AUTHOR=Xu Kun , Zhou Xiao-xia , He Run-cheng , Zhou Zhou , Liu Zhen-hua , Xu Qian , Sun Qi-ying , Yan Xin-xiang , Wu Xin-yin , Guo Ji-feng , Tang Bei-sha 

TITLE=Constructing Prediction Models for Freezing of Gait by Nomogram and Machine Learning: A Longitudinal Study

JOURNAL=Frontiers in Neurology

VOLUME=Volume 12 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2021.684044

DOI=10.3389/fneur.2021.684044

ISSN=1664-2295

ABSTRACT=Objectives: Although risk factors for freezing of gait (FOG) have been reported, there are still few prediction models based on cohorts that predict FOG. This one-year longitudinal study was aimed to identify the clinical measurements closely linked with FOG in Chinese Parkinson’s Disease (PD) patients and construct prediction models based on those clinical measurements using Cox regression and machine learning.
Methods: The study enrolled 967 PD patients without FOG in Hoehn and Yahr (H&Y) stage 1-3 at baseline. Development of FOG during follow-up was the end-point. Neurologists trained in movement disorders collected information from the patients on PD medication regimen and clinical characteristics. The cohort was assessed on the same clinical scales, and baseline characteristics were recorded and compared. After the patients were divided into the training set and test set by the stratified random sampling method, prediction models were constructed using Cox regression and random forests (RF).
Results: At the end of the study, 26.4% (255 / 967) patients suffered from FOG. Patients with FOG had significant longer disease duration, greater age at baseline and H&Y stage, lower proportion in Tremor Dominant (TD) subtype, higher proportion in wearing-off, LEDD, usage of L-Dopa and COMT inhibitors, higher score in scales of UPDRS, PDQ-39, NMSS, HDRS-17, PFS, RBD-HK, ESS, and lower score in scales of PDSS (P < 0.05). Risk factors associated with FOG included PD onset not being under the age of 50 years, lower degree of tremor symptom, impaired ADL, UPDRS item 30 posture instability, unexplained weight loss and higher degree of fatigue. The concordance index (C-index) was 0.68 for the training set (for internal validation) and 0.71 for the test set (for external validation) of the nomogram prediction model, which showed a good predictive ability for patients in different survival times. The RF model also performed well, the C-index was 0.74 for the test set, and the AUC was 0.74.
Conclusions: The study found some new risk factors associated with the FOG including lower degree of tremor symptom, unexplained weight loss and higher degree of fatigue through a longitudinal study, and constructed relatively acceptable prediction models.