Every person facing a torn anterior cruciate ligament eventually wonders whether they will return to the sport they love, and this study, drawn from a small group of 75 patients treated between 2016 and 2023, asks whether that answer might be glimpsed before the operation even begins. By feeding a machine learning model the measurable facts known the day before surgery, researchers found signals that pointed toward both a return to pivoting sports and a knee that feels livable one year later. The most striking clue lay not in the injured leg but in the healthy one, a reminder that recovery is
Machine Learning Model Predicts One-Year ACL Reconstruction Outcomes From Preoperative Testing
The healthy leg's strength helped predict how the injured knee would do
Mark
What is the headline here? A computer can predict who gets back to sport after an ACL repair?
Mimi
Roughly, yes, using information gathered before surgery. In 75 patients, a random forest model reached an AUC of 0.832 for return to pivoting sports and 0.815 for acceptable symptom state at one year.
Luke
Before we get excited, 75 patients is small, and the models weren't tested on any outside group. Those AUCs come from this one cohort.
Mark
Does cross-validation not cover that?
Luke
Partly. They used 3-fold cross-validation, which checks the model on held-out slices of the same data. That is not the same as external validation on different patients from a different clinic.
Mimi
The authors say as much. They call it exploratory and hypothesis-generating, not ready for clinical use.
Mark
What did the model lean on most?
Mimi
Younger age and greater flexor strength in the uninvolved limb. The healthy leg's strength mattered.
Luke
Worth noting that SHAP shows what drove this model's predictions, not what causes recovery. The source doesn't claim that strengthening the healthy leg would change outcomes.
Mark
And the two outcomes are different things?
Mimi
Yes. One is activity level, Tegner 6 or higher. The other is a symptom score of 75.9 or above on the IKDC, which reflects how the patient feels.
Luke
What we don't know is how it would perform on new patients. That is the open question.
Der Puls
- A torn ACL leaves patients asking a frightening question about whether their sporting life can return, and this study tries to answer it before surgery.
- Random forest models built on 75 patients reached AUCs of 0.832 for return to pivoting sports and 0.815 for an acceptable symptom state, well above chance.
- Younger age and stronger flexors in the uninvolved leg emerged as the strongest predictors, quietly tying the healthy limb to the injured one's fate.
- The sample is small and the models have never been tested on outside data, so the authors label them hypothesis-generating and not ready for the clinic.
- The path forward runs through larger, independent cohorts that must confirm whether these early signals hold for patients the models have never seen.
Every person facing a torn anterior cruciate ligament eventually wonders whether they will return to the sport they love, and this study, drawn from a small group of 75 patients treated between 2016 and 2023, asks whether that answer might be glimpsed before the operation even begins. By feeding a machine learning model the measurable facts known the day before surgery, researchers found signals that pointed toward both a return to pivoting sports and a knee that feels livable one year later. The most striking clue lay not in the injured leg but in the healthy one, a reminder that recovery is shaped by the whole body. The authors offer these findings as a beginning rather than a verdict, a question worth pursuing with larger and more diverse groups of patients.
Every ACL patient eventually asks the same question: will I get back to my sport? A research team has tried to answer it before the operation even happens, by giving a machine learning model what is already known about each patient on the day before surgery.
The study looked back at 75 people who had primary ACL reconstruction between 2016 and 2023. For each one, the team gathered preoperative information, including demographics, treatment details, isokinetic strength measured on a machine that controls movement speed, and patient-reported questionnaires. They then tracked two outcomes at one year: whether the person had returned to pivoting sports, and whether they reported an acceptable symptom state. One asks if the athlete is back on the field, the other whether they can live comfortably with their knee.
After screening variables for relevance, the team tested several algorithms, and the random forest performed best. Its results, with AUCs of 0.832 and 0.815, sit well above a coin flip. Analysis of which inputs mattered most highlighted younger age and greater strength in the uninjured leg. That healthy limb carried predictive weight for how the operated knee would fare, a quiet reminder that recovery involves the whole body.
The authors remain measured. They call the work exploratory and point to the small sample and the absence of external validation. These models are leads for researchers, not tools for the clinic, and they must prove themselves on patients they have never seen.
A group of researchers has taken a question that every ACL patient eventually asks, will I get back to my sport, and tried to answer it before the surgery happens. Their method was to feed a machine learning model what is already known about a patient on the day before the operation.
The study, published in Nature's Scientific Reports family, looked back at 75 patients who underwent primary anterior cruciate ligament reconstruction between 2016 and 2023. For each person, the team gathered a set of preoperative measures. These included demographic characteristics, treatment-related variables, isokinetic muscle performance (strength measured on a machine that controls the speed of movement), and patient-reported outcome measures.
The researchers then defined two outcomes at one year after surgery. The first was return to pivoting sports, counted as reaching a Tegner activity level of 6 or higher. The second was the patient acceptable symptom state, or PASS, defined as an International Knee Documentation Committee score of 75.9 or above. In plain terms, one outcome asks whether the person is back on the field, and the other asks whether they feel their knee is in a state they can live with.
To build the models, the team first ran univariable logistic regression to pick out the variables that appeared related to the outcomes, and only those variables went into the machine learning stage. They trained several algorithms and tested them with 3-fold cross-validation. Performance was judged by the area under the receiver operating characteristic curve (AUC), calibration plots, and Brier scores.
The random forest algorithm performed best. It reached an AUC of 0.832 for return to pivoting sports and 0.815 for PASS. An AUC of 1.0 would be perfect discrimination and 0.5 would be a coin flip, so both results sit well above chance.
The team also used SHapley Additive exPlanations, known as SHAP analysis, to see which inputs mattered most. Two stood out: younger age, and greater preoperative flexor strength in the uninvolved limb, meaning the healthy leg. Notably, the strength of the leg that was not injured carried predictive weight for how the operated knee would do a year later.
The authors are careful about what this shows. They describe the work as exploratory and say the models showed promising potential. They also point to two clear limits: the sample of 75 is small, and the models were not checked against an external dataset. For those reasons they say the models should be treated as hypothesis-generating rather than ready for clinical use.
The next step is validation in larger, independent cohorts. Until a model like this holds up on patients it has never seen, it remains a lead for researchers rather than a tool for the clinic.
Bemerkenswerte Zitate
The models should be considered exploratory and hypothesis-generating rather than ready for clinical implementation, and require validation in larger, independent cohorts.— Study authors, Scientific Reports
Möchten Sie die ganze Geschichte?
Das Original lesen bei Nature ↗
Technology