Multidimensional analysis of preoperative patient-reported outcomes identifies distinct phenotypes in patients booked for total knee arthroplasty: Secondary analysis of the SHARKS registry in a metropolitan hospital department

Abstract

AimsTraditional research on total knee arthroplasty (TKA) has predominantly utilised individual preoperative patient-reported outcome measures (PROMs) to predict postoperative satisfaction. The objective of this study was to leverage the power of machine learning to simultaneously analyse multiple PROMs, thus unveiling unique patient phenotypes within this heterogeneous population. This study aimed to identify phenotypes in patients scheduled for TKA using multidimensional analysis of preoperative PROMs and evaluate their correlation with patient characteristics.MethodsPatients booked for primary TKA between 2017 and 2021 at a metropolitan public hospital were enrolled in the clinical quality registry (SHARKS) maintained by the orthopaedics department. The registry collected demographic and clinical data, including PROMs such as the Veterans Rand 12 (VR-12) and Oxford Knee Score (OKS). The imputed data was used for the primary analysis, which involved clustering patients based on their responses to VR-12 and OKS subscale scores using a k-means clustering algorithm. A k value of 4 was chosen to identify four distinct phenotypes. ANOVA was used to compare average scores in each cluster. Relationships between phenotype and patient age, sex, body mass index (BMI), and laterality were determined using nominal logistic regressionResultsThere were 389 patients and 450 primary knees included in the study (49.6% females and 50.4% males). The cluster analysis categorised patients into four distinct phenotypes based on VR-12 PCS, VR-12 MCS, OKS-F, and OKS-P scores. Patients in phenotype 3 (Mild symptoms with good mental health) reported superior physical function, less pain, and higher mental and general health scores, while those in phenotype 2 (Severe symptoms with poor mental health) experienced the greatest knee pain, impairment, and poorer physical and mental health. Notably, phenotype 4 (Moderate symptoms with good mental health) patients reported high mental health scores despite significant knee pain and physical health impairment. Patient characteristics were significantly associated with phenotype. Patients in theSevere symptoms with poor mental healthphenotype were significantly more likely to be younger, female, have a higher BMI, and exhibit bilateral OA compared to those in theMild symptoms with good mental healthphenotype (P<0.05).ConclusionsThis preliminary study employed a multidimensional data analysis method to identify phenotypes in OA patients based on commonly used PROMs. These phenotypes demonstrated significant associations with patient demographics. A key advantage of this approach is the capability to identify patterns that may not be discernible when interpreting PROMs independently. The use of cluster analysis demonstrates potential for developing prognostic models in joint replacement and warrants further research to improve clinical decision-making and patient outcomes.