Opioid Cost Prediction Using PBM and SDOH: County-Level Machine Learning Framework, United States 2013–2023

This repository contains all data, code, and model artifacts for the study "Development and Feasibility Evaluation of a County-Level Machine-Learning Framework for Opioid Cost Surveillance Using Medicare Part D and Social Determinants of Health Data, United States, 2013–2023." It includes feature-engineered panel data (approximately 3,100 U.S. counties, 2013–2023), Random Forest model code, SHAP analysis scripts, counterfactual simulation outputs, and a README with full replication instructions. All data sources are publicly available from CMS, County Health Rankings, and the U.S. Census Bureau. Version 4 updated April 2026.

本代码仓库包含题为“美国2013-2023年基于医疗保险D部分（Medicare Part D）与健康社会决定因素数据构建县级阿片类药物成本监测机器学习框架的开发与可行性评估”的研究所需的全部数据、代码与模型工件。该数据集包含经过特征工程处理的面板数据（覆盖约3100个美国县级行政区，时间跨度为2013至2023年）、随机森林（Random Forest）模型代码、SHAP（SHapley Additive exPlanations）分析脚本、反事实模拟输出结果，以及一份完整载明复现步骤的README文档。所有数据来源均可从医疗保险与医疗补助服务中心（Centers for Medicare & Medicaid Services, CMS）、县健康排名（County Health Rankings）及美国人口普查局（U.S. Census Bureau）公开获取。本项目版本4更新于2026年4月。