AUTHOR=Li Mengjiao , Lan Lan , Luo Jiawei , Peng Li , Li Xiaolong , Zhou Xiaobo TITLE=Identifying the Phenotypic and Temporal Heterogeneity of Knee Osteoarthritis: Data From the Osteoarthritis Initiative JOURNAL=Frontiers in Public Health VOLUME=Volume 9 - 2021 YEAR=2021 URL=https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2021.726140 DOI=10.3389/fpubh.2021.726140 ISSN=2296-2565 ABSTRACT=Objective: Previous studies, discussing phenotypic and temporal heterogeneity of knee osteoarthritis (KOA) separately, have fatal limitation that either clustering patients with similar severity or assuming all subjects have a single common progression pattern which is unreliable. This study tried to uncover more reliable information on phenotypic and temporal heterogeneity of KOA. Design: Data were from Osteoarthritis Initiative database. We included 678 unilateral knees that have greater Kellgren and Lawrence (KL) grade than the contralateral knees at baseline and in all follow-up 48 months. Measurements of biomarkers at baseline were chosen. We applied Subtype and Stage Inference model (SuStaIn) as a subtype-progression model to identify subtypes, subtype biomarker progress sequences and stages of KOA. Results: This study identified three subtypes which account for 15%, 61%, and 24% of subjects respectively. Each subtype has distinct subtype biomarker progress sequence. For knees with KL grade 0/1, 2, 3 and 4, they have different distributions on stage and 26%, 53%, 89%, and 95% of them are strongly assigned to subtypes. When assessing whether a subject has KL (grade ≥ 2), subtypes and stages from subtypes-progression model (SuStaIn) are significantly better fitting than those from subtypes-only (mixture of Gaussians) (likelihood ratio = 105.59, p = 2.2×10-16) or stages-only (SuStaIn which setting c=1) (likelihood ratio = 58.04, p = 2.57×10-14) model. Stages in subtypes-progression model have greater β than stages-only model. Subtypes from subtypes-progression model have no statistical significance. Conclusions: For subtypes-progression model, stages contain more complete temporal information and subtypes are closer to real OA subtypes.