Table 6 A summary on the details of ML, strengths and limitations of the studies on pathogens of periprosthetic joint infection
First author [Ref.] | ML application | Algorithm | Statistical performance | Strengths | Limitations |
|---|---|---|---|---|---|
Davis, J.J. (2016) [14] | Used ML-based NGS-based pAST method to predict the antibiotic susceptibility status of Staphylococcus aureus on methicillin | Adaptive boosting | Area under the receiver operating characteristic curve: 0.991. Accuracy: 99.5%. F1 score: 0.995 | _ | No external validation |
Drouin, A. (2019) [15] | Used ML-based NGS-based pAST method to predict the antibiotic susceptibility status of: Staphylococcus aureus: Methicillin, Enterococcus faecium: Vancomycin, Escherichia coli: Amoxicillin/Clavulanic acid, Klebsiella pneumoniae: Gentamicin, Pseudomonas aeruginosa: Levofloxacin | Set Covering Machine | Accuracy: Staphylococcus aureus: 98.7%; Enterococcus faecium: 100.0%; Escherichia coli: 81.8%; Klebsiella pneumoniae: 95.0%; Pseudomonas aeruginosa: 93.9% | No prior knowledge of the genome was needed; some comprehensive tutorials were provided for visualization and annotation of the model | No external validation |
Moradigaravand, D. (2018) [16] | Used ML-based NGS-based pAST method to predict the antibiotic susceptibility status of Escherichia coli to 11 antibiotics | Gradient-boosting decision tree | Average accuracy: 91.0% (range: 81.0%–97.0%) | The ML model’s performance was compared with a rule-based model; ML outperformed the rule-based model; no prior knowledge of the biological mechanism was needed | No external validation |