This study was conducted in consecutive pediatric patients diagnosed with spastic cerebral palsy who received a single-level selective dorsal rhizotomy (SDR) procedure at the lumbar segment from January 2015 to January 2021. The post-operative rehabilitation program was applied to the children three days after the SDR, and they were required to have follow-up every 3-6 months. This study aimed to use unsupervised machine learning to effectively cluster pediatric patients with SCP for determination of optimal responders to SDR through correlation analyses and clustering of variables including age, pre-operative GMFCS level, GMFM-66 score, and number of target muscles, as well as MAS scores of bilateral hip adductors, hamstrings, gastrocnemius, and soleus.
The results of the study suggested that a hierarchical clustering approach was most effective for cluster the SCP patients and determining the optimal responders to SDR. The post-operative GMFM-66 score change was taken as an outcome measure to compare the effect of SDR amongst the three subgroups. The correlation analyses determined that GMFCS level had high correlation to GMFM-66 score and was thus eliminated as an input variable before the unsupervised machine learning cluster. The variables were then scaled before clustering, followed by the Elbow Method which was used to determine the “K” value of the variables before clustering. The clustering results were then visualized by dimensionality reduction through principal component analysis.
This study was approved by the Ethics Review Committee, Children’s Hospital of Shanghai, Shanghai Jiao Tong University (Approval No: 2020R069-E02), and the results are expected to lead to a better understanding of SDR and its effect on SCP patients. In addition, this study could provide a useful basis for assessing the best possible response to SDR in the selection of pediatric SCP patients who could potentially benefit most from the surgical procedures.
