Table 8 A summary on the details of ML, strengths and limitations of the studies on prognosis of periprosthetic joint infection
First author [Ref.] | ML application | Algorithm | Statistical performance | Strengths | Limitations |
|---|---|---|---|---|---|
Shohat, N. (2020) [20] | Preoperative prediction of the risk of debridement, antibiotics and implant retention failure | Random forest analysis | Area under the receiver operating characteristic curve: 0.74 | Important variables for the risk prediction were identified and expressed in a bar graph showing the relative importance value | No external validation |
Klemt, C. (2021) [21] | Preoperative prediction of the risk of recurrent periprosthetic joint infection following revision total knee arthroplasty | Artificial neural network | Area under the receiver operating characteristic curve: 0.84. Brier score: 0.053. Calibration intercept: 0.06. Calibration slope: 1.09 | Important variables for the risk prediction were identified and expressed in a bar graph showing the importance value | No external validation |