Prospective Forecasting of Quadriceps Performance Post-ACLR in Collegiate Athletes

Abstract

Background:

Accurate prognosis of quadriceps performance postanterior cruciate ligament reconstruction (ACLR) is valuable to clinicians to set expectations and guide interventions. Advanced probabilistic modeling, such as Bayesian frameworks, may enhance recovery forecasting, promoting more personalized clinical decisions. This study aimed to characterize recovery of quadriceps function throughout the initial 14 months post-ACLR in collegiate athletes, and to determine the influence of the number of previous assessments included on the accuracy of the model prediction

Hypothesis:

Predictive performance (as measured by root mean square predictive error) would improve significantly with each assessment added to the model.

Study Design:

Cohort study.

Level of Evidence:

Level 2.

Methods:

A total of 66 Division I collegiate athletes (317 assessments; 35 female athletes) completed serial isometric performance assessments at 1 to 14 months post-ACLR to quantify peak torque (PT) limb symmetry index (LSI), rate of torque development (RTD) LSI, and torque steadiness (TS) of the surgical limb. Bayesian hierarchical B-spline models including graft type and time post-ACLR, with varying degrees of complexity, were compared for PT, RTD, and TS recovery, separately. A novel cross validation design, mimicking the clinical use of this model, assessed the ability to forecast recovery trajectories based on an iterative increase in number of observations using root mean square error (RMSE).

Results:

The model that allowed the recovery curves’ shape to vary by graft type and included athlete-specific random intercepts produced the most accurate predictions, achieving RMSEs of 0.109, 0.173, and 0.536 for predicting PT, RTD, and TS after the first n = 2 assessments. It took at least 10, 12, and 6 months, respectively, on average, to reach the clinical targets utilized for PT (90%), RTD (85%), and TS (0.5% of PT) with 95% certainty.

Conclusion:

Bayesian hierarchical modeling offers a robust and flexible tool for forecasting quadriceps recovery post-ACLR, and demonstrated that RTD took the longest to achieve the clinical targets.

Clinical Relevance:

This predictive modeling approach serves as a “proof-of-concept” and has the potential to better individualize patient recovery trajectories.