Clinical Phenotyping Of End-Stage Osteoarthritis Comorbidities By Unsupervised Clustering Analysis Based On Machine Learning

Objectives Despite the great success of osteoarthritis in therapeutics,approximately 10-15%of patients are still dissatisfied with their outcome.For this reason,it is increasingly recognized that osteoarthritis is a complex and highly heterogeneous disease,consisting of different phenotypes with different characteristics at different stages of disease progression.Comorbidities,as components of these heterogeneous features,often coexist with osteoarthritis,and are particularly prevalent in end-stage osteoarthritis.Here,we attempted to identify the different clinical phenotypes of comorbidities in patients with end-stage osteoarthritis by cluster analysis.Methods A total of 421 hospitalized surgical patients with a confirmed diagnosis of end-stage knee osteoarthritis were included in this cross-sectional study,and 46 variables were collected,including age,sex,BMI,abdominal circumference,ABO blood group,EQ-VAS,VRS,NSAIDs medication use,pain duration,pain exacerbation time,pain frequency,and 35 comorbidities(obesity,hypertensive disease,type 2 diabetes,coronary artery disease,hyperlipidemia,stroke,sleep disorders,depression,anxiety,hyperuricemia,chronic kidney disease,liver disease,rheumatoid arthritis(excluding rheumatoid knee osteoarthritis),COPD,asthma,osteoporosis,migraine,viral hepatitis,prostate hypertrophy,kidney stones,gastritis and gastric ulcers,gallbladder disease,inflammatory bowel disease,cancer,hyperthyroidism,hypothyroidism,Parkinson’s disease,central arterial disease,peripheral vascular disease,tuberculosis,anemia,low back pain,cataract,hearing loss,and vision problems.Systematic clustering after factor analysis and separate cluster analysis for individual comorbid variables and all variables were performed to objectively identify the different clinical phenotypes of the study patients.Results Six clusters were finally identified.C1: Cluster 1 was the largest cluster,including94 patients,representing 22%,and was mainly characterized by osteoporosis(90%).Obesity(4%),gallbladder disease(9%),stroke(4%),coronary heart disease(5%)and anxiety disorders(4%)accounted for the smallest proportion of the six clusters,with no one suffering from asthma,depression,COPD or chronic kidney disease.C2: Cluster 2 was the second largest cluster,including 82 patients or 19%,with the largest proportion of patients with peripheral vascular disease(71%)and diabetes mellitus(78%).Gastrointestinal disease(15%)had the highest proportion of all clusters.C3: Cluster 3 was the third largest cluster and included 80 patients or 19%.All patients(100%)were depressed.Anxiety disorders were also common(78%).Sleep disorders(15%),Parkinson’s disease(4%)and cancer(6%)had the highest proportion of the six clusters.C4: Cluster 4 was the fifth largest cluster and included 54 patients or 13%.Hypertensive disorders(83%)had the largest proportion of patients.Depression also had a large proportion(72%).Migraine(19%),cataract(15%),vision problems(19%),stroke(24%)and rheumatoid arthritis(26%)had the highest proportions of all clusters.C5: Cluster 5was the fourth largest cluster and included 60 patients or 14% with no significantly high proportion of comorbidities,i.e.patients with isolated end-stage osteoarthritis combined with a few comorbidities.Diabetes(5%),gastrointestinal disorders(3%)and low back pain(8%)had the lowest proportions of all clusters.c6: Cluster 6 was the smallest cluster and included 51 patients or 12%,with the highest proportions of patients with obesity(96%)and hypertensive disorders(90%).In addition,hyperlipidemia(59%)and liver disease(65%)were also represented in high proportions.Hypothyroidism(10%),gallbladder disease(22%),and hyperuricemia(22%)had the highest proportions in the six clusters.Conclusions Patients with end-stage osteoarthritis may be classified into 6 different clinical phenotypes,namely "isolated end-stage osteoarthritis";"obesity disease";"depression +anxiety","osteoporosis","hypertension" and "peripheral vascular disease + diabetes",which exhibit different clinical outcomes.This study may provide a direction for personalized treatment strategies for patients with end-stage OA comorbidities,and future multi-omics analyses should be performed to elucidate the mechanisms of interaction between end-stage osteoarthritis comorbidities.