Study | Fontana et al. [2] | Harris et al. [27]* | Huber et al. [28]* | Katakam et al. [59] | Zhang et al. [60] | Kunze et al. [29] |
---|---|---|---|---|---|---|
Country of data origin | US | US | UK | US | Not reported | US |
Surgical procedure | THA/TKA | TKA | THA/TKA | TKA | TKA | THA |
PROMs/MCID values | HOOS JR: 17.7 KOOS JR: 13.6 SF-36 (MCS + PCS): 5.0 (both) | KOOS Total: 91.8 KOOS JR: 20.8 KOOS Pain: 25.0 KOOS Symptoms: 14.3 KOOS ADL: 24.6 KOOS Quality of Life: 12.5 KOOS Recreation: 17.5 | EQ VAS Hip: 11 EQ VAS Knee: 10 OHS: 8a OKS: 7a | KOOS: MCID value not reported PROMIS Global PF: MCID value not reported PROMIS Global MH: MCID value not reported NRS Pain: MCID value not reported | SF-36 PCS: 10.0 SF-36 MCS: 5.0 WOMAC: 15.0 | EQ VAS: Not reported |
MCID calculation method | Anchor-based Distribution-based | Anchor-based | Distribution-based (VAS) Anchor-based (OKS, OHS) | Distribution-based | Anchor-based | Distribution-basedb |
Time-difference surgery to post-surgery PROM collection (months) | 24 | 12 | 12 | 12 | 24 | 24 |
Number of observations | 7,239 (THA) 6,480 (TKA) | 587 | 30,524 (THA) 34,110 (TKA) | 744 | 2840 | 616 |
Number of featuresc | 66–97 | 6–106 | 81 (candidate predictors) | 24 (candidate predictors) | 18 (WOMAC); 19 (other PROMs) | 8 |
Applied machine learning methods | LASSO Random forest Support vector machine | LASSO Gradient boosting machine Quadratic discriminant analysis | Extreme gradient boosting machine Random forest Multistep adaptive elastic net Neural network Naive Bayes k-nearest neighbours Boosted logistic regression | Stochastic gradient boosting Random forest Support vector machine Neural network Elastic-net penalized logistic regression | Support vector machine LASSO Random forest Extreme gradient boosting | Stochastic gradient boosting Random forest Support vector machine Neural network Elastic net penalized logistic regression |
Ratio of training to test dataset | 80:20 | No test dataset | About 1:1 (dataset of the next year) | 80:20 | 80:20 | 80:20 |
Cross-validation applied in the training dataset | Yes | Yes | Yes | Yes | Yes | Yes |
Outlier detection and analysis performed? | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported |
Missing value management reported? | Yes | Not reported | Yes | Yes | Yes | Yes |
Feature preprocessing performed? | Yes | Not reported | Not reported | Not reported | Not reported | Not reported |
Imbalanced data adjustment performed? | Not reported | Not reported | Yes | Not reported | Yes | Not reported |
AUC/c-statistice | 0.89 (not reported) | 0.72 (not reported)d | Not reported on test data | 0.77 (0.74–0.79) | SVM: 0.95 (0.94–0.97) XGB: 0.95 (0.94–0.97) | 0.97 (0.94–0.99) |
J-statistic | – | – | 0.59 (not reported) | – | – | – |
F1-measure | – | – | 0.78 (not reported) | – | SVM: 0.85 (not reported) XGB: 0.86 (not reported) | – |
Sensitivity | – | – | 0.82 (not reported) | – | SVM: 93.1 (not reported) XGB: 95.6 (not reported) | – |
Specificity | – | – | 0.77 (not reported) | – | SVM: 86.8 (not reported) XGB: 84.9 (not reported) | – |
Accuracy |  | – | 0.79 (not reported), balanced accuracy | – | – | – |
Brier Scoree | Not reported | LASSO (KOOS Pain): 0.16 (not reported) LASSO (KOOS Symptoms): 0.17 (not reported) LASSO (KOOS ADL): 0.17 (not reported) GBM (KOOS PAIN): 0.16 (not reported) QDA (KOOS PAIN): 0.16 (not reported) | Not reported | 0.15 (0.12–0.19) | SVM: 0.12 (not reported) XGB: 0.11 (not reported) | 0.054 (0.047–0.062) |
Best predictive model | Logistic LASSO Random forest | LASSO, Gradient boosting machine, QDA (Pain) LASSO (KOOS Symptoms + KOOS ADL) | Extreme gradient boosting | Neural Network Elastic-net penalized logistic regression | Support vector machine (SVM) Extreme gradient boosting (XGB) | Random forest |
Best predictive PROM | SF-36 MCS | KOOS Pain KOOS Symptoms KOOS ADL | EQ VAS (hip) | KOOS | SF-36 MCS | EQ VAS |
Predictive task | Classification | Classification | Classification | Classification | Classification | Classification |