Page 24 - AIH-2-2

Artificial Intelligence in Health Machine learning in arthroplasty

Table 2. (Continued)
Author Study Performance (AUROC, Input Function Model type
population/ accuracy, sensitivity/
Training model specificity, etc.)
size
Aram et al. 59 430,455 AUROC: 0.71 Demographic, medical Revision Flexible parametric
survival,
random survival
forest,
parametric survival
model,
semiparametric cox
model
El-Galaly et al. 60 31,274 AUROC: 0.57-0.6 Demographic, medical Predict early revision TKA LASSO
Katakam et al. 61 12,542 AUROC: 0.76 Demographic, medical Predict prolonged post-op Stochastic gradient
opioid prescription boosting
Abbreviations: ANN: Artificial neural network; AUROC: Area under the receiving operating characteristic curve; DCNN: Deep convolutional neural
network; EHR: Electronic health record; GBM: Gradient boosting model; KOOS jr: Knee injury and osteoarthritis outcome score for joint replacement;
LASSO: Least absolute shrinkage and selection operator; LOS: Length of stay; MCS: Mental component scores; PCS: Physical component scores;
PROMS: Patient-reported outcome measures; RF: Random forest; SMV: Support vector machine; QDA: Quadratic discriminant analysis; TKA: Total
knee arthroplasties; VAS: Visual Analog Scale.
osteoarthritis outcome score (KOOS) and KOOS for joint predictions related to TKA. 59,60 Finally, an investigation by
61
replacement score, used for evaluating knee function from Katakam et al., found that the complication of extended
osteoarthritis and joint replacement with numerous studies opioid prescriptions could be predicted with an AUROC of
reporting AUROC between 0.75 and 0.88. 7,25,26 Table 2 0.76 based solely on pre-operative inputs. Table 2 displays
provides a summary of relevant studies examining the use the summaries of relevant studies investigating AI’s ability
of AI to predict PROMs in patients undergoing TKA. to predict complications in patients undergoing TKA.

3.2.2. Readmissions, LOS, and cost 3.3. Future direction
Several models have looked into using ML to predict ML has shown a wide range of uses and applications
common hospital-based issues surrounding TKA including regarding TKA in both diagnostic and planning modalities.
readmission, LOS, and cost to the patient. Mohammadi This technology holds a promising future, in regards to pre-
et al. found an AUROC of 0.82 from a neural network operative planning for both surgeons and patients. Patients
54
model using raw electronic health record data to predict often wonder what to expect when it comes to their first
30-day unplanned readmission following TKA. Other TKA, particularly regarding post-operative pain and the
studies were similarly efficacious in predicting LOS and rehabilitation process. Initial results suggest that there may
payment by the patients with AUROC ranging from 0.74 be potential for more tailored counseling for the patients
to 0.82 and 0.78 to 0.83, respectively. 55,56 Table 2 outlines on the duration of pain management prescriptions as well
key studies that have assessed the use of AI in predicting as insights into expected LOS and possible complications.
readmission, LOS, and cost for patients related to TKA. In addition, there is promise in using ML to predict TKA
success or PROM improvements after the surgery and
3.2.3. Complications proper rehabilitation.
Numerous complications associated with surgery For the surgeons, these AI tools and models may provide
have been examined using ML prediction techniques. better pre-operative planning regarding implant positioning
Renal complications and acute kidney injury have been and aid in tailored post-operative course counseling for
modeled with an AUROC of 0.78 in two studies, whereas patients. The ability of ML to accurately predict implants
cardiovascular complications and blood transfusions are within ±2 sizes may contribute to reductions in operating
predicted with an AUROC between 0.73 and 0.84. 34,57,58 room costs and streamline arthroplasty cases. This
In the context of revision, a challenging complication of improvement may reduce the burden on device companies
TKA, several studies have investigated this issue and found by decreasing the need for extensive supply and personnel.
an AUROC between 0.57 and 0.71. This indicates that it A further implementation of ML may be able to correlate
represents one of the weaker associations observed in ML patients, their comorbidities, and their disease burden

Volume 2 Issue 2 (2025) 18 doi: 10.36922/aih.3278