A New Clustering Method for Knee Movement Impairments using Partitioning Around Medoids Model

• Published in: Iranian journal of medical sciences 2020-11, Vol.45 (6), p.451-462
• Main Authors: Reza Farazdaghi, Mohammad, Razeghi, Mohsen, Sobhani, Sobhan, Raeisi Shahraki, Hadi, Motealleh, Alireza
• Format: Article
• Language: English
• Subjects: Classification; Cluster analysis; Knee; Methods; movement system impairment model; Pain; Patients; Physical therapists; Physical therapy; Rehabilitation; syndrome
• ISSN: 0253-0716; 1735-3688
• DOI: 10.30476/ijms.2019.82033.

The movement system impairment (MSI) model is a clinical model that can be used for the classification, diagnosis, and treatment of knee impairments. By using the partitioning around medoids (PAM) clustering method, patients can be easily clustered in homogeneous groups through the determination of the most discriminative variables. The present study aimed to reduce the number of clinical examination variables, determine the important variables, and simplify the MSI model using the PAM clustering method. The present cross-sectional study was performed in Shiraz, Iran, during February-December 2018. A total of 209 patients with knee pain were recruited. Patients' knee, femoral and tibial movement impairments, and the perceived pain level were examined in quiet standing, sitting, walking, partial squatting, single-leg stance (both sides), sit-to-stand transfer, and stair ambulation. The tests were repeated after correction for impairments. Both the pain pattern and the types of impairment were subsequently used in the PAM clustering analysis. PAM clustering analysis categorized the patients in two main clusters (valgus and non-valgus) based on the presence or absence of valgus impairment. Secondary analysis of the valgus cluster identified two sub-clusters based on the presence of hypomobility. Analysis of the non-valgus cluster showed four sub-clusters with different characteristics. PAM clustering organized important variables in each analysis and showed that only 23 out of the 41 variables were essential in the sub-clustering of patients with knee pain. A new direct knee examination method is introduced for the organization of important discriminative tests, which requires fewer clinical examination variables.