Machine learning algorithm to predict anterior cruciate ligament revision demonstrates external validity.

Area of Science:

• Orthopedic surgery
• Biomedical engineering
• Data science in medicine

Background:

• External validation of machine learning models is crucial for clinical translation but rarely performed.
• A machine learning tool was developed using the Norwegian Knee Ligament Register (NKLR) to predict anterior cruciate ligament (ACL) revision risk.
• The clinical utility of newly developed predictive models is often hindered by a lack of external validation.

Purpose of the Study:

• To externally validate the NKLR machine learning model for predicting ACL reconstruction revision.
• To assess the performance of the NKLR model when applied to a distinct patient cohort from the Danish Knee Ligament Registry (DKLR).

Main Methods:

• The NKLR model's predictive variables included graft choice, fixation device, KOOS QOL score, time from injury to surgery, and age at surgery.
• 10,922 patients from the DKLR with complete data were included for validation.
• Model performance was evaluated using concordance and calibration metrics, consistent with the original NKLR study.

Main Results:

• Despite dissimilarities in surgical techniques and injury characteristics between the NKLR and DKLR populations, the model achieved comparable concordance (DKLR: 0.68 vs. NKLR: 0.68-0.69).
• Calibration was found to be poorer at 1 and 5 years post-surgery in the DKLR cohort compared to the NKLR cohort.
• Calibration at 2 years post-surgery was similar between the two registries.

Conclusions:

• The NKLR machine learning algorithm demonstrates external validity, performing similarly when applied to the DKLR patient population.
• This study represents the first external validation of a machine learning model for predicting revision ACL reconstruction.
• The validated calculator can assist clinicians in estimating patient-specific revision risk pre-operatively, though performance in non-Scandinavian populations remains unassessed.