| 摘要: | 本研究探討了機器學習模型在輔助選擇抗生素處方以治療軟組織感染中的應用。通過分析大量臨床數據，構建了多個機器學習模型，包括SVM、Random Forest、Decision Tree和XGBoost。研究結果顯示，所有模型的交叉驗證平均值均達到0.9以上，其中XGBoost模型表現最佳，交叉驗證平均值達到0.996，並且在預測精度和穩定性方面也表現優異。
進一步的分析顯示，XGBoost模型在多分類問題上的精確率、召回率和F1分數也優於其他模型。在具體錯誤分類上，XGBoost模型的預測錯誤量最少，不合理的預測錯誤比例也最低。通過專家意見分析，發現錯誤預測主要集中在某些抗生素組合的分類上，這可能與臨床上同時使用這些藥品的開立頻率有關。另一類錯誤是申請單上多項適應症同時登載，所導致的研究偏差，反映了機器在處理複雜組合藥品時的挑戰。此外，本研究還進行了文獻比較，驗證了模型在不同文獻中的應用效果，並分析了模型的優勢與挑戰，以及在實際應用中的考量。
綜上所述，本研究顯示，機器學習模型，特別是XGBoost，在抗生素處方的準確性和有效性方面具有顯著優勢。這些結果為未來在臨床上應用機器學習輔助抗生素選擇提供了重要的參考依據，並期望能進一步提高臨床治療的精準度和有效性。
This study explores the application of machine learning models in assisting the selection of antibiotic prescriptions for treating soft tissue infections. By analyzing a large amount of clinical data, several machine learning models were constructed, including SVM, Random Forest, Decision Tree, and XGBoost. The study results show that all models achieved a cross-validation average of over 0.9, with the XGBoost model performing the best, reaching a cross-validation average of 0.996, and demonstrating excellent performance in prediction accuracy and stability.
Further analysis indicates that the XGBoost model also outperformed other models in terms of precision, recall, and F1 score for multi-class problems. Regarding specific misclassifications, the XGBoost model had the fewest prediction errors and the lowest proportion of unreasonable prediction errors.Through expert opinion analysis, it was found that mispredictions mainly concentrated on certain combinations of antibiotics, which might be related to the frequency of concurrent use of these drugs in clinical practice. Another type of error was due to multiple indications listed simultaneously on the application form, leading to research bias, reflecting the challenges faced by the machine in handling complex drug combinations. Additionally, this study conducted a literature comparison, verifying the model's application effects in different literature, and analyzed the model's advantages and challenges, as well as considerations in practical applications.
In summary, this study demonstrates that machine learning models, particularly XGBoost, have significant advantages in the accuracy and effectiveness of antibiotic prescription selection. These results provide important references for the future application of machine learning in assisting antibiotic selection in clinical practice, with the hope of further improving the precision and effectiveness of clinical treatments. |
