Predicting Outcomes after Arthroscopy for FAI Using a Machine-Learning Tool
A machine-learning algorithm has identified factors that significantly predict worse outcomes in patients who undergo hip arthroscopy to treat femoroacetabular impingement (FAI), including:
- Anxiety and depression
- Symptom duration of more than 2 years before surgery
- High preoperative score on patient-reported outcome surveys
- Preoperative steroid injections
RELATED: Register for the 12th Annual Detroit Regional Sports Medicine Symposium, July 16-17
“These findings are important from several perspectives. First of all, they demonstrate the value of integrating a machine-learning tool into clinical practice,” said lead author Benedict Nwachukwu, MD, MBA, a sports medicine surgeon and co-director of clinical research for the Sports Medicine Institute at Hospital for Special Surgery.
“From a clinical perspective, our results reinforce that we should be screening for anxiety and depression. The symptom duration finding should encourage more payers and insurance companies to cover hip surgery for patients sooner, rather than keep them in prolonged conservative treatment. Finally, the risk associated with the use of preoperative injections is a novel finding that has not been reported very often in the literature.”
Dr. Nwachukwu helped design the study and developed the machine-learning algorithm in collaboration with a data scientist at PatientIQ, the company that provided the data analysis platform. Researchers from Wake Forest Baptist Health, Winston-Salem, North Carolina, and Rush University Medical Center, Chicago, Illinois, collaborated on the study.
The machine-learning algorithm was built using electronic health record data for 898 patients with FAI who underwent hip arthroscopy between January 2012 and July 2016 at Rush University Medical Center. Senior investigator Shane J. Nho, MD, MS, performed all the hip arthroscopy surgeries and provided the patient data.
Preoperatively and 2 years postoperatively, patients included in the study completed 3 patient-reported outcome surveys that measured their overall hip condition and their ability to perform daily activities and sports activities:
- Hip Outcome Score-Activities of Daily Living Subscale (HOS-ADL)
- Hip Outcome Score-Sports Subscale (HOS-SS)
- Modified Harris Hip Score (mHHS)
The researchers established cutoff values for achieving a minimal clinically important difference (MCID) in the 2 years after surgery for each outcome measure, and then used the data to build predictive models with a LASSO (least absolute shrinkage and selection operator) algorithm. The PatientIQ platform was used to analyze the data.
About three quarters of patients achieved the threshold scores for MCID for each outcome survey:
- 74.0% for the HOS-ADL
- 73.5% for the HOS-SS
- 79.9% for the mHHS
In addition to identifying factors that predicted worse patient outcomes, the investigators found 2 variables that predicted better results: running, at least at the recreational level, and being female.
This study is one of few reporting on the negative impact of preoperative steroid injections on patient-reported outcomes after orthopaedic hip surgery. Patient response to steroid injections is a common diagnostic approach for confirming the hip joint as the location of the pain. Some insurance companies will not approve surgery for FAI unless a patient has had a diagnostic steroid injection first.
“Worse postoperative outcomes after the preoperative use of steroid injections may be due to poor healing of the labral tissue after exposure to steroid,” Dr. Nwachukwu said. “Our finding suggests that surgeons should consider the use of steroid injections judiciously. They have a role to play but using a purely anesthetic injection for diagnostic purposes may be a better option.”
Nwachukwu BU, Beck EC, Lee EK, Cancienne JM, Waterman BR, Paul K, Nho SJ. Application of machine learning for predicting clinically meaningful outcome after arthroscopic femoroacetabular impingement surgery. Am J Sports Med. 2020 Feb;48(2):415-423. doi: 10.1177/0363546519892905. Epub 2019 Dec 23.