate their prognostic value and deepen our understanding of the development of FoG in PD patients.

In summary, our findings determined the risk factors of FoG occurrence among a series of clinical, imaging, biological as well as genetic characteristics. Our results stress the importance of PIGD score, fatigue, SDMT performance and CSF Abeta₄₂ in predicting FoG onset. Combining these factors with further studies will assist patients, caregivers, and healthcare professionals to conduct early interventions as disease progresses. Physiotherapy, pharmacological treatments, or neuromodulation that improve the performance of these indicators will be of value in the early intervention of the debilitating symptom.