AJSM Webinar Series: ACL Injury and PTOA: Bridging Biology, Imaging, and Machine Learning to Improve Outcomes (Recording)-Lu Y et al. Establishing Clinically Distinct Patient Treatment Subgroups Following ACL Reconstruction: A Machine Learning Clustering Analysis. AJSM 2025. PMID: 40815848

Lu Y et al. Establishing Clinically Distinct Patient Treatment Subgroups Following ACL Reconstruction: A Machine Learning Clustering Analysis. AJSM 2025. PMID: 40815848

Back to course

Pdf Summary

This retrospective cohort study used machine learning to identify clinically distinct subgroups of patients after anterior cruciate ligament (ACL) injury and to compare outcomes of ACL reconstruction (ACLR) versus nonoperative treatment. The study drew from the Rochester Epidemiology Project and included 923 patients treated between 1990 and 2016 with at least 7.5 years of follow-up; 785 underwent primary ACLR and 138 were treated nonoperatively.<br /><br />An unsupervised random forest clustering algorithm identified two subgroups: an “optimal outcome” group (653 patients) and a “suboptimal outcome” group (270 patients). The suboptimal group was older, had higher body mass index, more smoking, more diabetes/inflammatory disease, lower activity levels, more laborer occupations, more medial meniscus injuries, and more proximal ACL tears. This group also had higher rates of symptomatic posttraumatic osteoarthritis (PTOA), total knee arthroplasty (TKA), and worse clinical stability on follow-up.<br /><br />Using targeted maximum likelihood estimation, the authors found that ACLR was protective in both groups against PTOA and progression to TKA. However, only the optimal subgroup showed a clear protective effect of ACLR against secondary meniscal injury and contralateral ACL injury. In multivariable analysis, being a student, participating in sports, and having a noncontact injury predicted optimal subgroup membership, while older age, higher BMI, and medial or bicompartmental meniscal injury predicted suboptimal membership.<br /><br />Overall, the study suggests that ACL injury patients are not a single homogeneous population. ACLR appears to reduce long-term degenerative complications in both subgroups, but its ability to prevent secondary injuries is less pronounced in older, heavier patients and those with concomitant medial meniscus injury.

Meta Tag

Concept Unsupervised Machine Learning Clustering

Concept Anterior Cruciate Ligament Injury

Concept ACL Reconstruction

Concept Patient Subset

Concept Nonoperative Management

Keywords

anterior cruciate ligament injury

ACL reconstruction

nonoperative treatment

machine learning clustering

retrospective cohort study

posttraumatic osteoarthritis

total knee arthroplasty

meniscal injury

patient subgroups

clinical outcomes

Unsupervised Machine Learning Clustering

Anterior Cruciate Ligament Injury

ACL Reconstruction

Patient Subset

Nonoperative Management