基于机器学习对WHO GLASS监测数据的细菌抗菌药物耐药趋势预测：一种用于政策决策支持的检索增强生成方法
Forecasting Bacterial Antimicrobial Resistance Trends Using Machine Learning on WHO GLASS Surveillance Data: A Retrieval-Augmented Generation Approach for Policy Decision Support
AI总结：
利用XGBoost模型预测全球抗菌药物耐药趋势，结合检索增强生成系统提供可溯源的政策建议，误差较基线降低85.3%。
摘要：
背景：抗菌药物耐药性（AMR）是全球健康威胁。尽管WHO全球抗菌药物耐药性与使用监测系统（GLASS）提供了标准化数据，但基于人群的机器学习耐药趋势预测仍然有限。将计算预测转化为政策需要透明的解释机制。方法：处理了2021-2023年的监测数据，包含44个国家和五个WHO区域的5909个观测值。采用严格的时间划分防止数据泄露。使用包括前一年耐药性和抗生素消耗在内的特征，对六种模型（Naive、Linear、Ridge、XGBoost、LightGBM、LSTM）进行基准测试，以预测一年后的耐药率。计算评估指标（MAE、RMSE、sMAPE），并给出MAE的95%自助法置信区间。实现了一个利用Gemma 4的本地检索增强生成（RAG）系统，将预测结果转化为基于检索到的WHO文件的政策指导。结果：XGBoost取得了最佳性能（测试MAE = 6.13% [95% CI: 5.83-6.44]），相比Naive基线（MAE = 41.79%）误差降低了85.3%。SHAP分析确定前一年耐药性为最主要的预测因子（贡献50.5%），证实了强自回归行为。区域预测误差与监测覆盖率密切相关，范围从欧洲区域的3.65%到东南亚区域的8.61%。RAG管道生成了准确、可溯源的政策响应，没有虚构引用。结论：短期AMR耐药率表现出强时间自相关性，可通过梯度提升准确预测。将这些预测与抗幻觉的RAG系统相结合，为AMR治理提供了一个可扩展、基于证据的决策支持框架。
英文摘要：
Background: Antimicrobial resistance (AMR) is a global health threat. While the WHO Global Antimicrobial Resistance and Use Surveillance System (GLASS) provides standardized data, population-level machine learning forecasting of resistance trends remains limited. Translating forecasts into policy also requires transparent interpretation. Methods: Surveillance data (2021-2023) comprising 5,909 observations across 44 countries and five WHO regions were processed. Six models (Naive, Linear, Ridge, XGBoost, LightGBM, LSTM) were benchmarked to forecast one-year-ahead resistance rates using prior-year resistance, antibiotic consumption, and related features. MAE, RMSE, and sMAPE were computed with 95% bootstrap confidence intervals for MAE. A local Retrieval-Augmented Generation (RAG) system (Gemma 4) translated forecast findings into policy guidance grounded in WHO documents. Results: XGBoost achieved the best predictive performance (test MAE = 6.13% [95% CI: 5.83-6.44]), an 85.3% error reduction versus the naive baseline (MAE = 41.79%). SHAP analysis identified prior-year resistance as the dominant predictor (50.5% gain). Regional forecast error tracked surveillance coverage, from 3.65% in the European Region to 8.61% in South-East Asia. The RAG pipeline generated source-attributed policy responses without fabricated citations. Conclusion: Short-term AMR resistance rates exhibit strong temporal autocorrelation that can be accurately forecasted using gradient boosting. Coupling these forecasts with a hallucination-resistant RAG system provides a scalable, evidence-based decision-support framework for AMR governance. v3 correction: a leakage audit found feature-construction defects; on the corrected panel a persistence forecast (MAE 5.79) significantly outperforms XGBoost (6.55), and the 85.3% claim is withdrawn (see Correction).
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
