統計モデルおよび機械学習モデルを用いた薬剤耐性（AMR）トレンドのグローバル予測：中断時系列分析による国家行動計画政策の影響評価

Antimicrobial resistance (AMR) is a pressing global health challenge, particularly affecting low- and middle-income countries (LMICs). This study aims to evaluate the spread of AMR both across time and across the different regions of the world.

We analysed the clinical isolates from 65 countries. A country specific time series forecasting i.e., SARIMA, LSTM, and SARIMA-LSTM hybrid models was performed for Acinetobacter baumannii in Argentina (2004-2030) as a case study to demonstrate model applicability for national level prediction. Moreover, Interrupted Time Series (ITS) regression was applied to predict antibiotic resistance trends and assess the global impact of National Action Plans (NAPs).

Southeast Asia and Africa exhibited the highest AMR burdens, with Indonesia (0.65), Egypt (0.52), and Malawi (0.49) with highest resistance scores. An income-based gradient was observed across key pathogens, third-generation cephalosporin and carbapenem-resistant Escherichia coli, Klebsiella pneumoniae, and Acinetobacter baumannii were significantly more prevalent in LMICs. Gender-wise analysis revealed significantly higher resistance rates in males across most antibiotics, especially levofloxacin. Age-stratified analyses revealed higher resistance in elderly populations, particularly to fluoroquinolones and β-lactams. Forecasting for A. baumannii in Argentina (2004-2030) indicated a continued upward resistance across β-lactam and fluoroquinolones, with LSTM achieving the lowest RMSE across five antibiotics. The ITS revealed a pre-NAP decline but no significant post-implementation change.

This study provides a comprehensive data-driven framework to monitor and forecast AMR, evaluate policy interventions, and hence suggest targeted intervention and strategies for each income group and region and moving beyond "one-size-fits-all" approach.