Summary

An unsupervised clustering algorithm successfully analyzed demographic and injury characteristics of 1194 anterior cruciate ligament (ACL) injuries, identifying distinct patient subgroups (optimal and suboptimal) based on outcomes following ACL reconstruction (ACLR) versus nonoperative treatment. Machine learning analysis assessed the treatment effects of ACLR within these groups on key outcomes.

Abstract

Background

Treatment decisions in patients with anterior cruciate ligament (ACL) injuries are dictated by the risks of recurrent instability, the age and activity level of the patient, concomitant meniscal or cartilage injury, and the goals of the patient. Identifying the differential treatment effect of ACL reconstruction (ACLR) on a patient-specific level can inform surgical decision making. We hypothesize an unsupervised clustering algorithm can identify distinct patient subgroups based on outcomes achievement following ACLR.

Methods

A geographic database identified patients with ACL injury between 1990 and 2016 with minimum 7.5-year follow-up. Variables collected include age, sex, body mass index (BMI), activity level, occupation, relevant comorbid diagnoses, radiographical findings, injury characteristics, and clinical course. An unsupervised random forest algorithm was utilized to develop and internally validate distinct patient subgroups. Treatment effect of ACLR on outcomes were analyzed using a machine learning causal inference estimator (Targeted Maximum Likelihood Estimation-TMLE), while controlling for confounders. Patient subgroup membership, along with a total of 30 additional variables were incorporated into a stepwise multivariable logistic regression to identify factors predictive of optimal outcomes achievement.

Results

A total of 1194 patients with a minimum follow up of 7.5 years were included, among them 974 underwent primary reconstruction while 220 underwent nonoperative treatment. The random forest algorithm arrived at an optimal partition of 2 subgroups, with 709 patients in the optimal outcomes subgroup (403 male [56.8%], Age 25.9±10.2, BMI 26.4±4.24), and 485 patients in the suboptimal outcomes subgroup (283 male [58.4%], Age 35.2±9.93, BMI 30.10±5.47). The latter group demonstrated significant increased rates of secondary meniscal injury, development of symptomatic osteoarthritis (OA), and progression to total knee arthroplasty (TKA) at final follow-up (all p<0.001). In the optimal outcomes subgroup, ACLR had significant protective treatment effects on the risk of secondary meniscus injury (53%, 95% CI: 51-55%), contralateral ACL injury (7%, 95% CI: 4.9-10%), symptomatic OA (7.9%, 95% CI: 5.5-10%), and progression to TKA (6%, 95% CI: 5-6.9%, all p<0.001). Conversely, in the suboptimal outcomes subgroup, ACLR only protected against symptomatic OA (14%, 95% CI: 9.9-17%), and progression to TKA (4%, 95% CI: 2-7.1%, both p<0.001). Negative predictors for membership in the optimal outcomes subgroup included older age at injury (OR: 0.89, 95% CI: 0.87-0.92), greater BMI (OR: 0.77, 95% CI: 0.72-0.82), previous arthroscopic knee surgery (OR: 0.04, 95% CI: 0.02-0.09), and concomitant medial meniscus injury (OR: 0.05, 95% CI: 0.02-0.01, all p<0.001). Positive predictors for optimal outcomes included sports participation (OR: 4.50, 95% CI: 1.96-10.35, p<0.01).

Conclusion

Clinically meaningful subgroups exist following ACL injuries. Reconstruction exerted a protective effect on the development of post-traumatic OA and TKA in both subgroups, but was not effective in preventing secondary meniscus injuries or contralateral ACL injuries in patients who were older, heavier, or had concomitant medial meniscus injuries.